Quantitative Risk Analysis Approach To Port Hydrocarbon Logistics
Quantitative Risk Analysis Approach To Port Hydrocarbon Logistics
Quantitative Risk Analysis Approach To Port Hydrocarbon Logistics
)
Subscripts
p
i
product i
b
j
berth j
terminals and shipyards are required to produce a contin-
gency plan for accidental marine hydrocarbon pollution,
including a study of the effects of possible spills and of their
evolution.
In view of these facts, a method is needed to standardise
risk assessment in port settings. We feel that this structured
procedure will help port system stakeholders (especially
port authorities and hydrocarbon terminals) to optimise the
performance of their investments in the elds of prevention
and safety, by helping them to reduce the most signicant
risks. For example, newly projected terminals might be
located by taking into consideration losses due to accident
scenarios. The method devised allows port authorities to
build an objective basis for making decisions about the
conditions to be required of hydrocarbon terminal dealers,
in order to guarantee safety.
Insights on different kinds of risk assessment for
HazMat handling at port terminals can be found in the
following:
Rao and Raghavan [4], Thomas [5] and Hartley [6], who
present the use of risk indexes specically devised for port
areas;
Kite-Powell et al. [7], who attempt to build a risk assess-
ment tool based on historical data for US ports;
Trbojevic and Carr [8], on the subject of safety manage-
ment systems (with several examples of risk assessment
techniques);
Cunningham [9], who provides a demonstration of a risk
matrix;
Ronza et al. [10], on simplied event trees for port acci-
dents;
Darbra et al. [11], who provide a historical analysis of
accidents in harbours.
Egidi et al. [12] briey explain how they dealt with the
problem of assessing HazMat accident risk at a sea-terminal,
while recognising the scarcity of literature on this topic.
Several risk assessment reports, made available to the
public via the Internet, proved to be a valuable source of
information. Some of these reports were taken into account
while carrying out the present project [13,14], despite the
fact that they are not actually complete QRAs. The Canvey
Reports [15,16] were the rst signicant contribution to
industrial port environment QRAs, and they are still relevant
today. What these works lack, however, is an attempt at
standardising the process of risk assessment of navigation
and (un)loading operations for a generic port/terminal.
This is what has been done in this project in the case of
hydrocarbons, with a special regard to accident frequency
estimation.
1990. The 1998 OPRC re-issue is now the principal legislation on counter
pollution from a harbour authority and oil handling facility perspective.
12 A. Ronza et al. / Journal of Hazardous Materials A128 (2006) 1024
2. Scope of the method
Only liquid hydrocarbons were considered. Moreover,
only bulk transportation and handling are included within
the scope of the research.
The analysis covers port waters (from port entrance to
berths) plus (un)loading terminals. Accidents occurring dur-
ing the (external) approach of the tankers to the port were
not taken into account, nor were land accidents, such as
those that can take place during storage and land transporta-
tion (within and outside the connes of the port). Finally,
possible sabotage-related scenarios and accidents likely to
occur during tanker maintenance operations were excluded
fromthe analysis. Instead, navigation through port waters and
(dis)charge were specically addressed. For a discussion on
the patterns of accidental events, such as the operation carried
out when the accident occurred, see [10] and [11].
Therefore, the operations considered are (1) tanker navi-
gating through the port, (2) tanker manoeuvring in the prox-
imity of berths, (3) tanker (un)loading bulk hydrocarbons
2
and (4) bunkering operations.
3. Description of the methodology
3.1. Collection of relevant information
The rst step to take is, of course, to gather the relevant
data that will be used during the analysis (see Fig. 1 for a
schematic representation of the method). This is an extremely
important phase and ensuring that it is carried out properly
can save a great deal of time and avoid rough approximations.
Critical data to be collected are as follows:
The geographical location of the port.
Adetailed map of the port (at least of port waters, berthing
lines and areas where hydrocarbon stevedoring companies
are located).
Climate data (average temperatures, humidity, wind roses
and atmospheric stability). The critical data that are neces-
sary for accident simulation models are (1) average ambi-
ent temperature, (2) average water temperature, (3) average
relative humidity, (4) wind speed, and (5) atmospheric sta-
bility distribution.
Technical data on berths and (un)loading locations. These
data can be obtained fromthe port authority, but it is easier
to collect them directly from the stevedoring companies
that make use of the loading arms and berthing facilities.
Critical data are (1) typical tank volumes for (un)loading
tankers, (2) product temperatures and pressures both for
ship transport and (un)loading, especially for refrigerated
2
From the point of view considered in this study there is no signicant
difference between loading and unloading operations, since the probability
of a loss of containment are the same for both situations, as are the physical
effects of the scenarios.
Table 1
Chart for the collection of trafc data
Number of (un)loading operations
LNG LPG Petrol Gasoil +kerosene Fuel oil . . .
Berth 1
Berth 2
Berth 3
Berth 4
.
.
.
Blanks should be lled in with the number of tanker (un)loading operations
per unit time.
or liqueed products (LPG, LNG), (3) the number of load-
ing arms per berth, (4) operational ow rates for loading
arms and hoses, and (5) loading arm and hose diameters.
Physical and chemical data for the hydrocarbon products
taken into account. Critical data used later in the simula-
tions are (1) density, (2) estimated molecular weight, (3)
vapour pressure, (4) thermal conductivity, and (5) heat of
combustion, etc.
Trafc data. These are critical to the calculation of the
frequencies of accidents. The best way to collect trafc
data is by organising them according to product type and
berth (see Table 1). They should be given by tanker visit
per unit time (e.g. per year). In order to estimate them, one
should refer to past data (for example the last 2 or 3 years),
but if more accurate data or reliable estimations on future
trade are available, these should be used instead. Likewise,
bunkering operations data should be taken into account
(operations per year). General trafc data (the number of
ship visits to the port per year, regardless of ship cargo)
are also needed, because they affect the frequency of ship
collisions (the busier the port, the more likely collision
events will be).
Duration of (un)loading operations. This is also necessary
for the estimation of the frequency of accidents. When
these data are not directly available, an estimate of an aver-
age duration for product p
i
and berth b
j
might be assessed
in the following way:
t
p
i
,b
j
=
total loaded and unloaded volume
p
i
,b
j
operational owrate
p
i
,b
j
no. operations
p
i
,b
j
Tanker hulls. As double hull tankers are much less likely
to give rise to releases when they undergo a collision or
groundingthansingle hull ships are, it is important toknow,
for every product, the ratio of single to double hull tankers.
Information that is not critical but is nonetheless useful
can be gathered about past accidents (spills, res, etc.) that
have occurred in the port involving the hydrocarbons under
analysis.
3.2. Scenario identication
From a general point of view, only two basic events can
cause a loss of containment during the aforementioned opera-
A. Ronza et al. / Journal of Hazardous Materials A128 (2006) 1024 13
Fig. 1. Diagram of the suggested method (n =number of hydrocarbon products handled; m=number of hydrocarbon products bunkered).
14 A. Ronza et al. / Journal of Hazardous Materials A128 (2006) 1024
tions: hull failure and loading arm/hose failure. In both cases
the approach described by TNO [17] was followed. This
means that, for every loss of containment, a two-fold pos-
sibility has to be considered:
in the case of hull failure, a minor as well as a massive
spill;
for loading arms, partial and total rupture.
For bunkering operations, only two scenarios are considered
(one for hull failure and one for hose rupture), as the amounts
spilled are generally small. In a general application, the num-
ber of scenarios will therefore be as follows:
number of scenarios = 4n +2m (1)
n being the number of hydrocarbon products traded and m
the number of products bunkered (normally m=2, diesel oil
and fuel oil being the bunkered fuels).
3.3. Frequency estimation
The approach that was followed is to estimate accident
frequencies on the basis of both trafc data and general fre-
quencies found in technical literature. Great efforts were
made to select appropriate general frequencies for the sce-
narios previously described. Table 2 summarises the gen-
eral frequencies that were selected and used. Many sources
were consulted, but none of them proved to be actually
focused on accidental events in a port environment (many
are rather general, related to open sea maritime accidents;
see Rmer et al. [21]). Apart fromproposing data specically
intended for ports and focussing on the most recent and/or
widely used frequencies, an additional criterion that was
followed in the selection of data is the intention not to com-
plicate excessively the calculations by introducing too many
scenarios.
An important remark must be made here. While loading
arm scenarios are of a purely punctual nature, hull ruptures
are both punctual and linear. In fact the latter may be caused
by any of the following:
an external impact (shipship or shipland) while the
tanker is moving towards the berth or from the berth to
the port entrance (linear operation);
by an external impact (shipland) during manoeuvres near
the (un)loading berth or a shipship collision while the
tanker is (dis)charging (punctual operations).
This dual nature must be taken into account, because,
while the physical effects of the accident are practically the
same, their consequences (on people and installations) may
be different. If a re or explosion takes place during the move-
ment towards/away from the berth, it will generally have
less severe consequences because the accident location is
further away from the docks. For this reason, it is impor-
tant to calculate separate frequencies for punctual and linear
scenarios.
3.3.1. Loading arm and hose failures
These events are purely punctual. Moreover, given the
failure (i.e. the rupture), the probability of spillage is 1. There-
fore, once proper literature data are selected, they are used
directly without further calculations.
As for loading arms, we suggest using the data proposed
in the Purple Book [17]. Different gures can be found in
DNV Technica [22], which are used, for instance, by the
Environmental Resources Management [14]. The approach
followed by [22] is to consider the possibility of three spill
sizes (instead of two, as in TNOs approach). The order of
magnitude of the data is the same. DNV suggests an overall
failure rate of 1.94 10
4
operation
1
; approximately 76%
of the spills is considered to be small, 18%is medium and
6% large. In order not to increase the number of scenarios,
TNOdata were preferred. The Rijnmond Report [23] presents
some loading arm failure frequency data as well. They are
smaller than those proposed by the Purple Book, and they are
expressed as number of failures per hour of operation. TNOs
gures were preferred here just because they are much more
recent.
For the same reason, in the case of hoses, the data found in
[19] are suggested, rather than those of the Rijnmond Report.
3.3.2. Hull failures, punctual events
Two initiating events are likely to provoke accidents at the
berths:
a. a shipland collision while the tanker is manoeuvring near
the berth;
b. a shipship collision during the (dis)charge, caused by
a ship running adrift and colliding with the (un)loading
tanker.
Literature data for these events are shown in Table 2.
The frequency of both classes of initiating event must be
expressed using consistent units. To do so, a frequency per
unit time of shipship collision during (dis)charge must be
estimated thus:
f
b
= F
b
T t
where f
b
is expressed in events per ship visit, F
b
is the fre-
quency of a shipship collision while a tanker is (dis)charging
at a terminal, expressed per ship passage (4.0 10
6
ship
passage
1
), T the ship trafc in the proximities of the berth
[ship passages h
1
] (this can be estimated from the general
trafc data for the port and the berth position), and t is the
duration of the discharge. Note that this can change according
to the berth and the product that is being discharged, depend-
ing on owrates, tank dimensions, and the number of loading
arms actually used.
Another aspect must be taken into consideration: given an
external impact, the probability of an actual spill occurring
must be identied. Several probability data have been found
in the literature. They are reviewed in Table 3.We suggest to
use TNOs data [17], because:
A. Ronza et al. / Journal of Hazardous Materials A128 (2006) 1024 15
Table 2
General frequencies for the initiating events
Operation or scenario Type of initiating event Initiating event General frequency Source
Transfer, loading arm (total rupture) Punctual External impact, mechanical failure 6 10
5
operation
1
[17]
Transfer, loading arm (partial rupture) Punctual External impact, mechanical failure 6 10
4
operation
1
[17]
Transfer, hose (total rupture) Punctual External impact, mechanical failure 4 10
3
operation
1
[18], after
data from
[19]
Tanker manoeuvre Punctual Shipland collision 2.2 10
3
ship visit
1
[20]
Tanker (dis)charge Punctual Shipship collision (passing ship) 4 10
6
passage
1
[20]
Tanker moving to/from berth Linear Shipland collision 1.5 10
4
visit
1
[15]
Grounding 0.3 10
4
visit
1
[15]
Shipship collision with passing ship 2.3 10
5
visit
1
[15]
Shipship collision with moored ship 0.5 10
4
visit
1
[15]
They allow for both a minor and a major spill scenario,
that is, for a greater detail in scenario denition.
In the case of liqueed gas carriers, TNOs data are much
more recent than the Canvey data [15]. Shipbuilding and
vessel trafc control have changed since then (which is
reected in a decrease of spill probability by an order of
magnitude).
In the case of oil tankers, Ref. [14] recommends gures
of the same order of magnitude as those of Purple Book,
but makes a distinction among various circumstances
(berthing impact, impact with jetty, shipship collision).
To use these data would complicate the method, without
signicant improvements.
Therefore, after the approach suggested by TNO [17], we
propose the following probabilities that, given an external
impact, a spill will take place:
p
M
=
2
=0
6
k=1
f RF
k,
(x, y)p()p
k
d (10)
where represents the wind direction, k stands for stability
class, f is the accident frequency, RF
k,
(x, y) the lethality
function estimated on the basis of the vulnerability criteria,
p() the probability that the wind will blow in the direction
and p
k
is the probability of the class of stability k.
Eq. (10) is solved by commercial software that discretises
the integral by way of a summation of 8, 12 or 16 radial
directions.
3.7. Estimation of overall risk for the population
By integrating the product of R by the local population
density over spatial coordinates, the global risk for a given
accident scenario is obtained. By adding up the several R
functions (one for each scenario), a global risk function is
obtained. In order to estimate the number of injured and
evacuated people, historical data have been used. The aver-
age ratios of injured people/evacuees to fatalities have been
estimated to be the following:
2.21 injured people for each fatality;
220 evacuees for each fatality.
The data used to obtain these gures are a subset of
the 1033 port-area accidents analysed in [11]. Of these
accidents, only the 428 that occurred during bulk hydro-
carbon (un)loading and tanker movement/manoeuvres were
retained. The data are taken fromthe MHIDAS database [28],
in which three elds are devoted to gauging the consequences
of the accidents on humans: KR, IR and ER, which represent
the number of people that were killed, injured and evacuated
as a consequence of the accident. Unfortunately, these elds
do not always give positive information. This means that KR
may be 0 or more or it might not be dened at all. The same
happens with IRand ER. In order to estimate the above IR/KR
and ER/KR ratios, the following assumptions were made:
1. whenever KR and ER are not dened, they are assumed
to be 0;
2. to obtain the IR/KR rate, only the accidents for which IR
is dened were used.
In fact, it is highly probable that an undened KR (ER)
simply means that there have not been any victims (evacuees)
as a consequence of an accident. This is certainly not true for
IR data, since many accidents have a high KR record while
the number of injured people remains undened. It is very
A. Ronza et al. / Journal of Hazardous Materials A128 (2006) 1024 19
Table 7
IR/KR and ER/KR ratios for different accident data subsets
Accident subset No. accidents IR/KR ER/KR
(Un)loading accidents 261 3.09 358
Tankers approaching or manoeuvring 167 0.67 1.66
General 428 2.21 220
unlikely that these accidents have not caused some people
to be affected other than fatally, so the average rate IR/KR
was estimated solely on the basis of the records for which a
positively dened IR was available.
Apart from the above rates (IR/KR=2.21 and
ER/KR=220), which are general, averaged for all the
accidents, more specic rates can also be estimated, as a
function of the operation that was carried out during the
accident. The values are shown in Table 7.
It is interesting to note how both IR to KR and ER to KR
ratios decrease dramatically when it comes to accidents that
occurred during the approach or manoeuvre of a tanker (on
average, IR is even smaller than KR in these circumstances).
This means that manoeuvre/approach do not have signicant
aftermaths other than in terms of human life loss. The reason
why the injured to killed ratio is so low is that these acci-
dents mainly involve tanker crews, who are often so close to
the accident that they are more likely to suffer death than
non-fatal injuries. Likewise, ER/KR is low because these
accidents normally happen farther from working and resi-
dential areas, and are consequently of less concern in terms
of people to be evacuated.
The general ratios should be used whenever the present
QRA conceptual approach is applied to a port, because the
scenarios, as they have been designed and structured, entail
both (un)loading and ship manoeuvre/approach operations.
Nevertheless, the operation-specic values can be used for
studies that focus on a particular stage in port hydrocarbon
logistics. Note that, however useful it is to estimate the con-
sequences of accident for humans, the gures in Table 7 only
represent historical averaged data.
4. Case study: the Port of Barcelona
The Port of Barcelona is one of the largest Mediterranean
ports in terms of the number of tonnes traded, and the largest
in Spain. Bulk liquid trade amounts to about 25% of the total
tradedgoods. Almost 9milliontonnes of bulkliquidenergetic
hydrocarbons were transported out of and (mainly) into the
port during 2003 [29], which constitutes the main part of the
hazardous material ux through the harbour.
The port is quite close to the city. The oldest, down-
town terminals have been reconverted in the last decade
and are now a tourist and commercial district. The main
hydrocarbon terminals are located in a separate section of
the port (Moll dInamables, Flammable Product Wharf),
which is more than 2 km away from the nearest residential
area.
Table 8
List of scenarios for the Port of Barcelona
Scenario no. Description
1 Major LNG spill from cargo tank rupture
2 Minor LNG spill from cargo tank rupture
3 Major LNG loading arm failure
4 Minor LNG loading arm failure
5 Major LPG spill from cargo tank rupture
6 Minor LPG spill from cargo tank rupture
7 Major LPG loading arm failure
8 Minor LPG loading arm failure
9 Major petrol spill from cargo tank rupture
10 Minor petrol spill from cargo tank rupture
11 Major petrol loading arm failure
12 Minor petrol loading arm failure
13 Major diesel oil/kerosene spill from cargo tank rupture
14 Minor diesel oil/kerosene spill from cargo tank rupture
15 Major diesel oil/kerosene loading arm failure
16 Minor diesel oil/kerosene loading arm failure
17 Major fuel oil spill from cargo tank rupture
18 Minor fuel oil spill from cargo tank rupture
19 Major fuel oil loading arm failure
20 Minor fuel oil loading arm failure
21 Fuel oil spill from cargo tank during bunkering operations
22 Fuel oil hose failure during bunkering operations
23 Diesel oil spill from cargo tank during bunkering operations
24 Diesel oil hose failure during bunkering operations
The following bulk hydrocarbon products are traded:
LNG,
LPG,
petrol,
kerosene and diesel oil,
fuel oil.
Crude oil is virtually absent as a bulk liquid. For practical
purposes, kerosene and diesel oil were grouped together, as
they present similar characteristics with regard to ammabil-
ity and general hazardousness issues.
The harbour, like most Mediterranean ports, is compact,
and not scattered over multiple locations. Nine private com-
panies carry out bulk liquid trade activities. Five of them
perform energetic liquid hydrocarbon stevedoring, one of
which is exclusively devoted to the unloading and distribution
of LNG cargo, and another trades in LPG. All the companies
but one are located on the Flammable Product Wharf, where
they make use of the berths and unloading facilities located
therein. One company currently holds the concession for a
separate bulk liquid jetty.
The bunkering service is performed by a specialised barge
held by one of the companies. As a result (see Eq. (1)), 24 sce-
narios were considered during the study, which are itemised
in Table 8.
By way of example, Scenario 1 (a major LNG spill from
cargo tank rupture) is presented and discussed below as a
particular application of the method.
Firstly, relevant data are collected for the scenario in step
(a). Apart from the physical conditions of LNG being stored
in the tankers (112 K, 120 kPa), it is necessary to consider
20 A. Ronza et al. / Journal of Hazardous Materials A128 (2006) 1024
Fig. 3. Iso-individual risk curves for scenario no. 1 (major LNG spill from cargo tank rupture).
the trafc ow of LNG tankers and the duration of unloading
operations, considering that the Port of Barcelona, when this
work was being carried out, had only one LNG unloading
berth, which is situated almost at the entrance of the port.
It was estimated that, on average, 169 LNG tankers entered
the Port of Barcelona in a 1-year period to discharge. A sin-
gle discharge operation, considering the average dimensions
of the LNG tankers usually in service at the Port, the num-
ber of loading arms (two), and their operational ow rate
(3000 m
3
/h), lasts 13.2 h on average. It is also necessary to
estimate the average number of ships passing the LNGtanker
while it is discharging; considering the position of the LNG
berth and the overall trafc data of the Port of Barcelona, it
is estimated that 3.7 ships pass that spot every hour.
The frequency of LNG spill events due to hull failure is
thenestimatedin(b). Because it is botha linear anda punc-
tual scenario, two frequencies must be taken into account.
The frequency of the accidents that are likely to happen in
the proximity of the berth is calculated using Eq. (4):
f
HF,p,M
= (f
a
+F
b
Tt)p
M
= (2.2 10
3
visit
1
+4.0
10
6
3.74 13.2) 0.00012
= 2.88 10
7
operation
1
where the probability of spill p
M
is specic to gas carriers,
in compliance with Eq. (2). Thus, the actual frequency of a
spill, considering the yearly LNGtanker trafc, is as follows:
f = 2.88 10
7
169 = 4.87 10
5
year
1
For the linear phenomenon, Eq. (8) must be used:
f
HF,l,M
= 2.5 10
4
0.00012 = 3.0 10
8
visit
1
which implies a frequency of
f = 3.0 10
8
169 = 5.07 10
6
year
1
A. Ronza et al. / Journal of Hazardous Materials A128 (2006) 1024 21
An event tree is drawn out (step c) in which it is made clear
that the spill can give rise to a pool re, a ash re followed
by a pool re or simply to the dispersion of a gas cloud.
The probabilities of these events are 0.065, 0.037 and 0.898,
respectively.
The consequences for people of each of these sub-events
(apart from cloud dispersion, which does not cause harmful
effects) must be calculated (step e).
When, for example, a pool re is considered, the steps
to be taken in order to calculate the radiated power are the
following:
calculation of the released ow rate,
estimation of pool diameter:
d = 2
Q
f
y
where y