MSC Thesis Steven Kox

S.A.J.
Kox
A Tool for Estimating Marine

Terminal Dimensions and Costs
in a Project’s Feasibility Phase
Taking into account uncertainties
A Tool for Estimating Marine Terminal Dimensions
and Costs in a Project’s Feasibility Phase
Taking into account uncertainties
by
S.A.J. Kox
in partial for the degree of
Master of Science
in Civil Engineering
at the Delft University of Technology,

to be defended publicly on Wednesday January 25, 2017 at 9:30 AM.
Supervisor: Prof. ir. T. Vellinga TU Delft

Thesis committee: Dr. ir. B.A. Pielage Witteveen+Bos
Ir. P. Quist TU Delft
Dr. ir. O. Morales-Nápoles TU Delft
An electronic version of this thesis is available at http://repository.tudelft.nl/.

Keywords
Marine terminal design, containers, dry bulk, liquid bulk, design rules, construction costs, unit
costs, probabilistic tool, Monte Carlo simulation, Expert Judgement Elicitation, uncertainty,
dependence, Cooke’s Classical model, conditional probability technique, EMO terminal.
Abstract
The main objective of this study is the realisation of a computer tool. The function of the tool is to
estimate the required dimensions and quantities of essential container, dry and liquid bulk
terminal elements, in an early design phase. As well as to give an estimate of the construction
costs of these marine terminals. Terminal element dimensions are for instance the quay length and
storage area. Examples of terminal element quantities are the number of berths and the number of
ship-unloading equipment. The spear point of the tool is the probabilistic approach, in which
uncertainties -concerning design rule variables- are taken into account. This approach results in
probability distributions for the dimensions and quantities. The aim of the tool is to aid terminal or
port designers by allowing them to easily consider a vast amount of input combinations. The
designer therefore does not have to make exact assumptions that could lead to certain important
combinations not being considered. The computations that the tool performs are based on research
in this study. This research concerns terminal design rules and guidelines, common values of
design rule variables and unit costs. For two variables -of which no common values could be
found- Expert Judgement Elicitation (EJE) on uncertainty is applied. The results are weighted
combinations of uncertainty distributions elicited from the experts. A different EJE on
dependence is applied to estimate the relationships between average import, export and
transhipment container dwell times. Finally, the tool is applied to the EMO terminal in Rotterdam.
The EMO terminal is the largest dry bulk terminal in Europe. This application makes it possible to
compare the tool’s results to the actual terminal properties. For the same terminal a sensitivity
analysis is performed on the estimated total construction costs, to certain variables.
i
Preface
Hereby I would like to present my master thesis. This thesis forms the conclusion of my education
at the faculty of Civil Engineering at Delft University of Technology. There, I followed the
Rivers, Ports, Waterways and Dredging Engineering track as part of the master Hydraulic
Engineering. In this track I specialised in the field of Ports.
I conducted my research at the consultancy firm Witteveen+Bos. Their workplace meant being
surrounded by a group of likable and inspirational people in an excellent environment and -not
unimportantly- provided an outstanding cup of coffee.
First of all, I would like to thank my supervisor Ben-Jaap Pielage from Witteveen+Bos for his
input and constructive feedback during the entire length of my research. I would also like to thank
the other members of my thesis committee: chairman Tiedo Vellinga, Peter Quist and Oswaldo
Morales-Nápoles. I am grateful for their time, trust and understanding during both the highs and
the lows. A special thanks goes out to Oswaldo for his enthusiasm, fast e-mail replies and all the
short-notice meetings.
Furthermore, I would like to thank Sip Meijer from Witteveen+Bos and Han Ligteringen from TU
Delft for their contributions to this study regarding marine terminals in general. Likewise, I want
to thank Robert Jan Smits van Oyen from Tebodin for his input concerning dry bulk terminals.
Another person I would like to express my gratitude to is Erik Schulte Fischedick, who helped me
a great deal with the part of my study concerning the subject of costs. And of course I cannot
refrain from thanking my fellow students at Witteveen+Bos for the regular chats, the occasional
drinks and their humour which surely have helped me along the way.
Last but not least I want to thank my parents and my sisters for their interest, constant support and
belief in me.
Steven Kox,
Delft, January 2017
ii
Summary
In the feasibility phase of a marine terminal design project dimension estimates are made.
Ordinarily they are based on few calculations, with many assumptions and therefore risk a high
level of uncertainty. This leads to the consideration of only a small number of possible input
combinations. Furthermore marine terminal design rules and guidelines are scattered over a large
number of publications. Specific information about when to use a particular value of a design rule
variable cannot easily be found. For the previously mentioned reasons Witteveen+Bos requires an
easy-to-use tool that can compute the main required terminal dimensions1. As well as a
construction costs estimate of container, dry bulk and liquid bulk terminal types. In this way many
input combinations can be computed in a timely fashion. Moreover they can be compared by
looking at the resulting dimensions, quantities and/or the costs. An instant cost estimate is in this
case very useful since the economic and financial feasibility are often of main concern for a client.
Realising such a tool is the main objective of this study. A distinctive feature of the tool is the
probabilistic approach. This approach allows the terminal designer/planner to easily consider a
great amount of input combinations. The designer therefore does not have to assume single, fixed
values for variables, which could lead to certain important input combinations not being
considered. The consideration of many input combinations can result in a more accurate and
realistic design. Moreover, when -in a project- certain necessary information is missing, not all
required values of the parameters may be known. Common values of these variables -researched
in the literature study- can be called upon when desired. The tool is applicable to modern
greenfield terminals -that are used for the handling and short-term storage of cargo- in developed
countries. The tool’s output is purely numerical. Consequently neither the position of the terminal
in a port nor the layout of the terminal are considered.
An extensive literature study covers information about the essential elements of the three terminal
types. As well as the available design guidelines and rules, common values of variables and unit
costs. The foremost goal of the literature study is to obtain information that can be used for the
development of the tool. A second intention of the literature study is to function as a manual for
terminal designers. For all main terminal elements design rules are found that are used for
dimensioning and quantification. For most variables common -or realistic- values are gathered
from literature. Unit costs for all essential terminal elements are researched as well. For some
variables common values could not be identified. Two of these variables, namely the total
terminal factor2 and the average storage occupancy3, are factors that can significantly influence
the required storage and terminal areas. Due to the importance of these variables common values
are determined by using the opinions of experts. Values for the remaining variables are
determined by analysing the properties of existing terminals.
A method to combine the opinions of experts is Expert Judgement Elicitation (EJE). With this
method experts are assessed in a structured way, allowing for the results to be treated as scientific
data. Four experts in the field of marine terminal design and/or port master planning have been
assessed. They have been asked to quantify their uncertainty of the total terminal factor and
average storage occupancy for the three terminal types. The assessments and the aggregation of
the experts’ estimates have been performed in accordance with Cooke’s Classical model. With
1
The main terminal dimensions being the quay length, total terminal area and dredging depth. Other dimensions
and quantities such as the storage yard area, number of (un)loading cranes, number of storage yard equipment,
etc. are determined as well.
2
The total terminal factor is the percentage of the storage area (including internal infrastructure) with respect to
the total terminal area.
3
The average storage occupancy is the percentage of the design storage capacity that is actually used, averaged
over a year.
iii
this method the experts are weighted based on their estimates of seed variables, of which the
answers are known to the researcher. The two, previously mentioned, target variables are
quantified by using the obtained weights to combine the individual experts’ uncertainty
distributions of these variables. The resulting cumulative probability distributions are depicted in
Figure 1.
Figure 1: Cumulative probability distributions resulting from the EJE on uncertainty
Since the tool uses random variables dependence between certain variables is of importance.
Dependence between the average dwell times of import, export and transhipment containers is
expected. The same four experts have been asked for their estimates of these dependencies using
the conditional probability technique. This elicitation method also uses seed variables to be able
to weight the experts’ estimates of the dependencies between target variables. The dependence
between random variables can be expressed by means of the rank correlation. The experts think
that there are moderately positive relationships between import & transhipment and export &
transhipment container dwell times. They think the relationship between import & export
container dwell times is weakly positive. A positive rank correlation, ranging from zero to one,
quantifies how well two random variables are described by a function that is entirely increasing.
The resulting rank correlations are depicted in Figure 2. The EJE method for dependence is a new
technique and is currently being researched at Delft University of Technology. A second aim of
the application of EJE on uncertainty and dependence in this study is to introduce these methods
to this field of expertise. These methods are therefore elaborated on in this study.
Figure 2: Rank correlations between the average container dwell times from the EJE on dependence
The tool is realised in Microsoft Excel using the programming language Visual Basic for
Applications. Users select the desired terminal type, where after uncertainty distributions can be
specified for all required variables. The tool then performs a Monte Carlo simulation in which
random values are drawn from the uncertainty distributions. A computation results in probability
distributions that are derived from the estimates of the terminal elements’ dimensions, quantities
and costs. The final exact dimensions, quantities and costs estimates are obtained when the
designer/planner or client specifies a desired quantile. The 70% quantile is recommended. The
iv
aim of the tool is to be easy-to-use, this reflects onto the design of the tool itself. The tool is
positively received by Witteveen+Bos and the Port of Rotterdam has shown interest as well.
The tool is applied to the actual coal and iron ore handling terminal EMO (Europees Massagoed
Overslag), which is the largest dry bulk terminal in Europe. The input for the tool is partly based
on terminal information and information about operations originating from EMO. Not of all
required variables information is available. For these remaining variables uncertainty
distributions, that are based on the studied common values, are used. The tool gives realistic
results, of which the 70% quantiles match the actual dimensions and quantities of the terminal.
When the assumption is made that the terminal was well designed in the past, it can be concluded
that the tool can be trusted. Nevertheless it is advised -based on experience from the case study- to
not use large uncertainties for variables. As these variables will dominate the resulting dimensions
and costs. As a rule-of-thumb for uniform and triangular distributions the upper limit should not
be more than twice the lower limit. The estimated total construction costs of the terminal amount
to € 358,853,000.4 To this sum the quay wall for sea-going vessels is the highest contributor, as is
depicted in Figure 3. From a sensitivity analysis of the variables with an uncertainty it can be
concluded that the average vessel length, and hence the quay wall, as well as the stockpile height
have the highest influence on the estimated total costs. These results are for the EMO case with its
specific input. The conclusions are therefore not per definition true for other terminals. They may
however be of assistance in other terminal design projects.
Figure 3: The contribution of the 70% quantiles of the cost elements to the total costs
4
Besides the quay wall for sea-going vessels this costs estimate considers gantry grab cranes for unloading,
loading equipment, stockpile pavement and stacking-reclaiming equipment. Terminal elements that are not taken
into account are roads, buildings, train and crane rails, belt conveyors, the quay wall for barges and
corresponding loading equipment.
v
List of abbreviations
AGV Automated Guided Vehicle

ASC Automated Stacking Crane
BN Bayesian Network
CFS Container Freight Station
DM Decision Maker
DWT Deadweight Tonnage
EJE Expert Judgement Elicitation
FEU Forty foot Equivalent Unit
LNG Liquefied Natural Gas
LPG Liquefied Petroleum Gas
MHC Mobile Harbour Crane
RMG Rail Mounted Gantry
RTG Rubber Tyred Gantry
SC Straddle Carrier
SSK Standaard Systematiek voor Kostenramingen (Dutch)
STS Ship To Shore
TEU Twenty foot Equivalent Unit
TTU Tractor Trailer Unit
ULCC Ultra Large Crude Carrier
VBA Visual Basic for Applications
VLBC Very Large Bulk Carrier
VLCC Very Large Crude Carrier
vi
List of symbols
Latin Description Unit

symbols
dd Required dredging depth m

fb Bulking factor of CFS -
farea Ratio gross over net storage area -
fs Storage factor TEU/year/m2 or
tonne/year/m2
fTEU TEU factor -
hdt Dredging tolerance m
hguc Gross underkeel clearance m
hpile Stockpile height m
hpile,max Maximum stockpile height m
Average stacking height m
llane Required stockpile lane length m
mb Estimated berth occupancy factor -
ms Estimated storage occupancy factor -
nb Number of berths -
nc Total required number of cranes -
ncb Number of cranes per berth -
nhy Number of operational hours per year hour/year
nlanes Number of stockpile lanes -
r Spearman’s rank correlation -
rc Capacity ratio -
rst Ratio of average stacking height over maximum stacking
-
height
Average dwell time of cargo days
wlane Stockpile lane width m
ACFS Required area for the container freight station m2
Agr Gross required storage area m2
At Required total terminal area m2
ATEU Gross storage density for maximum stacking height m2/TEU
C Throughput over the quay TEU/year or
tonne/year
Cb Berth productivity TEU/year or
tonne/year
Cequip Combined loading/stacking and unloading/reclaiming
tonne/hour
capacity of terminal equipment
Cin Number of incoming containers from the waterside per year TEU/year
Cout Number of outgoing containers to the waterside per year TEU/year
Ctrans Number of transhipment containers per year TEU/year
Cy Number of container movements over the yard per year TEU/year
D Silo or tank diameter m
Ds Draught of design vessel m
H Silo or tank height m
Hretain Retaining height of a sheet pile quay wall m
I Entropy -
Lml Additional quay length to account for mooring lines m
Average vessel length m
vii
Ls,max Length of the largest vessel frequently calling at the
m
terminal
Lq Quay length m
N20 Number of TEU’s -
N40 Number of FEU’s -
Average combined pump productivity per berth. tonne/hour
Average gross productivity per (un)loading crane moves/hour or
tonne/hour
Ts Service time of a vessel hour
Vc Volume of a TEU container m3
Vstorage Required total storage volume m3
W Terminal equipment weight tonne
Greek
symbols
α Total terminal factor -
θ Angle of repose of cargo degrees
ρ Pearson’s product moment correlation -
Average cargo density tonne/m3
µ Ratio of transhipped containers -
viii
List of figures
Figure 1: Cumulative probability distributions resulting from the EJE on uncertainty .................. iv
Figure 2: Rank correlations between the average container dwell times from the EJE on
dependence .................................................................................................................. iv
Figure 3: The contribution of the 70% quantiles of the cost elements to the total costs .................. v
Figure 4: Commodities with cargo types that are considered in this research. Source: Own work 21
Figure 5: Some examples of container types. Source: PIANC (2014a) ......................................... 24
Figure 6: Development of the container vessel capacity with milestones of container shipping
company Maersk. Source: Clarkson Research Services, 2013 ................................... 24
Figure 7: Schematic representation of typical container terminal elements. Based on: Monfort et
al. (2011) ..................................................................................................................... 26
Figure 8: Horizontal cross-sections of various combined sheet pile quay walls. Source: de Gijt &
Broeken (2005) ........................................................................................................... 27
Figure 9: Example of a combined sheet pile quay wall with concrete capping beam in Hansaport,
Hamburg. Source: de Gijt & Broeken (2005) ............................................................. 27
Figure 10: Development of ship-to-shore cranes. Source: Presentation Prof. ir. J.C. Rijsenbrij. TU
Delft course “wb3410 Large Scale Transportation Systems” .................................... 28
Figure 11: Single and Dual Hoisting systems of an STS crane. Source: Lind et al. (2007) .......... 29
Figure 12: Clockwise starting from the left: TTU, reach stacker, straddle carrier, AGV. Source:
Google......................................................................................................................... 30
Figure 13: Rubber tyred gantry crane (left), Rail mounted gantry crane (right). Source: Google . 31
Figure 14: Container storage yard layout per equipment type. Source: Monfort et al. (2011) ...... 31
Figure 15: Container flow over the quay and yard. Based on: Saanen (2004) ............................... 35
Figure 16: Container terminal with total terminal area (blue) and gross storage areas (red). S.L.
Port of Barcelona (Spain). Source: Google Earth ....................................................... 39
Figure 17: Example of dimensions of a Handymax and VLBC vessel. Source: Kleinheerenbrink
(2012).......................................................................................................................... 47
Figure 18: Schematic representation of typical dry bulk terminal elements. Based on: Monfort et
al. (2011) ..................................................................................................................... 48
Figure 19: Schematic overview of three different jetty types. Source: Own work ........................ 49
Figure 20: Clockwise starting from the left: Quadrant loader, linear loader and travelling loader.
Source: Google ........................................................................................................... 50
Figure 21: Clockwise starting from the top left: Gantry grab, level luffing cranes, bucket elevator,
pneumatic, spiral and screw conveyors (with screw detail). Source: Google ............ 51
Figure 22: Belt conveyor. Source: Google ..................................................................................... 52
Figure 23: Loader (top left), scraper reclaimer (bottom left) and stacker-reclaimer (right). Source:
Google......................................................................................................................... 52
Figure 24: Dry bulk wind-row stockpiles with stacker-reclaimer. Source: Google ....................... 53
Figure 25: Covered storage with a belt conveyor for stacking. Source: Google ............................ 54
Figure 26: Artist impression of silos on a terminal. Source: Google ............................................. 54
Figure 27: GSI grain silo. Source: GSI (2016) ............................................................................... 62
Figure 28: Eurosilo in comparison to storage shed. Source: Eurosilo (2016) ................................ 62
Figure 29: Histogram of the ratio of silo diameter and height ....................................................... 63
Figure 30: Liquid bulk vessel types with the typical cargo types and DWTs. Source: U.S. Energy
Information Administration ........................................................................................ 67
Figure 31: Spherical LNG carrier (upper), Membrane LNG carrier (lower). Source: PIANC (2012)
.................................................................................................................................... 68
Figure 32: Schematic representation of typical liquid bulk terminal elements. Based on: Monfort
et al. (2011) ................................................................................................................. 69
Figure 33: Schematic overview of an island berth type jetty. Source: Own work ......................... 70
Figure 34: Liquid bulk loading arm. Source: Agerschou (2004) ................................................... 70
ix
Figure 35: Liquid bulk tanks with external floating roofs. Source: Google ................................... 71
Figure 36: Satellite picture of a liquid bulk terminal with various tank sizes and earth bunds (green
squares). Source: Google ............................................................................................ 72
Figure 37: Full containment cryogenic LNG storage tank schematisation .................................... 72
Figure 38: Indication of LNG terminal components ...................................................................... 73
Figure 39: Bucket wheel stacker-reclaimer combined capacities versus machine weight (black).
And high (red) and low (blue) boundaries for investment costs. Based on Vianen, T.
van (2015) ................................................................................................................... 85
Figure 40: A schematic example of the determination of a DM with 1 seed and 1 target variable
by 3 experts. Source: Aspinall (2008) ........................................................................ 91
Figure 41: BN of four variables with conditional relations. Source: Morales-Napoles et al. (2007)
.................................................................................................................................... 92
Figure 42: TCB, S.L. Port of Barcelona (Spain). Source: Google Earth........................................ 93
Figure 43: Kinder Morgan, Norfolk (Virginia, USA). Source: Google Earth................................ 94
Figure 44: Vopak TTR, Port of Rotterdam. Source: Google Earth ................................................ 94
Figure 45: (Un)Conditional rank correlations between the random dwell times. Source: Own work
.................................................................................................................................... 95
Figure 46: Uncertainty distributions of the experts for seed question 8. Source: Own work ........ 98
Figure 47: Uncertainty distributions of the experts for seed question 4 (upper) and target question
12 (lower). Source: Own work ................................................................................... 99
Figure 48: Uncertainty distributions of all DM’s for seed question 3 .......................................... 101
Figure 49: Uncertainty distributions of the experts and the chosen DM for seed question 6.
Source: Own work .................................................................................................... 101
Figure 50: Cumulative probability distributions as a result of the EJE on uncertainty. Source: Own
work .......................................................................................................................... 103
Figure 51: Probabilities estimated by the experts, as answers to the seed and target questions. . 105
Figure 52: BN with unconditional rank correlations between the average container dwell times.
Source: Own work .................................................................................................... 107
Figure 53: The nine worksheets in Excel (upper). The relations between worksheets and VBA
code (lower). Source: Own work .............................................................................. 110
Figure 54: Part of container terminal input sheet (upper), uniform distribution input window
(lower left), information label of the throughput variable (lower right). Source: Own
work .......................................................................................................................... 111
Figure 55: Examples of the four possible uncertainty distribution types. Source: Own work ..... 112
Figure 56: Calculation properties. Source: Own work ................................................................. 112
Figure 57: A main tool result; histogram of the required quay length of an example project.
Source: Own work .................................................................................................... 114
Figure 58: Overview of the direct and total construction costs of an example project. Source: Own
work .......................................................................................................................... 115
Figure 59: Overview of a container terminal unit costs table. Source: Own work ...................... 116
Figure 60: Relations between het Excel worksheets and the VBA modules. Source: Own work 117
Figure 61: Schematic cross-section of sheet pile quay wall dimensions. Source: Own work ..... 119
Figure 62: Finger pier type jetty with lengths overview. Porto de Tabarao, Brazil. Source: Google
Earth.......................................................................................................................... 120
Figure 63: Island berth type jetty with lengths overview. Korsakov, Sakhalin, Russia. Source:
Google Earth ............................................................................................................. 121
Figure 64: Overview of the EMO dry bulk terminal in Rotterdam. Source: EMO (2016) .......... 124
Figure 65: Satellite image of the EMO terminal with gross storage areas; wind-row stockpiles
(red) and less structured stockpiles (purple) and the total terminal area (blue). Source:
Google Earth ............................................................................................................. 125
Figure 66: Expert Judgement Elicitation results that are used for the EMO case. Source: Own
work .......................................................................................................................... 127
x
Figure 67: Calculation results; total quay length and total number of berths. Source: Own work
.................................................................................................................................. 130
Figure 68: Calculation results; total terminal area and dredging depth. Source: Own work ....... 130
Figure 69: Scatter plots of three different input variables and the corresponding total terminal
area. Linear curves are fitted; their equation and coefficient of determination are
depicted as well. Source: Own work ........................................................................ 132
Figure 70: Total terminal area distribution resulting from a computation with an updated average
dwell time distribution. Source: Own work .............................................................. 133
Figure 71: Probability density distributions of the main direct cost elements. Source: Own work
.................................................................................................................................. 134
Figure 72: Calculation results; number of unloading equipment (left) and the gross storage area
(left). Source: Own work .......................................................................................... 135
Figure 73: The contribution of the 70th percentiles of the cost elements to the total direct costs.
Source: Own work .................................................................................................... 136
Figure 74: Probability density distributions of the total direct costs for various number of
iterations. Source: Own work ................................................................................... 138
Figure 75: Cumulative probability distributions of DMit_op for the three terminal types .......... 144
Figure 76: BN with rank correlations between the average container dwell times ...................... 144
Figure 77: The contribution of the 70th percentiles of the cost elements to the total costs .......... 145
Appendices
Figure III-1: BN of four variables (X1, ..., X4) with conditional relations. Source: Morales-Napoles
et al. (2007) ............................................................................................................... 162
Figure III-2: Relation between the unconditional rank correlation and probability for the Gaussian
copula. Source: Own work ....................................................................................... 164
Figure IV-1: TCB, S.L. Port of Barcelona (Spain). Source: Google Earth .................................. 167
Figure IV-2: Dry bulk terminal in Immingham (UK). Source: Google Earth .............................. 168
Figure IV-3: ENGIE RC, Port of Rotterdam. Source: Google Earth ........................................... 169
Figure IV-4: 5th, 50th and 95th percentile of a normal distribution. .............................................. 170
Figure IV-5: Relations between the average dwell times per cargo flow. ................................... 175
Figure V-1: Seed variable results per questionnaire item ............................................................ 185
Figure V-2: Target variable results per questionnaire item .......................................................... 187
Figure VI-1: Overview of tiles with one silo each. Source: Own work ....................................... 193
Figure VI-2: ‘Rectangle-like’ placement of tiles. Source: Own work ......................................... 193
Figure VI-3: Silo tiles with surrounding stroke (light gray). Source: Own work ........................ 194
Figure VI-4: External boundaries in red. Source: Own work ...................................................... 194
Figure VI-5: One of the four corner areas of a tile group. Source: Own work ............................ 195
Figure VI-6: Cross-section of a bund. Source: Own work ........................................................... 196
Figure VI-7: Overview of bunded tank group with centre-to-centre distance between bunds.
Source: Own work .................................................................................................... 196
Figure VI-8: Overview with waterline heff (red) on inner slope in bunded tank group. Source: Own
work .......................................................................................................................... 197
Figure VI-9: Simplification for the calculation of the bund volume. Source: Own work ............ 198
Figure VI-10: Internal boundaries in red. Source: Own work ...................................................... 198
Figure VI-11: Quadratic curve fitting in order to find relation between number of internal
boundaries and tiles. ................................................................................................. 199
Figure VI-12: Combination of two tank groups. Source: Own work ........................................... 199
xi
List of tables
Table 1: Container vessel classes and dimensions ......................................................................... 25

Table 2: Quay crane productivity ................................................................................................... 32
Table 3: Apron widths per quay crane type.................................................................................... 33
Table 4: Properties for equipment between quay and storage yard and on the yard itself ............. 33
Table 5: Required number of equipment per quay crane ............................................................... 34
Table 6: Berth capacity benchmarks .............................................................................................. 37
Table 7: Storage factor benchmarks ............................................................................................... 41
Table 8: Minimum gross underkeel clearance for certain wave conditions ................................... 41
Table 9: Parameter values of container terminal design rules ........................................................ 42
Table 10: Acceptable berth occupancy factor for container terminals for Tw/Ts = 0.10 ................ 44
Table 11: Material properties ......................................................................................................... 46
Table 12: Dry bulk vessel classes and dimensions ......................................................................... 47
Table 13: Typical rated capacities for unloaders ............................................................................ 55
Table 14: Through-ship efficiency factors ..................................................................................... 55
Table 15: Stacking and reclaiming capacities ................................................................................ 56
Table 16: Quay capacity factors for coal and iron ore terminals ................................................... 58
Table 17: Storage factors ................................................................................................................ 61
Table 18: Pile dimensions of coal and iron ore wind-row stockpiles............................................. 61
Table 19: Ratios between length and width of stockpiles for coal and iron ore............................. 61
Table 20: Minimum and maximum values of characteristics of GSI silos for free flowing materials
....................................................................................................................................... 62
Table 21: Minimum, maximum and average values of characteristics of Eurosilo silos for non-free
flowing materials........................................................................................................... 63
Table 22: Parameter values of dry bulk terminal design rules ....................................................... 63
Table 23: Acceptable berth occupancy factor for dry bulk terminals for T w/Ts = 0.50 ................. 65
Table 24: Liquid bulk oil/product vessel classes and dimensions .................................................. 67
Table 25: Liquid bulk LPG/LNG vessel classes and dimensions .................................................. 68
Table 26: Flammable liquid classes ............................................................................................... 76
Table 27: Minimum distances between objects and tanks with flammable products .................... 76
Table 28: Minimum distances between objects and tanks with flammable fuels .......................... 77
Table 29: Minimum distances from object to LPG tank ................................................................ 77
Table 30: Minimum distances from object to cryogenic gas tanks ................................................ 78
Table 31: Investigated liquid bulk terminals divided into groups .................................................. 78
Table 32: Minimum, maximum and average values of characteristics of liquid bulk tanks .......... 79
Table 33: Parameter values of liquid bulk terminal design rules ................................................... 79
Table 34: Acceptable berth occupancy factor for liquid bulk terminals for T w/Ts = 0.50 ............. 81
Table 35: Unit costs container terminal equipment ........................................................................ 83
Table 36: Weight as function of combined capacity for various dry bulk terminal equipment ..... 85
Table 37: Calibration and information scores and the total score of the four experts .................. 100
Table 38: Calibration and information scores for experts and Decision Makers calculated with
Excalibur. .................................................................................................................... 100
Table 39: Uncertainty distributions as a result of the Expert Judgement Elicitation ................... 102
Table 40: d-calibration scores per weighting method (combination of experts) .......................... 106
Table 41: Dependence calibration score and weight per expert ................................................... 106
Table 42: Actual EMO terminal characteristics ........................................................................... 124
Table 43: Terminal and operations information of EMO. Source: EMO (2016) ......................... 125
Table 44: Selected options in the tool .......................................................................................... 128
Table 45: Parameter values used as input for the computations .................................................. 128
Table 46: EMO case direct costs and construction costs corresponding to the 70% quantile ..... 136
xii
Table 47: Entropy quantities of the total direct costs for constant variables but a varying number
of iterations.................................................................................................................. 137
Table 48: The analysed variables with the base case and narrowed distributions per step. The
lower and upper limits of the uniform distributions are presented. ............................ 139
Table 49: Entropy quantities between the total direct costs distribution of the base case and the
total direct costs distributions per step ........................................................................ 141
Appendices
Table III-I: Correlation matrix with unconditional product moment correlations per expert. ..... 165
Table V-I: Calibration and global information scores of the seed variables for DMit_op ........... 187
xiii
Table of contents
Preface ....................................................................................................................................... ii
Summary .................................................................................................................................. iii
List of abbreviations................................................................................................................ vi
List of symbols ........................................................................................................................ vii
List of figures ........................................................................................................................... ix
List of tables ............................................................................................................................ xii
Table of contents..................................................................................................................... 14
1 INTRODUCTION .......................................................................................................... 17
1.1 Problem description..................................................................................................................... 17
1.2 Research objective....................................................................................................................... 18
1.3 Research scope ............................................................................................................................ 20
2 LITERATURE STUDY ................................................................................................. 23

2.1 Container terminals ..................................................................................................................... 23
2.1.1 Background information ..................................................................................................................... 23
2.1.2 Equipment properties .......................................................................................................................... 32
2.1.3 Design guidelines, rules and parameters ............................................................................................ 34
2.2 Dry bulk terminals....................................................................................................................... 45
2.3 Liquid bulk terminals .................................................................................................................. 66
2.4 Costs determination ..................................................................................................................... 82
2.4.1 Unit costs, construction costs and inflation ........................................................................................ 82
2.4.2 General unit costs ............................................................................................................................... 82
2.4.3 Container terminal unit costs .............................................................................................................. 83
2.4.4 Dry bulk terminal unit costs ................................................................................................................ 84
2.4.5 Liquid bulk terminal unit costs ............................................................................................................ 86
3 EXPERT JUDGEMENT ELICITATION ................................................................... 89

3.1 The method.................................................................................................................................. 89
3.1.1 Eliciting uncertainties ......................................................................................................................... 90
3.1.2 Eliciting dependence ........................................................................................................................... 91
3.2 The assessed variables................................................................................................................. 92
3.2.1 Total terminal factor ........................................................................................................................... 92
3.2.2 Storage occupancy factor .................................................................................................................... 95
3.2.3 Average dwell time .............................................................................................................................. 95
3.3 The elicitation.............................................................................................................................. 95
3.4 Analysis of the results ................................................................................................................. 97
14
3.4.1 Uncertainty ......................................................................................................................................... 97
3.4.2 Dependence ....................................................................................................................................... 104
3.5 Discussion ................................................................................................................................. 108
4 THE TOOL ................................................................................................................... 109

4.1 Tool function ............................................................................................................................. 109
4.2 Tool structure ............................................................................................................................ 109
4.2.1 Software ............................................................................................................................................ 110
4.2.2 Worksheets ........................................................................................................................................ 110
4.2.3 VBA code........................................................................................................................................... 116
4.3 Tool calculation methods and restrictions ................................................................................. 117
4.3.1 Monte Carlo simulation .................................................................................................................... 117
4.3.2 Container terminal calculation ......................................................................................................... 118
4.3.3 Dry bulk terminal calculation ........................................................................................................... 119
4.3.4 Liquid bulk terminal calculation ....................................................................................................... 121
4.4 Tool verification ........................................................................................................................ 122
5 CASE STUDY ............................................................................................................... 123

5.1 Case description ........................................................................................................................ 123
5.2 Actual terminal characteristics .................................................................................................. 124
5.3 Input parameters ........................................................................................................................ 125
5.3.1 Throughput........................................................................................................................................ 126
5.3.2 Vessel dimensions ............................................................................................................................. 126
5.3.3 Estimated berth occupancy ............................................................................................................... 126
5.3.4 Gross productivities .......................................................................................................................... 126
5.3.5 Average dwell time ............................................................................................................................ 127
5.3.6 Total terminal factor and estimated storage occupancy ................................................................... 127
5.3.7 Stockpile properties........................................................................................................................... 127
5.3.8 Summary of the input parameters ..................................................................................................... 128
5.4 Tool application ........................................................................................................................ 129
5.4.1 Computation results .......................................................................................................................... 129
5.4.2 Analysis of the results ....................................................................................................................... 130
5.4.3 Conclusions & construction costs estimate ....................................................................................... 135
5.5 Sensitivity analysis .................................................................................................................... 136
5.5.1 Comparing distributions ................................................................................................................... 136
5.5.2 Variables of interest .......................................................................................................................... 138
5.5.3 Analysis of the results ....................................................................................................................... 140
5.5.4 Conclusions ....................................................................................................................................... 141
6 CONCLUSIONS AND RECOMMENDATIONS ..................................................... 143

6.1 Conclusions ............................................................................................................................... 143
6.2 Recommendations ..................................................................................................................... 146
BIBLIOGRAPHY ................................................................................................................ 147
APPENDICES ...................................................................................................................... 151
Appendix I Unit costs CONFIDENTIAL ........................................................................ 152
Appendix II Interviews ................................................................................................... 153
1.1 Problem description 15

Appendix III How to perform an Expert Judgment Elicitation ................................... 159
Appendix IV Expert Judgment Elicitation questionnaires........................................... 167
Appendix V Expert Judgment Elicitation results ........................................................ 183
Appendix VI Calculation of silo and tank group dimensions ....................................... 193
16 INTRODUCTION
1 INTRODUCTION
In the introduction the reader is presented with the motivation for this study. Hereafter the main
research objective and research questions are specified. The chapter concludes with the limits that
are posed to this research.
1.1 Problem description
Marine port terminals form the interface between different modes of transport of cargo. The
design of ports and port terminals occurs in multiple phases. During the early design phase of a
terminal a basic design is made. This is done by analysing different scenarios of trade and traffic
for the port’s master plan period.
For the design of port terminals no official international regulations are available. Due to this lack
of regulation various guidelines or design rules for determining the required dimensions of port
terminals exist. Most design rules are in the form of equations, consisting of parameters. These
design rules are based on miscellaneous studies concerning a wide range of situations. As a
consequence it may become unclear when to use particular rules and especially which values for
the parameters. This leads to the demand for an elaboration on the main existing terminal design
rules in combination with a clear overview of the common values of the parameters.
Calculating the required main terminal dimensions by hand for the various scenarios is a time
consuming effort. For this process computer tools can be very helpful. Presently several tools 5
exist that use simulation for the design of terminals. These tools however require extensive input
and are thus more suited for later design phases. In the early design phase less data is available
and multiple scenarios have to be analysed. Therefore a tool is needed that can be used to compute
terminal designs with few data in a timely fashion.
Predictions for future situations usually have an increasing uncertainty over time. Port master
plans are designed for large periods of time and some variables that -among others- determine the
terminal dimensions may therefore be subject to uncertainty. In the present situation calculations
are made by assuming certain values for parameters that are uncertain, this can lead to under or
over dimensioning of terminals. In order to account for uncertainty we can incorporate a
probability distribution for specific variables. This way many input combinations are included in
one single calculation resulting in a more complete and realistic analysis.
Among the main concerns for parties that order the design of a terminal are the economic and
financial feasibility. In order to make a project feasible costs are an important aspect. It would
therefore be useful to be able to give a rough construction cost estimate for a terminal design in
the feasibility phase of a project. Similarly to the design rule parameter values uncertainties about
the prices or costs of terminal elements are common. This of course lends itself for a similar
probabilistic calculation approach.
5
For example the simulation software TIMESQUARE of TBA (2016) or Port Simulation Software of
Simio (2016) or results from studies like Vianen, T. van (2015).
17
According to PIANC (2014b) liquid-, dry bulk and containers are the main commodities in
waterborne transport. The worldwide container transport has an average forecasted annual
increase in cargo of 6% (from 2012 to 2017). This can mostly be attributed to the increase of
manufactured goods and the shift from bulk and general cargo to container cargo; PIANC
(2014b), Quist & Wijdeven (2014). Due to the increase in global energy and steel demand the
demand for coal and iron ore also increases according to Vianen, T. van (2015). This has as an
effect that the bulk transport volumes of these goods increase as well. Because of the importance
of these commodities in worldwide sea trade the focus in this research will be on these types of
cargo and their corresponding terminal designs.
1.2 Research objective
In this section the main objective of this research is defined. To fulfil this objective research
questions and sub-objectives are defined that each partly contribute to reaching the main
objective. Each research question/objective is individually elaborated on. This section can be seen
as an outline of the current report.
Main research objective

“The development and application of a tool for estimating the main required marine terminal
dimensions and corresponding construction costs with a limited available amount of data and
including uncertainties for the variables and costs.”
At Witteveen+Bos the demand for a tool that can be used to quickly determine marine terminal
dimensions exists. This thesis fulfils this demand by the realisation of a program written in
Microsoft Excel. With the tool the main terminal dimensions (quay length, terminal area and
water depth) can be calculated as well as the required amount of terminal equipment and storage
utilities and their dimensions. Also the direct costs are calculated for most of the main terminal
elements individually and in total. The tool is applied in a case study.
Research question I
“What are the main terminal elements and the existing terminal design guidelines and design
rules?”
Information about the different terminal types and their main elements is gathered, this supports
the more theoretical design guidelines and rules. These guidelines and rules are described in
literature and come in the form of equations, rules-of-thumb and recommendations. With these
rules the required terminal dimensions can be determined for a specific situation. These rules are
primarily studied so that they can be implemented in the tool. Another goal is to create a clear
overview of these rules so that terminal designers can consult it when needed. This research
question is treated in Chapter 2.
Research question II
“What are common values for the design rule parameters and what are common unit costs of
terminal elements?”
Since during the early design phases specific information is often missing assumptions have to
be made. It would therefore be useful to have an overview of parameter values 6 that are
commonly used. These common -or standard- parameter values are studied in literature. Often
6
Parameter values are the values that are required as input for the terminal dimensioning equations.
18 INTRODUCTION
literature proposes ranges of values, these can be seen as probability distributions. These
distributions function as standard input for the tool in case no project information is available to
the tool user. In Chapter 2 a clear overview is created by presenting the common values in a
table per terminal type. This way terminal designers can easily consult the considered
information. In order to be able to give a cost estimate of a terminal, terminal element prices are
required. These prices are treated in Section 2.4.
Research question III

“What are the uncertainty distributions of the total terminal factor7 and the average storage
occupancy factor8 and what are the correlations between average import, export and
transhipment container dwell times?”
The research in literature on common values of parameters did not result in values or
distributions for two important variables; the total terminal factor and average storage
occupancy. In this study experts are asked for their estimates of these common values by means
of a scientific method called Expert Judgement Elicitation. Expert Judgement Elicitation is a
statistical method to objectively combine the opinions of experts. Furthermore the presumption
existed by the writer that between the random average dwell times of import, export and
transhipment containers dependencies exist. This could however not be confirmed in literature.
Therefore another Expert Judgement Elicitation method is used to ask these dependencies from
experts and to scientifically combine them. A dependence between random variables can be
expressed by the rank correlation. This research question is treated in Chapter 3.
Chapter 4 concerns the creation of the tool. The tool computes the main required dimensions of a
container, dry or liquid bulk terminal. The input variable values can be chosen to be
deterministic or a probability distribution. The tool performs a number of iterations, as specified
by the user, in which realisations are drawn from the distributions. The tool uses the design rules
from the literature study for the calculations. The outputs are probability distributions for the
resulting terminal dimensions and for the other intermediate results. Construction costs are
calculated in the same way.
Research question IV
“What terminal elements contribute the most to, and what input parameters have the largest
influence on, the total construction costs of one specific terminal?”
In a case study an existing terminal is considered. Actual terminal data is used as much as
possible for the input parameters, for the unknown parameters common values are used. The
terminal elements that contribute the most to the construction costs are identified. Furthermore
an analysis is performed in order to determine the variables to which the total costs are sensitive.
This research question is treated in Chapter 5.
7
The gross storage area (net storage area incl. internal infrastructure) divided by the total terminal area.
8
Average volume of cargo in the storage yard divided by the design capacity of the yard.
1.2 Research objective 19

1.3 Research scope
The limits of the study are discussed in this section.
 Marine terminals for greenfield port development are considered as requested by

Witteveen+Bos.
 Only modern terminals for developed countries are taken into account. This choice is
made since most parameter values in literature of the terminal design rules are for these
kind of terminals.
 Functions of the considered terminals are the handling and short-term storage of cargo.
Therefore terminals that are used as strategic buffer for cargo or that include or only serve
local industry (e.g. refineries) are not included in the research.
 The research and the tool include the level of detail that is required for the early design
phases of a project, as requested by Witteveen+Bos.
 The output of the tool is purely numerical. The layout of a terminal or port is not
considered, the tool’s results can be used as a help in determining these.
 The main terminal dimensions are considered, intermediate results are presented as well.
The most important results are:
o Quay wall length or jetty length and the amount of jetties
o Water depth
o Total terminal area
o Gross storage areas
o Number of (un)loading equipment
o Number of equipment in between waterside and storage yard, and/or on the yard
itself.
o Storage utility (stockpiles, silos, tanks, etc.) dimensions and amounts.
 The main cost items are considered. The individual and total direct costs are presented as
well as the estimated construction costs. The main cost items are:
o Quay wall or jetties
o Storage utilities (pavement, storage sheds, silos, tanks, etc.)
o Ship (un)loading equipment
o Equipment in between waterside and storage yard, and/or on the yard itself
o Bunds for liquid bulk terminals
 The commodities and cargo types that are considered in this research are depicted in
Figure 4. The choice for coal, iron ore and grains is made since these are the three most
transported dry bulk goods according to PIANC (2014b). The choice for crude oil is made
since it is one of the most transported liquids according to Agerschou (2004) and rules-of-
thumb can be found in literature. Liquefied Natural Gas (LNG) is included as well since it
is more environmentally friendly compared to other fossil fuels and some parties therefore
expect a growth, according to Het Financieele Dagblad (2016).
The tool includes the option to determine the terminal dimensions for other kinds of
liquids as well by changing material properties. This also holds for dry bulk cargoes,
however not all dry bulk cargoes are handled by the equipment types and stored in the
storage utilities that are included in the tool.
20 INTRODUCTION
Figure 4: Commodities with cargo types that are considered in this research. Source: Own work
In the course of this study some smaller limitations are made in view of the limited duration of a
MSc thesis. When this is the case a reference is made to this chapter.
1.3 Research scope 21

2 LITERATURE STUDY
The literature study focuses on available design guidelines, rules, common values of parameters
and unit costs. The aforementioned related to the design of a marine terminal in the early phases
of a greenfield project. For each terminal type 9 this information is provided in this chapter. In
order to assist in explaining the treated terminal specifics; background information for each
terminal type is provided as well. In this study only modern terminals in developed countries are
considered. This narrowing of the scope is done since common values of parameters for terminals
in developing countries are not sufficiently treated in literature.
Goal of the literature study: To provide a clear overview of terminal dimension design rules,
common values of parameters and terminal element unit costs for the reader and as input for the
computer tool.
2.1 Container terminals
For container terminals first general background information is presented in order to give the
reader more insight in the most important aspects of a terminal. Then the relevant equipment
properties that are needed as input for the design rules are given. In the last section the design
guidelines, rules and common values of parameters are treated.
2.1.1 Background information

This section considers container terminals and is based on Agerschou (2004), PIANC (2014b),
PIANC (2014b), Quist & Wijdeven (2014) and Mohseni (2011). Information is provided about
container types, vessels and terminal elements.
2.1.1.1 Containers
Containers are a standardised form of cargo and exist in various types and sizes. In Figure 5 some
examples of container types are given that are all 40 ft except the Tank Container which is 20 ft.
However all these containers come in both sizes; PIANC (2014a).
9
The considered terminal types are container, dry bulk and liquid bulk terminals.
23
Figure 5: Some examples of container types. Source: PIANC (2014a)
For this research only the following container sizes are considered:
 Twenty foot Equivalent Unit or TEU (l = 6.10 m, b = 2.44 m, h = 2.60 m)

 Forty foot Equivalent Unit or FEU or 2 TEU (l = 12.20 m, b = 2.44 m, h = 2.60 m)
2.1.1.2 Vessels
Container vessels are categorised in different classes depending on the vessel size. Shipping lines
continuously increase the size of their vessels to make use of the economy of scale principle
according to Quist & Wijdeven (2014). In Figure 6 the development of container vessel capacities
(vertical axis) from 1981 to 2013 (horizontal axis) is depicted, in which the light blue line
represents the largest ship size and the dark blue line the average ship size.
Figure 6: Development of the container vessel capacity with milestones of container shipping company Maersk.
Source: Clarkson Research Services, 2013
In Table 1 an overview of the most important classes with corresponding vessel sizes is presented.
The Deadweight Tonnage (DWT) is the mass of the cargo, fuel, crew, passengers, fresh water and
24 LITERATURE STUDY
provisions on a ship. The draught is the maximum distance in meters between the waterline and
the keel of the ship and the beam is the maximum width of the ship; Ligteringen & Velsink
(2012).
Table 1: Container vessel classes and dimensions
Capacity DWT average Length Draught Beam

Class
[TEU] [tonne] [m] [m] [m]
1st generation 750 - 1100 14,000 180 - 200 9 27
2nd generation 1500 - 1800 30,000 225 - 240 11.5 30
3rd generation 2400 - 3000 45,000 275 - 300 12.5 32
4th generation 4000 - 4500 57,000 290 - 310 12.5 32.3
Post Panamax 4300 - 5000 54,000 270 - 300 12 38 - 40
Jumbo 6000 - 9000 90,000 310 - 350 14 43
New Panamax 13,000 151,000 366 15.2 49

Super-Post
14,000 - 18,000 157,000 - 194,000 400 14.5 - 15.5 56 - 59
Panamax
Source: Quist & Wijdeven (2014)
2.1.1.3 Terminal elements

Typical elements of a container terminal are (see Figure 7):
 Quay
 Apron
 Storage yard
 Landside area
2.1 Container terminals 25

Figure 7: Schematic representation of typical container terminal elements. Based on: Monfort et al. (2011)
2.1.1.4 Quay
For container terminals a vertical quay wall that is directly connected to the land is used. Jetties
are not commonly used since ship-to-shore cranes need much space and the storage yard is
preferably as close to the berth as possible. Quay walls exist in many forms with the main types
being sheet pile walls and gravity walls. For quay walls with large retaining heights often
combined sheet pile walls are used; these consist of heavy primary elements (such as tubular
piles) with intermediate sheet piles, as depicted in Figure 8 and Figure 9. Combinations between
and variants of these types exist as well, for more information reference is made to de Gijt &
Broeken (2005). Often containers are transported to and from barges from the quay for sea-going
vessel. This has as disadvantage that the capacity of the large STS-cranes designed for sea-going
vessels is not met. Also barges regularly collect containers at multiple terminals, this is time
consuming; Quist & Wijdeven (2014). To tackle these problems a separate quay for barges can be
implemented in a terminal. This is however not implemented as explained in the introduction.
26 LITERATURE STUDY
Figure 8: Horizontal cross-sections of various combined sheet pile quay walls. Source: de Gijt & Broeken
(2005)
Figure 9: Example of a combined sheet pile quay wall with concrete capping beam in Hansaport, Hamburg.
Source: de Gijt & Broeken (2005)
2.1.1.5 Apron
The apron area typically consists of the following elements with their specific functions and
equipment:
 Section between quay wall and crane rail

 Crane
o Crane rail gauge: area between the crane rails.
o Outreach: extending part of the boom on the backside of the crane.
 Access roadway
 Zone between roadway and storage yard
The choice of equipment usually depends on vessel sizes, economics and the desired density and
productivity.

Cranes
To load and unload containers between vessel and quay large cranes are used. There are two
different types of cranes that are used to transport containers from ship to shore and vice versa:
 Ship-to-Shore (STS) gantry crane: Consists of a frame that is mounted on rails running
parallel to the quay and of a boom that extents horizontally over the ship. Containers are
hoisted between the legs or at the outreach of the crane. Cranes usually are described by
the length of the boom (in number of containers) and the rail gauge.
 Mobile Harbour Crane (MHC): A crane that is mounted on rubber tyres and supported by
outriggers. The crane can be moved to any location, given that the terminal foundation can
withstand the pressure, and it can be used for multiple sorts of cargo. MHC cranes have a
lower capacity than STS cranes but have increased flexibility.
STS gantry cranes are the most used cranes on container terminals. These cranes have to adapt to
the increasing container vessels sizes, which is depicted in Figure 10.
Figure 10: Development of ship-to-shore cranes. Source: Presentation Prof. ir. J.C. Rijsenbrij. TU Delft course
“wb3410 Large Scale Transportation Systems”
Improvements to gantry cranes exist that increase the capacity of the cranes. Some examples of
such improvements are listed below; from PIANC (2014b) and Lind et al. (2007):
 Tandem forties spreader: can hoist two FEU’s side by side at the same time (see Figure
11):
o Single hoist: Both spreaders are lifted only together.
o Dual hoist: Spreaders can be lifted separately, horizontal motion is combined.
Doubles the productivity according to Saanen (2004).
 Dual trolley: Two independent trolleys. One trolley moves containers between ship and a
platform, the second trolley moves containers between platform and landside. Can
improve productivity with about 15 to 20% according to Saanen (2004).
 Remote controlled cranes.
28 LITERATURE STUDY
Figure 11: Single and Dual Hoisting systems of an STS crane. Source: Lind et al. (2007)
These improvements are not implicitly included in this research in order to not further complicate
the input. To still be able to take these improvements into account the user can manually change
the STS crane productivity in the tool.
Transport between crane and storage yard
Equipment types for the handling of containers from the apron to the storage yard are (Figure 12):
 Tractor Trailer Units (TTU) or chassis system: Using trailers specifically designed for
container transport. The containers are loaded to/from the trailer/chassis at the crane. The
container can then be (un)loaded by a crane located at the storage yard, or the chassis with
the container can be parked in the storage area. It is also possible to immediately transport
the container over the road network to its destination. The trailers/chassis are moved by
means of tractors or trucks.
 Reach stackers: Comparable with top loader but with a telescopic boom. It can therefore
reach further then one row so that stacks can be four containers wide, with access on both
sides.
 Straddle carriers: Can lift containers between its wheels. A straddle carrier is able to lift a
container 1 over 2 or 1 over 3. The straddle carrier is quite space efficient but can be
difficult to operate.
 Automated Guided Vehicle (AGV): Vehicles that transport containers automatically on
the terminal. Results in an efficient terminal but investment costs are high. This equipment
type is not included in the computer tool since no capacity properties are known.

Figure 12: Clockwise starting from the left: TTU, reach stacker, straddle carrier, AGV. Source: Google
Some of these equipment types do not need other machines to do the (un)loading. Others can only
transport the container and need cranes to be (un)loaded. These cranes are described in the
following section. In PIANC (2014a) more extensive information can be found about all container
terminal equipment.
2.1.1.6 Storage yard

The storage yard has to be able to accommodate different sorts of containers, in this research
regular and empty containers are considered. The storage yard itself can be divided into the
following main elements:
 Import
 Export: Positioned as close to the quay wall as possible will facilitate efficient loading of
vessels.
 Empties: Can have a dedicated storage area and can be stacked up to nine containers high
by means of Empty Container Handlers that are similar to Top loaders. All empty
containers do not have to be reachable by the equipment and can therefore be very densely
stacked. May be positioned outside of the terminal area.
 Container Freight Station (CFS): A covered area (building) that is used to strip and stuff
containers. The CFS may also be positioned outside of the terminal area.
An additional container terminal element can be a leaking container pit. This may be positioned
outside of the storage yard but inside the terminal area.
Equipment used only at the container yard is (see Figure 13):
 Rubber Tyred Gantry (RTG) crane: A portal crane mounted on rubber tyres. The most
common dimensions are a width of 6 + 1 or 7 + 1 (number of containers + driving lane)
and an ability to lift 1 over 5 or 6. These cranes are used in combination with tractor-
trailers. The optimum layout has the tracks parallel to quay.
 Rail Mounted Gantry (RMG) crane: Similar to RTG but mounted on rails. Can be used in
combination with tractor-trailers or AGV’s. Most used forms are perpendicular rails with
end loaded stacks or parallel rails with side loaded stacks. An RMG can be (partly)
automated and is then called Automated Stacking Crane (ASC).
30 LITERATURE STUDY
Figure 13: Rubber tyred gantry crane (left), Rail mounted gantry crane (right). Source: Google
Container stacks have different forms depending on the terminal equipment. The space in between
the containers and the maximum stacking height depends on the equipment type. As an example
RMG cranes have a much higher maximum yard density than reach stackers because of the large
differences in space needed between containers and the maximum stacking height. Some
container yard layouts corresponding to a certain equipment type are displayed in Figure 14.
Figure 14: Container storage yard layout per equipment type. Source: Monfort et al. (2011)
Gateway terminals are terminals that mainly import and export cargo. For terminals with a large
transhipment to gateway ratio parallel stacks have been found to be more efficient. For a larger
proportion of gateway transport perpendicular stacks are more efficient according to simulations;
PIANC (2014a).
The pavement and foundation of the apron and storage yard are also important factors for a
terminal design. The requirements depend on the type of terminal equipment.

2.1.1.7 Landside area
The landside area consists of:
 Landside traffic circulation system

o Gate
o Loading space
o Queuing space
o Equipment parking
 Buildings
o Offices
o Terminal equipment maintenance/repair facility
o Fuel station
Additional elements that may be required are:
 Rail terminal
 Container repair facility
 Container inspection facility
2.1.2 Equipment properties

The properties of the terminal equipment that are required for the dimensioning of a container
terminal are given in this section. A division is made between quay cranes and equipment used
between quay and storage yard and on the yard itself.
2.1.2.1 Quay cranes

The quay cranes that are mentioned in Section 2.1.1.4 are STS Gantry cranes and MHC cranes.
The average gross productivity is defined as the average number of moves per hour including
unproductive intervals such as crane repositioning, moving hatches to/from quay and time
between shifts; PIANC (2014b). These productivities are listed per crane type in Table 2.
Table 2: Quay crane productivity
Average gross
Crane
productivity
Low
Ship-to-shore
20 - 25 moves/hour
gantry crane
Medium
25 - 30 moves/hour
High
30 - 35 moves/hour
Mobile harbour
15 - 20 moves/hour
crane
Source: PIANC (2014b)
The required apron width per crane with a certain rail span is listed in Table 3.
32 LITERATURE STUDY
Table 3: Apron widths per quay crane type
Rail span Apron width

Crane
[m] [m]
Ship-to-shore
15 - 20 40 - 55
gantry crane
30.48 55 - 751
Mobile harbour
- 25 - 30
crane
Note 1: When a perpendicular RMG is used on the storage yard or an AGV system in between; the apron width
ranges between 100 - 120 meter.
Sources: Agerschou (2004), PIANC (2014a) and Quist & Wijdeven (2014)
2.1.2.2 Storage yard equipment

The equipment properties for transport between quay and storage yard and storage yard cranes are
given in Table 4. All these equipment types can place containers on the yard, only the RTG and
RMG need other equipment for the transport between quay and storage yard.
The gross storage density takes into account the internal roads and the stacking height.
Table 4: Properties for equipment between quay and storage yard and on the yard itself
Average Maximum Gross storage Average over

stacking stacking density maximum
Equipment
height height stacking height
[-] [-] [m2/TEU] [-]
Chassis system 1 1 40 - 66.7 1.00
Empty handler 6-9 9 6 - 11 0.67 - 1.00
Reach stacker 2-4 7 20 - 30 0.29 - 0.57
Straddle carrier1 1-2 4 10 - 13 0.25 - 0.50
RTG 3-4 6 7.5 - 16 0.50 - 0.67
RMG 3-4 7 7.5 - 11 0.50 - 0.57
Note 1: Maximum from Monfort et al. (2011) unrealistically high with respect to the other literature so an
alternate maximum is used.
Based on Böse (2011), Ligteringen & Velsink (2012), Monfort et al. (2011), PIANC (2014b) and PIANC
(2014a).

The required number of yard equipment per quay crane depends on the (combination of) used
equipment types on the terminal. The number of equipment per combination is presented in Table
5.
Table 5: Required number of equipment per quay crane
Required number Equipment between quay and yard

of equipm. per
TTU SC AGV
STS crane
3 – 6 TTU
RS
3 – 4 RS
Equipment at
SC 3 – 5 SC
yard
3 – 6 TTU
RTG
2 – 3 RTG
3 – 6 TTU 3 – 6 AGV
RMG
2 – 3 RMG 2 – 3 RMG
Notes: TTU = Tractor Trailer Unit, SC = Straddle Carrier, RS = Reach Stacker, RTG = Rubber Tyred Gantry,
RMG = Rail Mounted Gantry, AGV = Automated Guided Vehicle, STS = Ship-To-Shore crane.
Based on Böse (2011) and PIANC (2014a)
2.1.3 Design guidelines, rules and parameters

First the concept of transhipment is described. Then design guidelines, rules and parameters are
provided for determining the quay length, storage yard area and water depth. Values used in
practice of the parameters are listed in Section 2.1.3.5.
2.1.3.1 Transhipment
The terminal throughput is the sum of the incoming and outgoing number of containers over the
quay per year. Incoming and outgoing containers cannot be confused with import and export
containers, if there is transhipment of containers. Transhipment containers are containers that
enter the terminal via deep-sea or feeder vessels and leave the terminal via those same modes of
transport. Other containers are counted as import/export containers. When containers are -after
being imported- exported again to barges this does not count as transhipment according to Saanen
(2004). The ratio between the number of transhipment containers and incoming/outgoing
containers is denoted by µ. The ratio can be calculated with the following equation, based on
Saanen (2004) and adjusted for TEU’s instead of container moves:
(1)
Where,
µ Ratio of transhipped containers [-].
Ctrans Number of transhipment containers per year [TEU/year].
Cin Number of incoming containers from the waterside per year [TEU/year].
Cout Number of outgoing containers to the waterside per year [TEU/year].
C Throughput. Number of container movements over the quay per year [TEU/year].
When the terminal throughput is used to calculate the required storage yard area, transhipment
containers are counted twice. The number of containers over the yard (stack visits), that
determines the required yard area, is the number of containers over the quay (terminal throughput)
34 LITERATURE STUDY
corrected for transhipment containers. This is defined by the following equation, based on Saanen
(2004):
(2)
Where,
Cy Number of container movements over the yard per year [TEU/year].
C Throughput. Number of container movements over the quay per year [TEU/year].
Ctrans Number of transhipment containers per year [TEU/year].
µ Ratio of transhipped containers [-].
The previously mentioned cargo flows over the quay and over the storage yard are depicted in
Figure 15. To help clarify this; a 100% transhipment terminal requires half the storage yard
capacity of a 100% import/export terminal while the throughputs (movements over the quay) are
the same, from Saanen (2004).
Figure 15: Container flow over the quay and yard. Based on: Saanen (2004)
2.1.3.2 Quay length

Two quick estimation methods exist for determining the required quay length. Also empirical
rules-of-thumb exist that can be used as a check for calculated quay lengths. The methods are:
1. Berth productivity
2. Maximum service time
3. Empirical rules-of-thumb
Berth productivity
This method uses the annual throughput, crane productivity and number of cranes to determine
the required quay wall length. First the berth productivity is calculated with the following
equation from PIANC (2014b):
(3)

Where,
Cb Berth productivity [TEU/year].
Average gross productivity per crane [moves/hour]. Average number of containers moved
between berthing completed and de-berthing started. This variable therefore includes
unproductive intervals such as crane repositioning, moving hatches to/from quay and time
between shifts.
fTEU TEU factor; the ratio between 40 ft. and 20 ft. containers [-]. Defined in Equation (5),
from Böse (2011).
ncb Number of cranes per berth [-].
nhy The number of operational hours per year [hour/year]. Very much dependent on the local
situation.
mb Estimated berth occupancy factor [-]. The acceptable occupancy factor depends on the
allowable waiting time in terms of service time, the number of berths and the sort of
terminal (e.g. dedicated shipping line). The factor can be calculated with the following
equation, based on Saanen (2004):
(4)
Where,
Ls,i Length of vessel i of total of n vessels berthed in measurement period nhy [m]. This
length should include twice the Lml (see Equation (7)). Only vessel i = n should
include this additional length once.
Ts,i Service time (or effective berthing time) of vessel i of total of n vessels berthed in
measurement period nhy [hour]. Time the vessel is berthed outside of the
operational hours does not count.
Lq Quay length [m].
(5)
Where,
fTEU TEU factor; the ratio between 40 ft. and 20 ft. containers [-]. A value of 1.60
means 60% of the containers are FEU’s.
N20 Number of TEU’s [-].
N40 Number of FEU’s [-]. A factor of two is used in the equation since two TEU’s
equal one FEU.
The required number of berths is then calculated by the equation from PIANC (2014b):
(6)
Where,
nb Number of required berths [-].
C Number of container movements over the quay per year [TEU/year].
Cb Berth productivity [TEU/year].
36 LITERATURE STUDY
Then with the determined number of berths the required quay wall length is calculated by means
of the following equations from PIANC (2014b):
(7)
Where,
Lq Required quay wall length [m].
Ls,max The length of the largest vessel frequently calling at the terminal [m].
The average vessel length [m].
Lml The additional quay length required for and aft of the vessel to account for the mooring
lines [m].
nb Number of berths [-].
The factor 1.1 accounts for variability in the vessel length since an average value is used, results
from a study done by UNCTAD (1984).
Maximum service time

Another method to determine the required number of STS cranes, the respective crane
productivity and the number of berths is to pose a maximum allowable service time for a vessel. A
common maximum of 24 hours of time-in-port is used since the service time has a large influence
on a vessel’s operating costs; from Saanen (2004) and Quist & Wijdeven (2014). The required
quay length can then be calculated with the equations in the previous section.
Empirical rules-of-thumb
A few empirical rules exist that are mostly used for a quick check of the calculated quay length.
The first rule-of-thumb gives empirically determined benchmarks for the berth capacity in TEU
per metre of quay per year that are given in Table 6. The table originates from Drewry (2010).
The values are based on actual, non-estimated, throughput data of the year 2009 of 201 container
terminals with a throughput of more than 200,000 TEU. The quay length can be calculated by
dividing the annual container throughput by the berth capacity.
Table 6: Berth capacity benchmarks
North Latin Southeast

Region Europe Far East
America America Asia
TEU per
526 712 742 1,224 1,578
metre quay
Middle
Region South Asia Others World
East
TEU per
1,341 1,216 743 933
metre quay
Notes: Container terminals with throughput > 200,000 TEU in 2009.
Only terminals with STS cranes are considered.
Others includes Africa and Oceania.
Source: Drewry (2010)
Another berth capacity is proposed by Ligteringen & Velsink (2012) and is 300 to 1000 TEU per
year. A remark is made that the large uncertainty is due to the many types of equipment and the
number and type of ships.

The second rule-of-thumb is based on an average quay length per STS crane. According to
PIANC (2014b) on average one crane per 100 metre of quay is common on container terminals.
This can however increase to one crane per 75 metre for high capacity terminals. To calculate the
quay length the average metres per crane should be multiplied with the required number of cranes
(nc), that can be calculated with the following equation that is derived from Equation (3):
(8)
Where,
nc Required number of cranes [-].
C Number of container movements over the quay per year [TEU/year].
nhy The number of operational hours per year [hour/year].
Average gross productivity per crane [moves/hour]. Average number of containers moved
between berthing completed and de-berthing started. This variable therefore includes
unproductive intervals such as crane repositioning, moving hatches to/from quay, time
between shifts and simple repairs.
fTEU TEU factor; the ratio between 40 ft. and 20 ft. containers [-]. Defined in Equation (5).
mb Estimated berth occupancy factor [-]. This factor depends on the allowable waiting time in
terms of service time, the number of berths and the sort of terminal (e.g. dedicated
shipping line).
2.1.3.3 Terminal area

The terminal area can be calculated with:
1. General approach
2. Empirical rule-of-thumb
General approach
The total terminal area can be determined by a summation of the gross required storage areas for
the different flow directions (import, export and transhipment; see Figure 15) plus the required
area for empties and the container freight station, see Figure 16. The equations needed for these
elements are given in this section. The desirable total depth (from quay wall to landside terminal
boundary) of a container terminal lies between 400 and 500 metres depending on local conditions.
38 LITERATURE STUDY
Figure 16: Container terminal with total terminal area (blue) and gross storage areas (red). S.L. Port of
Barcelona (Spain). Source: Google Earth
The general equation for determining the gross required storage area per container flow direction
is based on the version of Quist & Wijdeven (2014) of an equation originating from UNCTAD
(1985), and is as follows:
(9)
Where,
Agr,i Gross required storage area per flow direction and/or stack type (including internal
roads)10 [m2]. For import, export and transhipment cargo flows and empties stack.
Cy,i Number of container movements over the yard per year per flow direction [TEU].
Average dwell time of containers in the stack [days]. Average time a container spends in
the yard.
ATEU Gross storage density for maximum stacking height [m2/TEU]. The required area per
container in a fully utilised stack including roads in between the stacks.
rst Ratio of average stacking height over maximum stacking height [-]. In order to limit the
amount of repositioning of containers in the stack, which is larger for higher stacks.
ms Estimated storage occupancy [-]. Average number of containers in yard divided by the
design capacity of the yard (taking into account re-positioning, so rst). To account for the
random arrivals and departures of cargo.
The gross required storage area plus additional area for infrastructure and buildings on the
terminal (see Section 2.1.1) is the required total terminal area, calculated by the following
equation (own work):
(10)
10
The gross required storage area includes the net storage area (purely container stacks)
and internal roads in between the stacks. The apron, other roads and buildings on the terminal are included in the
required total terminal area (At).

Where,
At Required total terminal area [m2]. Area including the stacks, storage yard, apron,
infrastructure and buildings.
Agr,i Gross required storage area per flow direction and/or stack type (including internal roads)
[m2]. For import, export and transhipment cargo flows and empties stack.
ACFS Required area for the container freight station [m2]. See Equation (11).
α Total terminal factor; the percentage of storage area and internal roads in the stack with
respect to the total terminal area [-].
Container freight station
To determine the required area for the container freight station the following equation is used,
based on a version of Quist & Wijdeven (2014) of an originating equation from PIANC (2014b),
and is as follows:
(11)
Where,
ACFS Required area for the container freight station [m2].
CCFS Number of container movements per year through the CFS [TEU].
Vc Volume of 1 TEU container [m3].
Average dwell time of containers in the CFS [days]. Time between arrival of vessel and
departure of container from the terminal for import containers, for export vice versa
according to Monday Nyema (2014).
farea Ratio gross over net area of CFS [-]. To account for containers being stored around the
CFS during transfer of cargo.
fb Bulking factor of CFS [-]. To account for cargo that needs special treatment or repairs.
Average stacking height in the CFS [m].
ms Estimated storage occupancy [-]. Average number of containers in yard divided by the
design capacity of the yard. To account for the random arrivals and departures of cargo.
Empirical rule-of-thumb
According to Saanen (2004) the required total terminal area can be determined by the annual
number of containers over the quay divided by the maximum annual container handling capability
per total area of the terminal (or container storage factor fs). The container storage factors
according to Saanen (2004) can be used as a check for a calculation, the factors are:
 2.3 TEU/year/m2 for gateway terminals (import/export terminals).

 5.0 TEU/year/m2 for hub terminals (transhipment terminals).
A terminal can be classified as a gateway terminal if the ratio of transhipped containers µ ≤ 0.5. A
terminal can be classified as a hub terminal if µ > 0.5.
In Table 7 average container storage factors are listed, based on Drewry (2010). The values are
based on actual, non-estimated, throughput data of the year 2009 of 201 container terminals with a
throughput of more than 200,000 TEU. The figures are averages, therefore variation because of
terminal types (gateway or hub) is not taken into account. The higher values for the regions in
Asia are due to the large amount of hub terminals, based on Heymann (2006).
40 LITERATURE STUDY
Table 7: Storage factor benchmarks
North Latin Southeast

Region Europe Far East
America America Asia
TEU/year/m2 0.74 2.09 1.74 3.37 5.66
Middle
Region South Asia Others World
East
TEU/year/m2 2.84 3.34 2.18 2.25
Notes: Container terminals with throughput > 200,000 TEU in 2009.
Only terminals with STS cranes are considered.
Others includes Africa and Oceania.
Based on Drewry (2010)
Ligteringen & Velsink (2012) proposes 0.6 to 1.0 TEU per year per terminal surface.
2.1.3.4 Water depth

When few data is available a quick estimation of the required depth to be dredged can be made.
The required depth is dependent on the static draught of the fully loaded design vessel, the gross
underkeel clearance and a dredging tolerance. The gross underkeel clearance consists of the net
underkeel clearance and vertical ship motions (due to swell, waves, squat and trim). The required
depth can be calculated with the following equation, based on PIANC (1995):
(12)
dd Required dredging depth [m].

Ds Draught of design vessel [m].
hguc Gross underkeel clearance [m].
hdt Dredging tolerance [m].
In PIANC (1995) a minimum gross underkeel clearance (hguc) is specified as a percentage of the
ship’s draught (Ds). Values are presented in Table 8 for various wave conditions. For container
vessel dimensions see Table 1.
Table 8: Minimum gross underkeel clearance for certain wave conditions
Wave conditions
Sheltered waters 10 %
Wave height ≤ 1.0 m 30 %
Higher waves with

unfavourable periods and 50 %
directions
These values apply to large ships (≥ 200,000 DWT), it is an overestimation for smaller vessels.
Source: PIANC (1995)

2.1.3.5 Common values of parameters
Values that are proposed in literature of the input parameters used in the design rules are listed in
Table 9. For various parameters minimum and maximum values are proposed, resulting in a range
of possible values. This can be represented by a uniform random distribution. When this is the
case the reader should make an estimate of which part of the distribution to use. The sources of
the values are numbered in square brackets.
Table 9: Parameter values of container terminal design rules
Symbol Unit Description and values

Ratio gross over net area of CFS.
farea -
A value of 1.40 is proposed by [2].
Bulking factor of CFS.
fb -
Ranges between 1.10 and 1.20 [2].
Storage factor.
Value of 2.3 for gateway terminals and 5.0 for transhipment
fs TEU/year/m2 terminals [9]. Global average of 2.2 regardless of terminal
classification. Between 0.6 and 1.0 [12]. For values per global
location see Table 7 (p.41) [10].
Ratio between 40 ft. and 20 ft. containers.
Typical value for a modern terminal is 1.50 [5] or 1.60 [1]. Can be
fTEU -
as high as 1.9 for developed countries or smaller than 1.5 for
underdeveloped countries [5].
Dredging tolerance.
hdt m
Minimum gross underkeel clearance.
hguc m
For values see Table 8 (p.41).
Average stacking height in CFS.
m
Height of 1 container is 2.60 m.
Estimated berth occupancy.
mb -
Estimated storage occupancy.
Determined in Expert Judgement Elicitation, see Section 0.
ms -
Literature: Typical values around 70% [5]. Between 65% and 70%
[2]. For CFS between 60% and 70% [2].
Number of cranes per berth.
ncb -
On average 3 cranes per berth [2]. Maximum of 7 [7].
Number of operational hours per year.
Very much dependent on local situation. A typical modern
nhy hour/year
terminal operates 24 hours a day for 360 days a year; this is 8640
hours per year [5].
Ratio average stacking height over maximum stacking height.
rst -
Average dwell time of containers in stack.
Typical values for full containers are 4 - 10 days [8]. They should
days not exceed 5 days [11]. Imported containers 6 - 7 days [8], export
containers 4 - 5 days [8]. Empty containers have dwell times from
7 - 20 days [5 & 8]. CFS’s have typical dwell times of 5 days [4].
Gross storage density.
ATEU m2/TEU
Draught of the design vessel.
Ds m
Lml m Additional quay length required for and aft of vessel to account
42 LITERATURE STUDY
for mooring lines.
Normally 15 metres [2] or 30 metres [5] is used.
m Average vessel length.
For regular vessel lengths see Table 1 (p.25).
m Length of largest vessel frequently calling at terminal.
Ls,max
For regular vessel lengths see Table 1 (p.25).
Average gross productivity per crane.
moves/hour
Volume of 1 TEU container.
Vc m3
The volume is 29 m3.
Total terminal factor.
Literature: Ranges between 55% and 80% [3], or between 60% and
α -
70% [8]. A typical value is 75% [1]. Between 50% and 70% when
CFS within terminal boundaries, between 60% and 80% without
CFS in terminal [9].
Sources:
[1] Personal communication with S. Meijer (19-04-2016)
[2] Quist & Wijdeven (2014)
[3] Monfort et al. (2011)
[4] Kersten (2010)
[5] PIANC (2014b)
[6] Agerschou (2004)
[7] Personal communication with B.A. Pielage.
[8] PIANC (2014a)
[9] Saanen (2004)
[10] Drewry (2010)
[11] Fourgeaud (2000)
[12] Ligteringen & Velsink (2012)
[13] PIANC (1995)
Berth occupancy factor
The acceptable berth occupancy factor depends on the number of berths and the allowable average
waiting time in terms of average service time of vessels (Tw/Ts). Based on various economic
feasibility studies a reasonable value for Tw/Ts for container terminals is found to be 0.10;
according to Agerschou (2004). The probabilistic Queuing Theory is used to determine the
occupancy ratios. This theory is not treated in this research since it is outside its scope, for a better
understanding of the general Queuing Theory reference is made to Sztrik (2012). An application
of the theory on terminal design can be found in Agerschou (2004).
The queuing system can be denoted by the Kendall notation: “inter arrival time
distribution”/”service time distribution“/”number of berths”. Arrivals of vessels can be random for
common user terminals, a negative exponential inter arrival time (denoted by M) is then assumed.
When arrivals are less random an Erlang-K distribution (denoted by EK) with K = 2 is assumed.
Service time distributions are also taken as an Erlang-K distribution with a value of K = 4 (the
higher the K-value the more constant the inter arrival or service times are). Realistic queues are
M/E4/n for common user container terminals and E 2/E4/n for dedicated shipping line container
terminals, according to Monfort et al. (2011) and Terblanche & Moes (2009). Acceptable berth
occupancy factors corresponding to these queues for Tw/Ts = 0.10 are given in Table 10.

Table 10: Acceptable berth occupancy factor for container terminals for T w/Ts = 0.10
Number of Acceptable berth occupancy

berths nb factor mb [%]
[-] Common-user Dedicated
M/E4/n E2/E4/n
1 14 31
2 36 53
3 49 63
4 57 70
5 63 73
6 or more 67 77
Source: Monfort et al. (2011)
To use this table an iteration must be performed. First a value for the occupancy has to be
assumed, then the number of required berths is calculated with Equations (3) and (6). The
occupancy corresponding to this calculated number of berths must then be checked with the
assumed occupancy. If they do not correspond the same steps have to be taken again. The berth
occupancy factor tables are not used in the tool since in some cases this iteration proves to be
divergent. A manual iteration by the user of the tool by using the information in the table is of
course still possible. Saanen (2004) states that most terminals try to keep the berth occupancy
below 60 to 65%.
44 LITERATURE STUDY
2.2 Dry bulk terminals
For dry bulk terminals first general background information is presented in order to give the
guidelines, rules and common values of parameters are treated.

This section considers dry bulk terminals and is based on Agerschou (2004), Ligteringen &
Velsink (2012), PIANC (2014b), UNCTAD (1985) and GreenPort (2016). Information is
provided about dry bulk commodities, vessels and terminal elements.
Dry bulk is characterised as cargo that is loaded or discharged in a loose form. In 2010 over one
third of the international seaborne trade consisted of dry bulk. The most transported commodities
are coal, iron ore, grain, phosphate rock, bauxite/alumina forming respectively 30.75%, 28.25%,
10.72%, 0.72% and 2% for bauxite of the total annual worldwide shipment of dry bulk in 2010
according to PIANC (2014b).
Only the three most important commodities (coal, iron ore and grains in general) are included in
the computer tool, this to make it less complex. In this section however more commodities are
treated in order to give the reader an overview of the different possibilities.
2.2.1.1 Commodities
Various properties of the cargoes should be taken into account when designing dry bulk terminals.
These properties of the cargoes are:
 Cargo density
 Angle of repose: maximum angle with respect to the horizontal to which a material can be
piled without slumping.
 Dust generation
 Hazardous properties
o Susceptibility to fire/explosion
o Spontaneous combustion
 Resistance to degradation by mechanical handling
 Handling properties
o Corrosiveness
o Abrasiveness
For this research however only the main dimensions of the terminal are analysed. Resistance to
degradation and the handling properties of the cargoes do not directly influence these dimensions,
therefore these properties are not taken into account. The commodity types that are included in the
research are listed below including some examples of cargoes:
 Minerals
o Iron ore
o Bauxite
o Phosphate rock
 Coal
o Thermal coal
o Anthracite coal
2.2 Dry bulk terminals 45

o Metallurgical coal
 Wood products
o Wood chips
o Wood pellets
 Agricultural products
o Grain
 Wheat
 Rye
o Maize
o Sugar
o Soybeans
 Others
o Alumina
o Cement
Properties of the mentioned goods are listed in Table 11.

Table 11: Material properties
Hazardous properties
Cargo Angle of Common
Dust Susceptibility
Commodity density repose storage Spontaneous
generation to fire/dust
[tonne/m3] [] location combustion1
explosion
Iron ore 2.13 - 3.03 30 -50 Outdoors Yes No No
Bauxite 1.09 - 1.19 28 – 49 Outdoors Yes No / yes No

Phosphate
1.02 - 1.09 30 -35 Both Yes No No
rock
Coal 0.52 - 0.93 30 - 45 Outdoors Yes Yes / yes Yes
Wood chips 0.15 - 0.40 42 - 50 Outdoors Yes Yes / yes Yes

Wood
0.60 - 0.70 32 - 39 Indoors Yes Yes Yes
pellets
Wheat 0.75 - 0.85 25 - 30 Indoors Yes Yes / yes No
Rye 0.71 30 Indoors Yes Yes / yes No
Maize 0.71 - 0.80 30 -40 Indoors Yes Yes / yes No
Sugar 0.80 - 0.90 40 Indoors Yes Yes / yes No
Soybeans 0.78 - 0.81 30 Indoors Yes Yes / yes Yes
Alumina 1.19 - 1.43 35 Indoors Yes No No
Cement 1.56 - 1.64 35 Indoors Yes No / yes No

Note 1: Sensitivity to spontaneous ignition requires maximum stockpile heights
Based on Agerschou (2004), Ligteringen & Velsink (2012), PCA Consultants (n.d.), Wu (2012) and material
safety data sheets.
46 LITERATURE STUDY
2.2.1.2 Vessels
In Table 12 an overview of the most important vessel classes with corresponding vessel sizes is
presented.
Table 12: Dry bulk vessel classes and dimensions
Number of
DWT Length Draught Beam
Class holds
[tonne] [m] [m] [m]
[-]
Handysize 10,000 - 35,000 115 - 170 7 - 10 14 - 27 3-5
Handymax 35,000 - 55,000 180 - 190 10 - 12 27 - 32 5-7
Panamax 55,000 - 80,000 200 - 290 12 - 15 32.2 7
Capesize 80,000 - 150,000 230 - 280 14 - 18 35 - 45 8-9
VLBC > 150,000 280 - 362 18 - 24 45 - 65 9 - 11
Source: Kleinheerenbrink (2012)
Handysize and Handymax vessels are used for the transport of different types of cargoes between
smaller ports. Capesize and VLBC (Very Large Bulk Carrier) vessels are used for the transport of
coal and iron ore and are too big to pass the Panama and Suez canals. For a comparison between
the Handymax and VLBC vessel classes see Figure 17.
Figure 17: Example of dimensions of a Handymax and VLBC vessel. Source: Kleinheerenbrink (2012)
2.2.1.3 Dry bulk terminal elements

Dry bulk terminals consist of the following elements (see Figure 18):
 Quay or jetty
 Loading/unloading equipment
 Horizontal transportation equipment
 Storage equipment
 Storage facilities
Differences exist between for example large-scale export terminals and import terminals. The
basic elements are similar but particulars are dependent on local conditions and the transported
commodities.

Figure 18: Schematic representation of typical dry bulk terminal elements. Based on: Monfort et al. (2011)
2.2.1.4 Quay/jetty
Dry bulk vessels can have large draughts. Especially vessels that transport ores can have draughts
of up to 24 metres because of the large cargo density and thus large DWT. For these kind of
vessels it can be more economical to realise a jetty (pier) instead of a quay wall.
Quay walls exist in many forms with the main types being sheet pile walls and gravity walls.
Combinations between and variants of these types exist as well, for more information reference is
made to de Gijt (2010).
Jetties can be positioned parallel or perpendicular to the terminal. Perpendicular placed jetties are
called finger piers. Parallel positioned jetties are called T-shaped or L-shaped jetties, see Figure
19. These types of jetties can have separate mooring dolphins on one or both ends, connected to
the main structure with a catwalk.
48 LITERATURE STUDY
Figure 19: Schematic overview of three different jetty types. Source: Own work
On the quay and on a pier loading and/or unloading equipment is positioned. The transport of bulk
to and from the jetty is done with conveyor belts or pipelines.
2.2.1.5 Loading/unloading equipment

In contrast to container STS cranes the equipment needed for the loading and unloading of dry
bulk differs per transport direction. The selection of equipment depends on the type and quantity
of the bulk material, space and environmental conditions and the intensity of operations. The
available equipment types are listed in this section.
Loading equipment
Loading equipment are often relatively simple machines since most bulk can be dropped in the
cargo holds making use of gravity. Basically the loader consists of a feed conveyor and a chute.
Typical loading equipment types are (see Figure 20):
 Quadrant loaders: Consist of a bridge that is supported by a pivot point landwards and a
circular track seawards and has a telescopic boom. A shuttle system provides full hatch
coverage. One or two quadrant loaders can be used on one vessel.
 Linear loaders: Consists of a bridge that is supported by a pivot point landwards and a
seawards track parallel to the ship. The bridge moves along its length over the pivot
causing the front side of the bridge to move parallel to the ship. A loading boom is
connected to the bridge.
 Travelling loaders: Boom moving parallel to the ship, bulk material is fed to the boom via
a conveyor.
For very dusty materials like cement pneumatic loader/unloaders are used. For small terminals
with low capacity requirements wheel mounted mobile installations can be used. Another option
is a fixed loading point where the ship has to move to distribute the cargo.

Figure 20: Clockwise starting from the left: Quadrant loader, linear loader and travelling loader. Source:
Google
Unloading equipment
Unloading of bulk carriers requires more activities and therefore the installations are more
complex. There are five main unloading systems (see Figure 21):
 Gantry grab unloaders: Rail mounted with a cantilever boom and grab that can move
perpendicular to the berth. Discharges through a hopper onto a conveyor.
 Grab rigged harbour cranes
o Level luffing cranes: Moves on rails and discharges into a hopper in front of the
crane avoiding the need the rotate during discharging.
o Mobile harbour crane: Moves on wheels. The movement takes longer than for rail
mounted cranes since the outriggers have to be moved every time.
 Continues unloaders: Rail mounted with a rotating boom of fixed length and a vertical
conveyor system with a steerable digging foot. The vertical conveyor system can be a
bucket elevator, a screw or a spiral conveyor. They can discharge onto belt conveyors,
trucks or rail wagons.
 Pneumatic unloaders: Exists in the form of suction or pressure types. They can be wheel
or rail mounted and are used to handle dry, low density bulk material (e.g. cement or
grain). They discharge into pipeline systems or silos on shore.
50 LITERATURE STUDY
Figure 21: Clockwise starting from the top left: Gantry grab, level luffing cranes, bucket elevator, pneumatic,
spiral and screw conveyors (with screw detail). Source: Google
2.2.1.6 Horizontal transport equipment

Transport within the terminal from ship to stacker or from reclaimer to ship is generally done by
belt conveyors. For dusty materials conveyor belts are covered over the full length, also water can
be used to reduce the creation of dust. Covered conveyer belts are also used for cargoes that can
suffer from the weather, such as grains. Transport by means of water and pneumatic transport is
done through pipelines.
For smaller terminals another option is bulldozers that move the cargo between storage and ship.
This is much cheaper but also dramatically reduces the throughput capacity.
For export terminals often a rail connection is used. At the terminal the wagons are emptied to a
belt conveyor or pneumatic discharge system. For import terminals the transport out of the
terminal is done by barges, road or rail. This can be done directly from the bulk carrier or via the
stockpile. For road and rail transport often loading silos are used.
Recommended are shared routes (a belt conveyer and corresponding equipment that can transport
cargo in both directions), unless for a specific reason a dedicated route is applied.

Figure 22: Belt conveyor. Source: Google
2.2.1.7 Storage equipment

The handling of material in the storage yard is generally done by bulldozers or by stackers,
reclaimers or a combination of both systems; a stacker-reclaimer.
Stackers are machines that move on rails parallel to a stockpile with a perpendicular boom
containing a belt conveyor. A stacker generally can make piles on either side of the rails. A belt
conveyor is positioned under the stacker along the stockpile length and is used to transport the
material from the stockpile to the (un)loading equipment or the inland transport loading system.
Reclaimers are similar machines but equipped with for example a bucket wheel in order to
reclaim the material. Both functions are combined in a stacker-reclaimer, only one function can be
performed at a time however. Other means of reclaiming are scraper reclaimers or underground
belt conveyors that transport material that is pushed on it with bulldozers. The mentioned
machines are depicted in Figure 23.
Figure 23: Loader (top left), scraper reclaimer (bottom left) and stacker-reclaimer (right). Source: Google
52 LITERATURE STUDY
2.2.1.8 Storage facilities
There are three kinds of storage facilities for dry bulk:
 Open storage
 Covered storage
 Silos
A general observation is that the more different products a terminal handles the more storage area
is required.
Open storage
The main type of storage is open storage in stockpiles. The dimensions of the stockpile are
dependent on the angle of repose of the material and the discharge of the storage equipment. Open
storage can be used for materials that do not impact the environment and do not suffer from
serious degradation by exposure to the elements. The height of a stack is determined by the
bearing capacity of the soil and the reach of the stacker/reclaimer. The most common way to store
bulk material is the wind-row arrangement where long stacks are formed. Another way is to make
circular piles with a stacker/reclaimer in the centre. This form is used at smaller terminals, it is not
included in the computer tool because of time limitations.
For dusty materials that cause pollution or dust explosions a protective foam layer can be used or
the stockpile can be sprayed with water.
Figure 24: Dry bulk wind-row stockpiles with stacker-reclaimer. Source: Google
Covered storage
Materials that can suffer from the weather or that impact the environment require covered storage.
The main types of covered storage are portal framed structures (horizontal storage or sheds) and
domes. Discharge into the horizontal storage usually takes place via a belt conveyor positioned in
the top of the structure. Reclaiming is done by a scraper reclaimer, an underground conveyor or
bulldozers. For an example of a covered structure see Figure 25. In a dome circular stackers and
scraper reclaimers or bulldozers are used. The use of covered storage increases over the years
because of the increasing attention at the environmental impact of dry bulk terminals. Domes are
not further treated in this study for they are not included in the computer tool because of time
limitations.

Figure 25: Covered storage with a belt conveyor for stacking. Source: Google
Silos
Vertical circular silos are a specific form of covered storage and are typically used for the storage
of free flowing dusty materials such as grains or cement. Moreover silos exist that can be used to
store non-free flowing materials; Eurosilo (2016). The silos are filled with covered conveyors
from the top and are emptied at the bottom to conveyors or pipes, or directly to trucks or train
wagons. An example of a terminal with silos is depicted in Figure 26. For non-free flowing
materials a central vertical screw conveyer is used to realise the outflow. Advantages of silos are
that no reclaiming equipment is needed (gravity is used) and a limited space is needed.
Disadvantages are the higher costs and the limitation to a specific commodity.
Figure 26: Artist impression of silos on a terminal. Source: Google

The properties of the terminal equipment that are required for the dimensioning of a dry bulk
terminal are given in this section. A division is made between ship (un)loading equipment on the
quay and stackers and reclaimers in the storage area.
54 LITERATURE STUDY
2.2.2.1 Ship loading and unloading equipment
Productivity rates of unloaders and loaders are dissimilar. Typical rated capacities for unloaders
are listed in Table 13. The rated capacity (or free digging rate) is the unloading rate based on the
cycle time of a full bucket or grab from the digging point in the ship to the hopper and back. Some
of these values are common minimum and maximum values, although lower or higher capacities
may exist. The reader should make an estimate of values that are appropriate to be used in a
specific case. For the coal and iron ore unloading process grab cranes or bucket elevators are used.
Grain can be unloaded by all listed unloaders.
Table 13: Typical rated capacities for unloaders
Typical rated capacity

Unloader type
[tonne/hour]
Gantry grab unloader 500 - 3000
Grab rigged harbour crane 200 - 1500
Continues unloader
Bucket elevator 1000 - 5000
Chain conveyor 200
Vertical screw conveyor 900
Spiral conveyor 75
Pneumatic unloader 200 - 500
Based on Ligteringen & Velsink (2012), UNCTAD (1985) and PIANC (2014b).
Normal load capacities vary between 500 and 7,000 tonne/hour according to Kleinheerenbrink
(2012) and Vianen, T. van (2015). Kleinheerenbrink even reports capacities as high as 20,000
tonne/hour.
The effective capacity of (un)loaders is the typical rated capacity including interruptions for e.g.
cleaning, moving between holds and small repairs. The ratio between the effective capacity and
the typical rated capacity is called the ‘through ship efficiency’, these factors are given in Table
14.
Table 14: Through-ship efficiency factors
Unloading Loading
Through-ship efficiency 0.5 0.7
Source: UNCTAD (1985)
Lodewijks et al. (2009) states that in reality the gross productivity of loaders and unloaders is
much less than the effective capacity. This is due to operational availability of the equipment,
which is around the 80%. The downtime of equipment can have a significant influence on the
berth productivity. According to Vianen et al. (2014) the installed unloading capacity of coal
and/or iron ore import terminals is 3 to 4.5 times the minimum required unloading capacity. For
export terminals, and thus loading equipment this factor is 1.5 to 2.5. In the tool this factor is not
taken into account since it is no specific design rule used in early phase terminal planning,
according to the studied literature.
2.2.2.2 Stackers and reclaimers

According to UNCTAD (1985) typical reclaim capacities for a stacker-reclaimer are between
1000 and 3000 tonne/hour while stacking capacities of 6000 tonne/hour and more are reached.
Scraper-reclaimers can achieve reclaim capacities of up to 1000 tonne/hour.

Vianen et al. (2014) assumes for coal and iron ore terminals as common gross hourly capacity of
stackers 2000 tonne/hour, for reclaimers 1500 tonne/hour, for stacking of a bucket-wheel stacker-
reclaimer 2000 tonne/hour and for reclaiming of the same machine 1500 tonne/hour.
Kleinheerenbrink (2012) gives the capacities for coal and iron ore terminals as listed in Table 15.
Table 15: Stacking and reclaiming capacities
Coal Iron ore

[tonne/hour]
Average Maximum Average Maximum
Stacking 3000 10,000 3500 10,000
Reclaiming 2000 6000 2500 15,000

The design guidelines, rules and parameters are provided for determining the number of berths
and storage yard area. The required dredging depth can be determined as in Section 2.1.3.4.
Values used in practice for the parameters are listed in Section 2.2.3.3.
2.2.3.1 Number of berths

First a method is treated with which the required number of berths can be determined. Also a
method is considered that can function as a check. The methods are:
 Berth productivity
 Empirical rule-of-thumb
Berth productivity
The method to calculate the required number of berths is similar to the berth productivity method
explained in Section 2.1.3.2. However the cargo is not expressed in TEU but in metric tonnes. The
berth productivity is therefore calculated by the following equation, based on PIANC (2014b):
(13)
Where,
Cb Berth productivity [tonne/year].
Average gross productivity per ship (un)loading equipment [tonne/hour]. Average number
of tonnes moved between berthing completed and de-berthing started. This variable
therefore includes unproductive intervals such as crane repositioning, moving hatches
to/from quay, time between shifts and simple repairs.
ncb Number of (un)loaders per berth [-].
mb Berth occupancy factor [-]. The acceptable occupancy factor depends on the allowable
waiting time in terms of service time, the number of berths and the sort of terminal (e.g.
dedicated shipping line). For berthing at a quay wall the factor can be calculated with the
following equation, based on Saanen (2004):
(14)
56 LITERATURE STUDY
For berthing at a jetty the berth is treated as a binary object; the berth is either free or
occupied. Lengths are therefore not taken into account. For this situation the factor can be
calculated with the following equation, based on Saanen (2004):
(15)
Where,
Ls,i Length of vessel i in [m] of total of n vessels berthed in measurement period nhy.
This length should include twice the Lml (see Equation (7)). Only vessel i = n
should include this additional length once.
Ts,i Service time (or effective berthing time) in [hour] of vessel i of total of n vessels
berthed in measurement period nhy. Time the vessel is berthed outside of the
Lq Quay length [m].
The required number of berths is then calculated by the following equation from PIANC (2014b):
(16)
Where,
nb Number of required berths [-].
C Mass of handled cargo per year [tonne/year].
Then with the determined number of berths the required quay wall length can be calculated
according to the following equation from PIANC (2014b):
(17)
Where,
Lq Required quay wall length [m].
Ls,max The length of the largest vessel frequently calling at the terminal [m].
The average vessel length [m].
Lml The additional quay length required for and aft of the vessel to account for the mooring
lines [m].
nb Number of berths [-].
The factor 1.1 accounts for variability in the vessel length since an average value is used, results
from a study done by UNCTAD (1984).
Instead of a quay wall a jetty can be used as well, the specific form has to be determined. The
dimensions are among other things dependent on the design ship length, required length for the
mooring lines and the required water depth. In the tool jetty dimensions are approximated in order
to be able to derive the costs of this terminal element. The method that is used is described in
Section 2.4.2.
Empirical rule-of-thumb
Vianen et al. (2014) gives empirically determined quay capacity characteristics in tonne per metre
of quay per year for coal and iron ore terminals, both for import and export. These figures are
based on his study of 49 dry bulk terminals. The resulting values are given in Table 16. The quay

length can be calculated by dividing the annual throughput by the quay capacity factor, this can be
used as a check for calculated quay lengths.
Table 16: Quay capacity factors for coal and iron ore terminals
Quay capacity factor

Commodity [x103 tonne/year/m1]
Import terminals Export terminals
Coal 10 - 30 10 - 30
Iron ore 25 - 75 50 - 150
Source: Vianen et al. (2014)
2.2.3.2 Storage area

To determine the required storage area for a dry bulk terminal a few methods exist. First there is a
conventional equation that uses the annual throughput and average dwell time of the cargo.
Secondly two rules-of-thumb exist that can be used to estimate the required storage area; these
rules are based on a capacity ratio and a storage factor.
General approach
This method makes use of the conventional equation to calculate the required gross storage area
for general cargo/multi-purpose terminals, based on PIANC (2014b):
(18)
Where,
Agr,i Gross required storage area per commodity type11 [m2].
Ci Mass of handled cargo per year per commodity type [tonne/year].
Average dwell time of cargo in the stockpile [days].
farea Ratio between gross and net storage area [-].
Average cargo density [tonne/m3].
Average stacking height [m].
ms Estimated storage occupancy [-]. Average mass of cargo in stockpile divided by the design
capacity of the stockpile. To account for the random arrivals and departures of cargo.
Instead of calculating the required area with Equation (18), the equation is rewritten in order to
calculate the required volume of the storage at a certain moment. In order to calculate a volume
both sides of the equation are multiplied by the average stacking height . The ratio between
gross and net storage area is used in a later calculation step. This results in the following equation:
(19)
Where,
Vstorage,i Required total storage volume per commodity type [m3].
Average dwell time of cargo in the stockpile [days].
11
The gross storage area includes the net storage area and internal roads, pipelines and/or conveyor belts and
equipment rails at the storage yard. Other buildings and infrastructure in between quay and yard on the terminal
are included in the required total terminal area (At).
58 LITERATURE STUDY
ms Estimated storage occupancy [-]. Average mass of cargo in stockpile divided by the design
capacity of the stockpile. To account for the random arrivals and departures of cargo.
In this study wind-row stockpiles, covered storage sheds and silos are considered. Wind-row and
covered storage use the same methods of calculating the required stockpile length and storage
area. For storage in silos the required number of silos must be determined, the required storage
area can then be calculated with a method that is described in Appendix VI . More information
about common dimensions of the three different storage types can be found under the heading
‘Storage types’ in this subsection. The calculation method of wind-row and covered stockpiles is
presented in this subsection.
Wind-row stockpiles are lanes consisting of a trapezoidal cross-section of a certain length. A lane
only consists of the pile of cargo, the adjacent infrastructure is not included. The area occupied by
internal infrastructure is taken into account by farea in Equation (22). This lane length can be
calculated with the following equation, derived from Vianen et al. (2012) and including the
number of lanes in the stockpile (assuming equal lane lengths for all lanes):
(20)
Where,
llane Required lane length [m].
Vstorage Required total storage volume [m3].
hpile Height of the pile [m].
wlane Width of a lane [m].
θ Angle of repose [degrees].
nlanes Number of lanes in stockpile [-]. The lanes are assumed to have equal length.
hpile,max Maximum height the pile can attain [m], defined in Vianen et al. (2012) as:
(21)
Where,
θ Angle of repose [].
The gross required storage area per commodity type can then be calculated with the following
equation (own work):
(22)
Where,
Agr,i Gross required storage area per commodity type [m2].
nlanes Number of lanes in stockpile [-].
llane Required lane length [m].
farea Ratio between gross and net storage area [-].
The gross required storage area plus additional area for other infrastructure and buildings on the
terminal is the required total terminal area (own work):

(23)
Where,
At Required total terminal area [m2]. Area including the stockpiles, terminal infrastructure
and buildings.
Agr,i Gross required storage area per commodity type [m2].
α Total terminal factor; the percentage of storage area and internal roads in the stockpile
with respect to the total terminal area [-].
Common in modern terminals is a bypass which makes it possible to transport cargo between ship
and hinterland without using the storage area. Typically less than 5% of the annual throughput is
bypassed while terminals want to reach about 20% according to Lodewijks et al. (2009).
Capacity ratio
The capacity ratio (rc) indicates the percentage of the annual throughput that must be able to be
stored on the terminal at a certain time. In the literature several ratios are proposed. Kraaijveld &
van Hemert (1984) gives a rough assumption of the required storage capacity of 2 months of the
annual throughput, for coal only. For a steady annual throughput this gives a capacity ratio of
16%. Lodewijks et al. (2009) proposes a value of 10% of the annual throughput. UNCTAD (1985)
proposes a capacity ratio of 16%. Vianen et al. (2014) distinguishes between import with capacity
ratios between 5% and 22% and export with ratios between 3% and 10%. These values are
however for coal and iron ore terminals. PIANC (2014b) proposes a ratio between 10% and 25%
for coal import terminals.
Vianen et al. (2012) notes that a dry bulk terminal should at least be able to store one full shipload
of a cargo type. So if a terminal can handle multiple cargo types the storage area should be
dimensioned accordingly.
The required storage volume per commodity type can be calculated with the following equation
that is rewritten from Equation (18) and is adjusted to fit the descriptions of the capacity ratio in
the literature:
(24)
Where,
Vsp,i Required storage volume of the stockpile per commodity type [m 3].
rc Capacity ratio [-]. Percentage of annual throughput that must be able to be stored on the
terminal at a certain time.
The total gross storage area can again be calculated by means of Equations (20) to (23).
Storage factor
The storage factor (fs) relates the annual throughput with the total terminal area. Ligteringen &
Velsink (2012) proposes storage factors for coal, iron ore and crude oil (liquid bulk) for import
terminals and states that export terminals have higher factors. According to Vianen et al. (2014)
these factors are too low, more material can be stored per square meter of storage area
accordingly. Vianen’s observations are based on a study of 49 dry bulk terminals. The information
from both sources has been combined in Table 17.
60 LITERATURE STUDY
Table 17: Storage factors
Storage factor fs
Commodity [tonne/year/m2]
Coal 15 - 75 60 - 185
Iron ore 30 - 80 70 - 210
Crude oil (liquid bulk) 40 - 50 -
Sources: Vianen et al. (2014) and Ligteringen & Velsink (2012)
Storage types
The considered storage types are the open wind-row storage, covered horizontal storage and silos
(see Section 2.1.1.6).
Coal and iron ore are usually stored in an open storage. Wind-row storage typically consists of
one or multiple trapezoidal or triangular shaped rows of cargo. Common widths are between 40
and 100 meters according to Vianen et al. (2014). Dimensions found in practice of coal and iron
ore lanes (or stockpiles) according to Kleinheerenbrink (2012) are given in Table 18. EMO
terminals use a maximum stacking height of 18 metres according to Kleinheerenbrink (2012).
Table 18: Pile dimensions of coal and iron ore wind-row stockpiles
Dimensions [m] Import terminals Export terminals

Lane length limits 300 - 1200 300 - 1300
Lane length average 665 800
Lane width limits 30 - 75 30 - 85
Lane width average 45 50
Ratios between the lane length and width, according to Vianen et al. (2014), are given in Table
19.
Table 19: Ratios between length and width of stockpiles for coal and iron ore

Length-width ratio limits 1.2 – 4.6 1.3 – 4.5
Length-width ratio average 2.5 2.6
Source: Vianen et al. (2014)
Covered horizontal storage in the form of a portal structure consists of the same trapezoidal or
triangular shaped row but is protected by an outer shell. It can be assumed that more area is
required with respect to wind-row storage, also depending on the type of stacking/reclaiming
equipment inside the shed. Extra space next to the stockpile on all sides can be assumed to be
between 5 and 10 metres approximately; personal communication with Smits van Oyen, R. J. (12-
05-2016).
Silos come in many different types and sizes. For free flowing materials, such as grains, a flat
bottom silo is a suitable storage facility. GSI manufactures these silos, that come in different sizes.
An example is shown in Figure 27.

Figure 27: GSI grain silo. Source: GSI (2016)
The unrelated minimum and maximum dimensions are listed in Table 20 to give a rough
indication.
Table 20: Minimum and maximum values of characteristics of GSI silos for free flowing materials
Minimum Maximum
Capacity [m3] 940 41,165
Diameter D [m] 7.3 41.2
Height H [m] 9.8 30.1
Data source: GSI (2016)
Eurosilo manufactures silos for non-free flowing materials ranging from 500 to 100,000 cubic
metres. Eurosilo states that a silo requires only one third of the area 12 required when using a
horizontal stockyard (open or covered storage), see Figure 28.
Figure 28: Eurosilo in comparison to storage shed. Source: Eurosilo (2016)
In order to give a rough indication of Eurosilo silo dimensions unrelated minima, maxima and
average values of silo characteristics data13 are given in Table 21.
12
The suggested area is the net required storage area; thus not including internal roads.
13
Silo characteristic data only from Eurosilo, regardless of cargo type, volume and continent.
62 LITERATURE STUDY
Table 21: Minimum, maximum and average values of characteristics of Eurosilo silos for non-free flowing
materials
Minimum Maximum Average

3
Capacity [m ] 144 100,000 13,305
Diameter D [m] 4.6 56.6 24.9
Height H [m] 4.7 50.0 17.5
D/H [-] 0.4 4.0 1.6
Data source: Eurosilo (2016)
The ratios between silo diameter and height are depicted in a histogram in Figure 29.
Data source: Eurosilo (2016)
Figure 29: Histogram of the ratio of silo diameter and height

Table 22. For various parameters minimum and maximum values are proposed, resulting in a
range of possible values. This can be represented by a uniform random distribution. When this is
the case the reader should make an estimate of which part of the distribution to use. The sources
of the values are numbered in square brackets.
Table 22: Parameter values of dry bulk terminal design rules

Ratio gross over net storage area.
farea - No common values found. Often a stroke with a width of 5 to 25
metres is used around the net storage area [5].
Storage factor.
fs tonne/year/m2
Dredging tolerance.
hdt m
hguc m
Maximum height of the stockpile.
hpile,max m
A common maximum stacking height for a stacker-reclaimer is

23 metres [1].
Required lane length.
llane m
mb -
ms - Determined in Expert Judgement Elicitation, see Section 0.
Literature: Common value is 80% [6].
Number of (un)loaders per berth.
ncb - Differs per equipment type. Usually 1 or 2 cranes per berth,
with a maximum of 4 [5].
Very much dependent on local situation. EMO terminal operates
nhy hour/year
24 hours a day for 365 days a year; this is 8760 hours per year
[7].
Capacity ratio.
rc % Ratios in general or for certain cargo types can be found at the
end of Section 2.2.3.2.
Average dwell time of cargo in the stockpile.
The average dwell time of dry bulk cargo should not exceed 2
days weeks [2], this value seems idealistic. Dwell time can also be
approximated by: .
Width of a lane.
wlane m
Silo diameter.
For values see Table 20 and in Table 21.
D m
Table 21 (p.62).
Draught of the design vessel.
Ds m
Silo height.
For values see Table 20 and in Table 21.
H m
Table 21 (p.62).
Additional quay length required for and aft of vessel to account
Lml m for mooring lines.
Average vessel length.
m
For normal vessel lengths see Table 12 (p.47).
Length of largest vessel frequently calling at terminal.
Ls,max m
For normal vessel lengths see Table 12 (p.47).
Average gross productivity per ship (un)loading equipment.
tonne/hour
For values see Section 2.2.2.1.
α -
Angle of repose of material.
θ 
Average cargo density.
tonne/m3
Sources:
[1] Vianen et al. (2014)
[2] Fourgeaud (2000)
64 LITERATURE STUDY
[3] PIANC (2014b)
[5] Various bulk terminals in Google Earth
[6] Kleinheerenbrink (2012)
[7] EMO (2016)
[8] PIANC (1995)

waiting time in terms of average service time of vessels (Tw/Ts). According to Monfort et al.
(2011) a reasonable value for Tw/Ts for dry bulk terminals is 0.50. The probabilistic Queuing
Theory is used to determine the occupancy ratios. This theory is not treated in this research since
it is outside its scope, for a better understanding of the general Queuing Theory reference is made
to Sztrik (2012). An application of the theory on terminal design can be found in Agerschou
(2004).
higher the K-value the more constant the inter arrival or service times are). A realistic queue is
M/E2/n for common user dry bulk terminals according to Monfort et al. (2011) and Terblanche &
Moes (2009). For dedicated shipping line dry bulk terminals E2/E2/n is proposed by Monfort et al.
(2011) and Vianen, T. van (2015). Acceptable berth occupancy factors corresponding to these
queues for Tw/Ts = 0.50 are given in Table 23.
Table 23: Acceptable berth occupancy factor for dry bulk terminals for T w/Ts = 0.50

M/E2/n E2/E2/n
1 41 55
2 64 73
3 73 81
4 78 84
5 82 87
6 or more 84 89
course still possible.

2.3 Liquid bulk terminals
For liquid bulk terminals first general background information is presented in order to give the
guidelines, rules and common values of variables are treated.

This section considers liquid bulk terminals and is based on Agerschou (2004), Ligteringen &
Velsink (2012), PIANC (2014b) and UNCTAD (1985). Information is provided about liquid bulk
commodities, vessels and terminal elements.
2.3.1.1 Commodities
The vast majority of liquid bulk trade is in crude oil and petroleum products. Other product
groups like chemicals or vegetable liquids form a smaller part of the world trade in liquid bulk;
Agerschou (2004).
Liquid bulk commodity types are:
 Crude oil
 Oil products
 Chemicals
 Liquefied gas
 Vegetable oils
 Bio-fuels
Liquid bulk materials often are hazardous substances. They can be flammable, explosive, toxic or
corrosive. Therefore safety measures are required in and around the port terminal. Liquid bulk
cargo can also be polluting which asks for protective measures.
Liquefied gasses are the most dangerous kind of liquid bulk. Examples are Liquefied Natural Gas
(LNG) and Liquefied Petroleum Gas (LPG). These gasses reach a liquid state when cooled and/or
pressurised. When the liquid reaches its boiling point gasses form that are highly flammable and
thus potentially very dangerous.
In the tool only crude oil and LNG are specifically implemented. This since crude oil is het most
transported cargo type and in literature rules-of-thumb can be found; Agerschou (2004). LNG is a
fossil fuel but it is cleaner than for example diesel, due to lower emission levels. Therefore -
although it still has a long way to go- LNG is a promising fuel; Het Financieele Dagblad (2016).
Because of this and in order to inform the reader about the very different requirements for these
kind of terminals LNG is included in this study. The tool also allows the user to define other types
of liquids, for these liquids some material properties have to be specified.
2.3.1.2 Vessels
Liquid bulk vessels are divided in different types, depending on the transported commodity.
Crude oil is typically transported in large tankers; Very Large Crude Carriers (VLCC) or Ultra
Large Crude Carriers (ULCC), of 200,000 DWT and greater. Refined oil products, chemicals and
66 LITERATURE STUDY
vegetable oils are transported in product tankers up to 100,000 DWT. Typical dimensions are
given in Table 24.
Table 24: Liquid bulk oil/product vessel classes and dimensions
DWT L Draught Beam

Class
[tonne] [m] [m] [m]
Product/chemical
3000 - 50,000 90 - 210 6.0 - 12.6 13.0 - 32.2
tankers
Tankers 60,000 - 175,000 217 - 300 13.0 - 17.7 36.0 - 52.5
VLCC 200,000 - 300,000 310 - 350 18.5 - 21.0 55.0 - 63.0
ULCC 350,000 - 500,000 365 - 415 22.0 - 24.0 65.5 - 73.0

Liquid bulk carriers with typical transported cargo type and DWTs are depicted in Figure 30.
Figure 30: Liquid bulk vessel types with the typical cargo types and DWTs. Source: U.S. Energy Information
Administration
2.3 Liquid bulk terminals 67

LNG is transported in liquefied gas carriers. LNG carriers are divided in Membrane or Prismatic
and Sphere or Moss type vessels. Typical dimensions are given in Table 25.
Table 25: Liquid bulk LPG/LNG vessel classes and dimensions
Approx.
DWT L Draught Beam
Class Capacity
[tonne] [m] [m] [m]
[m3]
LNG Carriers
51,000 - 75,000 249.5 - 288.0 10.6 - 11.5 40.0 - 49.0 90,000 - 145,000
(Membrane)
LNG Carriers
27,000 - 125,000 207.8 - 345.0 9.2 - 12.0 29.3 - 55.0 40,000 - 267,000
(Spherical, Moss)
Figure 31: Spherical LNG carrier (upper), Membrane LNG carrier (lower). Source: PIANC (2012)
2.3.1.3 Liquid bulk terminal elements

Liquid bulk terminals consist of the following elements (see Figure 32):
 Jetty
 Pipelines
 Storage tanks
Liquid bulk terminals are often linked to local industry, long term storage is therefore not needed.
However also stockholding terminals (that function as a long term storage), gateway terminals
(that import and export cargo), or transhipment terminals for liquid bulk exist. These different
types of liquid bulk terminals largely dictate the required capacity of the terminals.
68 LITERATURE STUDY
Figure 32: Schematic representation of typical liquid bulk terminal elements. Based on: Monfort et al. (2011)
2.3.1.4 Jetty
At marine liquid bulk terminals jetties are often of the island berth type. These consist of an
approach bridge (trestle) from the shore to the loading platform with access roadway and
pipeway. The loading platform houses a loading arm, pipelines, service building, fire fighting
equipment, spillage tank and possibly a jetty crane (depends if resupplying of the vessel is
accommodated). Furthermore separate mooring and breasting dolphins are used, since vessel
access is only required at the centre of the vessel. The dolphins are connected to the loading
platform by catwalks. An overview of an island berth type jetty is depicted in Figure 33.

Figure 33: Schematic overview of an island berth type jetty. Source: Own work
The loading arm can be in the form of rigid pipes connected to the vessel and allowing for ship
motion by means of swivel joins. Or in the form of a more flexible rubber or fabric hose.
Figure 34: Liquid bulk loading arm. Source: Agerschou (2004)
2.3.1.5 Pipelines
At import terminals the liquid cargo is pumped with ship pumps through the unloading arm into
pipelines positioned on the pier. The unloading capacity is therefore determined by the ships
pumps and not by the terminal equipment. The pipelines go to the storage tanks on the terminal.
When pipelines will be used for different commodities or when the liquid is likely to solidify
during transport; equipment is needed for cleaning the pipes. This equipment is normally located
at the (un)loading platform and near the storage tanks.
For some materials heating (e.g. some vegetable oils) or cooling (e.g. LNG or ammonia gas) has
to be applied in order to keep the material in a liquid form. This affects the pipeline design; for
example insulation and expansion loops have to be applied. Pipelines are usually combined in
large pipeways that run through the terminal area.
70 LITERATURE STUDY
Cooled or pressurised liquids require a vapour-return-system that transports vapour, that forms
when boiling occurs while loading the ship, back to the shore.
Methods for the handling of liquefied gas are similar to those of oil and petroleum products.
Exceptions are the cooling and/or pressurisation (including the loading arm) of the cargo. Strict
regulations apply for positioning the tanks with respect to other buildings, this is discussed in
Section 2.3.3. Also the position of ships with respect to other vessels is prone to regulations, these
are however not in the scope of this study.
2.3.1.6 Storage tanks

The storage of liquid bulk happens in large cylindrical steel tanks. The (floating) roofs of the
tanks prevent contamination from weather and prevent the evaporation of the liquid into the
atmosphere. Crude oil and oil product tanks usually have a capacity between 500 and 20,000
cubic metres, however larger tanks exist. Examples of tanks are depicted in Figure 35. Vegetable
oil tanks are generally smaller since shiploads are smaller, the tank capacity is normally about
1,000 tonnes or less.
Figure 35: Liquid bulk tanks with external floating roofs. Source: Google
Two options for storage exist; switch tanks or dedicated tanks. Switch tanks allow for different
types of cargo to be stored, this requires however cleaning and degradation costs but has large
impact on the size of the terminal. Dedicated tanks store one type of cargo only.
On liquid bulk terminals bunds are used to capture spills. These bunds can be concrete or earth
walls surrounding a single tank or tank group and must be designed so that at least a full tank load
can be captured within these walls.

Figure 36: Satellite picture of a liquid bulk terminal with various tank sizes and earth bunds (green squares).
Source: Google
LNG
Gasses that are liquefied by cooling require tanks with insulation and a refrigeration plant. LNG is
stored in cryogenic tanks that have to be manufactured from special alloys on account of the very
low temperature (and consequently brittleness of steel). These tanks also consist of a double wall
as a safety measure, see Figure 37.
Figure 37: Full containment cryogenic LNG storage tank schematisation 14
An LNG terminal requires additional components in order to convert the liquefied gas to a natural
gas. As an indication these components are depicted in. They are however not a part of this
research as explained in the introduction and because of the goal to maintain the general nature of
this research.
14
Source: http://www.epd.gov.hk
72 LITERATURE STUDY
Figure 38: Indication of LNG terminal components15

The only specific equipment on a liquid bulk terminal are the pumps and attachments. Important
aspects are the pump capacity and pipeline length and diameter. The properties and the selection
of pumps and pipelines is another field of expertise, this is therefore not included in this research.
Liquid bulk unloading performance depends on the vessel size and the corresponding pump
capacity of the vessel. It also depends on liquid properties such as viscosity, temperature and
safety regulations; Fourgeaud (2000).
Ligteringen & Velsink (2012) states that tankers smaller than 250,000 DWT typically have a
combined net pump capacity (from ship to shore) of about 10% of their DWT. And PIANC
(2014b) states that for edible oil carriers the pump capacity from shore to ship is 100 - 150
tonne/hour.
Fourgeaud (2000) states that most liquid bulk carriers are operated within one day. Throughput
varies from 300 to 1000 m3/hour for small and regular sized vessels to 15,000 m 3/hour and higher
for VLCC and ULCC.

The design guidelines, rules and parameters are provided for determining the number of berths
and storage yard area. The required dredging depth can be determined as in Section 2.1.3.4.
Values used in practice of the parameters are listed in Section 2.3.3.3.
15
Source: https://www.edf.fr

2.3.3.1 Number of berths
The number of berths can be determined using the berth productivity and/or maximum service
time method.
Berth productivity
This method is similar to the one for dry bulk terminals in Section 2.2.3.1, and is based on PIANC
(2014b). It is however adjusted so that the average combined pump productivity per berth (or
equivalently per vessel, since jetties are used) is used instead of the productivity per crane or
pump.
(25)
Where,
Average combined pump productivity per berth [tonne/hour]. Average number of tonnes
moved between berthing completed and de-berthing started by means of the combined
pump capacity.
mb Berth occupancy factor [-]. The acceptable occupancy factor depends on the allowable
waiting time in terms of service time, the number of berths and the sort of terminal (e.g.
dedicated shipping line). For berthing at a jetty the berth is treated as a binary object; the
berth is either free or occupied. Lengths are therefore not taken into account. For this
situation the factor can be calculated with the following equation, based on Saanen (2004):
(26)
Where,
Ts,i Service time (or effective berthing time) of vessel i of total of n vessels berthed in
measurement period nhy [hour]. Time the vessel is berthed outside of the
The required number of berths is calculated with Equation (16).
Then with the determined number of berths the number and form of the jetties has to be
determined. In this study only island berth type jetties for liquid bulk terminals are considered.
The jetty can be divided into the trestle structure and the berthing structure, an island berth jetty
only has one berth. The length of the trestle is purely dependent on the local situation. The
berthing structure length is a summation of the design vessel length (Ls,max) and an additional
length for and aft of the vessel (Lml).
Maximum service time

Another method to determine the required number of pumps, the pump productivity and the
number of berths is when a maximum allowable service time is posed. A common time-in-port is
1 to 1.5 days according to Ligteringen & Velsink (2012) and PIANC (2014b). The required
number of berths can be calculated with the equations of the previous section.
2.3.3.2 Storage area

For the calculation of the storage area for liquid bulk terminals the same general approach can be
used as for dry bulk terminals. Furthermore some rules-of-thumb for the storage capacity exist in
74 LITERATURE STUDY
literature. Safety requirements such as bunds and minimum distances are in order for some
commodities. These are treated in this section.
General approach
The required storage volume per cargo type can be calculated with the following equation that is
equal to Equation (19). It is rewritten from an equation in PIANC (2014b) in order to result in a
storage volume instead of a storage area. The equation is as follows:
(27)
Where,
Vstorage,i Required storage volume of the tanks per commodity type [m3].
Average dwell time of cargo in the tanks [days].
ms Estimated storage occupancy [-]. Average mass of cargo in tanks divided by the design
capacity of the tanks. To account for the random arrivals and departures of cargo.
When the required storage volume is determined the number of required tanks can be calculated.
The net storage area is the area covered by the storage tanks. The gross storage area consists of
the net storage area, the area surrounded by bunds and the internal roads. The gross storage area
for tank storage can be calculated with a method that is described in Appendix VI . When the
gross storage area is calculated and the total terminal factor (α) is known the total terminal area
can be calculated with Equation (23).
Rules-of-thumb
The rules-of-thumb consist of percentages that determine the required number of tonnes of cargo
that should be stored at a certain point of time.
PIANC (2014b) proposes a storage capacity ratio (rc) of 3 or 4 times the DWT of the largest
vessel that calls at the terminal. They also propose a capacity of 3% to 5% of the annual terminal
throughput. Agerschou (2004) complements this by stating that 3 or 4 times the largest shipload
per cargo type must be able to be stored for dedicated storage tanks. For switch tanks this is 3 or 4
times the largest shipload regardless the cargo type.
Ligteringen & Velsink (2012) recommends the ability to store one month of consumption on the
terminal. In other words the total output of one month should be able to be stored on the terminal
at a certain point of time. Furthermore for crude oil they propose a storage factor that is given in
Table 17.
Bunds
VROM (2008) gives as a guideline that the volume between bunds should have a minimum
capacity of the volume of the largest tank surrounded by the bund plus 0.25 metres for wind
waves plus an additional height for subsidence. If there are other tanks surrounded by the same
bund then 10% of the volume of these tanks should be able to be contained too.
Minimum distances
For non-hazardous materials there are no rules for the distance between tanks. Using Google
Earth a practical minimum is found of 5 metres, this is of course not a binding value.

In the Netherlands flammable products are categorised as listed in Table 26.
Table 26: Flammable liquid classes
Flashpoint boundaries
0 C < flashpoint < 21 C
Class 1
Boiling point > 35 C
Class 2 21 C < flashpoint < 55 C
Class 3 55 C < flashpoint < 100 C

Source: VROM (2008)
For minimum distances between objects and tanks -for the storage of flammable products of
classes 1, 2 or 3- see Table 27. The table is based on VROM (2008) which in turn is based on
Institute of Petroleum Code 19, Model Code of safe Practice.
Table 27: Minimum distances between objects and tanks with flammable products
Tank type Object Minimum distance

Between groups of
15 metres
small tanks1
Smallest distance of:
1. Half of largest diameter of the
Between a group of
tanks
small tanks1 and a
2. Diameter of the smallest tank
Tanks with fixed tank outside the
3. 15 m.
roofs, including group
But not smaller than 10 m or larger
tanks with internal
than 15 m.
floating roofs.
Between tanks that
are not part of a 15 metres
group of small tanks
Between a tank and
a filling point or 15 metres
building.
10 m if tank diameter ≤ 45 m.
15 m if tank diameter > 45 m.
Between two tanks For crude oil: 30% of tank diameter

with floating roofs. but larger than 10 m.
*Tank with largest diameter

determines distance.
Tanks with floating Smallest distance of:
roofs. 1. Half of largest diameter of the
Between tank with tanks
floating roof and 2. Diameter of the smallest tank
tank with fixed roof. 3. 15 m.
But not smaller than 10 m or larger
than 15 m.
Between a tank and
building.
76 LITERATURE STUDY
Tank type Object Minimum distance
Between a tank and
15 metres
terminal border.
Note 1: Small tanks are defined as tanks with a diameter < 10 m and a height < 14 m. A group of small tanks
may be considered a single tank with respect to spacing or bunding. A small tank group can have a
combined capacity of maximum 8000 m3.
Note 2: For tanks with diameter larger than 18 m. it may be necessary to enlarge the distances.
Source: VROM (2008)
For minimum distances between objects and tanks -for the storage of flammable fuels of classes 1,
2 or 3- see Table 28. The table is based on VROM (2008) which in turn is based on Institute of
Petroleum Code 2, Marketing Safety Code.
Table 28: Minimum distances between objects and tanks with flammable fuels
Object Minimum distance

Smallest distance of:
4. Half of largest diameter of the
Between tanks with
tanks
diameter > 10 m. or
5. Diameter of the smallest tank
height > 14 m.
6. 15 m.
But not smaller than 10 m.
Between a tank and
building.
Between a tank and
15 metres
terminal border.
Note: For tanks with diameter larger than 18 m. it may be necessary to enlarge the distances.
Source: VROM (2008)
The failure of an LPG tank can lead to the evaporation of a large amount of LPG. With air an
explosive mixture is formed that can cause a boiling liquid expanding vapour explosion when an
ignition source is present. To avoid fires or explosions minimum distances to near objects are
listed in Table 29. A threshold for the maximum permissible heat intensity is taken as 10 kW/m 2
in VROM (2013). The source does not mention specific international regulations.
Table 29: Minimum distances from object to LPG tank
Tank with flammable Distance corresponding to 10 kW/m2

liquid (flashpoint < 60 C) in Figure 4.1 of VROM (2013)
Tank with flammable

3 metres
liquid (flashpoint > 60 C)
Flammable material or Distance corresponding to 10 kW/m2
other objects inside the in Figure 4.3a, 4.3b or 4.3c of VROM
bund (2013)
Source: VROM (2013)

The Dutch guideline for the storage of cryogenic gasses (such as LNG) is VROM (2014), this
however only applies to tanks from 0.125 to 100 m3. The minimum distances to objects near the
tank are listed in Table 30. The source does not mention specific international regulations.
Table 30: Minimum distances from object to cryogenic gas tanks
Tank with flammable Distance corresponding to 10 kW/m2

liquid (flashpoint < 60 C) in Figure 3.1 of VROM (2014)
Tank with flammable

Half the tank diameter1
liquid (flashpoint > 60 C)
Tank with liquefied
flammable gasses (volume 15 metres
> 150 m3)
Note 1: VROM (2014) proposes 3 metre for not-soil-covered tanks, the minimum distance for soil-covered tanks
is used in the table.
Source: VROM (2014)
Tanks sizes
Liquid bulk tanks come in many different types and sizes. The DWT varies per cargo type. For
example crudes are transported by large carriers while vegetable oils are transported by smaller
vessels. The required storage space is therefore smaller for the latter, this results in smaller tanks.
To be able to give a rough indication of tank dimensions the liquid bulk cargoes are divided into
four groups:
1. Crude oil, petroleum products, diesel, fuels

2. Chemicals, bio fuels, mineral oils
3. Vegetable oils
4. LNG
The following liquid bulk terminals located in Rotterdam (unless indicated otherwise) are
investigated:
Table 31: Investigated liquid bulk terminals divided into groups
Group Terminals
Vopak Europoort, Maasvlakte Olie Terminal, Euro Tank

1
Terminal, Nova Terminals
Vopak TTR, Koole Tank Storage, Odfjell Terminals,
2 Rubis Terminals, Noord Natie Odfjell Terminals
(Antwerp)
Koole Tank Storage, Nova Terminals, Noord Natie
3
Odfjell Terminals (Antwerp)
4 Gate Terminal, Enagas (Barcelona), Sagas (Sagunto)
78 LITERATURE STUDY
Per terminal on average the smallest, medium and largest tanks are measured. The tank diameter
is approximated by means of Google Earth. The tank height is approximated by means of data
from Actueel Hoogtebestand Nederland (2016). The tank capacity is approximated with the
diameter and height. The resulting minimum, maximum and average characteristics per group are
listed in Table 32.
Table 32: Minimum, maximum and average values of characteristics of liquid bulk tanks
Group 1 Group 2
Minimum Maximum Average Minimum Maximum Average
3
Capacity [m ] 1,154 118,236 58,252 142 45,696 8,997
Diameter D [m] 12.0 84.7 55.0 3.0 51.9 20.9
Height H [m] 10.2 33.7 20.2 12.5 25.0 17.3
D/H [-] 1.2 4.4 2.8 0.2 2.4 1.2
Group 3 Group 4
Minimum Maximum Average Minimum Maximum Average
Capacity [m3] 142 21,220 5,874 53,766 180,000 152,233
Diameter D [m] 3.0 42.3 18.9 75.8 88.0 82.0
Height H [m] 10.2 25.0 16.1 23.9 33.2 28.9
D/H [-] 0.2 2.8 1.3 2.3 3.4 2.9

Table 33. For various parameters minimum and maximum values are proposed, resulting in a
range of possible values. This can be represented by a uniform random distribution. When this is
the case the reader should make an estimate of which part of the distribution to use. The sources
of the values are numbered in square brackets.
Table 33: Parameter values of liquid bulk terminal design rules

Storage factor.
fs tonne/year/m2
Dredging tolerance.
hdt m
hguc m
mb - For values see Table 34 (p.81)
Optimal value is between 50% and 65% [1].
ms -
nhy hour/year
No values in literature.
Capacity ratio.
rc %
Ratios can be found in Section 2.3.3.2.
Average dwell time of cargo in the tanks.
days
No values in literature.
Tank diameter.
D m
Ds m Draught of the design vessel.

For values see Table 24 and Table 25 (p.67).
Tank height.
H m
Additional quay length required for and aft of vessel to account
Lml m for mooring lines.
Length of largest vessel frequently calling at terminal.
Ls,max m
For normal vessel lengths see Table 24 and Table 25 (p.67).
Average combined pump productivity per berth.
tonne/hour
For general pump capacities see Section 2.3.2.
α -
Average cargo density.
tonne/m3
Crude oil: 0.790 and 0.973 and LNG: between 0.41 and 0.50.
Sources:
[1] Personal communication with R. J. Smits van Oyen (13-05-2016)
[2] PIANC (2014b)
[4] PIANC (1995)

waiting time in terms of average service time of vessels (Tw/Ts). According to Monfort et al.
(2011) a reasonable value for Tw/Ts for (liquid) bulk terminals is 0.50. The probabilistic Queuing
Theory is used to determine the occupancy ratios. This theory is not treated in this research since
it is outside its scope, for a better understanding of the general Queuing Theory reference is made
to Sztrik (2012). An application of the theory on terminal design can be found in Agerschou
(2004).
higher the K-value the more constant the inter arrival or service times are). A realistic queue is
M/E2/n for common user (liquid) bulk terminals according to Monfort et al. (2011) and
Terblanche & Moes (2009). For dedicated shipping line bulk terminals E2/E2/n is proposed by
Monfort et al. (2011). Acceptable berth occupancy factors corresponding to these queues for Tw/Ts
= 0.50 are given in Table 34.
80 LITERATURE STUDY
Table 34: Acceptable berth occupancy factor for liquid bulk terminals for T w/Ts = 0.50

M/E2/n E2/E2/n
1 41 55
2 64 73
3 73 81
4 78 84
5 82 87
6 or more 84 89
course still possible.

2.4 Costs determination
In order to give a rough estimation of the construction costs of a greenfield marine terminal prices
for certain terminal elements are required, these are referred to as unit prices or unit costs. The
prices of those elements are covered in this section and are based on literature review and
interviews. First general prices, that can be used for multiple container types, are given. Then for
each terminal type prices are given for the specific terminal elements. Prices are often not easily
obtained since manufacturers see it as competition sensitive information.
2.4.1 Unit costs, construction costs and inflation

Direct costs are defined as costs directly related to the production or supply of a product or
service; from the Dutch Standaard Systematiek voor Kostenramingen (SSK) definitions. A
construction costs estimate consists of direct costs (labour, equipment and materials), indirect
costs (e.g. site organisation, site management, general costs etc) and contingencies (unforeseen
costs). An elaboration on how to calculate these costs can be found in Appendix I , since they are
confidential.
The direct costs are calculated by using unit costs and quantities. Unit costs are a price per unit
metrics. These unit costs are presented in this section.
Prices are adjusted for inflation in the tool itself, therefore prices in this section are given in their
original price level. All prices are in Western European rates. Prices that were given in USD are
converted to EUR with the conversion rate of 1 USD = 0.896 EUR, from Google on 30-08-2016.
Prices of some elements go back to as far as 1998. In the period between then and now not only
inflation plays a role but also technological improvements. These improvements may have an
impact on the price, for this study this is however discarded due to time limitations.
2.4.2 General unit costs

Terminal elements that apply to more than one terminal type are quay wall and jetty structures.
2.4.2.1 Quay wall

In de Gijt (2010) quay walls from around the world are analysed. De Gijt looked at the quantities
of material (concrete, steel) in the quays and the amount of man hours needed to build the quays.
In his study direct costs for certain types of quay walls per metre of quay are presented. The
studied quay wall type in the current research is limited to sheet pile walls, due to time limitations.
De Gijt’s research led to the following direct cost formula for sheet pile quay walls per metre of
quay:
(28)
Where,
C Quay wall direct costs [EUR/m1]. In 2008 price level.
Hretain Retaining height [m].
Obviously this is an approximation of real costs since the total costs are dependent on aspects like;
soil conditions, crane loads, construction materials, wave climate, tidal variation and the country
de Gijt (2010).
82 LITERATURE STUDY
De Gijt’s formula is checked by means of data of quay walls of different terminal types, from a
port master plan project of Witteveen+Bos. This data is confidential, the results are therefore
presented in Appendix I .
2.4.2.2 Jetty
In de Gijt (2010) many quay walls were analysed in order to derive a formula for the costs per
metre of quay as a function of the water depth. No similar literature can be found concerning the
costs of jetties. Also not enough data is available to determine a similar relation for jetty costs as
part of this study. Therefore the decision is made to approximate the jetty costs per metre of jetty
based on a few projects performed by Witteveen+Bos. For this analysis a distinction is made
between jetties of the finger pier and liquid bulk type. An elaboration can be found in Appendix I
.
2.4.3 Container terminal unit costs

Container terminal cost elements can be divided into civil works and equipment. Civil works
consists of the quay wall, STS crane rails, apron and storage yard pavement and the construction
of the container freight station (CFS). The equipment consists of STS cranes, tractor trailer units,
straddle carriers, automated guided vehicles, reach stackers, rubber tyred and rail mounted gantry
cranes.
2.4.3.1 Equipment
Saanen (2004) gives prices for these equipment types. The prices are among others based on
Drewry (1998), it is assumed that these prices are according to the Dutch 1998 price level. In
Saanen (2004) no prices are given for a rubber tyred gantry crane. In Böse (2011) however
investment costs of both RMG and RTG cranes are qualitatively described. As a scale low,
medium, high and very high investment costs is used. Costs of RMG cranes are described as high
and RTG cranes as medium. Therefore it can be assumed that there is a 25% difference between
the costs of both crane types.
Table 35: Unit costs container terminal equipment
Unit costs Unit costs

Equipment (1998 price level) (2016 price level)
[EUR/pc] [EUR/pc]
STS crane 5,000,000 – 7,500,000 7,140,000 – 10,710,000
Tractor trailer unit 90,000 130,000
Straddle carrier 500,000 715,000

Automated guided
350,000 500,000
vehicle
Reach stacker 325,000 465,000
Rail mounted gantry 1,100,000 (6 wide) – 1,600,000
1,570,000 – 2,285,000
crane (9 wide)
Rubber tyred gantry
825,000 – 1,200,000 1 1,180,000 – 1,715,000
crane
Note 1: No price given but determined by own calculation
Source: Saanen (2004)
2.4 Costs determination 83

2.4.3.2 Civil works
The quay wall costs are considered in Section 2.4.2. Apron and storage yard pavement is assumed
to be the same sort of (paving stones) pavement. The CFS is assumed to be a warehouse structure.
Unit prices for these last two elements are provided in Appendix I , since they are confidential.
Crane rails are not included in the tool since the costs are minor compared to the costs of a quay
wall per metre.
2.4.4 Dry bulk terminal unit costs

Dry bulk terminal cost elements can be divided into civil works and equipment. Civil works
consists of the quay or jetty, open stockpile foundation, conveyors and pipelines, storage shed and
silos. The equipment consists of gantry grab crane, bucket elevator, chain-, vertical screw-, spiral-
and pneumatic conveyors, ship loader, bucket wheel stacker-reclaimer and scraper-reclaimer.
2.4.4.1 Equipment
In Vianen, T. van (2015) investment costs for bucket wheel stacker-reclaimers are determined by
the weight of the machines. These weights are in turn related to the machines combined (sum of
stacking and reclaiming) capacity. In order to derive an equation for the bucket wheel stacker-
reclaimer weight as a function of the combined capacity data points from the research are used
and a linear curve is fitted:
0 (29)
Where,
W Stacker-reclaimer weight [tonne].
Cequip Combined stacker-reclaimer capacity [tonne/hour]. Sum of the stacking and reclaiming
capacity.
No difference is made between the various boom lengths. According to the same study several
experts use the rule-of-thumb that the machine’s investment costs in Euros are 6 to 8 times the
weight in kilograms. Since this is a rule-of-thumb no price level is known. The previous
information is depicted in Figure 39.
84 LITERATURE STUDY
Figure 39: Bucket wheel stacker-reclaimer combined capacities versus machine weight (black). And high (red)
and low (blue) boundaries for investment costs. Based on Vianen, T. van (2015)
In Chapter 2.2.2.2 the average combined capacity of this type of machine is 5,000 tonne/hour.
This leads to costs ranging from € 3,121,800 to € 4,162,400 per piece of equipment.
Of the other dry bulk terminal equipment no price information can be found. The contacting of
different equipment manufacturers in order to ask for price information would take up too much
time and results would not be guaranteed. Therefore the choice is made to use the relationship
between capacity and weight of stacker-reclaimers also for the other equipment types. Obviously
the machines do not consist of exactly the same parts and do not come in the same sizes, but they
are used for the same purpose executed in a different way. By means of Google pictures of
different machines of the same type are inspected and compared to the base case; that is the
stacker-reclaimer. The average difference in weight is then visually estimated with a percentage.
This leads to a capacity-weight function for each equipment type. With the equipment capacities
found in the literature study the corresponding weights can be determined. For the bucket wheel
stacker-reclaimer the combined (un)loading capacity is used, the unloading machines however do
not have a loading capacity. Unloading equipment capacities have to be multiplied by 2 in order
to match the required combined capacity for the equation. The same method applied for a loader
results in too high prices so the loading capacity is not doubled. The percentages and functions are
listed in Table 36. Multiplying the weights with a factor of 6 to 8 EUR/Kg results in the
equipment costs. It is stressed that this is a rough estimation of the investment costs of dry bulk
terminal equipment.
Table 36: Weight as function of combined capacity for various dry bulk terminal equipment
Percentage of machine
Weight W [tonne] as a
weight with respect to
Equipment type function of combined capacity
stacker-reclaimer
Cequip [tonne/hour]
[%]
Bucket wheel stacker-reclaimer 100 W= 0.07690 Cequip + 135.80
Gantry grab crane 95 W= 0.07306 Cequip + 129.01
Bucket elevator 110 W= 0.08459 Cequip + 149.38
Chain conveyor 95 W= 0.07306 Cequip + 129.01

Percentage of machine
Weight W [tonne] as a
weight with respect to
Equipment type function of combined capacity
stacker-reclaimer
Cequip [tonne/hour]
[%]
Vertical screw 95 W= 0.07306 Cequip + 129.01
Spiral conveyor 95 W= 0.07306 Cequip + 129.01
Pneumatic conveyor 80 W= 0.06152 Cequip + 108.64
Loader 85 W= 0.06537 Cequip + 115.43
Scraper-reclaimer 5 W= 0.00385 Cequip + 6.79
Note: percentages based on visual comparison between photos.
According to Vianen, T. van (2015) continuous unloading equipment is more expensive than grab
crane unloaders because of the higher technological complexity, this is however not considered
when only looking at the machine weights. Therefore further research is recommended for
increasing the reliability of these costs.
2.4.4.2 Civil works

The quay wall and jetty costs are considered in Section 2.4.2. The open stockpile foundation is
assumed to be a pavement. Storage sheds are assumed to be warehouse structures and silos are
assumed to have a conical or flat roof and conical bottom. Unit costs for these last two elements
are provided in Appendix I , since they are confidential. Belt conveyors and pipelines that are
used to transport dry bulk cargo across the terminal are not included in the cost estimates in the
tool. The length of these elements depends on the layout of the terminal which is not determined
by the tool.
2.4.5 Liquid bulk terminal unit costs

Liquid bulk terminal cost elements consists of jetty for (un)loading (see Section 2.4.2), pipelines,
LNG tanks, regular tanks and bunds. This subsection also covers a rule-of-thumb for the total
costs of the landside of the terminal.
2.4.5.1 Pipelines
The length of the pipelines that transport cargo across the terminal depend on the layout of the
terminal. The terminal layout is not determined by the tool and pipelines are therefore not
included in the cost estimates in the tool.
2.4.5.2 Tanks and bunds

LNG tank costs are much higher than regular steel tanks; Ligteringen & Velsink (2012). It turned
out to be impossible to find decent cost information for LNG tanks. Only one price is found in an
online version of a LNG market research report from The McIlvaine Company 16. The report gives
construction costs ranging from 49.3 mln to 67.2 mln EUR for an LNG tank of 160,000 m3. From
the text it can be derived that the price level is from around the year 2000. Obviously the
downside is that the costs of only one specific tank volume are given. To still be able to calculate
the costs of LNG tanks with other capacities the unit costs are determined by dividing the two cost
limits by the tank volume and by subtracting the additional direct costs, indirect costs and
contingency costs. This leads to direct unit costs between 181.1 EUR/m3 and 246.9 EUR/m3. In
reality the unit costs will decrease for increasing tank volume (the economy of scale principle) but
16
Information about the report (such as title or year) cannot be found.
URL: http://www.mcilvainecompany.com/industryforecast/LNG/overview/Otofc.htm
86 LITERATURE STUDY
this cannot be quantified due to lack of data. The relations found for regular tanks also cannot be
used since LNG tanks are very different than regular storage tanks.
Unit costs for regular tanks and bunds are provided in Appendix I , since they are confidential.
2.4.5.3 Rule-of-thumb total landside area

Robert Jan Smits van Oyen (Manager Logistics and Asset Management & Maintenance at
Tebodin Consultants & Engineers) mentioned in an interview (12-05-2016) that as a rule of thumb
for the total landside17 the investment costs range between € 500 and € 1,000 per cubic metre of
storage volume on the terminal. The subtraction of the additional direct costs, indirect costs and
contingency costs results in the direct unit costs from € 295 to € 590 per cubic metre of storage
volume. Terminals with large tanks are on the lower side of the range while terminals with
multiple small tanks are on the higher side. Approximately 50% of the total costs can be attributed
to the tanks according to Smits van Oyen.
17
Landside includes: Storage tanks, pipelines, equipment and other infrastructure.

3 EXPERT JUDGEMENT
ELICITATION
In Chapter 2 equations are provided that make it possible to determine the required container, dry
and liquid bulk terminal dimensions. Common values for the variables, or parameters, in these
equations are studied as well. Some parameters like for example the throughput are project
specific and for which no useful standard values can be found. For other variables like for
example cargo dwell time or crane capacities similar values or values -within certain ranges-
occur at terminals globally. However not for all parameters -that are considered in this study-
common values can be found. In order to be able to scientifically obtain values for these variables
Expert Judgement Elicitation (EJE) is used. In this study EJE is used for the variables total
terminal factor18 (α) and average storage occupancy19 (ms). These variables are studied because
they are expected to have a large influence on the dimensions, since they are both factors. By
using this method the opinions of expert are combined with a structured performance-based
approach which fills in the missing information from literature. The results of the expert
assessment can be treated as scientific data. This data is an addition to the variable values in
Sections 2.1.3.5, 2.2.3.3 and 2.3.3.3, and is used as input in the computer tool.
Furthermore another method using the same principle is used to estimate the dependence between
the import, export and transhipment container dwell times when they are assumed to be random
variables. The uncertainty distributions of these random variables are based on the common
values of parameters found in the literature study (Section 2.1.3.5). Both methods are presented
and applied in this chapter.
The use of Expert Judgement Elicitation in this chapter is to introduce this type of research to the
port or terminal design sector as an excellent way to combine the opinion of experts. Aligning
with this purpose the theory is explained in this chapter and in detail in Appendix III .
Goal of the application of Expert Judgement Elicitation: Obtaining the uncertainty

distributions of the total terminal factor and average storage occupancy factor. As well as the
dependence between the import, export and transhipment container dwell times, represented as
random variables. And bridging the gap between practice and theory of Expert Judgement
Eliciting in this field of expertise.
3.1 The method
Using the advice of experts for uncertainties is obviously not a new practice. However usually this
happens in an informal way resulting in an unstructured decision making process. Expert
Judgement Elicitation is a structured process with transparent rules that has the goal of treating the
18
The total terminal factor is the percentage of the storage area (including internal infrastructure) with respect to
the total terminal area.
19
The average storage occupancy is the percentage of the design storage capacity that is actually used, averaged
over a year
89
judgements as scientific data; Aspinall (2008). Currently at Delft University of Technology
research is being performed on Expert Judgement Elicitation.
In this study expert opinions are asked for the uncertainty of six different parameters and the
dependence between three other variables. Both these applications are elaborated on in Section
3.1.1 and Section 3.1.2.
3.1.1 Eliciting uncertainties

Various methods for eliciting expert uncertainties exist, for this study the Classical model from
Cooke (1991) is used. Cooke’s model applies a performance-based combination of expert’s
uncertainty distributions. Multiple experts20 are assessed individually about their uncertainty of
the value of a measurement or observation from within their field of expertise. They are asked to
give their opinion about values of variables that are the target of the research (target variables) and
about values of variables that are not of direct interest (seed variables). The uncertainty
distributions given by the experts for these seed variables are scored since the exact actual values
are known by the analyst. From these scores a weight is determined for each expert. These
weights are attributed to the experts’ uncertainty distributions for the target variables. Therefore
resulting in a performance-based determination of uncertainty by experts. This method results in a
cumulative probability distribution, according to the chosen weighting scheme, for each variable.
This subsection is entirely based on Aspinall (2008), unless indicated otherwise.
Most common is to ask the expert for his/her opinion on the representation of the 5th, 50th and 95th
percentile of the expert’s subjective uncertainty distribution. The calibration, or weighting, of the
experts happens by means of two characteristics. On the one hand calibration, which measures the
statistical likelihood that a set of actual values corresponds to the expert’s assessments. In other
words this tests to what extent the expert’s estimates can be trusted to approximate the
(uncertainty of the) variables. On the other hand information is used. Information measures the
degree to which the uncertainty distribution of an expert is concentrated compared to other experts
in the same group. In other words this tests how sure an expert is about his estimates.
A certain weighted combination of the experts’ uncertainty distributions is often called a Decision
Maker (DM). Usually ten seed variables are used for scoring the experts. The expert’s weight is
determined by the product of the calibration and information scores. Good expertise means good
calibration and superior information, this is rewarded by a high weight and thus a larger influence
on the DM. The decision making process is schematised in Figure 40.
In 2008 already 45 expert elicitations with the unequal weighting of experts had been performed
and reviewed under contracts, and often the results were published. This experience shows that in
the majority of the cases the individual weighting of experts results in more accurate and
informative results then assigning equal weights to experts. Therefore the Classical Model
approach is well established and an excellent method to assess the judgement of experts in a
structured fashion. For the theory concerning Expert Judgement Elicitation on uncertainty
reference is made to Section III.1 of the appendices.
20
The minimum number of experts to assess is four according to Cooke & Goossens (2008).
90 EXPERT JUDGEMENT ELICITATION

Figure 40: A schematic example of the determination of a DM with 1 seed and 1 target variable by 3 experts.
Source: Aspinall (2008)
3.1.2 Eliciting dependence

In this study dependencies between random variables are quantified by means of correlations.
Correlation is a measure of, on one hand, the strength; quantifying the degree to which two
random variables are functionally related. And on the other hand the direction (positive or
negative) of the relationship between two random variables. In this study both Spearman’s rank
correlation coefficient (r) as Pearson’s product moment correlation coefficient (ρ) are used.
Ranking means the assignment of positions to the observations of a variable, in this case, in
ascending order. When the rank correlation between two variables is wanted the rankings of the
two are compared. The degree of similarity determines Spearman’s rank correlation coefficient,
which ranges between -1 and 1. A positive rank correlation quantifies how well two random
variables are described by a function f(x) that is increasing for increasing x. In other words, f(x) is
a positive monotonic function. The rank correlations r = -1 and r = 1 are assigned to two variables
that are perfectly described by respectively a negative and positive monotonic function. A value
of 0 means there is no monotonic relation, however this does not imply there is no relation at all.
Pearson’s product moment correlation coefficient assesses the positive or negative linear
relationship between two variables and also ranges between -1 and 1. The previous is based on
Udny Yule et al. (1950).
To visualise relationships between variables a Bayesian Network (BN) can be created, an example
is depicted in Figure 41. The dependencies are represented by the rank correlation. Where r4,3
stands for the unconditional rank correlation between random variables 4 and 3. r4,2|3 represents
the conditional rank correlation between random variables 4 and 2, given random variable 3.
r4,1|3,2 represents the conditional rank correlation between random variables 4 and 1, given random
variables 2 and 3.
3.1 The method 91

Figure 41: BN of four variables with conditional relations. Source: Morales-Napoles et al. (2007)
Multiple experts21 are assessed individually about the dependency between different
measurements or observations from within their field of expertise. For the dependence assessment
at least one seed dependence must be used to be able to score the experts, here holds the more
seeds the better.22 For this study one seed variable is used due to reasons elaborated on in the
introduction. The expert’s estimates of the target correlations are weighted by means of these
scores. Various methods for eliciting rank correlations exist, Kraan (2002) gives an overview. For
this study the conditional probability technique is used. This method makes experts give an
estimate of what they think is the probability that some variable X1 is observed above its median
when it is given that variable X2 is observed above its median. From these probabilities the rank
correlations can be determined. Combining the weighted correlation estimates of the experts
results in a correlation matrix. This paragraph is based on Morales-Napoles et al. (2007). For the
theory concerning Expert Judgement Elicitation on dependence reference is made to Section III.2
of the appendices.
3.2 The assessed variables
For two important variables no common values could be found during the literature study. These
variables are the total terminal factor (α) and the storage occupancy (ms). These parameters are
both factors used to calculate the total terminal area. They can therefore have a large impact on
the terminal dimensions. The dependence between average dwell times for different cargo flows is
determined by the conditional probability technique.
3.2.1 Total terminal factor

The total terminal factor (α with unit %) is the percentage of all the gross storage areas with
respect to the total terminal area; . For container terminals the gross storage area only
includes the stacks and internal infrastructure inside the storage area (including import, export,
transhipment, empties stacks and container freight station). The total terminal area includes the
total landside (including the gross storage area, other buildings, other infrastructure on the
terminal, terminal gates, etc.) and waterside area (the apron). The uncertainty of the total terminal
factor is assessed for container, dry bulk and liquid bulk terminals. See Figure 42 for an example
of a container terminal, with in red the gross storage areas and in blue the total terminal area.
21
Dr. Morales-Nápoles stated in personal communication that the minimum number of experts to assess is 4.
22
From personal communication with Dr. Morales-Nápoles.

Figure 42: TCB, S.L. Port of Barcelona (Spain). Source: Google Earth
For dry bulk terminals the gross storage area includes the area covered by stockpiles and
stackers/reclaimers and conveyor belts in between the stockpiles. As well as the area covered by
storage sheds and silos including the surrounding area that is kept clear of objects. The total
terminal area also includes the quay or jetty area used for berthing. For an example see Figure 43.
3.2 The assessed variables 93

Figure 43: Kinder Morgan, Norfolk (Virginia, USA). Source: Google Earth
For liquid bulk terminals the gross storage area includes the area covered by the storage tanks.
Also the surrounding area that is kept clear of objects or the area that is surrounded by and
including the bunds (earth walls) is included. The jetty area used for berthing is included in the
total terminal area. For an example see Figure 44.
Figure 44: Vopak TTR, Port of Rotterdam. Source: Google Earth

3.2.2 Storage occupancy factor
The average annual storage occupancy (ms with unit %) -or storage or yard utilisation- is defined
as the number of occupied storage slots or occupied storage volume divided by the total number
of storage slots or storage volume according to the design capacity. This factor takes into account
the fluctuations in required storage capacity due to random arrivals and departures of cargo. The
storage occupancy factor is assessed for container, dry bulk and liquid bulk terminals. The
uncertainty is assessed by experts.
3.2.3 Average dwell time

The definition used for dwell time in this research is: the time between the moment that a
container is unloaded from a vessel and the moment that same container leaves the terminal
boundaries, and vice versa. The average dwell time is taken as the yearly average of dwell times
of all containers within a certain cargo flow. The cargo flows that are taken into account in this
research are import, export and transhipment. For these average dwell times common values are
found in literature. However there might be a dependence (relationship) between them when they
are considered as random variables. The dependencies, denoted by r, are depicted in Figure 45.
Where r2,1 stands for the unconditional rank correlation between random variables 2 and 1. r3,1|2
means the conditional rank correlation between random variables 3 and 1, given variable 2. The
three dependencies are assessed by the experts.
Figure 45: (Un)Conditional rank correlations between the random dwell times. Source: Own work
3.3 The elicitation
The uncertainty elicitation method the Classical model is applied to the total terminal factor and
the average annual storage occupancy. The dependence elicitation method conditional probability
technique is applied to the average container dwell times. For this study four experts have been
elicited. These four experts all specialise in the field of terminal and port planning. The experts
were personally assessed in approximately one hour. The assessments were face-to-face since the
probabilistic approach is likely to raise questions. All experts completed two questionnaires; one
regarding uncertainty and one dependence. The two questionnaires are included in Appendix IV .
The uncertainty questionnaire consists of ten seed variables (or calibration variables) and six
target variables. These ten seeds consist of six questions concerning the total terminal factor and
four concerning the average storage occupancy, and thus are related to the target variables. In
these questions the expert is asked for his/her estimate of the variable in an actual situation, in this
case a marine terminal. The seed variables are selected such that they resemble as much as
possible the variables of interest, when also the required information is available. The dependence
questionnaire consists of one seed dependency and three target dependencies. Only one seed is
used because data -used to determine the dependence- is very hard to come by. As well as for the
3.3 The elicitation 95

reasons stated in the introduction. The seed dependency variables are the quarterly container
throughput and quarterly number of container vessels in the Port of Rotterdam in between January
1997 and December 2014. These seed variables do not match the target variables which are
average container dwell times for import, export and transhipment containers. The experts are
however expected to be knowledgeable about all these variables since they occur in the same field
of expertise. The seed variable realisations (or actual values) can be found in Section IV.3 of the
appendices. The assessment results are presented and discussed in Section 3.4.

3.4 Analysis of the results
This section provides an analysis of the elicitation results and presents the conclusions. These are
again divided into uncertainty and dependence. The complete Expert Judgement Elicitation is
performed as is described in Appendix III .
3.4.1 Uncertainty
This section first presents a summary of the elicitation results. Then observations of the
performance of the experts are provided. Hereafter the uncertainties of the experts are combined
by various Decision Makers. As explained earlier a Decision Maker (DM) is a certain weighted
combination of experts’ uncertainty distributions. The best DM is then chosen and the
corresponding resulting probability distributions for the target variables are determined.
3.4.1.1 Elicitation results summary
This subsection presents general observations that can be made regarding the assessment results
of the seed and target variables. The experts’ uncertainty distributions are given in Section V.1 of
the appendices. All experts’ conclusions, and motivations for these conclusions, are summarised
per questionnaire item in Section V.2 of the appendices.
Seed variables
For most seed variables multiple experts tend to arrive at the same conclusions. However they do
propose different distributions for all questions. Also their motivations for the conclusions are not
always similar; they have their own angle from which they assess the problem. To show this, one
of the ten seed question is elaborated on. The mentioned differences in conclusions and
distributions is demonstrated by seed question eight, of which the results are depicted in Figure
46. The graph represents the experts’ opinions about this single questionnaire seed variable. Each
horizontal bar represents the uncertainty between the 5 th and 95th percentiles provided by the
corresponding expert. The white vertical bars represent their 50th percentile. The true value (or
realisation) of the seed variable is depicted by a vertical green line.
The conclusions and motivations for this specific seed question are:
 Experts two and three (E2 and E3) have the same conclusions and motivations: they think
that this terminal is near its maximum capacity since they know an expansion is planned
(motivation), therefore utilisation will be high (conclusion).
 E3 also states that for container terminals an occupancy of 70% is high while for dry and
liquid bulk terminals the percentages are lower. Eventhough E2 and E3 have similar
conclusions their uncertainty distributions are not the same.
 E4 has the same conclusion but a different motivation; he thinks that dwell times in Africa
are larger (motivation), so that probably leads to a higher yard utilisation (conclusion). He
also thinks that in Africa strange situations can occur so the expert uses a large
uncertainty.
For the seed questions in seven of the ten cases at least two experts give the same motivation for
their shared conclusion. For only one question two experts (E2, E4) give contradicting
conclusions. The concerning seed variable is the total terminal factor for the Vopak TTR liquid
bulk terminal in Rotterdam. E1 provides a similar distribution as E4. The consequences are
discussed in 3.4.1.3.
3.4 Analysis of the results 97

Notes: Bar limits are 5th and 95th quantiles, white stripes are medians and green line is the actual value.
Figure 46: Uncertainty distributions of the experts for seed question 8. Source: Own work
Target variables
This paragraph discusses the general observations that can be made from the elicitation results of
the target variables.
 Expert one (E1) provides asymmetrical uncertainty distributions. The medians of his
estimates for the total terminal factor target variables -for the three terminal types- are
similar.
 E2 uses for his target variables similar medians as for his estimates of comparable seed
variables. He chooses his uncertainty interval larger for the target questions.
 E3 chooses the target total terminal area distributions exactly the same as for the (first in
the sets of two) corresponding seed variables. His estimated interval widths for all target
variables are equal to his seed variable estimates.
 E4 uses for his target variables similar medians as for his estimates of comparable seed
variables. His interval widths are much smaller for the target total terminal factor
variables than for corresponding seed variables, this in contrast to E2. For the storage
occupancy target variables his interval widths are similar to the those of the corresponding
seed questions.
A demonstration of some of these observations is depicted in Figure 47. Here the upper plot
shows the results of the experts for the seed variable concerning the Kinder Morgan dry bulk
terminal in the USA. The lower plot shows corresponding target variable results for the total
terminal factor of a dry bulk terminal in general.

Figure 47: Uncertainty distributions of the experts for seed question 4 (upper) and target question 12 (lower).
Source: Own work
3.4.1.2 Experts performances

As is explained in Section III.1 of the appendices, the uncertainties of the experts are judged by
means of their calibration and information scores (considering seed variables only). Calibration
quantifies the extent to which the estimates of the expert concur with the actual values (or
realisations), in a statistical sense. The calibration score is an absolute score and lies in the interval
(0, 1), in which 1 equals perfect calibration. Information quantifies the extent to which an expert
is certain. The information score is determined relative to another distribution. This distribution is
dependent on the uncertainty estimates of the other experts. The information score scale is
therefore not absolute. Since information does not take realisations into account, information
scores can be determined for only the seed or both seed and target questions.
The total score (or un-normalised weight) of an expert is the product of the calibration and
information scores. Here holds the higher the weight the better. The calibration score has the most
influence on the total score since the calibration score increases more rapidly. Table 37 presents
the mentioned scores and the resulting total score per expert. Two different information scores are
computed by considering only seed variables or both seed and target variables. Large differences
between both information scores can indicate that an expert answers the seed and target questions
differently. This is not desired in an elicitation. In this assessment this is however not the case, as
can be observed.
Experts 2 and 3 both have low calibration scores. Expert 2 however has a high information score.
Expert 3 is a little bit less certain and has slightly better calibration results. However the
calibration score is still low compared to experts 1 and 4, therefore resulting in a low weight. This,
as calibration has a larger effect on the un-normalised weight. Expert 1 and 4 both have good
calibration scores as well as information scores, this results in good un-normalised weights (or
total scores) for both. Very low calibration scores for the seed variables may originate from the
expert not understanding the method or the meaning of the questions themselves. Even though the
expert may have good estimates for the target variables, as a result of their calibration scores their

opinions are only slightly or not used at all in the Decision Maker. This is further discussed in
Section 3.5.
Table 37: Calibration and information scores and the total score of the four experts
Information Information
Calibration Total score
seed variables all variables
Expert 1 0.3136 0.5786 0.5604 0.18145
Expert 2 2.39E-08 1.016 0.8417 2.428E-08
Expert 3 5.60E-05 0.7015 0.6602 3.928E-05
Expert 4 0.2895 0.2807 0.3658 0.08126
3.4.1.3 Decision Makers

As is already explained the uncertainties of the experts are weighted on account of their
calibration and information scores, this is done by means of the Excalibur23 software. A Decision
Maker combines the uncertainty distributions of the experts per item by using a certain weighting
method. The considered weighting methods are:
 Global weights: Uses the un-normalised weights. Per expert the weights are the same for
all items.
 Item weights: A variation of global weights where each item can have a different set of
weights. Uses the individual information scores of the experts per item and the calibration
scores to determine a different set of weights for each item.
 Equal weights: The weights are the same for all experts and for all items.
The global and item weights can be optimised. When optimising; an iteration is performed where
the minimum allowable calibration score is increased. When experts do not meet the minimum
score their uncertainties are not taken into account in the DM. The optimisation ultimately
chooses the calibration score for which the un-normalised weight of the DM is maximum. The
goal is to choose the DM with the best calibration and information scores. In Table 38 the
calibration scores and information scores (considering only seed variables) are given per expert
and Decision Maker (with given weighting method). A DM can be seen as a virtual expert.
Table 38: Calibration and information scores for experts and Decision Makers calculated with Excalibur.
Information
Calibration
seed variables
Expert 1 0.3136 0.5786
Expert 2 2.39E-08 1.016
Expert 3 5.60E-05 0.7015
Expert 4 0.2895 0.2807
DMgl 0.5503 0.2379
DMgl_op 0.3136 0.5786
DMit 0.6827 0.2658
DMit_op 0.6827 0.2660
DMeq 0.1135 0.1187
Notes: DM = Decision Maker, gl = global weights, it = item weights, eq = equal weights, op = optimised.
23
Originally developed at TU Delft, now maintained by Lighttwist Software.
URL: http://www.lighttwist.net/wp/excalibur

The DM with the highest calibration and total information scores is desired. In principle all DM’s
are acceptable to be used since the calibration scores are higher than the 0.05 limit. However
equal weights (DMeq) can be discarded since the scores are lower than those of the best experts.
Global optimised weights (DMgl_op) does not combine distributions but only uses the
distributions of expert 1, this is not preferable. Global weights (DMgl) scores a little bit lower on
calibration and information than item weights (DMit) and item optimised weights (DMit_op). The
latter two have almost identical scores, DMit_op is however a little bit better on estimating the
distributions with more certainty (smaller intervals). It can be observed that DMit_op has more
than twice the calibration score of the highest scoring expert. Using this Decision Maker therefore
highly increases the accuracy of the resulting uncertainty distributions compared to for example
using only the best scoring expert or taking the average of all the experts. Figure 48 demonstrates
some of the mentioned observations for the total terminal factor seed variable; concerning a dry
bulk terminal in Shanghai. The distributions of the various Decision Makers for the specific
variable are depicted.
Figure 48: Uncertainty distributions of all DM’s for seed question 3
As discussed in Section 3.4.1.1; E2 and E4 have contradicting conclusions for the question about
the Vopak TTR liquid bulk terminal in Rotterdam. However due to the choice for DMit_op, E1
and E4 receive by far the highest weights so the DM is not affected by this disagreement. Which
expert gives the right conclusion can of course not be determined in this study. The uncertainty
distributions of the experts and the DMit_op are depicted in Figure 49. Similar graphical
representations of the distributions for all questionnaire items are given in Section V.1 of the
appendices.
Figure 49: Uncertainty distributions of the experts and the chosen DM for seed question 6. Source: Own work

3.4.1.4 Resulting distributions
Using the Decision Maker with the optimised item weighting method (DMit_op) results in
probability distributions for the six target variables. The 5th, 50th and 95th quantiles are calculated
and the lower and upper 5% of the distributions can again be estimated by the k% overshoot rule
with k = 10 (see Section III.1 of the appendices). This results in the quantiles presented in Table
39.
Table 39: Uncertainty distributions as a result of the Expert Judgement Elicitation
Quantiles
q0 q5 q50 q95 q100
Containers 46.00 52.24 65.60 74.80 94.00
factor [%]
terminal
Total
Dry bulk 45.50 50.41 64.40 88.75 99.50
Liquid bulk 45.50 50.62 63.22 75.00 99.50
Containers 40.50 60.34 73.09 80.00 94.50

occupancy
Average
storage
[%]
Dry bulk 41.00 60.80 79.73 85.00 89.00
Liquid bulk 41.00 48.71 65.00 81.29 89.00
From these distributions cumulative probability distributions can be made by linear interpolation.
The cumulative distributions are used for the common values of parameters in Chapter 2 and as
standard values in the computer tool. Random values can be drawn from these cumulative
distributions by using the Random Sampling method which is described in Section 4.3.1. The
cumulative probability distributions are depicted in Figure 50.

Figure 50: Cumulative probability distributions as a result of the EJE on uncertainty. Source: Own work
These distributions are constructed from the different opinions of experts. They give an indication
of what the values of the variables can be. This study (especially Chapter 2) aims to provide a
general overview of the different terminal types with their main elements, design rules and
common values of variables. In some situations other values may be more adequate to be used.

The tool allows the user to specify custom distributions, this also holds for the total terminal
factor and estimated storage occupancy. Even though the distributions are only based on the
estimates of experts one and four they are valid results and can therefore be used.
3.4.2 Dependence
For the elicitation the four experts were asked the following target questions:
1. Suppose the dwell time of import containers is observed above its median value. What is
the probability that the dwell time of export containers will also be observed above its
median value?
2. Suppose the dwell time of export containers is observed above its median value. What is
the probability that the dwell time of transhipment containers will also be observed above
its median value?
3. Suppose the dwell time of import and export containers are observed above their median
values. What is the probability that the dwell time of transhipment containers will also be
observed above its median value?
This section first presents a summary of the elicitation results. Then observations of the
performance of the experts are provided. The best pooling method is chosen and the
corresponding resulting correlation matrix for the target variables is determined.
3.4.2.1 Elicitation results summary

The experts give quite similar probabilities for the seed question, only expert three thinks the two
random variables are independent. In fact experts one and four exactly predict the actual
correlation of the two random variables. In Figure 51 the true seed probability is depicted in
green. Due to the fact that only one seed is used these two experts dominate when the experts’
estimates are combined. Of course this does not mean that experts two and three are by definition
wrong about the target questions.
For the target questions it cannot be concluded that certain experts give similar estimates. The
most notable are the estimates of expert four since they seem to differ quite much compared to the
other experts’ estimates. In Figure 51 also the target estimates of the experts are depicted. A
summary of the motivations of the experts is presented:
 Experts one and two (E1 and E2) think that import and export dwell times are both
dependent on customs, while transhipment is not. They therefore choose a medium
positive dependence. Their estimated probability for target question two is smaller, since
they argued that transhipment is not influenced by customs. Therefore the correlation
between export and transhipment dwell times is smaller. For the last target question the
experts think that in this situation (when import and export dwell times are high) the
terminal does not work efficiently. So the transhipment dwell time will also be higher,
resulting in a larger probability. Despite having the same motivations they use different
probabilities.
 E3 thinks that export and import dwell times are intertwined, he therefore chooses a large
probability for target question one. He thinks that export and transhipment dwell times are
not much related since export flows can be planned while for transhipment this is much
more difficult. Despite this he chooses a small positive dependence for question two. The
expert states that when the import dwell time is also taken into account -next to export and
transhipment- there is no dependence any more. For target three he therefore chooses
independence.

 E4 thinks that import and export dwell times are independent since they are two different
modes of transport. For the second target question the expert argues that for example
when an STS crane is impeded this will affect both export and transhipment dwell times.
He therefore chooses a relatively high dependence. For the last question he uses the same
motivation as E1 and E2, his chosen probability is however much larger.
For their full motivations reference is made to Section V.2 of the appendices.
Figure 51: Probabilities estimated by the experts, as answers to the seed and target questions.
3.4.2.2 Experts performances

As described in Section III.2 of the appendices, the experts are scored by means of a dependence
calibration score (d-calibration score with symbol D). This d-calibration score is introduced in
Morales-Napoles & Worm (2013) and makes use of the Hellinger distance. The experts are
scored on their estimate of a correlation between two seed variables; namely the quarterly
container throughput (Cq) and quarterly number of container vessels (nships) in the Port of
Rotterdam in between January 1997 and December 2014. The expert’s estimates are in the form
of a probability, this probability can be converted to a rank (r) or product moment correlation (ρ)
for a normal copula24 by using the theory provided in the appendix. The correlations are
represented by a correlation matrix. In these matrices each item represents a dependence between
two variables, the main diagonals are therefore one. Since for the seed question only one
dependency between two variables is considered the matrices are 2 x 2. The seed correlation
matrix therefore has the following form:
The symmetrical product moment correlation matrix of the seed variable realisation (or actual
value) is:
24
A copula is a multivariate probability distribution of which the marginal probability distributions are uniform;
Genest & Favre (2007).

The symmetrical product moment correlation matrices of the experts (E1, ..., E4) are represented
by . Expert 3 thinks the two variables are independent, his correlation ρ is therefore
zero.
To pool the experts’ estimates the equal and global weights methods are considered. For the equal
weights method every expert gets the same weight. For the global weights method a calibration
threshold is used that is increased in an iterative process. When the individual d-calibration score
of an expert is below this threshold the weight of that expert is zero. The remaining experts their
weights are based on the contribution of the individual calibration scores to the total. The
weighted combination of each iteration step (each increase of the threshold) is treated as a virtual
expert of which the d-calibration is determined as well. The weighted combination with the
highest calibration score is selected. The resulting d-calibration scores for the two different
weighting methods are listed in Table 40. It can be concluded that the use of global weights
results in the best calibration of the pooled experts’ estimates.
Table 40: d-calibration scores per weighting method (combination of experts)
d-calibration
score
Equal weights 0.8635
Global weights 0.9994
The optimum calibration threshold for the global weights is 0.9130 which leads to the use of
experts one and four only, see Table 41 for the individual d-calibration scores. These two experts
almost exactly predicted the true correlation, the d-calibration is therefore practically 1. It is
therefore also quite obvious that the combined calibration score would be lower when expert two
and three would be included. The d-calibration scores of experts two and three are good as well,
this indicates that their target estimates are also valuable. Due to the near perfect seed estimates of
experts one and four they are however not taken into account in the pooled dependencies. This
situation is exceptional and can be prevented by using more complex seed variables; such as seeds
with conditional dependencies (like the target dependencies in this study). The experts are less
likely to perfectly predict the seed dependencies in that case.
Table 41: Dependence calibration score and weight per expert
d-calibration
Weight
score
Expert 1 0.9994 0.5
Expert 2 0.9125 0
Expert 3 0.6816 0
Expert 4 0.9994 0.5
Due to the global weights scheme, the average of the estimates is taken per dependency of experts
one and four. As can be seen in Figure 51; E1 and E4 give the most opposing estimates for the
target dependencies, while both perfectly agree on the seed variable. Only for the first target
question E3 instead of E1 gives an opposing estimate to E4. E3 however has the lowest d-
calibration score, so this is ignored for this matter. The global weights scheme, by taking the

average of E1 and E4, therefore gives a good representation of all the individual experts’
estimates.
3.4.2.3 Resulting correlation matrix

For combining the target dependency estimates of the experts, their correlation matrices and
weights are required. Since for the target questions three dependencies between three variables are
considered the matrices are 3 x 3. The weights are calculated in Section 3.4.2.2, the symmetrical
matrices are presented here:
Pooling the experts’ opinions by means of the global weights method results in the target
correlation matrix . This matrix is used in the tool when drawing random numbers for the
import, export and transhipment dwell times. As already explained the correlations are a measure
of strength and direction of the relationship between two variables, ranging from -1 to 1. The
target rank correlation matrix and product moment correlation matrix are:
These matrices are equivalent since the normal copula is used. The relation between the rank
correlation (r) and product moment correlation (ρ) is: , reference is made to
Equation (36). The product moment correlation matrix is used in the tool, as the RiskAMP add-in
requires this. The rank correlations are depicted in Figure 52.
Figure 52: BN with unconditional rank correlations between the average container dwell times. Source: Own
work
It can be concluded that there are dependencies between the dwell times of the three different
container flows. There are moderate positive relationships between import & transhipment and
export & transhipment container dwell times. As well as a weakly positive relationship between
import and export container dwell times.

3.5 Discussion
Some of the four elicited experts know each other. This could raise doubt about the results being
biased. This is however not the case since the experts did not know the questionnaire questions
beforehand and the elicitation documents were collected afterwards.
EJE on uncertainty
For the EJE four experts are assessed. Including more experts may lead to a more complete
Decision Maker, since more motivations are taken into account. Most of the assessed experts
claim to have more knowledge about container terminals than about dry and liquid bulk terminals.
Even though the calibration and information scores of the DMit_op of all the questionnaire items
are quite good, ultimately they are only based on the opinions of two experts. It can be observed
from the assessment data that on average the information and calibration scores are the best for
the dry bulk terminal variables, the scores for the container and liquid bulk terminal variables are
similar but lower. Creating three separate questionnaires, for the three terminal types in
combination with a sufficient number of experts only on those specific terminals, may improve
the results. This way experts are scored only on their own field of expertise, which can improve
their weights for the corresponding target variables. Then the chance that ‘good’ knowledge is
lost, because of low weights, is smaller. The number of seed variables in this elicitation is
sufficient.
EJE on dependence
For the dependence assessment only one seed is used. This choice is made since suitable data is
very hard to come by. One goal of the EJE on dependence is finding the correlations between the
average container dwell times of the different cargo flows. Another goal is to introduce this
method in this field of expertise. For the latter goal the importance is that the technique is
presented, regardless the number of seed variables. Moreover in this study two experts perfectly
estimated the seed correlation and therefore have dominating weights. This is an exceptional
situation and can be prevented by asking the experts about conditional dependencies. Accordingly
it is recommended to use seed variables with a similar number of (conditional) dependencies as
for the target variables. Nevertheless the results in this study are a good representation of the
individual experts’ opinions.
In the study Morales-Napoles et al. (2016) it was concluded that experts who perform well in an
uncertainty assessment do not necessarily show good performance in a dependence assessment
with similar variables. Calibration scores therefore have to be determined separately. The results
of the current study show that experts one and four score best in both the uncertainty and
dependence assessments. This highlights that the same experts scoring well in an uncertainty and
dependence assessment is possible but is not a certainty, according to Morales-Napoles.

4 THE TOOL
This chapter is dedicated to the computer tool that is created from the ground up as part of this
study. The tool is used to compute terminal dimension and costs, this is elaborated on in the first
section. Hereafter the structure of the tool is considered. By the structure is meant: information
about the programming language and the framework of the main elements of the front-end and
back-end of the tool.25 The tool is the property of Witteveen+Bos and is therefore not included in
this report.
Chapter goal: Providing information about the application, assumptions & restrictions,
computations and software structure of the tool.
4.1 Tool function
The tool determines the required dimensions, quantities and construction costs of a marine
terminal, which are listed in Section 4.2.2. It can function as a quick help in estimating the main
terminal dimensions as well as for more detailed aspects like the number of vessel (un)loading
equipment or yard handling equipment. An estimate of the direct costs is made by using the
calculated dimensions and quantities in combination with unit prices (Section 2.4).
When certain information is missing in the early design phases of a project not all required
parameter values may be known. Therefore standard (or common) values for these input
parameters are presented in the literature study (Sections 2.1-2.3) and by means of Expert
Judgement Elicitation (Chapter 3). Thus when not all required input parameters are obtained for
the project of the terminal designer, the standard values can be used to give a proper estimate of
the terminal dimensions and costs. The tool helps the terminal planner by making it possible to
insert the uncertainties (in the form of uncertainty distributions) into the tool instead of having the
planner assume a single value. Consequently the tool performs a probabilistic calculation which
results in probabilities of occurrence for the terminal dimensions and costs.
Growth scenarios of for example the estimated throughput over a certain number of years cannot
specifically be inserted in the tool. However, it is possible to perform separate computations with
input corresponding to certain moments in the growth path. The results of the different
computations can then be used to develop a phased master plan for the terminal.
4.2 Tool structure
This section covers the choice of the programming language used for the tool and it defines the
main elements of the tool and their relations.
25
By front-end the user interface is meant. By back-end the underlying software code is meant.
109
4.2.1 Software
The tool is realised in Microsoft Excel. Excel is chosen since it is a widely used program; at
Witteveen+Bos similar modelling tools are often made in Excel. The tool makes use of a lot of
user input and standard values of parameters, the cells in an Excel worksheet can function
perfectly as a way to input and output values. Moreover when insights change or minor errors are
discovered it is quite easy to adjust the tool by people with minor programming skills.
The input and output of information takes place in the worksheets that are directly accessible to
the user. The calculations and logics however take place in coded modules; the code is written in
Visual Basic for Applications26 (VBA). VBA is based on the object oriented standalone
programming language Visual Basic. VBA however is not standalone and requires a host
application like Excel. By separating the input and output from the operations the tool in itself is
more transparent. The choice to implement the calculations and logics in VBA is made because
using the cells in Excel for this purpose would lead to a chaotic collection of code of which the
overview is easily lost and which is very hard to debug.
For the probabilistic calculations that are performed in the tool the Monte Carlo method is used,
this method is elaborated on in Section 4.3. To be able to perform a Monte Carlo simulation in
Excel the tool uses a paid Excel add-in called RiskAMP27. The tool only uses RiskAMP’s
functions to generate random numbers from given random distributions and its functionality to
include dependence between random variables. For the writer it is possible to create these
functions himself however due to time limitations the choice is made to use RiskAMP.
4.2.2 Worksheets
The Excel tool consists of nine worksheets, each with its own function. In general the sheets are
the input/output, standard data, unit costs and calculation sheets. These worksheets interact by
means of the VBA code, elaborated on in Section 4.2.3. This is depicted in Figure 53. A summary
of the most important aspects of the different sheets is given in this subsection.
Figure 53: The nine worksheets in Excel (upper). The relations between worksheets and VBA code (lower).
Source: Own work
26
For more information about VBA, URL: https://msdn.microsoft.com/en-us/library/office/gg264383.aspx
27
For more information about RiskAMP, URL: https://riskamp.com/
110 THE TOOL

4.2.2.1 Input/output sheets
The Containers, Dry bulk and Liquid bulk sheets form the input/output sheets. These sheets
consist of separate input and output parts.
Input
The input section of these sheets consists of all the calculation variables that are divided into
categories; general, cargo flows, quay, storage yard and vessels/basin. A small part of the
Containers input sheet is depicted in Figure 54 (upper). For each variable one of four uncertainty
distributions can be chosen via a dropdown menu, examples of these distribution types are
depicted in Figure 55. For the chosen distribution the corresponding parameters have to be
specified, this is done with a separate window that can is opened by clicking the '...' button. The
uniform distribution input window is displayed in Figure 54 (lower left). Right from the '...' button
the values that are used for the computation are displayed. A small red triangle notifies the user
that more information about the corresponding variable or option is available, this information is
shown when the mouse is hovered above the triangle, see Figure 54 (lower right).
Figure 54: Part of container terminal input sheet (upper), uniform distribution input window (lower left),
information label of the throughput variable (lower right). Source: Own work
As mentioned in the previous paragraph the four distribution types that can be used for the input
parameters are depicted in Figure 55.
4.2 Tool structure 111

Figure 55: Examples of the four possible uncertainty distribution types. Source: Own work
For many variables common values of parameters are included in the tool. These values can be
used for a specific variable by using the 'Reset' button in the uncertainty window. If the user wants
to reset all variables in the selected input sheet the 'Reset all variables' button can be used. Some
non-numerical variables (e.g. yard equipment type) are also required, these can be chosen via a
dropdown menu. For dry bulk en liquid bulk terminals the cargo type(s) have to be selected. Most
of the input parameters have to be provided for each cargo type separately. The tool performs the
calculations for each cargo type separately and finally sums them up. For container terminals the
standard berthing object is a quay wall since jetties are almost never used at container terminals.
For dry bulk terminals a choice can be made between a quay wall or jetty because both are used in
practice. For liquid bulk terminals the standard berthing object is a jetty, since quay walls are not
common. These limitations are made to reduce the tool’s complexity.
The number of iterations of the Monte Carlo simulation must be specified in the input section as
well. The inflation parameters must be specified as well. These parameters are the year until
which the inflation has to be calculated and the average annual inflation rate. These options are
depicted in Figure 56. A recommended number of iterations -based on the case study- is a 1000.
Too quickly see the effect on the results, of changes to the input, a smaller number of iterations
(e.g. 250) can be used.
Figure 56: Calculation properties. Source: Own work
Output
The output section displays the results of the calculations. Since probabilistic calculations are
performed the results are probability density distributions (in the form of histograms). These
112 THE TOOL

histograms are in other words relative frequency plots, an example is depicted in Figure 57. The
relative frequency (or probability of occurrence) that a result falls in a certain range (or bin) of
results is represented on the vertical axis. Each value on the horizontal axis represents the upper
boundary of a bin, the previous value represents the lower boundary of the same bin. The lower
limit of the very first bin is not given, but this bin has the same width as the others. All the
computation results are given below, for dry and liquid bulk terminals the other results are shown
for each cargo type that is selected.
 Main results:
o Quay length (zero when a finger pier jetty is selected. For an island berth jetty this
length is the required berthing structure length).
o Total terminal area
o Dredging depth
 Other results for a container terminal:
o Number of berths
o Annual berth productivity
o Number of STS cranes
o Number of equipment between quay and yard
o Quay length from berth capacity check
o Quay length from crane spacing check
o Total gross storage area
o Gross storage areas for import, export, transhipment, empties and CFS
o Number of storage yard equipment
o Total terminal area from storage factor check
 Other results for a dry bulk terminal, results are divided per cargo type:
o Number of berths
o Number of jetties
o Berth productivity of unloading and loading equipment
o Total number of unloading and loading equipment
o Required capacity (in weight and volume) of storage
o Total number of stacking and reclaiming equipment
o Lane length of the stockpile or stockpile length in the storage shed or number of
silos.
o Terminal area
o Terminal area from storage factor and capacity ratio and from a single shipload
 Other results for a liquid bulk terminal, results are divided per cargo type:
o Number of jetties
o Jetty trestle length
o Berth productivity of unloading and loading equipment
o Required capacity (in weight and volume) of storage
o Number of tanks
o Number of tank groups
o Volume of sand required for the bunds
o Terminal area
o Terminal area from storage factor and capacity ratio and from a single shipload
For the ‘other results’ only the value of a user-specified quantile is provided. The probability
distribution of these additional results can be called for individually. When for dry and/or liquid
bulk terminals multiple cargo types are selected the main results show the sum of the individual
results.

Figure 57: A main tool result; histogram of the required quay length of an example project. Source: Own work
Another output are the estimated terminal construction costs that are computed by means of the
unit costs presented in Section 2.4. The main cost items -per terminal type- are:
 Container terminal
o Quay wall: Costs including the supply of material and the required man-hours.
o Pavement: Costs for the supply and construction of the terminal pavement. This
includes the stabilising sand layer underneath the pavement.
o STS cranes: Investment costs, excluding crane rails. The foundation is not directly
taken into account, however a quay wall partly bears the weight of the equipment.
The foundation is therefore partly included in the unit costs of a quay wall; if the
equation of de Gijt is used.
o Terminal equipment: Investment costs of equipment between the quay and storage
yard and on the yard itself.
o CFS: Costs for the construction of a warehouse structure.
 Dry bulk terminal
o Quay wall/jetties: Costs including the supply of material and the required man-
hours.
o (Un)Loading equipment: Investment costs of the loading and unloading
equipment, excluding rails. Regarding the foundation, see STS cranes.
o Storage facilities: Costs including the construction of:
o Terminal pavement, for wind-row stockpiles.
o Warehouse structure, for storage sheds.
o Silos.
o Storage equipment: Investment costs of the stacking and reclaiming equipment,
excluding rails and foundation.
 Liquid bulk terminal
o Jetties: Costs including the supply of material (concrete, steel, tubular steel piles,
bollards, fenders, loading arms and quick release hooks) and the required man-
hours.
o Tanks: Costs including the supply and construction of the tanks.
o Bunds: Costs including the supply and handling of the soil.
114 THE TOOL

Cost elements that are not included are roads, buildings, train and crane rails, belt conveyors and
pipelines. The main direct cost items, that can consist of multiple terminal elements, are listed and
a pie chart displays this information graphically. The costs are given for the user-specified
percentile; as the costs are also represented by a probability distribution. The total construction
costs are computed from the total direct costs using the SSK definitions (reference is made to
Section 2.4.1). When for dry or liquid bulk terminals multiple cargo types are selected the
presented costs are a summation of the costs per cargo type. An example is depicted in Figure 58
in which the direct, indirect, contingency and resulting construction costs of a dry bulk terminal
are shown. These costs correspond to the 80% quantile.
Figure 58: Overview of the direct and total construction costs of an example project. Source: Own work
4.2.2.2 Standard data and unit costs sheets

The standard data sheet contains literature study results (see Chapter 2) that can be used for
calculations. The sheet contains tables with common values for the calculation variables per
terminal type. Per variable the distribution with corresponding value is listed. The distribution
parameters can all be changed by the user if desired, this is however not required. Parameters of
multiple distribution types (deterministic, uniform, triangular and normal) can be inserted. Also a
table with constants exists in this sheet, these constants are used for the computations and are all
deterministic. The data sheet also contains equipment properties and benchmarks that are sorted in
tables and that are used by the tool.
The unit costs sheet has the same function as the data sheet but it stores unit prices/costs. The unit
costs that are used for the current calculation are stored in this sheet, these can be changed by the
user -in the same way as the common values of variables in the data sheet- if desired. Also
standard values for these costs are stored in this sheet, these standard costs are treated in Section
2.4. Per terminal type the costs used for calculations can be changed to the standard costs by using
the 'Reset to standard' button. For each cost item the year of the corresponding price level is given.
Over the period between that year and the year specified in the input sheet the inflation is
calculated; but only if 'Yes' is selected in the table for the specific cost item. For some cost items
the costs are determined with a formula. This can be selected -if applicable- at the same place the
distribution type is specified. Additionally the tool requires the user to fill in the value -1 in the
deterministic cell if the costs should be computed by an equation. An example of a unit costs table
for a container terminal is depicted in Figure 59. The last column cannot be changed by the user, it

holds all random values drawn from the specified distribution when a new calculation is
performed.
Figure 59: Overview of a container terminal unit costs table. Source: Own work
4.2.2.3 Calculation sheets

A calculation sheet contains the values of the input variables (the parameters of the specified
uncertainty distribution) of the specific terminal type. During a calculation randomly drawn
instances of the distributions (both input variables and unit costs) are written to this sheet. The
results (both dimensions and costs) of a calculation are written to the sheet as well. The
calculation sheets are not meant to be adjusted by the user.
Since the drawn values and results are written to the sheet on the same row, the data can be used
to create scatter plots. These plots can indicate dependence between two selected variables.
4.2.3 VBA code

The VBA code handles the ‘communication’ between Excel cells, it changes the layout of the
input/output sheets, it makes logical decisions and it calculates the terminal dimensions and costs.
The VBA code is written inside functions that are in turn positioned inside Modules, Sheet
Objects and Forms. Forms (one for every uncertainty distribution type) are used by the user to
insert the distributions. Also the standard values of parameters can be obtained from here. Sheet
Objects (one for every worksheet) change the lay-out of the input/output sheets when certain
events, triggered by the user, happen. The modules contain the remaining -and largest part- of the
code. All these elements including their relations are depicted in Figure 60. The used modules are:
 SheetCore: Controls all the functions that have to do with the reading and writing of
values/content. Passes values between cells of the different sheets. Also contains the
function that starts a calculation.
 CalcCore: Contains the general calculation function, which in its turn delegates
calculation functions in calc_C, calc_Db and calc_Lb. The Monte Carlo simulation is
performed in this function. This module also draws and reads random values and creates
histogram data.
 calc_C, calc_Db, calc_Lb: Contain the equations to calculate the terminal dimensions and
costs per iteration per terminal type.
 calc_Gen: Contains general equations that can be used for multiple terminal types.
116 THE TOOL

Figure 60: Relations between het Excel worksheets and the VBA modules. Source: Own work
4.3 Tool calculation methods and restrictions
This section covers equations not treated in the literature study (Chapter 2), choices and
restrictions that are used for the terminal calculations in the tool. First the Monte Carlo simulation
method is treated, then each terminal type is elaborated on individually.
4.3.1 Monte Carlo simulation

Monte Carlo calculations are used by many people for many applications and are a scientifically
accepted method. Reasons for the use of Monte Carlo simulation are the convenience, ease and
directness of the method; Whitlock & Kalos (2009).
In a Monte Carlo simulation a given number of N realisations of a given number of random

variables I are sampled. In each simulation iteration one realisation for the I variables are
sampled. These realisations are used as input for a deterministic calculation. Therefore each
iteration has one calculation outcome. The total simulation then has N outcomes. In this study the
probability density of the outcomes is determined by creating bins of possible outcome values and
dividing the number of outcomes in each bin by the number of iterations N. The probability
density can therefore also be called the relative frequency of the outcomes.
4.3 Tool calculation methods and restrictions 117

Sampling
For Monte Carlo simulations often Random Sampling is used. For each sample using this method
a uniform random value is drawn. The sample’s realisation is obtained by using the random value
as input for the inverse of the cumulative distribution function of the variable. The previous is
based on Whitlock & Kalos (2009).
The Excel add-in RiskAMP uses Latin Hypercube Sampling. In Latin hypercube Sampling the
probability distribution range is divided into a number of bins K. The width of each bin is chosen
such that the bin represent a probability of size 1/K. Then per bin Random Sampling is used to
obtain a specific number of realisations per bin. The previous is based on M. D. Mckay &
Conover (2000). Using Latin Hypercube Sampling leads to every part of the distribution being
equally represented during the sampling. This results in less required samples (Monte Carlo
iterations) and thus less computing time according to RiskAMP (2016).
4.3.2 Container terminal calculation

For the container terminal design calculation all important equations and calculation methods are
described in Section 2.1.3. The following clarifications can however be made.
Cargo flows
All cargo flow percentages (import, export, transhipment, empties and CFS) are with respect to
the total throughput (containers over the quay).
Quay wall
When for the calculation of the quay wall costs the formula of de Gijt (2010) is used, the retaining
height H is calculated. This is done by summing up the water depth (assumed to be equal to the
required dredging depth, calculated by the tool) and a value for the freeboard resulting in Lupper.
To calculate H (H = Lupper + Llower) the sheet pile length under the sea bed Llower needs to be
determined. This is in this case done by defining the linear relation Lupper : Llower. In the tool this is
assumed to be 1 : sheet pile length factor. These dimensions are depicted in a simplified cross-
section of a sheet pile quay wall in Figure 61. The freeboard and sheet pile length factor are given
as a constant in Table 1 Additional parameters (Constants) of the data sheet in the tool.
118 THE TOOL

Figure 61: Schematic cross-section of sheet pile quay wall dimensions. Source: Own work
4.3.3 Dry bulk terminal calculation

The most important dry bulk terminal design equations and calculation methods are described in
Section 2.2.3. The remaining dry bulk calculations are treated in this subsection and are divided
into general aspects, quay or jetty and terminal area calculations.
4.3.3.1 General
 In order to limit the complexity of the tool the cargo type has to be selected. A choice can
be made between: coal, iron ore, coal and iron ore, one or two grain types. For each
selected cargo type a separate calculation is made. The results of these separate
calculations are displayed in the output section of the worksheet. The main terminal
dimensions are of course dependent on all cargo types and thus are a summation of the
individual results.
 One type of berth can be chosen for all cargo types; quay wall or jetty. In the tool this
basically only leads to differences in costs since the required number of berths is assumed
to be independent of the type of berth. For dry bulk terminals the considered jetty type is a
finger pier only; this to not further complicate the tool input and the realisation of the tool
for that matter.
 For the costs calculation of the loading, unloading, stacking and reclaiming equipment the
capacity is required of each machine. The tool however only uses the gross productivity as
input. The capacity is computed by assuming it is equal to the typical rated capacity and
using the factors presented in Section 2.2.2.1.

4.3.3.2 Quay/jetty calculation
 For a dry bulk quay wall the note in Section 4.3.2 also holds.
 For a terminal with jetties it must be selected if the jetties (finger piers) have berths on
both sides.
 Finger piers often have an additional length (Ladd) next to the required length for berthing
( ). This length has to be specified. For an overview of these lengths
see Figure 62.
 When jetties are selected the overall results include a summation of all the jetty lengths,
this is to give an indication. The individual jetty lengths and the required number of jetties
are listed per cargo type separately.
When a quay wall is selected the required quay lengths are summed up, taking into
account the additional lengths for mooring that are counted double.
 When in the input parameters the incoming cargo is not set to 100% of the total
throughput there is also an outgoing cargo flow. Outgoing cargo flow requires loading
equipment. The number of loading and unloading equipment is determined and the
number of required berths is determined separately. The quay lengths or number of jetties
corresponding to the summation of the number of required berths for both flow directions
are given as results in the output.
Figure 62: Finger pier type jetty with lengths overview. Porto de Tabarao, Brazil. Source: Google Earth
4.3.3.3 Terminal area calculation

 The storage utility type can be selected. For coal and/or iron ore a choice can be made
between wind-row and storage shed. The difference, in the tool, between wind-row and
storage shed is the price per square metre. For grains a choice can be made between
storage shed and silos, for these the calculations differ.
 The calculation of the required area for a group of silos is thoroughly described in Section
VI.1 of the appendices.
 The possible stacking and/or reclaiming equipment depends on the storage utility type. A
bucket-wheel stacker-reclaimer has both stacking and reclaiming capabilities, a unit price
is included. A scraper-reclaimer only has reclaiming capabilities. The stacking in storage
sheds is often done by conveyors mounted right under the roof of the shed. Only a unit
price for the reclaiming part is included.
 Conveyor belts are neither included in this study nor in the tool.
120 THE TOOL

4.3.4 Liquid bulk terminal calculation
The most important liquid bulk terminal design equations and calculation methods are described
in Section 2.3.3. The remaining liquid bulk calculations are treated in this subsection and are
divided into general aspects, jetty and terminal area calculations.
4.3.4.1 General
 In order to limit the complexity of the tool the cargo type has to be selected. A choice can
be made between: LNG, crude oil, one or two other liquids, crude oil and one or two other
liquids. For each selected cargo type a separate calculation is made. The results of these
separate calculations are displayed in the output section of the worksheet. The main
terminal dimensions are of course dependent on all cargo types and thus are a summation
of the individual results.
 The liquid bulk berth type is a jetty (island berth type).
4.3.4.2 Jetty calculation

 Island berth type jetties consist of a trestle and the berthing structure.
 The berthing structure length (Lberthing; centre-to-centre distance between outer mooring
dolphins) is determined by the length of the design vessel (Ls) plus an additional length for
the mooring lines for as well as aft of the vessel (Lml). The length of the trestle (Ltrestle) is
an input variable that is dependent on the local situation. For an overview of these lengths
see Figure 63.
 For the overall results a summation of all the jetty lengths is made, this is to give an
indication. Only the trestle lengths are summed up. The individual jetty lengths and the
required number of jetties are listed per cargo type separately.
 Pipelines are neither taken into account in this study nor in the tool.
Figure 63: Island berth type jetty with lengths overview. Korsakov, Sakhalin, Russia. Source: Google Earth
4.3.4.3 Terminal area calculation

 The calculation of the required area for one or multiple groups of storage tanks is
thoroughly described in Section VI.2 of the appendices.
 For certain cargo types safety distances between tanks or tank groups are recommended,
more information can be found in Section 2.3.3.2. These minimum distances overrule the
tank distance chosen by the user if applicable.
 Pipelines are neither taken into account in this study nor in the tool.

4.4 Tool verification
Models that describe physical processes such as river morphology or fluid dynamics require
calibration and validation to tweak certain parameters or model schematisations in order to make
sure they give reliable results. The tool -that can be seen as a model- does not describe physical
processes. It mainly uses capacity/productivity equations with values of parameters that are
provided by the user to calculate quantities such as required dimensions or certain units.
Calibration and validation are therefore not required. The tool however has to be checked to make
sure it works correctly. By this is meant that the interface, equations and coding logic have to be
tested. This is done by choosing realistic input, based on real terminals, and doing a manual
calculation of the dimensions and costs. This same completely deterministic input is used for the
tool and one iteration is performed resulting in a single solution per calculation output (e.g. quay
length, gross storage area, etc). When both the manual and tool outcomes are equal the tool is
verified.
For container terminals one input combination is calculated. For dry and liquid bulk terminals
three combinations each are calculated. This difference is made because for container terminals
each calculation uses the same equations and logic. For dry and liquid bulk terminals this depends
on the selected (combinations of) cargo type(s). Combinations can be for example ‘Coal & Iron
ore’ for dry bulk or ‘Crude oil’ for liquid bulk terminals. The verification is completed
successfully.
122 THE TOOL

5 CASE STUDY
The case study is an application of the tool on one existing marine terminal. The design rules,
guidelines and common values of parameters found in the literature and in the expert judgement
studies are therefore applied. The case study shows what the tool can be used for and at the same
time it is checked if it produces realistic results. In addition to this a sensitivity analysis is
performed for this specific case.
Goal of the case study: Application of the tool on a real scenario and the identification of items28
that greatly affect the construction costs.
5.1 Case description
For the case study the EMO terminal in Rotterdam, the Netherlands is selected. EMO (Europees
Massagoed Overslag BV) is the largest European dry bulk handling terminal. The terminal serves
the energy and steel sector by handling coal and iron ore. EMO itself was established in 1975 but
its roots date back to 1954. The terminal is situated at the Maasvlakte, a man-made extension of
the Port of Rotterdam, built in the sixties. The largest existing sea-going vessels are able to berth
at the terminal due to the large water depth. EMO distinguishes itself because of the high level of
automation. Currently the terminal has fully automated stacker-reclaimers, a sea-vessel loader, a
coal wagon loader and sea-vessel unloaders. These machines are operated from a central control
room. Information in this chapter originates from EMO (2016) unless indicated otherwise.
The EMO terminal is chosen for the case study since dry bulk terminals are underexposed in
literature, compared to container terminals. For liquid bulk terminals it is very difficult to find
sufficient information so this is only an option if the terminal management itself provides
information. The big advantage of EMO is that there is a lot of information available on the
website.
The terminal consists of a sea-going vessel quay and a separate barge quay. The sea-going vessel
unloaders are gantry grab cranes. The storage yard consists of multiple wind-rows. These rows
may or may not be divided by retaining walls to accommodate multiple material types in a single
row. The handling of cargo at the stockpiles happens by means of stacker-reclaimers. The
terminal has a railway connection providing a link to the vast hinterland. Services that are offered
on the terminal are: the blending of different grades, processing, washing (removal of impurities)
and screening (separation of cargo into different grades). Coal on the terminal can be compacted
in order to reduce the oxygen in between the coal to avoid spontaneous combustion. A power
plant is situated near the terminal, coal can directly be transported to the plant. An aerial picture of
the terminal is presented in Figure 64.
28
Items being variables and terminal elements.
123
Figure 64: Overview of the EMO dry bulk terminal in Rotterdam. Source: EMO (2016)
5.2 Actual terminal characteristics
The actual terminal properties are listed in Table 42. Some of these characteristics are based on
information provided by EMO (2016), others are determined using Google Earth.
Table 42: Actual EMO terminal characteristics
Terminal property Actual value Source

Total quay length [m] 1,793 EMO
Total terminal area [ha] 171.5 Google Earth
Total gross storage area [ha] 121.8 Google Earth
Max. vessel draught [m] 23.0 EMO
Number of berths [pcs] 6 EMO
Number of unloaders [pcs] 5 EMO
Number of loaders [pcs] 1 EMO
Number of stacker-reclaimers [pcs] 7 EMO
Avg. stockpile length [m] 1,200 Google Earth
An overview of the terminal is depicted in Figure 65, with in blue the total terminal area and in
red the gross storage area. The purple area represents a storage area in which the stockpiles are
placed in a less structured way, no stacker-reclaimer is present in this area. The area is most likely
used to store cargo which has to be transhipped to sea-going vessels. East of the terminal the
Engie power plant is located. The storage area north-east of the power plant belongs to the plant,
not to the terminal. The largest part of the southern quay wall is used to unload vessels. The quay
wall south of the purple area in Figure 65 is used to load sea-going vessels. The western quay wall
is used to load inland waterway vessels. The apron area of the western quay wall (including the
loading equipment) is not taken into account, this is elaborated on in Section 5.3.
124 CASE STUDY

Figure 65: Satellite image of the EMO terminal with gross storage areas; wind-row stockpiles (red) and less
structured stockpiles (purple) and the total terminal area (blue). Source: Google Earth
5.3 Input parameters
This section covers the tool input that is used to estimate the dimensions, quantities and costs of
the EMO terminal. For the variables for which no information could be found the standard values
of parameters of the literature study are used. The tool does not have the function to differentiate
between sea-going and inland waterway vessels and the corresponding berths and equipment. The
choice is therefore made to not include the barge quay -located on the Westside of the terminal-
while still accounting for the transhipped cargo volume. Due to this simplification the amount of
transhipped cargo is still used to compute the required storage area, only the loading equipment
and quay length are not considered. The transhipment to other sea-going vessels is taken into
account. Public information about the terminal and its operations is presented in Table 43. In the
following subsections the tool input is specified.
Table 43: Terminal and operations information of EMO. Source: EMO (2016)
Variable Value
Operational hours 7 days a week, 24 hours a day
Design throughput capacity 42 million tonnes
Actual throughput coal1 20 million tonnes (61%)
Actual throughput iron ore1 13 million tonnes (39%)
Actual transhipment coal1 56% (sea shipping 1%)
Actual transhipment iron ore1 56% (sea shipping 8%)
2
Design vessel - unloading berths VLBC with DWTmax = 400,000 tonne, Ls = 362 m
Design vessel - loading berths DWTmax = 150,000 tonne, Ls = 232 m
Number of cranes - unloading 5 for 4 berths
Number of cranes - loading 1 for 2 berths
Grab capacity unloading equipm. 2 x 50 tonne and 3 x 85 tonne
Capacity loading equipm. 3,000 tonne/hour
Stacker-reclaimer - stacking capacity 6,000 tonne/hour
Stacker-reclaimer - digging capacity 4,500 tonne/hour
Note 1: Data from 2015
Note 2: VLBC = Very Large Bulk Carrier
5.3 Input parameters 125

5.3.1 Throughput
To compute the required terminal dimensions and number of equipment, etc. the throughput
capacity has to be used instead of the actual achieved throughput. The ratio of achieved coal and
iron ore throughput to the total throughput of 2015 is used to approximate the design coal and iron
ore throughputs. This leads to the design throughputs of 25.5 million tonnes coal and 16.5 million
tonnes iron ore. The terminal does not export cargo so the amount of outgoing cargo is equal to
the transhipped cargo. Therefore the incoming cargo is 100% minus the transhipment percentages
to sea-going vessels, from Table 43.
5.3.2 Vessel dimensions

According to the information from EMO the unloading and loading berths have different design
vessels. To compute the required quay length the tool uses the average vessel length ( ); if the
required number of berths is larger than one, see Equation (7). The average vessel length is taken
from typical vessels that moor at the unloading and loading quays. Typical lengths of vessels that
are unloaded are derived from Google Earth satellite images of 20-12-2016. The typical length of
vessels at the loading quay (181 metre) is determined by using the inverse of Equation (7), with a
quay length of 443 metres and 2 berths. The average is 250 metres, using an assumed uncertainty
interval of ±10% results in the uniform distribution UNI[225; 275]. To compute the required
water depth the draught of the VLBC is taken, this is 23 metres.
5.3.3 Estimated berth occupancy

Since in principle two different quays are taken into account the estimated berth occupancy is
determined accordingly. The actual value will be somewhere in between the values corresponding
to two and four berths, in Table 23. This leads to the distribution UNI[65; 80].
5.3.4 Gross productivities

EMO provides maximum capacities for their loading cranes and stacker-reclaimers, but provides
no capacity for their unloading cranes. The unloading capacity is estimated by using public
information from Ertsoverslagbedrijf Europoort C.V. (EECV) (2013) about their unloading
capacities. The EECV cranes with 60 tonne grabs have a capacity of 2300 tonne/hour, their 65
tonne cranes have a capacity of 2600 tonne/hour. Using the average of the EMO grab sizes and
linear extrapolation results in an estimated maximum capacity of 2960 tonne/hour of the EMO
unloading equipment.
Instead of maximum capacities the tool requires gross productivities 29 for its computations. In
Section 2.2.2.1 factors are used to determine the gross productivity from the typical rated
capacity. The typical rated capacity is in theory not equal to the maximum capacity but they do
not differ by much and no other information is available. The through-ship efficiency factor (for
unloading 0.5 and for loading 0.7) and operational availability factor (0.8) are used. These factors
are in theory only for (un)loading equipment, since no other factors could be found in literature
they are also used for the storage yard equipment. (Un)Loading cranes and storage yard
equipment are connected via belt conveyors, so if a crane is not working the stacker reclaimer is
not working either. However cargo is sometimes repositioned by the stacker-reclaimers, which are
then in-use while the cranes are not. On the other hand small downtimes can occur for the yard
29
Gross productivity: Average amount of cargo moved between berthing completed and de-berthing started.
This variable therefore includes unproductive intervals such as crane repositioning, moving hatches to/from quay
and time between shifts.
126 CASE STUDY

equipment while the cranes are working fine. This assumption can therefore be considered as
reasonable. The gross productivity of an unloader is 1,185 tonne/hour, of a loader 1,680
tonne/hour, stacking of a stacker-reclaimer 2,400 tonne/hour and reclaiming of the same machine
1,800 tonne/hour. For all gross productivities an uncertainty interval of ±25% is assumed.
5.3.5 Average dwell time

The average dwell time of cargo on the terminal is unknown. It can however be approximated by
using the capacity ratios that are presented in Section 2.2.3.2. The capacity ratio indicates the
amount of cargo that has to be able to be stored at a given moment of time. So if a steady annual
throughput is assumed the capacity ratio percentage of the number of days in a year is an estimate
of the average dwell time. Various ratios are proposed, they can be approximated by the triangular
uncertainty distribution TRI[5; 16; 22]. The limits are not used for the dwell time since they
probably are exceptions, therefore the mean value of 14.3% is used. This results in an average
dwell time of 52 days with an assumed uncertainty interval of ±50%; to capture the large
uncertainty of the dwell time. The resulting uniform distribution is UNI[26; 78].
5.3.6 Total terminal factor and estimated storage occupancy

Since common values do not occur in literature the distributions of the total terminal factor and
estimated storage occupancy have been determined by Expert Judgement Elicitation (see Section
0), they are depicted in Figure 66.
Figure 66: Expert Judgement Elicitation results that are used for the EMO case. Source: Own work
5.3.7 Stockpile properties

The horizontal stockpile lane dimensions (average lane length, width and additional width of the
area surrounding the stockpiles) are estimated using Google Earth. The design pile height for a
coal stockpile should be close to the maximum of 23 metre (according to the standard values of
parameters). An assumption of a uniform distribution in between 21 and 23 metre is made.
According to data from Actueel Hoogtebestand Nederland (2016) iron ore stockpiles are less
high; a maximum of 19 metres is found. A somewhat larger uncertainty of the iron ore pile height
is used ranging from 15 to 19 metres. Google Earth is used to estimate the number of lanes for the
coal and iron ore stockpiles. The iron ore storage consists of approximately three lanes. The main
terminal area counts about four lanes for coal. The northeast area is approximated to be 1.25 of an
average lane. The area used for storage of sea-going vessel transhipment cargo is estimated to be
5.3 Input parameters 127

0.75 of an average lane (purple area in Figure 65). The derived average lane length is 1200
metres.
The tool computations do not include the conveyor belts (with a total length of 47 km) and the
structures required for the additional services. However an approximation of the required area for
the additional structures is made by using the total terminal factor.
5.3.8 Summary of the input parameters

The options that are selected in the tool are listed in Table 44.
Table 44: Selected options in the tool
Item Selection
General
Berth type Quay wall
Wave conditions Sheltered waters
Coal & Iron ore
Unloading equipment Gantry grab crane
Yard equipment Bucket wheel stacker-reclaimer
Storage type Open storage: Wind-row
The probability distributions that are used as input for the computations of the case study are
summarised in Table 45. If the values for both coal and iron ore are the same one value is listed in
the middle. For the variables additional distance stockpile30, freeboard quay and sheet pile length
factor it is not possible in the tool to specify an uncertainty distribution, the values are therefore
deterministic.
Table 45: Parameter values used as input for the computations
Probability distributions
Variable Unit Source
Coal Iron ore
Operational hours per year DET[8,760] hour EMO
Total terminal factor See Figure 66 % Section 5.3.6
Draught design vessel DET[23] metre EMO
Add. quay length mooring UNI[15; 30] metre Ch. 2
Dredging tolerance DET[0.60] metre Ch. 2
Throughput DET[25.5x106] DET[16.5x106] tonne/year Section 5.3.1

Incoming cargo DET[99] DET[92] % Section 5.3.1
Transhipment DET[1] DET[8] % Section 5.3.1
Average cargo density UNI[0.52; 0 .93] UNI[2.13; 3.03] tonne/m3 Ch. 2
Average vessel length UNI[225; 275] metre Section 5.3.2
Design vessel length DET[362] metre EMO
DWT design vessel DET[400,000] tonne EMO
Avg. gross prod. unloader UNI[890; 1,480] tonne/hour Section 5.3.4
Avg. gross prod. loader UNI[1,260; 2,100] tonne/hour Section 5.3.4
Est. berth occupancy UNI[65; 80] % Section 5.3.3
Average dwell time UNI[26; 78] 1 days Section 5.3.5
Est. storage occupancy See Figure 66 % Section 5.3.6
30
The width of the area surrounding the net storage area; so including internal infrastructure.
128 CASE STUDY

Probability distributions
Variable Unit Source
Coal Iron ore
Avg. gross prod. stacking UNI[1,800; 3,000] tonne/hour Section 5.3.4
Avg. gross prod. reclaiming UNI[1,350; 2,250] tonne/hour Section 5.3.4
Pile height UNI[21; 23] UNI[15; 19] metre Section 5.3.7
Lane width UNI[78; 93] metre Section 5.3.7
Number of lanes DET[6] DET[3] - Section 5.3.7
Angle of repose UNI[30; 45] UNI[30; 50] Degrees Ch. 2
Add. distance stockpile 15 metre Section 5.3.7
Storage factor UNI[15; 25] UNI[30; 40] tonne/year/m2 Ch. 2
Capacity ratio TRI[5; 16; 22] % Ch. 2
Freeboard quay 3 metre Ch. 2
Sheet pile length factor 1 - Ch. 2
Note 1: In Section 5.4.2.2 the uncertainty distribution is changed to UNI[45; 80]
Other notes: DET = deterministic, UNI = uniform, TRI = triangular distributions
Computation properties
The number of iterations that is used for the Monte Carlo simulations is 1000. 31 The tool uses
Latin Hypercube sampling (as is explained in Section 4.3.1) which allows for less iterations for
the same accuracy compared to the standard Monte Carlo sampling. The prices are corrected for
the 2016 price level with an assumed annual average inflation of 2%.
5.4 Tool application
A computation is performed to estimate the EMO terminal dimensions and quantities. The
values/distributions that are used as input are specified in Section 5.3. This section presents the
computation results. These results are compared to the actual dimensions (see Section 5.2).
Finally conclusions are drawn and the resulting construction costs estimate is presented.
5.4.1 Computation results

The main results are presented by means of probability density plots. These plots are in other
words relative frequency plots. The frequency that a result falls in a certain range (or bin) is
determined. Each value on the horizontal axis represents the upper boundary of a bin, the previous
value represents the lower boundary of the same bin. The lower boundary of the very first bin is
not shown, yet the first bin has the same bin size as the other bins and includes the smallest
outcomes.
The main results concerning the quay are presented in Figure 67. The left plot represents the
required quay length for coal and iron ore combined. The 65 th percentile is 1,797 metre and best
represents the actual value of 1,793 metre. The right plot represents the combined number of
required berths. The actual value is 6, the best representation by the tool is the 71 st percentile.
31
Dr. Morales-Nápoles stated in personal communication that 1000 iterations for this application is reliable.
5.4 Tool application 129

Figure 67: Calculation results; total quay length and total number of berths. Source: Own work
The total terminal area results are depicted on the left in Figure 68. The actual area is about 171.5
hectare. This is best approximated by the 85th percentile with 171.4 hectare. The required
dredging depth is dependent on the wave climate, design vessel draught and the dredging
tolerance. Since the variables are chosen to be deterministic the computation presents one result; a
required depth of 25.90 metre. This depth is needed to allow a vessel with a 23 metre draught to
access. No actual value of the dredging depth or water depth is available.
Figure 68: Calculation results; total terminal area and dredging depth. Source: Own work
5.4.2 Analysis of the results

From the comparison between the computation results and the actual terminal properties in
Section 5.4.1 it can be noted that the best-estimate quantiles are in between 65% and 85%.
According to Wolfert (2014) the 70% quantile is generally used for budgeting purposes. The
percentiles of the best-estimates of the quay length and number of berths are very close to the
proposed 70% quantile. The best-estimate of the storage area deviates a bit from the proposed
quantile, but is not per definition incorrect. Possible reasons for the actual area being larger than
the area estimate of the proposed percentile can be:
 The storage yard is dimensioned in a risk aversive way. In other words more area -than
theoretically is required for the design capacity- is allocated to function as a buffer.
 The sea-going vessel transhipment storage yard (purple area in Figure 65) is not used for
the typical storage of cargo. This since no stacker-reclaimers are used and the handling of
cargo on the yard happens with bulldozers. The yard may be used for short or very long
storage or functions as an intermediate place to store cargo that has to be transhipped. In
130 CASE STUDY

this case the tool predicts a smaller required area for the same throughput because it does
not consider a buffer area.
 Estimated input values are incorrect. Variables that are uncertain in this case study (see
Table 45) and that affect the storage area are: cargo dwell time, stockpile properties
(height, lane width, number of lanes), storage occupancy, cargo density, angle of repose of
cargo and the total terminal factor.
5.4.2.1 Variables of importance

This subsection takes a closer look at the influence of the mentioned variables on the total
terminal area. The stockpile height, lane width, number of lanes, cargo density and angle of
repose are represented by uncertainty distributions. These distributions are however based on
best-estimates of information from Google Earth or trusted literature and therefore cannot really
be improved. Consequently, this section focuses on the cargo dwell time, storage occupancy and
the total terminal factor. The influence on the total terminal area is investigated by plotting the
values of these variables against the related total terminal area for each iteration; in Figure 69 each
point represents an iteration. It has to be noted that these input variables are only for the coal
calculation.32 However the same input distributions are used for both cargo types thus they give
comparable scatter plots. The variable with the highest influence can be identified by the scatter
plot that has a linear fitted curve that is the closest to a horizontal line and has a good goodness of
fit33. As an increase or decrease of the random input variable on the y-axis then also corresponds
to a large increase or decrease of the total terminal area. It can be observed that varying the
average cargo dwell time has the largest effect on the total area. The slope of the curve is similar
to that of the total terminal factor scatter plot, however the goodness of fit (quantified with the
coefficient of determination R2) of the linear curve for the dwell time is higher. The estimated
storage occupancy scatter plot has a linear curve that is the most horizontal but it has a large error
(low goodness of fit). From the scatter plot it can be observed that varying the storage occupancy
does not per se lead to a larger total terminal area.
32
The presented total terminal area is the sum of the total terminal areas of the separate coal and iron ore
calculations.
33
Goodness of fit is a measure of how good a curve fits the data. For quantification the coefficient of
determination R2 is used in this application; ranging from 0 to 1 where 1 is a perfect fit.

Figure 69: Scatter plots of three different input variables and the corresponding total terminal area. Linear
curves are fitted; their equation and coefficient of determination are depicted as well. Source: Own work
Based on the analysis of the linear relationships between the random input variables and the
resulting total terminal areas a conclusion can be made. This conclusion is that varying the
average dwell time and the total terminal factor can impact the total terminal area. Another way of
assessing dependence between random variables is by means of copulas (see Section III.2.1 of the
appendices). Due to reasons that are given in the introduction copulas are not applied here.
5.4.2.2 Applying different values

Changing the original average dwell time (in days) from UNI[26; 78] to an assumed UNI[45; 80]
distribution results in the actual terminal area (171.5 ha) being best approximated by the 75 th
132 CASE STUDY

percentile (171.4 ha); see Figure 70. Changing the original total terminal factor (in %) -which is
determined by experts- to an assumed UNI[45; 70] distribution also results in a best
approximation by the 75th percentile (171.6 ha). These resulting quantiles better relate to the
proposed 70% quantile.
Of the three reasons -for the total terminal area not being estimated well enough- the risk aversion
and non-typical storage yard reasons are discarded since they cannot be proven. Of the remaining
possible incorrectly estimated variables the total terminal factor is discarded. This, since the dwell
time is more likely to differ from the estimate than the terminal factor, of which the latter has been
estimated by experts. Therefore, for the remainder of this case study, in the input parameters, the
average dwell time is changed to its new estimated uniform distribution UNI[45; 80]. This
distribution is used both for coal and iron ore.
Figure 70: Total terminal area distribution resulting from a computation with an updated average dwell time
distribution. Source: Own work
5.4.2.3 Direct costs analysis

So far the main terminal dimensions have been treated, the direct costs 34 are considered in this
subsection. Direct costs are determined by multiplying the computed dimensions and quantities by
the unit costs. Since these quantities are represented by probability density distributions the direct
costs have the same form. The costs quantile is determined by taking the average of the quantiles
of the main tool results (now ranging between 65% and 75%) as specified in Sections 5.4.1 and
5.4.2.2; this results in the 70th percentile. This value is exactly the value that is proposed by
Wolfert (2014). The main cost elements for a dry bulk terminal are:
o Quay wall: Costs including the supply of material and the required man-hours.
Determined by the empirical equation of de Gijt.
o (Un)Loading equipment: Investment costs of the loading and unloading
equipment, excluding rails. The foundation is not directly taken into account,
however a quay wall partly or fully bears the weight of the equipment. The
foundation is partly included in the unit costs of the quay wall; since an empirical
equation is used to derive the quay wall costs.
o Storage facilities: Costs including the supply and construction of terminal
pavement, which is assumed to be used for stockpiles. This includes the stabilising
sand layer underneath the pavement.
o Storage equipment: Investment costs of -in this case- the stacker-reclaimers,
excluding rails and foundation.
34
Direct costs are defined as costs directly related to the production or supply of a product or service.
Construction cost estimates consists of direct costs, indirect costs and contingencies.

Terminal elements that are not included in the costs are belt conveyors, roads and buildings. For
the EMO terminal with input described in Section 5.3 and with the updated average dwell time
distribution the probability density distributions of the different main cost items are depicted in
Figure 71. The lines of the quay wall and storage equipment seem to be oscillating, this is
however due to the fact that the lines are based on histogram data with a limited resolution. The
differences are in the order of one percentage.
Figure 71: Probability density distributions of the main direct cost elements. Source: Own work
Aspects that immediately stand out are the narrow distributions with relatively low costs of the
(un)loading and storage yard equipment, the similar width of the quay wall and storage facilities
distributions and the quay wall being by far the highest contributor to the total direct costs. Both
the distributions of (un)loading and storage equipment are narrow. This is due to the fact that the
distribution widths of the computed number of equipment are relatively small; the result with the
largest width is depicted in Figure 72 (left). Another reason are the relatively small widths of the
investment costs distributions. The largest costs range corresponds to stacker-reclaimer
investment costs ranging between 4.45 and 9.16 million euro per piece, depending on the
machine’s capacity. The similar width of the quay wall and storage facility distributions is merely
a coincidence since the only variable that they have in common is the throughput, they are
dependent on many more different variables and different unit costs. The storage facilities
distribution has a relatively long tail, this is due to a similar tail in the gross storage area
distribution as depicted in Figure 72 (right).
134 CASE STUDY

Figure 72: Calculation results; number of unloading equipment (left) and the gross storage area (left). Source:
Own work
5.4.3 Conclusions & construction costs estimate

From Sections 5.4.1 and 5.4.2.2 it can be concluded that the best-estimate quantiles are in between
65% and 75%. As already is explained the costs of the terminal are estimated by taking the
average of the best-estimate quantiles; the 70% quantile. According to Wolfert (2014) the same
70% quantile is generally used for budgeting purposes. This can lead to the following two
conclusions:
1. The tool gives realistic design estimates for (coal and iron ore) dry bulk terminals when
realistic assumptions for the values of the parameters are made.
2. The terminal itself was well designed.
The first conclusion can only be made when the second conclusion is true. In this case the fair
assumption is made that Europe’s largest dry bulk terminal is well designed. Since the tool gives
realistic estimates of the terminal dimensions, other quantities and costs it can be concluded that it
functions correctly in this case.
For the application of the tool in general the following recommendation for choosing the values of
parameters must be made. It is advised to not use large uncertainties for variables since these
variables will then dominate the resulting dimensions and costs. As a rule-of-thumb for uniform
and triangular distributions the upper limit should not be more than twice the lower limit. Based
on the current case this gives credible results.
Construction cost estimate
As can be observed from Figure 71 the quay wall has by far the largest contribution to the costs,
followed by the storage facilities, the storage equipment and the (un)loading equipment. The
construction costs -according to the SSK method- consist of the total of the direct costs, indirect
costs and contingencies. To determine the latter two cost aspects certain percentages (see the
confidential Appendix I ) of the total direct costs are used. An overview of the costs is presented
in Table 46.

Table 46: EMO case direct costs and construction costs corresponding to the 70% quantile
Direct cost elements

Quay wall € 91,935,000
(Un)Loading equipment € 30,270,000
Storage facilities € 52,275,000
Storage equipment € 36,610,000 +
Total direct costs € 211,090,000
Total construction costs € 358,853,000
The mentioned costs -corresponding to the 70% quantiles of the direct cost elements- are
visualised in Figure 73; together amounting to the total direct costs.
Figure 73: The contribution of the 70th percentiles of the cost elements to the total direct costs. Source: Own
work
5.5 Sensitivity analysis
The goal of the sensitivity analysis is to identify the variables that have a large impact on the total
costs. In a terminal design project costs are an important factor. In order to be able to decrease the
costs terminals can be optimised. It is not practical to optimise many terminal elements so the
knowledge of what element influences the costs the most is valuable. The sensitivity analysis is
performed on the EMO case.
5.5.1 Comparing distributions

Since the cost outputs are probability distributions of different types, results cannot be compared
in a straightforward way. A method to determine the degree to which two densities disagree is
called the Kullback-Leibler divergence, as presented in Kullback (1997). This method computes
the so called Entropy quantity (I) between two continuous probabilistic distributions (f1 and f2).
The Entropy is defined as:
(30)
Bluntly stated Entropy quantifies the amount of information that is lost when f2 is used to
approximate f1. Entropy is in fact already used in this research, namely in Equations (31) and (32)
136 CASE STUDY

of the appendices. The Entropy is not an absolute score but a relative measure, with a different
scale for every application. For this sensitivity analysis the quantity is calculated between a base
case (with input values as described in Section 5.3.8) and a case in which the distribution of a
single variable is changed. When multiple cases -in which the distributions are changed- are
considered, the effect of that variable on the costs result can be quantified. In order to put things
into perspective the Entropy is also calculated for the tool’s total direct costs output with exactly
the same input as for the before mentioned base case but with various number of iterations. The
results are listed in Table 47. Here for f1 the density function of the total direct costs for i = 1000
is used. For f2 also the total direct costs are used, which is however computed using a different
number of iterations. A quantity of 0 resembles two perfectly matching distributions. Of course
the lower the number of iterations the more information is lost with respect to a thousand
iterations, and so the larger the Entropy will be.
Table 47: Entropy quantities of the total direct costs for constant variables but a varying number of iterations
Number of Entropy I
iterations [-]
1000 vs. 50 56.1260
1000 vs. 100 13.4339
1000 vs. 250 10.4004
1000 vs. 500 0.1467
1000 vs. 1000 0
To give more insight into these values the differences between the probability distributions of the
total direct costs of a thousand iterations and respectively 500, 250 and 50 iterations are depicted
in Figure 74.
5.5 Sensitivity analysis 137

Figure 74: Probability density distributions of the total direct costs for various number of iterations. Source:
Own work
5.5.2 Variables of interest

The cost elements that are computed in the tool are the quay wall, (un)loading equipment, storage
facilities and storage equipment costs. The costs are determined by multiplying the specific output
dimension or quantity with a unit cost. In this analysis the unit cost distributions are constants, the
distributions of the variables are varied. Not all variables are taken into account due to reasons
that are given in the introduction of this subsection. The variables that are provided by EMO are
taken as constants. The values of variables that are uncertain and are therefore represented by a
distribution are used for the sensitivity analysis. Per cost element the influencing variables that are
considered for the sensitivity analysis are listed:
1. Quay wall costs:

a) Average vessel length
b) Estimated berth occupancy
c) Gross productivity per loading equipment
d) Gross productivity per unloading equipment
2. (Un)Loading equipment costs:
a) Estimated berth occupancy
b) Gross productivity per loading equipment
c) Gross productivity per unloading equipment
138 CASE STUDY

3. Storage facilities costs:
a) Cargo dwell time
b) Estimated storage occupancy
c) Stockpile height
d) Stockpile lane width
4. Storage equipment costs:
a) Gross productivity per stacking equipment
b) Gross productivity per reclaiming equipment
For the analysis the distribution intervals of the variables are made smaller, thus smaller
uncertainties are created. The boundaries are moved inwards a certain percentage (5%, 10% and
15% per boundary per step) of the base case interval length. The resulting probability distributions
of the total direct costs f2 are compared to the distribution for the base case f1, in agreement with
Equation (30). For this analysis all variables are assumed to have uniform distributions since
narrowing a triangular distribution can also shift the mean value. This can have an extra impact on
the resulting total costs distribution which makes the comparison of the Entropies less fair. The
original distributions from Section 5.3.8 are used, in Table 48 the lower and upper limits of the
base case uniform distributions and the narrowed uniform distributions are listed.
Table 48: The analysed variables with the base case and narrowed distributions per step. The lower and upper
limits of the uniform distributions are presented.
Step 1 Step 2 Step 3

Variable Base case
10% smaller 20% smaller 30% smaller
Average vessel length
225.0 - 275.0 227.5 - 272.5 230.0 - 270.0 232.5 - 267.5
[m]
Estimated berth
65.00 - 80.00 65.75 - 79.25 66.50 - 78.50 67.25 - 77.75
occupancy [%]
Gross prod. loading
1,260 - 2,100 1,302 - 2,058 1,344 - 2,016 1,386 - 1,974
equip. [tonne/hour]
Gross prod. unloading
890.0 - 1,480.0 919.5 - 1,450.5 949.0 - 1,421.0 978.5 - 1,391.5
equip. [tonne/hour]
Cargo dwell time [days] 45.00 - 80.00 46.75 - 78.25 48.50 - 76.50 50.25 - 74.75
Estimated storage
60.00 - 85.00 61.25 - 83.75 62.50 - 82.50 63.75 - 81.25
occupancy [%]
Coal 21.0 - 23.0 21.1 - 22.9 21.2 - 22.8 21.3 - 22.7
Stockpile
height [m]
Iron ore 15.0 - 19.0 15.2 - 18.8 15.4 - 18.6 15.6 - 18.4
Stockpile lane width [m] 78.00 - 93.00 78.75 - 92.25 79.50 - 91.50 80.25 - 90.75
Gross prod. stacking
1,800 - 3,000 1,860 - 2,940 1,920 - 2,880 1,980 - 2,820
equip. [tonne/hour]
Gross prod. reclaiming
1,350 - 2,250 1,395 - 2,205 1,440 - 2,160 1,485 - 2,115
equip. [tonne/hour]

5.5.3 Analysis of the results
The sensitivity analysis does not test if the total costs increase or decrease given a certain change
in the input but it tests how much the costs distribution changes for each adjustment to the input.
These changes in the costs distribution are quantified by the Entropy quantity. Section 5.5.1
provides a reference for the Entropy quantities.
The Entropy is calculated per variable between the base case distribution and distributions of the
different narrowing steps (10%, 20% and 30%). These quantities are presented in Table 49. To
determine if a variable has a small, medium or large effect on the costs its Entropy quantities are
compared to the Entropies of all variables. For the ratio of a variable’s Entropies to the total the
following three ranges -of equal size- are considered, in which 0.057 is the minimum occurring
ratio and 0.167 the maximum. The overall effects are listed in Table 49 as well.
 Small effect when 0.057 ≤ Qi < 0.093

 Medium effect when 0.093 ≤ Qi < 0.130
 Large effect when 0.130 ≤ Qi ≤ 0.167
Where,
Where,
Qi The ratio of a variable’s Entropies to the total.
i The considered variable in the set, with i = 1, 2,..., 9, 10.
j The considered narrowing step, with j = 1, 2, 3.
By using these criteria it can be concluded that the average vessel length has the largest impact on
the total costs. In other words a change of the average vessel length input results in a relatively
large change of the costs. This can be explained by the fact that there is a linear relation between
the vessel length and the quay length -Equation (7)- and because the quay wall costs by far have
the highest contribution to the total costs; see Figure 71. The effect of the gross loading
productivity is small, this can be explained by the small number of loading equipment and
therefore the limited contribution to the total costs. The gross unloading productivity has a
medium effect, compared to the loading equipment this is due to the larger required number of
equipment. The cargo dwell time appears to have a small effect on the total costs. The estimated
storage and berth occupancies both have a medium effect. This is somewhat unexpected since for
the application of the Expert Judgement Elicitation it was expected that the storage occupancy is
an important factor, this of course does not influence the EJE results. The height of the stockpiles
has a large influence on the total costs. If the height decreases the cross-section of a stockpile lane
also decreases resulting in a larger required lane length, this in turn increases the required storage
area and thus the storage facility (or in this case pavement) costs. The stockpile lane width also
determines the stockpile cross-section. Given the input values all possible occurring cross-sections
are trapezoidal and not triangular. In this case the cross-section area is quadratically dependent on
the height and linearly dependent on the width. The width therefore has a smaller effect. Both the
gross stacking and reclaiming productivities have a small effect. Even though the storage
equipment costs (thus the total stacker-reclaimer costs) are higher than the (un)loading equipment
costs in Figure 71; especially the gross unloading productivity can have a larger effect since it also
influences the required number of berths and thus the quay wall costs.
To put things into perspective; the effect of the average vessel length and stockpile height on the
total costs can be compared to the effect on the costs distribution when the number of iterations is
140 CASE STUDY

changed from 1000 to somewhere in between 250 and 500, according to the results from Section
5.5.1.
Table 49: Entropy quantities between the total direct costs distribution of the base case and the total direct costs
distributions per step
Entropy I [-] Overall

Variable Base case vs. Base case vs. Base case vs. effect on the
10% smaller 20% smaller 30% smaller total costs
Average vessel length 3.2052 4.8603 3.8454 Large
Estimated berth
1.1763 1.8447 5.2367 Medium
occupancy
Gross prod. loading
0.5369 3.0130 0.4930 Small
equip.
Gross prod. unloading
6.4252 0.5055 0.0517 Medium
equip.
Cargo dwell time 2.9465 1.6049 0.5127 Small
Estimated storage
3.4070 0.9424 2.5306 Medium
occupancy
Stockpile height 2.9572 0.5613 5.8776 Large
Stockpile lane width 0.0729 4.5865 1.6286 Small

Gross prod. stacking
0.5182 2.7433 3.1863 Small
equip.
Gross prod. reclaiming
1.8455 3.0149 1.1688 Small
equip.
It is expected that the smaller the widths of the uncertainty distributions become, the larger the
Entropy quantities are. From the results this cannot be concluded since for most variables the
Entropies are not increasing over all three narrowing steps. A cause could be that a too small
number of bins is used for the histograms of the total cost distributions. However, after increasing
this number from 50 to 100 the conclusion does not change. Another reason could be that the used
number of iterations (1000) is too small. Computing the Entropy between the total cost
distributions, resulting from two runs with exactly the same input, results in a very low Entropy (I
= 0.075). This means that multiple runs give similar results and the number of iterations is
therefore sufficient.
5.5.4 Conclusions
Variables that affect the direct cost elements and that are uncertain in the EMO case study are
considered in the sensitivity analysis. The intervals between the minima and maxima of these
uniform distributions are made smaller in steps of 10%, 20% and 30%. The total direct costs
distributions resulting from the computations with the adjusted distributions are compared to the
costs distribution corresponding to the base case. They are compared using Entropy. According to
the analysis the average vessel length and the stockpile height have the most impact on the total
direct costs. The typical differences in the cost distributions due to varying the input can be
compared to the difference that occurs when a computation is run with a number of iterations in
between 250 and 500, compared to a 1000. Due to the large effect on the costs distribution,

optimising these two variables can lead to relatively large costs savings. The variables that have a
mediocre effect on the total costs are the estimated berth occupancy, gross productivity of
unloading equipment and the estimated storage occupancy. Variables with little effect are the
gross productivity of loading equipment, cargo dwell time, stockpile lane width and the gross
stacking and reclaiming productivities. It must be stressed that the considered variables are
variables of which the values are uncertain. It can be expected that when more variables are
included in the analysis the results will be different.
The conclusions that are drawn here are only applicable to this case; the EMO terminal. When
some constant variables would differ considerably their influence on the total costs would too; this
means that the impact of other variables also changes. As an example the gross productivity of
loading equipment can be taken. When the percentage of outgoing cargo would be much larger
(than the current 1% of the throughput) more loading equipment would be required. That number
of course also depends on the productivity of the machines and dry bulk (un)loading equipment is
quite expensive. Therefore the total costs, loading productivity and outgoing cargo flow are
interrelated. Despite this the insights from this case study may well be of help in other design
projects.
142 CASE STUDY

6 CONCLUSIONS AND
RECOMMENDATIONS
The main research objective of this study is:
The development and application of a tool for computing the main required marine terminal
dimensions and corresponding construction costs with a limited available amount of data and
including uncertainties for the variables and costs.
In pursuance of this objective multiple research questions were posed. These questions are
considered in the various chapters of this report. In the conclusions the main answers to the
research questions are presented. Parts of the research that may require further study are specified
in the recommendations.
6.1 Conclusions
The very first research question was: What are the main terminal elements and the existing
terminal design guidelines and design rules?
Of all the essential terminal elements background information is provided. Moreover many design
guidelines and rules for container, dry and liquid bulk terminals were collected. Only for the
estimation of the required area for dry bulk silo and liquid bulk tank groups a calculation method
was developed by the writer. Since so many rules have been collected, they are not listed here and
reference is made to Chapter 2. The main design rules and guidelines -for dimensioning the
terminal elements- are featured in an overview per terminal type. The overviews have been made
clear and complete with the intent that they can aid terminal and/or port designers.
The second research questions was: What are common values for the design rule parameters and
what are common unit costs of terminal elements?
Ultimately for all variables common values have been determined. As well as unit costs for all
terminal elements. Again, since there are so many variables and cost items, reference is made to
Chapter 2. In most cases the common values of variables are based on literature. The consulted
literature does not provide common values for all parameters, examples are project specific
parameters such as the throughput. For some other variables, that concern stockpile, silo or tank
dimensions, existing terminals were investigated. Realistic common values could therefore be
derived. For two important parameters a special method has been used to determine common
values, this is elaborated on in the next research question. The first of the concerned variables is
the total terminal factor, which is the percentage of the area required for storage with respect to
the total terminal area. The second variable is the average storage occupancy, which is the
percentage of the design capacity of the storage area that is actualy used. Unit costs can hardly be
found in literature since construction companies and manufacturers are not very generous in
supplying this kind of information. The considered unit costs are therefore predominantly based
on information provided by Witteveen+Bos.
143
The third research question was: What are the uncertainty distributions of the total terminal factor
and the average storage occupancy factor and what are the correlations between average import,
export and transhipment container dwell times?
To be able to answer these questions the opinions of experts were scientifically combined by
means of Expert Judgement Elicitation on respectively uncertainty and dependence. First four
experts have been assessed by Cooke’s Classical model in order to combine their uncertainty
about common values of the total terminal factor and the storage occupancy of the three terminal
types. The resulting cumulative probability distributions are depicted in Figure 75.
Figure 75: Cumulative probability distributions of DMit_op for the three terminal types
The experts were weighted based on their estimates of a sufficient number of seed variables.
Various weighting schemes -or Decision Makers (DM’s)- were considered. The item optimised
DM gave the best scores. Its resulting probability distributions -consisting of combinations of the
distributions of the four experts- were used.
To answer the second part of the research question; the same four experts have been assessed
about their estimates of the correlations between container dwell times. For this the conditional
probability technique was applied. From the combination of the experts’ opinions it can be
concluded that import, export and transhipment container dwell times are related. It can be
observed that there are moderate positive relationships between import & transhipment and export
& transhipment container dwell times. As well as a weakly positive relationship between import
& export container dwell times. The rank correlations of these dependencies are depicted in a
Bayesian Network (BN) in Figure 76.
Figure 76: BN with rank correlations between the average container dwell times
The experts were again weighted based on their estimates of an unconditional dependency
between two port-related variables, in other words the seed question. Two experts exactly
estimated the actual rank correlation of the seed question. The global weights weighting scheme
was used since it produced the highest d-calibration score of practically 1.0. This is due to the fact
144 CONCLUSIONS AND RECOMMENDATIONS

that the method discarded the estimates of two -slightly lower scoring- experts and gave the two
selected experts the individual weights of 0.5. A measure to avoid this is presented in the
recommendations. Although the estimates of two experts were not used, the final result of this
case is a good representation of the individual estimates.
The answers to the previous three research questions provided the information required to be able
to realise the tool. The tool satisfies the requirements as stated in the main research objective. It
has been positively received by the ‘client’ Witteveen+Bos and the Port of Rotterdam has also
shown interest in it. The writer therefore is confident it will be used in practice.
The fourth research question was: What terminal elements contribute the most to, and what input
parameters have the largest influence on, the total construction costs of one specific terminal?
The considered terminal is Europe’s largest dry bulk terminal; the EMO (Europees Massagoed
Overslag) terminal at the Maasvlakte in Rotterdam. The terminal handles coal and iron ore. It can
be concluded that for this specific case the quay wall has by far the highest contribution to the
total costs, followed by the storage facilities. This can be observed in Figure 77, which depicts the
contribution of the main terminal element costs 35 to the total construction costs, being €
358,853,000.
Figure 77: The contribution of the 70th percentiles of the cost elements to the total costs
In reality not the complete stockpile area is paved. Moreover the quay wall section used for ship-
loading has a smaller retaining height than the section for ship-unloading, while in this case the
latter is assumed for the entire quay for sea-going vessels. The estimated costs are simplified
model outputs and will therefore vary from the actual costs. On average the actual main terminal
dimensions correspond to the 70th percentiles of the probability distributions, resulting from the
tool computation. This quantile is therefore also used to estimate the total costs. Generally the 70 th
percentile is used for budgeting purposes. Therefore, when it is assumed that the terminal was
well designed in the past, it can be concluded from these realistic main dimensions that the tool
gives credible results. It is advised -based on the case study- to not use large uncertainties for
variables since these variables will then dominate the resulting dimensions and costs. As a rule-of-
thumb for uniform and triangular distributions the upper limit should not be more than twice the
lower limit.
To answer the second part of the research question a sensitivity analysis is performed. The
distributions of variables with an uncertainty were adjusted and the effects on the total direct costs
were analysed. The average vessel length, and consequently the quay wall, as well as the stockpile
height appeared to have the highest influence on the costs. The total costs showed medium or low
35
This cost estimate considers the quay wall for sea-going vessels, gantry grab cranes for unloading, loading
equipment, stockpile pavement and stacking-reclaiming equipment. Terminal elements that are not taken into
account are roads, buildings, train and crane rails, belt conveyors, the quay wall for barges and corresponding
loading equipment.
145
sensitivity to the other considered variables. It must be stressed that these results are only for this
specific case, other cases may give different results. It is therefore recommended to do a
comparable analysis for every case when more insight is desired.
6.2 Recommendations
During this study certain assumptions were made due to for example lack of data or time. The
research aspects that can be investigated further are presented in this section. The aspects are
listed per chapter of this report.
Literature study
Terminal equipment manufacturers do not easily share information, such as capacities or weights,
regarding their products. The dry bulk unloading, loading, reclaiming and stacking capacities of
equipment that are listed in this study therefore have a high uncertainty. More detailed
information would result in a more accurate overview and tool results. The same holds for the
investment costs of these machines. In this study an estimate of dry bulk equipment costs is made
by using a relation from Vianen, T. van (2015); between the capacity and the costs of stacker-
reclaimers. A similar study for the other equipment types would give more accurate costs
indications.
As already mentioned unit costs for terminal elements are hardly present in literature.
Nevertheless this study presents unit costs for all main terminal elements. Some of these costs are
rough estimates since they are based on little data and may therefore be less accurate or reliable.
Further research into these costs is recommended, especially for: jetties, silos and the different
types of liquid bulk tanks.
Expert Judgement Elicitation

Four experts were assessed for the Expert Judgement Elicitation. Although this number
theoretically is sufficient more sources of information may give more complete results. The
experts gave different motivations for certain questions. Including more experts may lead to even
more different views on a problem and can therefore make a Decision Maker more complete. The
field of expertise of most of the assessed experts are specifically container terminals. An option
for a future study is to create separate questionnaires for container, dry and liquid bulk terminals
and elicit corresponding experts. This way experts are scored purely on their field of expertise,
which may improve their weights for the corresponding target variables. Then the chance that
‘good’ knowledge is lost, because of low weights, is smaller.
The weights of the experts for the dependence elicitation are based on only one seed question.
Only one seed is used since these questions require data which is hard to come by. This was
acceptable in this case since -although two experts got zero weight- all experts their opinions are
well represented. Besides, another goal of the application of EJE in this study is to bridge the gap
between theory and practice and to introduce the method to this field of expertise. For future
applications of this method it is advised to include a more complex combination of seed variables.
Accordingly it is recommended to use seed variables with a similar number of (conditional)
dependencies as for the target variables. This decreases the chance that multiple experts give
exactly similar estimates and dominate the pooling of the correlations.
146 CONCLUSIONS AND RECOMMENDATIONS

BIBLIOGRAPHY
Actueel Hoogtebestand Nederland (2016), Website. http://ahn.arcgisonline.nl/ahnviewer/
Agerschou, H. (2004), Planning and Design of Ports and Marine Terminals, Thomas Telford.
Aspinall, W. (2008), Seventh session of the statistics and risk assessment section’s international
expert advisory group on risk modeling, in W. Aspinall, ed., ‘Expert Judgment Elicitation
using the Classical Model and Excalibur’, number Round IV.
Böse, J. W. (2011), Handbook of Terminal Planning, Springer.
Cooke, R. (1991), Experts in uncertainty: opinion and subjective probability in science, Oxford
University Press on Demand.
Cooke, R. M. & Goossens, L. L. (2008), ‘Tu delft expert judgment data base’, Reliability
Engineering & System Safety 93(5), 657–674.
dace (2015), ‘Price booklet edition 31’.
de Gijt (2010), A History of Quay Walls, PhD thesis, Delft University of Technology.
de Gijt, J. & Broeken, M. (2005), Handbook of Quay Walls, CRC Press.
Drewry (2010), ‘Global container terminal operators 2010: Annual review and forecast’.
EMO (2016), Website. http://www.emo.nl/en/
Ertsoverslagbedrijf Europoort C.V. (EECV) (2013), Terminal Information Book, 1.06 edn, EECV.
Eurosilo, E. (2016), ‘Storage solutions for non-free flowing bulk materials’, Website.
http://eurosilo.com/
Fourgeaud, P. (2000), ‘Measuring port performance’, The World Bank .
Genest, C. & Favre, A.-c. (2007), ‘Everything you always wanted to know about copula modeling
but were afraid to ask’, Journal of Hydraulic Engineering .
GreenPort (2016), ‘Dry bulk covered storage options’, Website.

http://www.greenport.com/news101/Projects-and-Initiatives/dry-bulk-covered-storage-
options
GSI (2016), ‘Stiffened grain bins’, Website. http://www.grainsystems.com/products/storage/grain-

bins/4032.html
Het Financieele Dagblad (2016), ‘Lng heeft nog een lange weg te gaan’, Het Financieele Dagblad
. Article in Dutch.
Heymann, E. (2006), ‘Container shipping. overcapacity inevitable despite increasing demand’,

Deutsche Bank Research 2006, 1–10.
Idrus, M., Samang, L., Adisasmita, R., Sitepu, G. & Ramli, M. I. (2012), ‘A study on the
container yard utilization of the major ports in indonesia eastern region’.
147
igg (2016), ‘Bouwkostenkompas: Woning- en utiliteitsbouw’.
Kersten, M. K. (2010), Master plan port romano bay, albania, Master’s thesis, TU Delft, Delft
University of Technology.
Kleinheerenbrink, A. (2012), A design tool for dry bulk terminals, Master’s thesis, TU Delft,
Delft University of Technology.
Kraan, B. C. P. (2002), Probabilistic inversion in uncertainty analysis: and related topics, TU

Delft, Delft University of Technology.
Krol, R. (2007), Next generation storage tanks, Master’s thesis, TU Delft, Delft University of
Technology.
Kullback, S. (1997), Information theory and statistics, Courier Corporation.
Ligteringen, H. & Velsink, H. (2012), Ports and Terminals, VSSD.
Lind, D., Hsieh, J. K. & Jordan, M. A. (2007), Tandem-40 dockside container cranes and their
impact on terminals, in ‘Proceedings of 11th triannual international conference on American
society of civil engineers’, pp. 1–9.
Lodewijks, G., Schott, D. & Ottjes, J. (2009), ‘Dry bulk terminal expansion or redesign?’, Port
Technology International .
M. D. Mckay, R. J. B. & Conover, W. J. (2000), ‘A comparison of three methods for selecting

values of input variables in the analysis of output from a computer code’, Technometrics
42(1).
Mohseni, N. (2011), Developing a tool for designing a container terminal yard, Master’s thesis,
TU Delft, Delft University of Technology.
Monday Nyema, S. (2014), ‘Factors influencing container terminals efficiency: A case study of
mombasa entry port’, European Journal of Logistics Purchasing and Supply Chain
Management .
Monfort, A., Aguilar, J., Vieira Gonçalves de Souza, P., Monterde, N., Obrer, R., Calduch, D.,
Martín, A. M. & Sapiña, R. (2011), Sea port capacity manual: application to container
terminals, Technical report, Fundación VALENCIAPORT.
Morales-Napoles et al. (2016), Calibration and combination of experts’ dependence estimates.

Unpublished.
Morales-Napoles, O., Hanea, A. & Worm, D. (2014), Experimental results about the assessments
of conditional rank correlations by experts: Example with air pollution estimates, in ‘ESREL
2013: Proceedings of the 22nd European Safety and Reliability Conference" Safety,
Reliability and Risk Analysis: Beyond the Horizon", Amsterdam, The Netherlands, 29
september-2 oktober 2013’, CRC Press/Balkema-Taylor & Francis Group.
Morales-Napoles, O., Kurowicka, D. & Roelen, A. (2007), ‘Eliciting conditional and

unconditional rank correlations from conditional probabilities’, Reliability Engineering &
System Safety 93(5), 699–710.
Morales-Napoles, O. & Worm, D. (2013), Hypothesis testing of multidimensional probability

distributions, Technical report, TNO.
Mwasenga, H. (2012), ‘Port performance indicators: A case of dar es salaam port’.
148 BIBLIOGRAPHY
PCA Consultants (n.d.), Technical pollution prevention guide for dry bulk terminals in the lower
fraser basin, PCA Consultants.
PIANC (1995), Report no. 30: Approach channels - preliminary guidelines, Technical report, The
World Association for Waterborne Transport Infrastructure (PIANC).
PIANC (2012), Report no. 116: Safety aspects affecting the berthing operations of tankers to oil
and gas terminals, Technical report, The World Association for Waterborne Transport
Infrastructure (PIANC).
PIANC (2014a), Report no. 135: Design principles for small and medium marine container
terminals, Technical report, The World Association for Waterborne Transport Infrastructure
(PIANC).
PIANC (2014b), Report no. 158: Masterplans for the development of existing ports, Technical
report, The World Association for Waterborne Transport Infrastructure (PIANC).
Quist, P. & Wijdeven, B. (2014), Container Terminals, VSSD, chapter 7, p. 276.
RiskAMP (2016), ‘Latin hypercube sampling’, Website. https://riskamp.com/kb/Latin-

Hypercube-Sampling
Saanen, Y. A. (2004), An approach for designing robotized marine container terminals, TU Delft,
Delft University of Technology.
Simio (2016), ‘Port simulation software’, Website. http://www.simio.com/applications/port-

simulation-software/
Sztrik, J. (2012), Basic Queueing Theory, Sztrik, J.
TBA (2016), ‘Timesquare: proven, valid, and accurate’, Website.

https://www.tba.nl/en/simulation/timesquare/
Terblanche, L. & Moes, H. (2009), ‘Maritime highways to a port - modelling of maximum

capacity’.
The Tioga Group (2010), ‘Container port capacity study’.
Udny Yule, G., Kendall, M. G. et al. (1950), ‘An introduction to the theory of statistics.’, An
introduction to the theory of statistics. (14th ed).
UNCTAD (1985), Port development: A handbook for planners in developing countries, second
edition edn, United Nations.
Vianen, T. v., Ottjes, J. & Lodewijks, G. (2012), ‘Stockyard dimensioning for dry bulk terminals’.
Vianen, T. v., Ottjes, J. & Lodewijks, G. (2014), ‘Dry bulk terminals characteristics’, Bulk Solids
Handling .
Vianen, T. van (2015), Simulation-integrated Design of Dry Bulk Terminals, PhD thesis, TU
Delft, Delft University of Technology.
VROM (2008), Publicatiereeks gevaarlijke stoffen 29: Richtlijn voor bovengrondse opslag van
brandbare vloeistoffen in verticale cilindrische tanks, Technical report, VROM. Language:
Dutch.
149
VROM (2013), Publicatiereeks gevaarlijke stoffen 18: Lpg: depots, Technical report, VROM.
Language: Dutch.
VROM (2014), Publicatiereeks gevaarlijke stoffen 9: Cryogene gassen: opslag van 0,125 - 100
m3, Technical report, VROM. Language: Dutch.
Werner, C., Bedford, T., Cooke, R. M., Hanea, A. M. & Morales-Nápoles, O. (2016), ‘Expert
judgement for dependence in probabilistic modelling: A systematic literature review and
future research directions’, European Journal of Operational Research .
Whitlock, P. A. & Kalos, M. H. (2009), Monte Carlo Methods, second revised and enlarged
edition edn, Wiley.
Wolfert, A. (2014), ‘Projects risk management. introduction probabilistic costing & planning.’,
Lecture notes. Lecture from TU Delft course CIE4130.
Wu, M. (2012), A Large-scale Biomass Bulk Terminal, PhD thesis, TU Delft, Delft University of
Technology.
150 BIBLIOGRAPHY
APPENDICES
Appendix I Unit costs CONFIDENTIAL ........................................................................ 152
Appendix II Interviews ................................................................................................... 153

II.1 Sip Meijer | Witteveen+Bos ................................................................................................ 154
II.2 Robert Jan Smits van Oyen | Tebodin................................................................................. 157
Appendix III How to perform an Expert Judgment Elicitation ................................... 159

III.1 Uncertainty assessment | Classical Model (Cooke) ............................................................ 159
III.1.1 Elicitation ................................................................................................................................ 159
III.1.2 Calibration............................................................................................................................... 159
III.1.3 Information .............................................................................................................................. 160
III.1.4 Aggregation ............................................................................................................................. 161
III.2 Dependence assessment | Conditional probability technique ............................................. 162
III.2.1 Elicitation ................................................................................................................................ 162
III.2.2 Aggregation ............................................................................................................................. 165
Appendix IV Expert Judgment Elicitation questionnaires........................................... 167

IV.1 Expert Judgment questionnaire | Uncertainty ..................................................................... 167
IV.1.1 Total terminal factor ................................................................................................................ 167
IV.1.2 Storage occupancy factor ........................................................................................................ 169
IV.1.3 How it works ............................................................................................................................ 169
IV.1.4 Calibration variables ............................................................................................................... 171
IV.1.5 Variables of interest ................................................................................................................. 172
IV.2 Expert Judgment questionnaire | Dependence .................................................................... 173
IV.2.1 Dependence ............................................................................................................................. 173
IV.2.2 Average dwell time .................................................................................................................. 173
IV.2.3 How it works ............................................................................................................................ 173
IV.2.4 Calibration dependency ........................................................................................................... 174
IV.2.5 Target dependencies ................................................................................................................ 175
IV.3 Seed variable realisations ................................................................................................... 177
IV.3.1 Uncertainty .............................................................................................................................. 177
IV.3.2 Dependence ............................................................................................................................. 182
Appendix V Expert Judgment Elicitation results ........................................................ 183

V.1 Uncertainty elicitation results ............................................................................................. 183
V.1.1 Seed variables .............................................................................................................................. 183
V.1.2 Target variables ........................................................................................................................... 186
V.1.3 Decision Maker seed item scores ................................................................................................. 187
V.2 Experts motivations ............................................................................................................ 188
V.2.1 Uncertainty ................................................................................................................................... 188
V.2.2 Dependence .................................................................................................................................. 190
Appendix VI Calculation of silo and tank group dimensions ....................................... 193

VI.1 Dry bulk silo groups ........................................................................................................... 193
VI.2 Liquid bulk tank groups ...................................................................................................... 195
151
Appendix I Unit costs CONFIDENTIAL
Intentionally left out in this version.
152
Appendix II Interviews
153
II.1 Sip Meijer | Witteveen+Bos
Interview Date:
Time:
Location:
19-04-2016
14:00 – 15:20
Deventer
Sip Meijer
Senior Port Specialist at Witteveen+Bos
1. What steps do you perform during the design of a terminal in the feasibility phase of a port
master plan?
Important are the costs, for the client a cost overview is the goal of a feasibility study. The
type of soil is important for e.g. the quay wall structure and the yard pavement and thus the
costs. Also terminal equipment forms a large part.
2. What literature do you use for design rules and guidelines?

Containers: An addition to the literature list shown is Chapter 2 from ‘Handbook of terminal
planning’, 2011 written by Birgitt Brinkmann. This contains information about the required
number of equipment needed per STS crane and useful properties of the equipment.
Dry bulk: the paper ‘Dry bulk terminal characteristics’ by Vianen et al.
Liquid bulk: Meijer says he doesn’t know much about this subject, normally Tebodin takes
care of the calculations for these kind of terminals.
3. Do unwritten rules of thumb exist for the calculation of terminal dimensions?

A common value for the depth of a container terminal is 500 meters. Furthermore capacity
benchmarks exist for container terminals for example TEU/m/year for quay walls or
TEU/ha/year for the storage yard (dependant on the type of storage; import, export,
transhipment). Often used for import/export is 25000 TEU/ha/year and for transhipment is
35000 TEU/ha/year. These values are more used for checking the calculated values. They
differ considerably across different regions in the world. Also existing ports are used as
reference for values of variables.
Master planning in this phase is working with a lot of uncertainties; for example the efficiency
of a crane driver or the real working hours per year. You don’t know this until the terminal is
operational. Important is that every assumption you make you have to write down.
4. By determining the total terminal area, what values do you use for the roads and additional
buildings at the terminal?
For container terminals 75% of the total terminal area is the area of the stack. The required
dimensions of dry and liquid bulk terminals are calculated by Tebodin so Meijer does not
know the rules of thumb for these kind of terminals. They have more knowledge about what
equipment is needed and how much space it requires to store the cargo. Tebodin has
specialised dry bulk (in Deventer) and liquid bulk (in Den Haag) groups.
5. What variables do you normally use during the calculation of terminal dimensions?
Variables, values and much more valuable information about (the design of) container
terminals can be found in ‘Sea port capacity manual: application to container terminals’ by
Monfort Mulinas et al. (2011).
Common variables are: throughput, dwell time, peaking factor, occupancy rate storage yard,
crane productivity, number of cranes, hours per year, occupancy rate quay, average and
maximum ship length, throughput per call, arrival and service distributions, TEU factor,
stacking height. Recently the most common value of the TEU factor worldwide is 1.60. The
154 Appendix II Interviews

arrival and service distribution is almost never known at this stage. The stacking height for
empties is usually 7 to 8 boxes high. Normally stacks are not made higher than 2 or 3 boxes
since otherwise the number of internal moves becomes too high. An average value of 2.5 for
the stacking height is common, regardless the stacking equipment.
6. Do you calculate with different values for the same variable in order to take into account the
uncertainty?
Meijer does a sensitivity analysis for, for example, the number of moves per crane, the TEU
factor and the dwell time. Terminal operators can influence the dwell time by changing the
price to store a container. “We don’t calculate with standard deviations”.
7. Do you use standard values for variables of which the values are unknown?
Yes, they are taken from literature and from experience. Clients expect the consultancy firm to
know these things. When values are taken from surrounding ports they are used as a minimum
since the new port must be better than the competitor’s.
8. Do you take the growth path of a port into account when calculating dimensions for a
terminal?
The throughput is almost always provided by the client, including a low, average and high
growth scenario. The growth path is split up into smaller parts of for example 5 years if the
growth is significant. For these periods different calculations are made so that the investment
for the entire terminal does not happen at once. Equipment is easier to acquire during the
lifetime of the terminal.
The advancement of regular cranes (per trolley) with respect to moves per hour is very low.
Optimisations exist that more trolleys are used or that in one cycle a container is imported and
exported, this however requires extensive planning.
9. Do you use software during designing in the feasibility phase?

No, it is all done by hand. For some minor calculations excel is used. Multiple calculations are
made but usually one typical outcome is presented in the report. Meijer: “It is a puzzle every
time again.”
10. Do you know projects of Witteveen+Bos that include a container, liquid and dry bulk terminal
that I can use for validation of the tool and for a case study?
For the Filyos and Taman projects only “vlekkenplannen” were made. So only the total
required area and quay length was calculated. Benin is another project that is being done at
this moment, the required storage area and quay length are calculated per transport direction.
Meijer thinks that in the mentioned projects no liquid bulk terminal is included. Another
option is a project in Cyprus (contact Johan de Boer), Witteveen+Bos did not determine the
dimensions but we possibly do have documents that include this.
11. Are in the feasibility phase costs determined?

Yes as already is discussed. The cost department of Witteveen+Bos does this. But also similar
older projects are used as reference by the master planner. In the feasibility phase for, for
example, a quay wall a similar wall with depth and terminal equipment is taken and the costs
per meter wall are determined (including fenders, boulders etc.). In the (financial) feasibility
phase of a project the goal is to determine if the project is viable. Therefore a cost estimate is
made for different variants in order to see what the return on investment is.
12. a) Do you try to minimise the costs during the design?

In the feasibility phase you should not be too optimistic with costs. It is better to be on the
safe (higher) side.
155
b) How do you get prices for e.g. equipment, dredging, structures, etc?
As was already mentioned the costs are determined at the cost department of
Witteveen+Bos or older projects are used as a reference. Meijers does not know about the
existence of an overview with costs, but says that maybe the cost department has
something like that. Costs for dredging are usually determined by the dredging group of
Witteveen+Bos, for this I can contact Niels Borgers or Marijn Huijsmans. Equipment
costs are rather easy to acquire. There can however be large differences between for
example European and Asian built equipment (e.g. a factor 2).
13. Design rules

Capacity ratios can probably be found in Monfort Mulinas et al. (2011). Always the same
standard rules are used. The queuing theory is almost never used in the feasibility phase since
the distributions are not known and some variable values neither. It is an option to include the
queuing theory in the tool as a supporting calculation. But when much information is not
known a lot of assumptions have to be made, so what is then the value of the result?
14. What would you like to have included in the tool?

Meijer has at this moment no wishes. He is content with the current ideas for the tool. Maybe
when the tool is finished he will think of additions/changes.
15. Are there more persons that you recommend me to interview?

People at Tebodin from the dry and liquid bulk group. Meijer will send me the contact
information.

II.2 Robert Jan Smits van Oyen | Tebodin
Interview Date:
Time:
Location:
12-05-2016
09:00 – 10:15
Den Haag
Robert Jan Smits van Oyen

Manager Logistics and Asset Management & Maintenance
at Tebodin Consultants & Engineers B.V.
1. What literature do you use for design rules and guidelines?

PGS 29 exists, this document is about tanks with flammable content. These are guidelines but
clients want them to be used in order to have safe facilities.
Liquid and dry bulk terminals are very project specific. The terminal design is dependent on
the type of terminal (i.e. gateway, storage and blending) and the business they are in. So some
have multiple small tanks for various liquids and some have a few very large tanks for a few
products.
Smits van Oyen did a study on liquid bulk terminals in Antwerp and Rotterdam for the storage
capacity and quay length.
Smits van Oyen states that it is impossible to contain all terminals (dry bulk but especially
liquid bulk) in a few design rules, there is too much variability.
2. Do unwritten rules-of-thumb exist for the calculation of terminal dimensions?

At Tebodin they determine the layout of the terminal, distances, infrastructure (pipelines),
tanks. Smits van Oyen focuses more on the logistic side of terminal design. For example using
quay occupancy, the serving of vessels and required pump capacity.
Smits van Oyen gives an example of a recent project where not the total throughput increased
but the number of parcels per vessel with different liquids. This increases the complexity.
There are no real rules-of-thumb for these terminals since they differ so much. For the berth
occupancy a value between 50% and 65% is optimal says Smits van Oyen.
For the percentage of the area of roads and buildings with respect to the total area there are no
rules-of-thumb. Smits van Oyen says that most of these terminals are automated for a large
part so not much personnel is needed.
3. What variables do you normally use during the calculation of terminal dimensions?
The presented formulas are indeed used. But also ship characteristics such as parcel size for
liquid bulk (because ships can transport multiple products). This leads to the question if all
these goods must be imported/exported from the same berth or that the ship has to reposition to
another berth. Also local conditions (such as restriction on number of ships in channel) play a
role, this is however more suitable for a later design phase.
The service time and arrival rate distributions are used in simulations. The arrival rate
distribution is determined by randomly spreading the annual throughput over the year. This is
delivered by the client. The service time distribution is dependent on het product (parcel size,
viscosity or pump capacity), this is taken from historical data.
4. Do you calculate with different values for the same variable in order to take into account the
uncertainty?
Most or all data is provided by the client. Tebodin does sensitivity analyses so this takes into
account uncertainty and is a form of risk management.
157
5. How do you determine the terminal area of dry bulk terminals?
The client usually expresses the need for an open stockpile, shed or silo. When a shed is
dimensioned the size of the pile is determined. Some dimensions have a maximum because of
lack of space on the terminal. The area inside the shed, next to the pile is for equipment and
access to the pile and structure. Usually 5 to 10 meters is used.
6. How do you determine the terminal area of liquid bulk terminals?

For liquid bulk pipeways are used, these are often underestimated in size (with respect to
future expansion). No rule-of-thumb exists since it is very much dependent on the sort and
number of handled products and pipelines.
A bund is only needed for safety reasons, so not all products need a bund surrounding the tank.
Also tanks exist with a double wall, reducing the required area.
The maximum pump capacity is determined by the type of product. A certain loading rate is
required on a terminal.
7. Do you use software during the feasibility phase?

Tebodin does static analysis by hand or by means of simple Excel worksheets for e.g. the
stockpile dimensions. But no software like the product of this Thesis (the design tool).
8. Where do you get costs of terminal elements for a cost analysis of a terminal project?
A rule-of-thumb in the feasibility phase for a liquid bulk terminal is between 500 and 1000
€/m3 storage capacity. These values are costs for the whole landside of the terminal, so
including tanks, pipelines and equipment. Terminals with large tanks are on the lower side of
the cost range, terminals with multiple smaller tanks are on the higher side. Approximately
50% of these costs are for the tanks, the rest is for the other landside terminal elements.

Appendix III How to perform an Expert Judgment Elicitation
This appendix provides the reader with the probabilistic theories that are required to perform an
Expert Judgment Elicitation. Separate theories are given for uncertainty and dependence
assessments.
III.1 Uncertainty assessment | Classical Model (Cooke)
This section covers the steps to take when using performance-based elicitation of experts in order
to determine (an) uncertainty distribution(s), specifically using Cooke’s Classical model; Cooke
(1991). To be able to understand this theory some knowledge about statistics is required. The
theory described in this section is implemented in the free software Excalibur36. This section is
entirely based on Cooke & Goossens (2008) unless indicated otherwise.
III.1.1 Elicitation
The 5th, 50th and 95th percentiles from each expert’s subjective uncertainty distribution are elicited.
This is done for a number of seed variables (a common number is ten questions), of which the true
value is known to the researcher, and for a number of target variables. Using the three percentiles
results in the intervals: (0; 0.05], (0.05; 0.50], (0.50; 0.95], (0.95; 1.00] with probability vector:
If N seed variables are assessed all true realisations of these variables must be in one of
the percentile intervals. The total number of realisations in each interval divided by N is the
sample distribution si of each interval i. All four sample distributions combined results in the
sample distribution s(e) of expert e.
III.1.2 Calibration
“Calibration measures the statistical likelihood that a set of experimental results correspond, in a
statistical sense, with the expert’s assessments”, Cooke & Goossens (2008). The calibration score
of expert e is:
(31)
Where,
N Number of seed variables.
s(e) Sample distribution of expert e.
p Probability vector. In this case a 1 x n matrix with n = 4.
is the relative information of distribution s with
respect to p. I is χ2-distributed with n degrees of freedom. Where,
36
Originally developed at TU Delft, now maintained by Lighttwist Software.
URL: http://www.lighttwist.net/wp/excalibur
159
si(e) Sample distribution of interval i of expert e.
pi Probability of interval i.
r Realisation of the variable (or true value).

He Hypothesis: “the inter-quantile interval containing the true value for each
variable is drawn independently from probability vector p”. Is in this case
assumed to be true since it is used to measure the degree to which the data
supports the hypothesis.
The calibration score is the probability under hypothesis He that a deviation at least as great as r
could be observed on N realisations.
III.1.3 Information
“The information in a distribution is the degree to which the distribution is concentrated”, Cooke
& Goossens (2008). Information cannot be measured absolutely; the expert’s distribution is not
compared to the realisation (or true value). Instead it is measured relative to another distribution;
the background distribution. A commonly used background distribution is the uniform
distribution.
The background distribution requires a so called intrinsic range. This range is defined as the
interval with as lower boundary the smallest 5% quantile (q5) of all experts per variable, and as
highest boundary the largest 95% quantile (q95) of all experts per variable. This results in interval
Int = [q5, q95]. To also include the lowest and highest 5% of the expert’s distribution the interval is
extended to a wider interval. The Classical Model uses the k% overshoot rule, typically a value of
k = 10 is used according to Aspinall (2008). The k-value is specified by the researcher. A higher
value tends to make all experts look more informative. The extended interval according to the
mentioned rule is:
Int* = [qL, qH]
Where,
qL
qH
With this information the cumulative distributions with minimum information with respect to the
uniform distribution of all the experts can be determined. This since the distribution boundaries
(qL, 0) and (qH, 1) are determined in the previous step. The rest of the distribution is determined
by linear interpolation between the estimated quantiles of the expert. This should be done for each
variable for each expert.
The information score of expert e then is:
(32)
Where,
N Number of seed variables.
fe,i Probability density of expert e for percentile interval i.

gi Probability density of the background distribution for percentile interval i.
160 Appendix III How to perform an Expert Judgment Elicitation

III.1.4 Aggregation
The Decision Maker (DM) combines the uncertainties of the experts per item (variable) by a
pooling method. A pooling function is a weighted combination of individual judgments. The
resulting distribution function of the DM is a normalised weighted linear combination of the
experts’ distributions per item, defined as:
Where,
Gn Distribution function of the DM for item n.
Ne Number of experts.
we Weight of expert e and/or per item n, depending on the weighting method.
fe,n Distribution function of expert e for item n.
This linear pooling is done for all four probability intervals (p) of the uncertainty distribution of
an item. The DM is then scored the same way as the experts resulting in a calibration and
information score of all items. Different DM’s can therefore be compared. The un-normalised
weight is the product of the calibration and the average information score over all the seed items
per expert. An expert with a high un-normalised weight can be said to have “good expertise”;
Aspinall (2008). DM weights can be determined by the following methods:
 Global weights: Uses the un-normalised weights. Per expert the weights are the same for
all items.
 Item weights: A variation of global weights where each item can have a different set of
weights. Uses the information scores of all experts per item separately instead of the
average score over all items.
 Equal weights: The weights are the same for all experts; .
The global and item weights can be optimised. When optimising an iteration is performed where
the minimum allowable calibration score is varied. When experts do not meet the minimum score
their uncertainties are not taken into account in the DM. The optimisation ultimately chooses the
calibration score for which the un-normalised weight of the DM is maximum. The goal is to
choose the DM with the best calibration and information scores.
161
III.2 Dependence assessment | Conditional probability technique
This section covers the steps to take when using performance-based elicitation of experts in order
to determine rank correlations between variables, specifically using the conditional probability
technique. To be able to understand this theory some knowledge about statistics is required.
III.2.1 Elicitation
Dependence can be visualised by a Bayesian Network (BN), where the objects are the random
variables. As an example and in order to explain the theory the BN of Figure III-1 is considered
which consists of four variables with (un)conditional correlations. The eliciting theory used for
unconditional rank correlations is first summarised by Kraan (2002). The theory for conditional
rank correlations is presented in Morales-Napoles et al. (2007). This section is entirely based on
the latter source unless indicated otherwise. During the elicitation experts are asked for certain
probabilities. An example of such a question is:
Suppose X1 and X2 and X3 are each observed above their median values. What is the probability
that X4 will also be observed above its median value?
The probabilities, corresponding to the example BN, that are asked to the experts are given below.
The example question above resembles question 3.
Question 1:
Question 2:
Question 3:
Where,
The cumulative probability of variable Xi. Therefore means that
variable Xi is observed above its median.
Figure III-1: BN of four variables (X1, ..., X4) with conditional relations. Source: Morales-Napoles et al. (2007)
Definitions
At this stage the expert has given probabilities as answer to the questions, these probabilities have
to be ‘converted’ to rank correlations. After selecting a certain copula the rank correlation (r)
corresponding to the expert’s probability (P) can be found. A copula is a distribution of which the
marginal probability distributions are uniform. Every continuous multivariate probability
distribution can be represented by a copula, copulas can be used to analyse the dependence
between random variables; the previous is based on Genest & Favre (2007). Random variables X
and Y are joined by copula C if their joint distribution can be written as:

Important statistical properties of a multivariate probability distribution are the Pearson’s product
moment correlation and the Spearman’s rank correlation. The unconditional product moment
correlation (ρX,Y) of random variables X and Y with expectations E(X) and E(Y), and variances
var(X) and var(Y) is defined as:
(33)
The rank correlation is the product moment correlation between the ranks of random variables X
and Y. The unconditional rank correlation (rX,Y) between random variables X and Y with
cumulative distribution functions FX and FY is defined as:
(34)
Many different copula’s exist, for this application however the Gaussian copula (or normal
copula) is used. The normal copula is used since it has computational advantages, according to
Morales-Napoles et al. (2014). The normal copula can be defined as the bivariate standard normal
cumulative distribution (Фρ) with correlation ( ) and with the inverse of the univariate standard
normal distribution function (Ф-1), then:
When using the normal copula a conditional rank correlation is equal to a partial correlation. The
conditional correlation between two random variables is thus related to the unconditional
correlation between the same variables. The partial correlation of X1 and X2 with respect to X3 is
given by:
(35)
A correlation matrix (see Section III.2.2) has to be positive definite. In order for it to be positive
definite unconditional correlations must be used. The relation between the rank correlation (r) and
the product moment correlation (ρ) of the normal copula is given by:
(36)
Unconditional correlation of question 1

The exceedance probability P1 can be calculated by taking the double integral of the bivariate
standard normal density function . So taking from the example of Figure III-1; X3, X4
and the corresponding unconditional product moment correlation ρ4,3 results in the exceedance
probability:
(37)
This exceedance probability, as a function of 4,3, is equal to the probability given by the expert,
thus:
163
Solving this equation results in ρ4,3. Using the inverse of Equation (36) for the normal copula
results in the rank correlation r4,3 corresponding to the expert’s probability. The relation between
the probability of exceedance and the unconditional rank correlation is depicted in Figure III-2.
Figure III-2: Relation between the unconditional rank correlation and probability for the Gaussian copula.
Source: Own work
Conditional correlation of question 2
The conditional rank correlations r4,2|3 and r4,1|3,2 of the example require a different calculation
approach. When an expert gives as probability P1 = 0 or P1 = 1 (complete positive or negative
correlation) to the question from the beginning of this subsection then X4 would be completely
described by X3, so probability P2 does not matter. When the expert thinks that X2 and X4 are
independent then probability P2 is equal to P1. When the expert chooses P2 to be anything but 0,
0.5 or 1 then he/she thinks that X3 only partly explains X4, so X2 can only partly explain X4. The
domain of P2 is therefore dependent on the expert’s answer to P1.
The exceedance probability P2 can be calculated by taking the triple integral of the three variate
standard normal density function . Of which 4,3 is fixed in the previous
step and , thus:
(38)
This exceedance probability, as a function of 4,2|3, is equal to the probability given by the expert,
thus:

Solving this equation results in ρ4,2|3. Using the inverse of Equation (36) for the normal copula
results in the rank correlation r4,2|3 corresponding to the expert’s probability.
Conditional correlation of question 3

The computation of r4,1|3,2 happens in the same way as for question 2. Now a quadruplet integral
of the four variate standard normal density function has to be
taken. Of which 4,3 and ρ4,2|3 are fixed in the previous step. This results in the exceedance
probability P3:
(39)
III.2.2 Aggregation
The combining of the dependencies resulting from the experts’ opinions can be done by pooling,
similarly to the uncertainty elicitation. In this study the following linear pooling methods are
considered: equal weights and global weights, where the latter is performance-based. The best
pooling method is determined by the highest calibration score. These aspects are considered in
this subsection which is entirely based on Werner et al. (2016) unless indicated otherwise.
For the aggregation the product moment correlations are inserted into a correlation matrix for each
expert, see Table III-I which is also based on the example BN in Section III.2.1. A correlation
matrix has to be positive definite.
Table III-I: Correlation matrix with unconditional product moment correlations per expert.
X1 X2 X3 X4
X1 1 0 0 ρ1,4
X2 0 1 0 ρ2,4
X3 0 0 1 ρ3,4
X4 ρ1,4 ρ2,4 ρ3,4 1
Calibration score
A calibration score from uncertainty or dependence elicitation cannot be used for both according
to Morales-Napoles et al. (2016). When conducting both assessments separate calibration scores
have to be determined. The calibration score for multivariate assessments used in this research is
the d-calibration score introduced in Morales-Napoles & Worm (2013), which makes use of the
Hellinger distance. To be able to determine scores for the experts seed variables have to be used.
This is similar to Cooke’s Classical model, as presented in Section III.1 . For Gaussian copulas
the Hellinger distance H is defined as:
165
Where is a correlation matrix with the correlations of the seed variables and a matrix that
consists of the correlations derived from the expert elicitation. The d-calibration score is defined
as:
(40)
The d-calibration score has the following properties, according to Morales-Napoles et al. (2016):
 When the assessment of an expert corresponds perfectly to the seed then D = 1.

 A score of D = 0 means that either at least two seed variables are linearly dependent and
the expert does not express this. Or the expert expresses perfect linear dependence
between two variables while the seed variables are not.
 In order for an expert to be highly calibrated he/she has to sufficiently approximate the
dependence structure for each entry.
Equal weights method

For the equal weights method each expert has the same impact on the pooled correlation matrix.
The d-calibration score has to be calculated from the seed variable correlation matrix and the
averaged expert correlation matrix.
Global weights method

For the global weights method the d-calibration score is determined per expert. Then an iteration
is performed where a calibration score threshold is increased in each step. The experts that are
below this threshold are not taken into account. For the experts above this threshold the average of
their weighted37 correlation matrices is determined. For this weighted average matrix the d-
calibration score is determined using the same seed correlation matrix. When the iteration is
complete the calibration threshold that resulted in the largest calibration score of the weighted
average calibration matrix is selected. Finally for the experts with calibration score above the
threshold; the ratio of their score to the sum of the d-calibration scores of these experts determines
their weight.
Pooling
Combining the target dependencies that are estimated by the experts leads to the resulting target
correlation matrix , which is defined as:
(41)
Where,
Resulting target correlation matrix.
W(i) Weight of expert i.
Target correlation matrix of expert i.
N Number of experts.
37
In this case weighted means the product of an expert’s d-calibration score and his correlation matrix resulting
from the assessment. This differs from the weights per expert that are the result of this method.

Appendix IV Expert Judgment Elicitation questionnaires
This appendix includes the uncertainty and dependence questionnaires that are used in this study.
Hereafter the anonymised questionnaires completed by the experts are included. In the last section
the realisations of the seed variables of both questionnaires are given.
IV.1 Expert Judgment questionnaire | Uncertainty
This questionnaire focuses on the percentage of the storage area with respect to the total terminal
area and on the storage occupancy factor. The purpose of this questionnaire is to combine the
opinions of experts in order to determine the uncertainty distribution of the variables of interest.
IV.1.1 Total terminal factor

The total terminal factor (α with unit %) is the percentage of all the gross storage areas with
respect to the total terminal area; . The gross storage area only includes the
stacks/stockpiles and internal infrastructure inside the storage area (including import, export,
transhipment, empties stacks and container freight station). The total terminal area includes the
total landside and waterside area (including quay, apron, other buildings, other infrastructure on
the terminal, etc). The total terminal factor will be asked for container, dry bulk and liquid bulk
terminals. This is denoted by subscript. See Figure 1 for an example of a container terminal, with
in red the gross storage areas and in blue the total terminal area.
Figure IV-1: TCB, S.L. Port of Barcelona (Spain). Source: Google Earth
167
For dry bulk terminals the gross storage area includes the area covered by stockpiles and
stackers/reclaimers and conveyor belts in between the stockpiles. As well as the area covered by
storage sheds and silos including the surrounding area that is kept clear of objects. The total
terminal area also includes the quay or jetty area used for berthing. For an example see Figure 2.
Figure IV-2: Dry bulk terminal in Immingham (UK). Source: Google Earth
For liquid bulk terminals the gross storage area includes the area covered by the storage tanks and
the surrounding area that is kept clear of objects or the area that is surrounded by and including
the bunds (earth walls). The jetty area used for berthing is included in the total terminal area. For
an example see Figure 3.
168 Appendix IV Expert Judgment Elicitation questionnaires

Figure IV-3: ENGIE RC, Port of Rotterdam. Source: Google Earth
IV.1.2 Storage occupancy factor

The average annual storage occupancy factor (ms with unit %) -or storage or yard utilisation
factor- is defined as the number of occupied storage slots or occupied storage volume divided by
the total number of storage slots or storage volume according to the design capacity. This factor
takes into account the fluctuations in required storage capacity due to random arrivals and
departures of cargo. The value asked is the average occupancy factor over 1 year of operation.
The storage occupancy factor will be asked for container, dry bulk and liquid bulk terminals.
IV.1.3 How it works

First you will give estimates of 10 calibration variables. The true values of these variables are
known to the researcher and the uncertainties of all the experts can therefore be scored. Of course
your contribution is always valued, regardless of the score! You probably do not know the exact
answers but you have an idea of the true value, this method is about specifying and comparing
uncertainties.
After the calibration variables, estimates of 6 additional target variables whose true values are not
known are asked.
You are asked to give 3 percentiles of your subjective uncertainty distribution per variable:
 5% quantile means: In 5% of the cases the true value will be lower than your estimate. Or
in other words in 95% of the cases the true value will be higher than your estimate.
 50% quantile means: In 50% of the cases the true value will be lower/higher than your
estimate.
 95% quantile means: In 95% of the cases the true value will be lower than your estimate.
Or in other words in 5% of the cases the true value will be higher than your estimate.
169
To make it more clear these three percentiles of a normal distribution are depicted in Figure 4.
Figure IV-4: 5th, 50th and 95th percentile of a normal distribution.
You are asked to specify the estimates of the quantiles in a table like the one below. As an
example a fictional expert’s estimate, about the length that a Beech tree (Beuk in Dutch) on
average grows every year in Zuid Holland, is given:
5% quantile 50% quantile 95% quantile

0.1 mm 2.5 mm 6 mm
Again; even if you do not know what to answer to the following questions, provide an estimate of
your uncertainty distribution.

IV.1.4 Calibration variables
1. What is your estimate of the αcontainers of the Global Gateway South container terminal in
Los Angeles (California, USA)? Containers are transported with chassis, the terminal has
2640 ground slots and it has a rail connection.
% % %
2. What is your estimate of the αcontainers of the MSC container terminal in Valencia (Spain)?
Containers are transported with tractor-trailer units and RTG’s, the terminal has 2700
ground slots.
% % %
3. What is your estimate of the αdry bulk of the SIPG Luojing dry bulk terminal in Shanghai
(China)? Stored goods are: iron ore and coal. The storage capacity is 1.15 million tonnes.
% % %
4. What is your estimate of the αdry bulk of the Kinder Morgan dry bulk terminal in Newport
News (Virginia, USA)? Stored goods are coal. The storage capacity is 1.4 million tonnes
and the terminal has a rail connection.
% % %
5. What is your estimate of the αliquid bulk of the Vopak Eurotank liquid bulk terminal in
Antwerp (Belgium)? Stored goods are: petroleum products, chemicals and gasoil. The
storage capacity is 454,492 m3 in 173 tanks and the terminal has a rail connection.
% % %
6. What is your estimate of the αliquid bulk of the Vopak TTR liquid bulk terminal in Rotterdam
(Netherlands)? Stored goods are: petroleum products, chemicals and biofuels. The
storage capacity is 318,736 m3 in 89 tanks and the terminal has a rail connection.
% % %
7. What is your estimate of the average annual storage occupancy of the container terminal
in Port Numbay in Jayapura (Indonesia, Southeast Asia) in 2010?
% % %
8. What is your estimate of the average annual storage occupancy of the container terminal
in Dar es Salaam Port (Tanzania, Africa) in 2011?
% % %
171
9. What is your estimate of the average annual storage occupancy of the container terminals
in Miami (Florida, USA) in 2008?
% % %
10. What is your estimate of the average annual storage occupancy of the container terminals
in New York and New Jersey (USA) in 2008?
% % %
IV.1.5 Variables of interest

The variables of interest are for the design of new modern terminals in developed countries
anywhere in the world in 2016.
1. What is in your opinion a realistic value for αcontainers for the design of a new modern
container terminal?
% % %
2. What is in your opinion a realistic value for αdry bulk for the design of a new modern dry
bulk terminal?
% % %
3. What is in your opinion a realistic value for αliquid bulk for the design of a new modern
liquid bulk terminal?
% % %
4. What is in your opinion a realistic value for the average annual storage occupancy for the
design of a new modern container terminal?
% % %
design of a new modern dry bulk terminal?
% % %
design of a new modern liquid bulk terminal?
% % %
Thank you for your time!

IV.2 Expert Judgment questionnaire | Dependence
This questionnaire focuses on the average dwell time of containers on a terminal. The purpose of
this questionnaire is to combine the opinions of experts in order to determine the dependence
between the average dwell times of import, export and transhipment containers.
IV.2.1 Dependence
The (rank) correlation between two random variables quantifies their linear relation. Since no
datasets of dwell times are available for this research the judgment of experts is used to give an
estimate of these correlations.
IV.2.2 Average dwell time

The definition used for dwell time in this research is: the time between the moment that a
container is unloaded from a vessel and the moment that same container leaves the terminal
boundaries, and vice versa. The average dwell time is taken as the yearly average of dwell times
of all containers within a certain cargo flow. The cargo flows that are taken into account in this
research are import, export and transhipment since they can have different dwell times.
IV.2.3 How it works

First you will give an estimate of a dependency between two variables. The true value of this
dependency is known to the researcher and the judgment of the experts can therefore be weighted.
Of course your contribution is always valued, regardless of the score! You probably do not know
the correct answer but you will have an idea of what the value could be.
After the calibration dependency, estimates of three additional target dependencies whose true
values are not known are asked.
For the researcher to determine the dependency you are asked to give a probability that you think
a certain situation will occur.
Again; even if you do not know what to answer to the following questions, please provide your
estimate of the asked probability.
173
IV.2.4 Calibration dependency
Consider the quarterly container throughput38 and quarterly number of container vessels39 in the
Port of Rotterdam for each quarter in between January 1997 and December 2014 (total of 72
quarters).
QUESTION:
1. What is in your opinion the probability that the quarterly number of visiting container
vessels is larger than its median value (being 1,505), when it is given that the quarterly
container throughput is larger than its median value (being 2,331,481 TEU)?
Probability [0 - 100%] = ...............
38
Considering incoming and outgoing, loaded and empty containers.
39
Considering all dead weight tonnages. Based on inward declarations from customs.

IV.2.5 Target dependencies
The dependencies between the average dwell times for the different cargo flows are depicted by
the arrows in Figure 2.
Figure IV-5: Relations between the average dwell times per cargo flow.
The probability distributions of the average dwell times are for modern terminals in developed
countries. These distributions are based on literature and are as follows:
 Import containers; Uniform(4, 8) distribution:
 Export containers; Uniform(3, 7) distribution:
 Transhipment containers; Uniform(3, 7) distribution:
175
QUESTIONS:
1. What is in your opinion the probability that the dwell time of export containers is larger
than its median value (being 5), when it is given that the dwell time of import container is
larger than its median value (being 6)?
Probability [0 - 100%] = ...............
2. What is in your opinion the probability that the dwell time of transhipment containers is
larger than its median value (being 5), when it is given that the dwell time of export
containers is larger than its median value (being 5)?
Probability [0 - 100%] = ...............
3. What is in your opinion the probability that the dwell time of transhipment containers is
larger than its median value (being 5), when it is given that the dwell time of import
containers is larger than its median value (being 6) and it is given that the dwell time of
export containers is larger than its median value (being 5)?
Probability [........... - ...........] = ...............
Thank you for your time!

IV.3 Seed variable realisations
This section covers the realisations of the seed variables from the uncertainty and dependence
questionnaires.
IV.3.1 Uncertainty
1. What is the αcontainers of the Global Gateway South container terminal in Los Angeles
(California, USA)?
Equipment Chassis
Terminal area Agr,t [x104 m2] 120
Storage area ∑Agr,i [x104 m2] 76.83
Gross storage coefficient α [%] 64
Source: Google Earth
2. What is the αcontainers of the MSC container terminal in Valencia (Spain)?
177
Equipment TTU + RTG
Terminal area Agr,t [x104 m2] 34.7
3. What is the αdry bulk of the SIPG Luojing dry bulk terminal in Shanghai (China)?
Storage type Open stockpiles


4. What is the αdry bulk of the Kinder Morgan dry bulk terminal in Newport News (Virginia,
USA)?
Storage type Open stockpiles

5. What is the αliquid bulk of the Vopak Eurotank liquid bulk terminal in Antwerp (Belgium)?
Petroleum products,
Stored liquids
chemicals, gasoil
179
6. What is the αliquid bulk of the Vopak TTR liquid bulk terminal in Rotterdam (Netherlands)?
Petroleum products,
Stored liquids
chemicals, biofuels

7. What was the average annual storage occupancy of the container terminal in Port Numbay
in Jayapura (Indonesia, Southeast Asia) in 2010?
The storage yard occupancy rate (YOR) for 2015 is calculated by using an estimate for
the container throughput of 2015, it can therefore be > 100%.
ms = 82%, value taken from the diagram.
Utilisation of container yard of various ports. Source: Idrus et al. (2012)
8. What was the average annual storage occupancy of the container terminal in Dar es
Salaam Port (Tanzania, Africa) in 2011?
ms = 63%, value taken from the diagram.
Utilisation of container yard Dar es Salaam Port. Source: Mwasenga (2012)
9. What was the average annual storage occupancy of the container terminals in Miami
(Florida, USA) in 2008?
ms = 53%, from The Tioga Group (2010).
10. What was the average annual storage occupancy of the container terminals in New York
and New Jersey (USA) in 2008?
ms = 75%, from The Tioga Group (2010).
181
IV.3.2 Dependence
1. What is in your opinion the probability that the quarterly number of visiting container
vessels is larger than its median value (being 1,505), when it is given that the quarterly
container throughput is larger than its median value (being 2,331,481 TEU)?
Spearman’s rank correlation r = 0.6909; from analysis of Eurostat data, see graph below.
The corresponding probability P ≈ 0.75, assuming a Gaussian copula.
Quarterly number of container vessels versus quarterly container throughput in Rotterdam. Data source:
Eurostat

Appendix V Expert Judgment Elicitation results
This appendix first graphically presents the results of the expert uncertainty assessments. The
calibration and information scores of the chosen Decision Maker are presented as well. Hereafter
the reasoning of the experts is provided for both the uncertainty and dependence assessments.
V.1 Uncertainty elicitation results
To be able to compare the estimates of the experts and the Decision Maker(s) their distributions
are depicted in graphs. Each graph represents a question from the questionnaire, in the graphs
each horizontal bar represents the uncertainty between the 5 th and 95th percentiles of an expert.
The white vertical bars represent their 50th percentile. For the seed variables the true value (or
realisation) is depicted by a vertical green line. The resulting distribution of the chosen Decision
Maker (with item optimised weighting scheme; see Section 3.4.1.3 of the report) is included in the
graphs as well. First the seed variable results are given, followed by the target variable results.
V.1.1 Seed variables
183
184 Appendix V Expert Judgment Elicitation results
Figure V-1: Seed variable results per questionnaire item
185
V.1.2 Target variables

Figure V-2: Target variable results per questionnaire item
V.1.3 Decision Maker seed item scores

The Decision Maker of the chosen weighting scheme (DM item optimized) can be seen as a
virtual expert. The DM estimates are also scored on calibration and information (with respect to
the seed and target variables) to be able to compare the different DM’s. The calibration and
information scores of DMit_op are presented in Table V-I.
Table V-I: Calibration and global information scores of the seed variables for DMit_op
Information
Calibration
Nr. Questionnaire item score
score
(seed variables)
1 Total terminal factor
0.5710 0.2151
Containers - GGS terminal, Los Angeles (USA)
0.5925 0.2754
Containers - MSC terminal, Valencia (ES)
0.7305 0.5720
Dry bulk - SIPG Luojing terminal, Shanghai (CN)
0.7305 0.5866
Dry bulk - Kinder Morgan terminal, Newport News (USA)
0.5710 0.2120
Liquid bulk - Vopak Eurotank terminal, Antwerp (BE)
0.5710 0.2131
Liquid bulk - Vopak TTR terminal, Rotterdam (NL)
7 Average storage occupancy
0.5710 0.2834
Containers - Port Numbay, Jayapura (ID)
0.4048 0.2964
Containers - Dar es Salaam port (TZ)
9 Average storage occupancy 0.5710 0.2451
187
Information
Calibration
Nr. Questionnaire item score
score
(seed variables)
Containers - Miami ports (USA)
0.5925 0.2824
Containers - New York & New Jersey ports (USA)
V.2 Experts motivations
The reasoning, opinions and remarks of the experts with respect to their answers and the
questionnaire itself are presented in this section.
V.2.1 Uncertainty
During the elicitation the experts (E1, E2, E3, E4) gave their motivations for certain choices or
assumptions that they made. The main motivations are presented in this section, they are given per
variable (questionnaire question).
Seed questions
General remarks for the total terminal factor: E3 notes that if empty containers are stored on the
terminal itself it increases the required space. In this assessment it is assumed that empties are
stored on the terminal.
1. E2/E3 think that the difference between storage and total area is small due to the use of
trailers. The alpha is therefore large.
E3 subtracts 10% from the value he originally had in mind, since a rail connection is
present.
E4 thinks that because of the required manoeuvring space relatively more space is needed
outside of the storage area, so that leads to a smaller alpha.
2. E2/E3 note that RTG’s are space efficient and therefore lead to a smaller storage area with
respect to the total area.
E4 thinks that because the gross storage area also includes internal roads (and thus space
efficiency does not matter) the answers to questions 1 and 2 are quite similar.
3. E2 thinks that for dry bulk terminals the alpha is similar to container terminals.
E3 thinks that because the environment in China is less of an issue water basins are not
present on the terminal, this decreases the total area and increases alpha. He thinks the
factor is still relatively high since on dry bulk terminals not much other space is required.
E4 notes that dry bulk terminals are highly mechanised, so high percentages of storage
area with respect to total area are realistic. He also notes that compared to a container
terminal the percentage of storage area is large but no large apron is needed.
4. E2/E3/E4 think that the alpha is smaller than for question 3 because of the presence of a
rail connection.
E3 also notes that dry bulk trains are (un)loaded while driving so a waiting area is not
required. A smaller alpha is chosen.
5. E1/E3/E4 think that alpha is high because bunds are part of the storage area and therefore
lead to a large storage area.
E1 also notes that not all liquid bulk terminals have bunds so this can make a big
difference and that due to safety distances the storage area will also be larger.
E3 also thinks that not much space is required for buildings and that liquid bulk terminals
with a rail connection do need a waiting area which takes up space.

6. E2 notes that this terminal is smaller than for question 5, for smaller terminals the alpha is
relatively large.
E4 thinks that less area is covered by storage due to the smaller number of tanks, therefore
the alpha is lower.
General remarks for the storage occupancy:
 E1 notes that for newer terminals there still is overcapacity and thus a lower storage
occupancy (or utilisation); therefore a terminal’s age matters. Also for some terminals
the occupancy is asked in or around the financial crisis, this may have led to lower
throughputs and therefore lower occupancies.
 E3 notes that the occupancy is determined by the throughput, when it is near the
maximum capacity the utilisation will be high and inefficiencies will occur. It also
depends on the management of the terminal. When supply and thus utilisation are
large the dwell time increases but the management does not want high dwell times but
also does not want to deny customers.
7. E1/E3 think that the remote location of this terminal reduces the throughput and thus
occupancy.
E2 thinks the utilisation is certainly not larger than 90%.
E4 thinks that Asian terminals are well organised which leads to a lower occupancy.
8. E2/E3 think that this terminal is near its maximum capacity since an expansion is planned,
therefore utilisation will be high.
E3 also states that for container terminals an occupancy of 70% is high while for dry and
liquid bulk terminals the percentages are lower.
E4 thinks that dwell times in Africa are larger so that probably leads to a higher yard
utilisation. In Africa strange situations can occur so the expert has a large uncertainty.
9. E1/E4 think that the efficiency between this and the terminal of the next question does not
differ much, so the occupancy factors are similar.
E2 thinks that the utilisation is less than in other countries because the storage area can be
larger since land is more available here. He notes that he is not very certain.
E3 thinks this terminal is also near its maximum capacity.
10. E2 states that New York is more compact and crowded, therefore less space is available
and the terminal has to use its space more efficiently, so he chooses a high utilisation but
again with a large uncertainty.
E3 thinks this terminal is also near its maximum capacity.
Target questions
General remarks for the target variables:
 E1 thinks the values for the target variables are higher than for the seed variables, because
the seed variables are actual situations while the target variables are a little bit more
idealistic.
 E3 assumes that the terminals are without a rail connection.
 E4 notes that modern terminals are space efficient so they have a relatively high terminal
factor. However they can also use advanced equipment like AGV’s, these require a large
apron and therefore lower the terminal factor. Therefore the expert thinks that modern
terminals do not really differ much from other terminals with respect to the terminal
factor. The expert also notes that he chooses the uncertainty intervals smaller since here
you don’t try to capture a value but you try to direct the terminal to a certain state by
means of your design. He uses the same median value as for the seed variables but takes a
smaller uncertainty interval.
189
1. E2 assumes that a modern terminal is medium to large in size and he therefore thinks it
has a lower alpha compared to smaller terminals.
2. E2 thinks that the factors for dry and liquid bulk (next question) are comparable. The
expert thinks that the additional area for container terminals is smaller than for the other
terminal types, therefore leading to a larger alpha. He thinks that 30% of the total area is a
maximum for this additional area.
3. E2: see previous question.
4. E2 notes that this is dependent on import and export quantities, so also peak throughputs.
He thinks that at a utilisation of 90% peaks cannot be handled anymore, so the average
utilisation should be lower. He also thinks that very low utilisation can occur, it is all
dependent on the type of terminal.
E3 takes 55% as median value, he uses this low value since shipping companies demand
low dwell times.
5. E2/E4 state it depends on the type of terminal (storage, import/export).
E2 notes that it also depends on the predictability of the vessel calls. If these are well
manageable the occupancy will be higher. Also if large vessels call at the terminal a large
buffer is required therefore decreasing the occupancy.
E3 states that it is very much dependent on the cargo. Different gradations of a single
cargo type or multiple cargo types on the terminal have to be separated and therefore take
up more space, this lowers the occupancy.
E4 compares the buffer of a container and dry bulk terminal and thinks that bulk trade is
less predictable so a lower occupancy is required.
6. E2 thinks that a liquid bulk terminal has a more constant supply of cargo in comparison to
dry bulk but finally he assumes that dry and liquid bulk terminals are quite similar.
E3 uses the same motivation as for question 5.
E4 thinks that there are more fluctuations in the liquid bulk trade, so he uses a larger
uncertainty interval.
V.2.2 Dependence
During the elicitation the experts (E1, E2, E3, E4) gave their motivations for certain choices or
assumptions that they made. The main motivations are presented in this section, they are given per
questionnaire question.
Seed question
1. E1/E4 think that there also is a dependency with the call size of the vessels. Because the
higher the call size the higher the throughput for a given number of vessels.
E2 thinks that vessel sizes have increased over the years so this has a negative influence
on the correlation, he still thinks that the correlation is positive.
E3 also thinks that the ship size has increased over the time period so for constant
throughput the number of ships decreases; so a negative relation. He reasons that the
throughput increased over the years so this is a positive relation. Therefore they cancel
each other out which leads to a probability of 50%.
E4 also states that the container vessels travel in circles between some ports. Therefore an
increase in throughput does not necessarily mean an increase in number of vessels. He
thinks the relation is positive but not too much due to the call sizes.
Target questions
General remarks for the dependence between the three different dwell times: E3 says that
transhipment is a different business compared to import and export. The development of
transhipment depends on the location and the development of a terminal. He states that the

realisation of Greenfield transhipment terminals is not often done. He further remarks that he
thinks the provided dwell time distributions seem high for a modern terminal in a western country.
1. E1/E2 state that import and export both are impacted by customs.
E1 also thinks the dwell times will be dependent on the costs of a container slot, which
will be more or less the same for import and export. Expert therefore thinks that the
correlation is positive.
E2 also thinks that it depends on the type of country, agreements with clients and
regulations. It also depends on the main cargo flow, if import percentage is the highest
then import dwell times will be lower. He reasons that if there is something wrong in the
port this will probably be the case for both cargo flows.
E3 states that export and import dwell times are intertwined.
E4 thinks that import and export are independent since they use two different
transportation modes. A jam in the inland transport does not necessarily mean an increase
of dwell times for both export and import.
2. E1/E2 think that transhipment containers are not influenced by customs. They choose a
smaller dependence, but still a positive one.
E3 thinks they are not much related since export can be planned pretty well but
transhipment is more difficult to schedule. He still chooses a small positive correlation.
E4 thinks the relation is much stronger than from question 1. As an example he says that
when the capacity of an STS crane is impeded this affects both export and transhipment.
3. E1/E2/E4 think in this scenario the terminal does not work in a efficient way. The experts
think that in this case the transhipment dwell times are also higher.
E1 also thinks that transhipment is not affected by customs which does affect import and
export, a combination of the arguments results in a medium positive correlation.
E2 thinks that this results in a higher probability.
E3 thinks because transhipment is not correlated to export nor import that these variables
are independent.
E4 chooses a very high correlation.
191
Appendix VI Calculation of silo and tank group dimensions
This appendix contains separate calculations for the required area of dry bulk silo groups and the
required area of liquid bulk tank groups.
VI.1 Dry bulk silo groups
For this calculation silos are imagined to each be positioned in tiles. These tiles are square areas in
which the silo is located with an ‘empty’ area surrounding it. A group of silos therefore is a group
of tiles.
Figure VI-1: Overview of tiles with one silo each. Source: Own work
In the calculations the following assumptions are made:
 A tile has a square shape.

 A silo has a cylindrical shape.
 One silo is positioned in the centre of a tile.
 The centre-to-centre distance (ssilos) between silos is equal to the distance between the
edge of a silo and the edge of a tile.
 All silos in a group are of equal diameter and equal height. And all groups that store the
same commodity are identical.
 The tiles are positioned in the most efficient form. This means that tiles are combined in
such a way that the tile group form corresponds as much as possible to a rectangle.
Figure VI-2: ‘Rectangle-like’ placement of tiles. Source: Own work
193
The total area of the tiles of a silo group can be calculated by multiplying the number of tiles with
the tile area. Then for all external boundaries of the tile group a stroke with a width of 0.5 ssilos
must be added in order to comply to the assumption that the distance between silo edge and tile
edge is equal to ssilos.
Figure VI-3: Silo tiles with surrounding stroke (light gray). Source: Own work
In order to determine the outer stroke area first the number of external boundaries (neb) must be
calculated of a given number of tiles. The following equation is empirically determined:
Figure VI-4: External boundaries in red. Source: Own work
The area of the outer stroke consists of parts with length D + ssilo and width 0.5 ssilo. The area of
the 4 remaining corners40 is determined with . The total area of the outer stroke can
then be calculated with:
40
This is the case for a square tile group. For irregularly shaped groups (like in Figure VI-4) this holds as well
since one corner area is counted double and one corner area is not counted, therefore they cancel each other out.
194 Appendix VI Calculation of silo and tank group dimensions

Figure VI-5: One of the four corner areas of a tile group. Source: Own work
This leads to a total area for the tiles plus stroke that can be calculated with:
VI.2 Liquid bulk tank groups
When a tank group is not surrounded by a bund the calculation of the required area is identical to
the calculation for silos. For a bunded tank group the calculation of the required area and the
required bund volume is given in this section.
All assumptions that hold for silo groups are used for tank groups as well, and additionally:
 A tank has a cylindrical shape.

 A tile can consist of one or multiple tanks; and thus represents a tank group.
 Between two tank groups a single bund is used.
 A bund wall surrounding a tank group is simplified to have a square form.
 The total area required for a tank group is specified as the area between the outer toes of
the bunds.
The effective bund height (heff) is the level to which the liquid can be contained; this is also called
waterline. The real crest height (hcrest) should be 0.25 metre higher because of possible wind
waves according to VROM (2008). Also the settlement of the soil may be taken into account.
195
Figure VI-6: Cross-section of a bund. Source: Own work
The effective height is therefore:
The unknown centre-to-centre distance between opposite bunds is wpit,CtC, see the figure below.
This width will be determined in the following calculation.
Figure VI-7: Overview of bunded tank group with centre-to-centre distance between bunds. Source: Own work
To calculate the storage capacity of the pit (the area or volume between the bunds) first the area
must be computed. Here a simplification is made by neglecting the four corner parts, this is
conservative. For the area see the figure below.

Figure VI-8: Overview with waterline heff (red) on inner slope in bunded tank group. Source: Own work
The storage volume can be calculated by multiplying this area with the effective height subtracted
with the bunds’ slope volumes and the effective tank volumes (volume of a tank for the effective
height). The effective slope width (wslope,eff) can be calculated, using slope angle β, with:
The width of the bund crest is wcrest and the width of the slope with maximum crest height can be
calculated with:
The pit width (wpit) can be calculated with:
The effective pit width (wpit,eff) is the width between two opposing waterlines and can be
calculated with:
The effective tank volume of tank with diameter D is:
The effective volume of the pit (Vpit,eff) is:
197
The slope volume (Vslope,eff) of one bund that is beneath the waterline is:
The number of tanks in a tank group is ntanks. The real pit volume or capacity (Vpit) then is:
In order to determine the unknown required wpit,CtC the real pit capacity must be larger than or
equal to the full capacity of one tank (Vtank) plus 10% of the capacities of other tanks in the same
pit, according to VROM (2008).
for ntanks ≥ 1
After this step it must be checked if the tanks fit in the pit. If this is not the case wpit,CtC should be
increased accordingly. The cross-section of a bund is calculated with:
For a less complex calculation the bund volume per tank group is simplified as in the figure
below. The volume can then be calculated with:
Figure VI-9: Simplification for the calculation of the bund volume. Source: Own work
In order to calculate the total bund volume of multiple adjacent tank groups the number of bunds
that function as a dike for both sides must be determined first. A function is found that determines
the number of internal boundaries given a number of tiles.
Figure VI-10: Internal boundaries in red. Source: Own work

The function is determined by fitting a quadratic curve to related sets of number of tiles (up to 12)
and number of internal boundaries, that are determined by hand. This results in the following
dataset and curve.
Figure VI-11: Quadratic curve fitting in order to find relation between number of internal boundaries and tiles.
The equation for the number of internal boundaries (nib) therefore is:
The situation when multiple tank groups are combined is depicted in the following figure.
Figure VI-12: Combination of two tank groups. Source: Own work
The equation for the total bund volume (Vbund,total) of multiple tank groups is:
The number of external boundaries neb is considered in Section VI.1 . The total area of all groups
(Agr) is then:
199

MSC Thesis Steven Kox

Uploaded by

Copyright:

Available Formats

MSC Thesis Steven Kox

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

MSC Thesis Steven Kox

Uploaded by

Copyright:

Available Formats

S.A.J.

A Tool for Estimating Marine

in partial for the degree of

at the Delft University of Technology,

Supervisor: Prof. ir. T. Vellinga TU Delft

An electronic version of this thesis is available at http://repository.tudelft.nl/.

Delft, January 2017

Figure 1: Cumulative probability distributions resulting from the EJE on uncertainty

AGV Automated Guided Vehicle

Latin Description Unit

dd Required dredging depth m

Table 1: Container vessel classes and dimensions ......................................................................... 25

Summary .................................................................................................................................. iii

List of symbols ........................................................................................................................ vii

List of figures ........................................................................................................................... ix

List of tables ............................................................................................................................ xii

2 LITERATURE STUDY ................................................................................................. 23

3 EXPERT JUDGEMENT ELICITATION ................................................................... 89

4 THE TOOL ................................................................................................................... 109

5 CASE STUDY ............................................................................................................... 123

6 CONCLUSIONS AND RECOMMENDATIONS ..................................................... 143

BIBLIOGRAPHY ................................................................................................................ 147

APPENDICES ...................................................................................................................... 151

Appendix I Unit costs CONFIDENTIAL ........................................................................ 152

Appendix II Interviews ................................................................................................... 153

1.1 Problem description 15

Appendix IV Expert Judgment Elicitation questionnaires........................................... 167

Appendix V Expert Judgment Elicitation results ........................................................ 183

Appendix VI Calculation of silo and tank group dimensions ....................................... 193

1.1 Problem description

1.2 Research objective

Main research objective

Research question III

1.2 Research objective 19

The limits of the study are discussed in this section.

 Marine terminals for greenfield port development are considered as requested by

1.3 Research scope 21

2.1 Container terminals

2.1.1 Background information

 Twenty foot Equivalent Unit or TEU (l = 6.10 m, b = 2.44 m, h = 2.60 m)

Capacity DWT average Length Draught Beam

2nd generation 1500 - 1800 30,000 225 - 240 11.5 30

3rd generation 2400 - 3000 45,000 275 - 300 12.5 32

4th generation 4000 - 4500 57,000 290 - 310 12.5 32.3

Post Panamax 4300 - 5000 54,000 270 - 300 12 38 - 40

Jumbo 6000 - 9000 90,000 310 - 350 14 43

New Panamax 13,000 151,000 366 15.2 49

2.1.1.3 Terminal elements

2.1 Container terminals 25

 Section between quay wall and crane rail

2.1 Container terminals 27

Transport between crane and storage yard

2.1 Container terminals 29

2.1.1.6 Storage yard

Equipment used only at the container yard is (see Figure 13):

2.1 Container terminals 31

 Landside traffic circulation system

Additional elements that may be required are: