Building Research & Information

ISSN: 0961-3218 (Print) 1466-4321 (Online) Journal homepage: https://www.tandfonline.com/loi/rbri20

Analyzing building stock using topographic maps

and GIS

Gotthard Meinel , Robert Hecht & Hendrik Herold

To cite this article: Gotthard Meinel , Robert Hecht & Hendrik Herold (2009) Analyzing building
stock using topographic maps and GIS, Building Research & Information, 37:5-6, 468-482, DOI:
10.1080/09613210903159833

To link to this article: https://doi.org/10.1080/09613210903159833

Published online: 02 Dec 2010.

Submit your article to this journal

Article views: 705

View related articles

Citing articles: 23 View citing articles

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=rbri20
BUILDING RESEARCH & INFORMATION (2009) 37(5-6), 468– 482

RESEARCH PAPER

Analyzing building stock

using topographic maps and GIS
Gotthard Meinel, Robert Hecht and Hendrik Herold

Leibniz-Institute of Ecological and Regional Development,Weberplatz 1,

D-01217 Dresden,Germany
E-mails: [email protected], [email protected] and [email protected]

Building stock is an extremely cost-intensive and long-lived component of the settlement structure. Considering its
importance, the information available on building stock and its development is often inadequate. Data such as the
age and structure of buildings, the number of dwelling units and floors are not comprehensively available nor
sufficiently processed. In order to fill this gap, this paper presents a method for the calculation of settlement structure
parameters by means of a spatial disaggregation through a combination of building data obtained from topographic
maps 1 : 25 000, block boundaries taken from authoritative digital maps, and statistical data at municipality level.
The inexpensive analogue maps are available nationwide and also for earlier time slices due to defined update
intervals. The maps are scanned at high resolution and geo-referenced. The building stock, which is shown
interspersed with traffic, vegetation and lettering, is filtered out by means of digital image processing. Buildings are
then surveyed, relations to neighbouring buildings interpreted, and the general physical structure and block
boundaries classified. Density can then be calculated either for the block boundaries or a reference geometry of one’s
choice (e.g. regular raster cells). By combining building type with typical reference density values for floor number
and building area-related dwelling and resident densities, further settlement structure parameters can be calculated.
Comparing the building stock on maps representing different time slices enables the development of the built
environment to be analysed. The method has been implemented as a toolbox in Geographical Information System
(GIS) software ESRI ArcGIS and is intended for use as a nationwide monitoring system of settlement and open space
development.

Keywords: building stock, Geographical Information System (GIS), settlement analysis, settlement structure,
topographic raster maps, Germany

Le parc bâti est un élément de la structure du peuplement extrêmement coûteux et à la durée de vie extrêmement longue.
Compte tenu de son importance, les informations disponibles sur le parc bâti et son développement sont souvent
insuffisantes. Les données telles que l’âge et la structure des immeubles, le nombre de logements et d’étages ne sont
pas disponibles dans leur ensemble, ni suffisamment traitées. Afin de combler cette lacune, cet article présente une
méthode permettant de calculer les paramètres de la structure du peuplement en recourant à une ventilation spatiale
obtenue grâce à une combinaison de données sur le bâti provenant des cartes topographiques au 1/25000ème, des
limites de pâtés de maisons tirées de cartes numériques faisant autorité, et des données statistiques au niveau
municipal. Des cartes analogiques bon marché sont disponibles pour tout le pays, ainsi que pour les tranches de
temps antérieures du fait des intervalles de mise à jour définis. Les cartes sont numérisées en haute définition et
géoréférencées. Le parc bâti, qui apparaı̂t entrecoupé par la circulation, la végétation et les écritures cartographiques,
est filtré en utilisant le traitement d’image numérique. Il est ensuite procédé à un relevé des immeubles, les relations
avec les immeubles voisins sont interprétées, et la structure physique générale ainsi que les limites des pâtés de
maisons sont classées. La densité peut alors être calculée pour les limites des pâtés de maisons ou bien pour une
géométrie de référence de son choix (par ex., des cellules de quadrillage régulières). En combinant le type
d’immeubles avec les valeurs types de la densité de référence pour le nombre d’étages et les densités de peuplement et
résidentielles relatives au secteur bâti, il est possible de calculer d’autres paramètres de la structure de peuplement. En
comparant le parc bâti sur des cartes représentant des tranches de temps différentes, il est possible d’analyser le

Building Research & Information ISSN 0961-3218 print ⁄ISSN 1466-4321 online # 2009 Taylor & Francis
http: ⁄ ⁄www.informaworld.com ⁄journals
DOI: 10.1080/09613210903159833
 Analyzing building stock using topographic maps and GIS

développement de l’environnement bâti. Cette méthode a été mise en œuvre sous forme d’une palette d’outils dans le
logiciel du Système d’Information Géographique (SIG) ESRI ArcGISw et est destinée à être utilisée comme un système
de suivi, à l’échelle nationale, du peuplement et du développement des espaces ouverts.

Mots clés: parc bâti, Système Géographique d’Information (SIG), analyse du peuplement, structure du peuplement,
cartes matricielles topographiques, Allemagne

Introduction by the individual states at varying levels of spatial

Although the building stock represents a very high and substantive detail. This is insufficient for the pur-
value and is extremely long-lived compared with poses of settlement and infrastructural spatial plan-
other assets, only small amount of information is avail- ning. For example, the data are available in rough
able on it. Research in urban morphology has recently substantive classification by building, with grouped
increased not least due to the requirements of energy numbers of dwelling units and usually only at the
and infrastructure efficiency (e.g. Salat, 2007; Schiller, local authority level. Information about the number
2007). Remote sensing imagery and light detection and of floors, type of structure (slab block housing, row/
ranging (LIDAR) data offer very up-to-date mapping terraced houses, closed perimeter block, etc.), and
and analyses, while the development of a consistent age is lacking in most cases.
time series and a high-quality building-based settle-
ment analysis are still hard to accomplish. Building Another problem is the difference between the urban
recognition rates only reach from 64% to about 81% and rural data situation. Big cities more often have
(e.g. Lee et al., 2003; Shan and Lee, 2005; Lari and their own survey offices, which are increasingly devel-
Ebadi, 2007; Koç San and Turker, 2007; Lefèvre oping three-dimensional city models linking address
et al., 2007). data with building volume. Such detailed information
is not retrospective, nor does it cover the areas sur-
The most important sources of information are cadas- rounding the cities or rural regions. The regional and
tral surveys, registers, and maps, which record not only supra-regional contexts are playing an increasingly
the boundaries of properties, but also building group important role in view of spatially dispersed develop-
plans. In recent years analogue cadastral data have ment patterns, demographic change, and the growing
been digitized by the surveying authorities in competition between regions.
Germany. However, access to the cadastral register is
extremely restricted for data protection reasons. Ultimately, the information that is available on the
Therefore, the data are not available to everyone and building stock is not analytically processed in a way
where access is permitted, it is expensive (E1.80 per that can be useful for supra-local infrastructure
building; ADV, 2007). Furthermore, information on planning. This requires collecting settlement structure
earlier time periods is not available since digital data parameters in Geographical Information System (GIS)
have only recently been introduced. that can show physical densities and population distri-
bution at high spatial resolution. Only then can suitable
Digital spatial base data and topographic maps are a databases be established for focusing spending on the
good alternative. Frischknecht and Kanani (1998) new construction, maintenance, expansion, and –
present examples for approaches to automatic increasingly – downsizing of infrastructural facilities.
interpretation raster maps, and Maderlechner and
Mayer (1994) describe an approach to recognize topo-
graphic features by means of semantic networks and
frames and they applied it on cadastral maps at Objectives
1 : 1000 and 1 : 5000 scales. Whereas the extraction The first aim of this work was to develop a method for
of roads and lettering is a frequent subject of research, remedying the information deficit on the building stock
little literature exists on the extraction of buildings discussed above. To achieve this, an appropriate data-
from maps for settlement structure analysis. Steiniger base needed to be prepared that allows the comprehen-
et al. (2008), for instance, describe an approach for sive application of the method, building-based display,
the classification of urban building structures based and analysis of settlement development. Apart from
on discriminant analysis techniques for generalization extracting building information from the data, build-
purposes. While the cadastral map only shows building ings were classified in terms of building type and settle-
size and location, which are indispensable for the ment structure parameters such as physical density at
analysis of settlement structure, official statistics high spatial resolution are calculated. Combining
provide additional information on buildings. Statistics this information with typical dwelling and resident
on buildings and dwellings in Germany are published density figures for the given building types could

469
Meinel et al.

provide indicators to the distribution of dwelling units limitations in accuracy due to the generalization of
and population. building plans, there is little if any distinction between
mixed domestic and non-domestic uses, and the build-
A second important objective was to develop an exten- ing height as well as the number of floors can only be
sion of the method to provide information about the estimated implicitly with the building type.
settlement development and the sustainability aspects
for time series. In recent years, sustainability goals The calculation of settlement structure parameters
have been repeatedly formulated by sectoral planners based on building ground plans uses analogue topo-
and politicians at various spatial levels, although nor- graphic maps (E0.04/km2) or digital ground plan
mally their attainment has proved difficult to assess. layers of topographic maps (E0.44/km2) and the settle-
One example is the implementation of guidelines for ment layer of the Digital Landscape Model (DLM) of the
sustainable urban development such as ‘development Amtliches Topographisch-Kartographisches Informa-
inside before development outside’, which has tionssystem (Authoritative Topographic-Cartographic
already been put into practice in some initiatives Information System, ATKIS1) in the basic version
(Federal Government of Germany, 2003). The policy (E2.63/km2) (ADV, 2008a, 2008b). Although the
aim is to achieve a 3 : 1 ratio between brownfield recording of buildings from aerial and satellite photo-
sites and greenfield sites in terms of consumed land. graphs has been the focus of recent photogrammetric
This requires every new building to be recorded, its research, data requirements are extremely stringent and
location in the inner or outer zone to be identified, reliable automatic building digitization (see above) can
and physical development to be reviewed and evalu- still involve a great deal of manual finishing.
ated while taking all factors into account.
There is much to be said for the use of topographic
The third aim was to implement the methods as a tool maps, since they are very inexpensive and are also
within a GIS environment and the results can be pro- available for earlier periods. A comparison of possible
vided in the form of layers as a basis for settlement scales led to selection of the topographic map with a
and infrastructure planning support. The obtained scale of 1 : 25 000 (TK25) as the optimal database
data can also be associated with statistical key (ADV, 2008c). This map series shows building
figures. In contrast to countries such as Scandinavia, ground plans almost completely and only slightly gen-
Austria, and Switzerland (e.g. European Commission, eralized. Partially generalized buildings (minimum
2008; Nippel and Klingl, 1998; Steinnocher, 2006; size, simplification of ground plans, inclusion of subsidi-
Wonka, 2006), such approaches of spatial disaggrega- ary buildings in the main building where the distance
tion have not yet been fully developed in Germany. between them is minimal) and slight displacement (dis-
placement through broader marking of the road
With this in mind, the present work can contribute to network) pose no problem for medium-scale studies.
the valorization of geographic base data that have A smaller-scale map with a scale of 1 : 10 000 shows
been previously used only for visualization and which building ground plans with greater precision, but build-
are insufficient for analysis purposes. Furthermore, ing extraction proves to be much more difficult, since
the presented method is intended to be used for a the buildings are shown in outline and not in full tone
nationwide application within a monitoring system as in the TK25. Meinel et al. (2008a) and Meinel
of settlement and open space development (Meinel et al. (2007) go into greater detail about the choice of
and Siedentop, 2007). the database, including a comparison of data costs.

The TK25 is available nationwide in Germany for

historical time periods as an analogue map and in
Base data cartographic form with relatively few changes with
The presented method is designed for users who do not regards to the building stock in maps that date back
have access to spatial building databases. Therefore, to the early 20th century. Present time slices, and
the procedure should be applicable nationwide and increasingly earlier slices, are now being provided by
for any time series. Some existing more precise data land surveying offices in the form of digital topo-
were excluded as input data because they either do graphic raster maps (DTK25-(V)) at an image resol-
not comprehensively exist or are available only at ution of 508 dpi (1.25
1.25 m in nature). More
high cost (three-dimensional city models, LIDAR detailed studies on the quality and availability of
data) or are not publicly available (such as the real these data are described in Meinel et al. (2008b).
estate register). Spatial base data like topographic
maps as input data are either inexpensive or freely Apart from the representation of buildings, it is also
available for management and planning. They allow a useful to show the spatial demarcation of blocks. For
broad and comprehensive application of the procedure. example, the parameters calculated on the basis of
However, when using topographic maps, a number buildings can be aggregated with the next highest
of limitations must be acknowledged. There are spatial level. The smallest spatial unit is the block,

470
 Analyzing building stock using topographic maps and GIS

which is bounded by a set of streets, open space, or a Urban Affairs (BMBVS) in 2006 (Meinel et al. 2008a).
land-use change along a parcel border. Within blocks Buildings are selected automatically by means of
of buildings there is generally a dominant use and digital image processing in a multistage procedure.
type of building, therefore it is appropriate to use the Traffic and boundary lines are then removed by means
block as a spatial unit for the calculation of settlement of morphological operations (opening). Commonly
structure parameters. used optical character recognition (OCR) software is
unable to remove lettering because fonts, font styles,
The automatic demarcation of blocks on the basis of orientation, and sizes differ from symbol to symbol
topographic maps is still being researched. Methods and the contracted names and abbreviations do not con-
to extract the street network (Bin and Cheong, 1998) stitute closed texts. A special object-recognition system
and road intersections (Chiang, 2009) from raster for lettering and symbols has been developed that oper-
maps might be helpful to obtain the geometry of the ates with image pyramids (hierarchies of image resol-
building block. Further work needs to be done. As a utions). Lettering is removed by automatic parameter
result, the boundaries of blocks and the block use analysis of all features in the segmented binary image
types are taken from the Digital Landscape Model with respect to morphological properties like compact-
(Basic DLM) 1 : 25 000 of the ATKIS, which is among ness, convexity, anisometry, and orientation. Special
one of the most important sources of spatial base data structural element parameters are used at various
in Germany (ADV, 2008b). Using the boundaries of image resolutions for analysing symbol objects, since
blocks has the advantage that building extraction can many symbols are contained only in fragmented form
be limited to this area and thus can save processing in the result layer, which prevents differentiation.
time and the block use types helps to differentiate
non-residential buildings in the later classification Remaining features like churches, towers, mines, trans-
process. In the object group ‘built-up area’ (2100), the former stations, etc. are subtracted from the original
basic DLM distinguishes between the following use image. The varying quality and peculiarities of the
types: ‘residential area’ (2111), ‘industrial and commer- DTK25-V produced by the different states in Germany
cial area’ (2112), ‘mixed-use area’ (2113), ‘buildings of cause problems in determining image-processing para-
particular importance’, such as schools, hospitals and meters. The few buildings that are not, or only partly,
office buildings (2114), and leisure amenities (2202). mapped for cartographic reasons, as a result of the
All the buildings within these areas have to be extracted, removal of background for lettering in the DTK25,
surveyed, and classified. cannot necessarily be reconstructed. Since the advanced
image processing operations used are not implemented
In order to determine dwelling and resident numbers in GIS (e.g. ArcGIS1) or remote sensing image processing
typical for buildings, existing, high-resolution (block- software (such as ERDAS IMAGINE1), the image analy-
related) intra-municipal data from Dresden and Bonn sis software HALCON1 (http://www.mvtec.com) was
were used. For the regional adjustment of the reference used as an alternative. Buildings are extracted with an
density values to communities of varying physical average reliability of about 96%. Four per cent of build-
density, recourse was taken to statistics on dwellings ings are not recognized and features that do not represent
and buildings. This permitted the adjustment of refer- buildings are incorrectly classified. This level of extrac-
ence density values for the settlement structure area tion quality, which depends strongly on the quality of
types introduced by the Bundesamt für Bauwesen und the digital raster maps, suffices to determine settlement
Raumordnung (BBR, Federal Office for Building and structure parameters at a medium resolution.
Regional Planning).

Building classi¢cation
Building extraction from the 1: 25 000 After extraction, buildings are shown as unclassified
topographic map polygons. The aim is to assign the entire building
The heart of the method is the building extraction from stock to defined types of building. Table 1 shows the
the ground plan layer (also called the settlement layer, hierarchically structured, two-level building typology
one of four layers, the others being vegetation, water selected. At the main level (level 1), three classes of
bodies, and isohypses) of the DTK25-V. In this raster, residential building and one class of non-residential
all the substantive elements of the TK25 are shown in building are roughly distinguished in terms of their
black, displaying buildings, traffic, vegetation symbols physical characteristics: perimeter and linear develop-
and lettering elements merged binarily. The building ment in the multi-family housing stock, smaller-scale
extraction was developed within the framework of the structures in the single-family housing stock, and
research project ‘Automatic Derivation of Urban irregular structures in non-residential uses. At level 2,
Land-Use Pattern and Integration in a Geographical more detailed differentiation takes account of the
Information System (GIS)’, which was funded by the structural size of the individual building complexes.
German Federal Ministry of Transport, Building and This pragmatic building typology is adapted to the
471
Meinel et al.

Table 1 Building typology

Level 1 Level 2 Pattern of extracted Median number of Median number of £oors

buildings £oors in communities in larger core cities in
in counties of low heavily built-up
density in rural areas agglomerations

Multi-family houses Multi-family houses (closed 2.2 3.8

(in perimeter coverage type)
development)

Detached multi-family 2.3 3.8

houses (open coverage
type)

Multi-family houses Multi-family house (linear 2.9 3.8

(open coverage slab development)
type)

Multi-family building (linear 5.1 5.3

prefab slab development)

High rise . 50 m ^ 11.4

Single-family Detached and semi- 1.6 2.0

housing detached family houses

Terraced houses 1.6 2.0

(Table continued)

472
 Analyzing building stock using topographic maps and GIS

Table 1 Continued

Level 1 Level 2 Pattern of extracted Median number of Median number of £oors

buildings £oors in communities in larger core cities in
in counties of low heavily built-up
density in rural areas agglomerations

Village traditional house 1.6 2.0

Non-residential uses Industry/commerce 1.0 1.0

Buildings of special 2.2 3.8

importance (e.g.
administration, health/
social services,
education/research,
culture, etc.)

information content of the maps and permits relatively The building classification proceeds in two stages. The
high-quality automatic classification. first involves classification of single buildings in terms
of form (five form types: very small, small, slab
Using automatic, rule-based classification, the entire shaped, very wide or complex buildings) using building
building stock was surveyed with regard to form, attributes such as area, circularity, and aspect ratio of
location, and orientation (e.g. building area, circum- the outlined rectangle. At this level no neighbourhood
ference, compactness, distance from neighbouring or block-related parameters are used. The form type of
buildings, distance from block boundary, etc.). These a building is defined simply by building-related par-
attribute values were then related to the building ameters. However, this does not allow the unambigu-
geometry. ous assignment of buildings to the defined class
(Table 1). Only in the second stage of classification
A rule-based decision-making support network was can conclusions about the building type be drawn on
developed for classifying buildings. It showed great the basis of neighbourhood relations (e.g. distances,
efficiency in data processing with good flexibility densities), block-related parameters (e.g. number of
in modifying the rule base. The elements of a buildings), the land use from ATKIS object class (e.g.
rule-based system are rules comprising conditions residential area, commercial area) and the calculated
(premises) and actions (conclusions). Buildings and parts of the building form types from the first stage
their characteristic parameters are classified by within each block. A total of 61 indicators are used
means of ‘if – then’ rules in conjunction with statistical in the decision support network, composed of 17 build-
threshold values. An example of a very simple rule in ing-related indicators and 46 block-related indicators.
classifying single-family houses is to select all buildings
with a footprint of under 200 m2. The thresholds for Figure 1 shows an excerpt of the results of building
the classification tree could be determined by statisti- extraction and classification for a section of Dresden.
cal analysis of the building measurements (mean, Non-residential buildings (industry/commerce and
standard deviation of the features) by use of training buildings of particular importance) are defined by the
data. Afterwards the thresholds were optimized with underlying use of the block (object classes 2112
the aid of training data by means of an error analysis and 2114), whereas all residential buildings in residen-
(trial and error). The training data were obtained from tial and mixed use areas (ATKIS object classes 2111
either digital building data or visual inspections of and 2113) are classified in eight residential building
aerial photographs. types.

473
Meinel et al.

Currently, additional machine-learning methods are

tested as in the studies of Steiniger et al. (2008) in
order to derive urban building structures for generaliz-
ation purposes.

Calculation of settlement structure

parameters
The classified building stock is the basis for calculating
base data on settlement structure. The following par-
ameters are fully and automatically mapped onto a
reference geometry of choice, e.g. a regular raster or
a given block geometry.

Urban block type

Block type (eight residential and two non-residential)
is determined on the basis of the buildings making
up the block. The structure type is determined by
the dominant building type, which is also a criterion
in visual aerial photograph interpretation. Mixed-use
blocks are found relatively often and they can be
identified and labelled by evaluating the number of
different building types (as heterogeneity criteria).
Figure 2 shows the principle of automated block
typification.

For the cities under study, Dresden and Bonn, the accu-
racy of classification was validated. These cities have
been chosen because of their exceptional reference
data (70 000 reference buildings) obtained from
digital building data and visual interpretation of ortho
imagery. In comparison with the reference, a classifi-
cation accuracy of 76% for urban blocks was attained.
While a good classification quality was reached for
multi-family houses (closed coverage type; 94%),
single-family and semi-detached housing development
(91%), and high-rise development (80%), the rate of
misclassification for morphologically similar structures
like terraced and row housing and slab block housing is
still high. Research is being conducted in order to solve
this problem, which primarily deals with the quality of
building representation.

Number of buildings/density of development

The number of buildings and where they belong can be
clearly determined by means of calculated centroids.
The density of development (number of buildings per
hectare) provides quantitative information on physical
density. However, ‘attached’ buildings like row and
terraced housing, slab blocks, and closed block devel-
opment are represented by only one centroid. By
taking into account individual total building lengths
Figure 1 (a) Ground plan layer of the DTK25-V; (b) result of in conjunction with characteristic distances between
building extraction; and (c) result of building classi¢cation in a building entrances, conclusions could be drawn about
section of Dresden the number of single buildings.
474
 Analyzing building stock using topographic maps and GIS

Building footprint/building footprint density

The building footprint (m2) is the sum of all building
footprints within the reference areas. In relation to
block area, which describes the share of the building
footprint in total block area, this can be described as
building footprint density (m2/m2). It can be compared
with the planning law concept of plot ratio, which
relates to plot size. Standard additions for sealed sur-
faces like roads and courtyards permit a first approxi-
mate estimate of the sealing rate for building footprint
density.

Floor area/£oor area ratio

Gross floor area is the sum of all building footprints
multiplied by the median number of floors (m2) for a
given reference area. For each building class a regional
defined constant number of floors was ascribed (Table
1), deduced by building statistics. Due to the strong
correlation of the number of floors with the density
and centrality of land-use patterns, 17 different types
of municipalities were chosen to determine their corre-
lation. All different types of municipalities were con-
tinuously classified by the Bundesamt für Bauwesen
und Raumordnung (BBR) (2007) according to their
central-place function (low- and middle-order centre),
their population (large, medium-sized, small towns,
rural municipality) and their land-use pattern.
Figure 3 shows the result of calculated block-related
floor area ratio for a section of Dresden. In addition,
dwelling size can be estimated on the basis of floor
area ratio in combination with the estimated number
of dwelling units.

Median number of £oors

The median number of floors represents the weighted
mean of the number of floors of all buildings taking
into account the area occupied by each building
type.

Building volume/building volume density

Building volume (m3) is the sum of the volume of all
buildings and is calculated on the basis of the building
footprint and regionally differentiated building type-
related floor numbers (see floor area ratio) and
height. Building volumes are especially interesting for
material flow modelling. Building volume density
(m3/m2) describes the ratio of building volume to the
reference area and is an indicator of the quantification
of development densities.

Dwelling number/dwelling density

Dwelling number can be estimated with building
type-related density parameters. To calculate these
Figure 2 (a) Classi¢ed buildings; (b) classi¢ed blocks; and (c) typical density parameters, the blocks that displayed
ortho-photograph of a section of Dresden a largely homogeneous building type stock were
475
Meinel et al.

Figure 3 Map of the £oor area ratio for a section of Dresden

chosen from the block-related reference data inven- area of reference area. A detailed study can be found
tory where the number of dwellings was known. It in Meinel et al. (2008a).
proved possible to establish a relation between the
number of dwellings and the sum of building foot-
prints in the block for each building type, and to Resident number/resident density
derive statistically secured building type-dependent Along with estimation of dwellings, the building-
density parameters. The resulting building-specific related and, subsequently, block-related number of
reference value for dwellings per footprint area residents can be estimated on the basis of residents
density (dwellings/m2) is linked in the calculation per footprint area density (residents/m2) and corrected
process to typified buildings. Summing all dwellings with the aid of the next known statistical unit. Local
in the block gives the block-related number of units dwelling vacancies are, however, extremely difficult
and the dwelling density (1/ha). Furthermore, the esti- to model and can only be done with highly accurate
mated values for dwelling and resident population can regionalized parameter adjustment. Deviation from
be subsequently combined with the next known area the initial estimate (without adjustment) for residents
unit with available statistics (generally the local auth- was þ10.4% compared with the reference value.
ority or possible parts thereof). Through a comparison Notably, the vacancy rate, which is implicit in the
of the initial estimate with the reference, a factor can reference data, will be equally distributed to all build-
be obtained in order to correct the building type- ings in a statistical unit.
related density parameters directly. Differences
within the building classes and sizes can thus naturally There are a number of limitations to the current
not be resolved. For the study areas, block-based stat- method. The accuracy of the settlement parameters dis-
istics of dwelling and resident number were available cussed above is strongly dependent on the accuracy of
as reference data from KOSTAT (http://www. the building classification. While previously the evol-
staedtestatistik.de/kostat.html). KOSTAT provides ution of residential building stock was in focus, the
cooperation between municipal statistical offices in industrial and commercial building stock is not yet dif-
Germany as well as data for 101 cities in Germany. ferentiated in the classification process. All buildings
For the Dresden study area, deviation from the located in ATKIS blocks (object class 2112 or 2114)
initial estimate (without adjustment) for the number are assigned to non-residential buildings. For buildings
of dwellings was þ1.1% compared with reference with great building depth in a mixed-use area (object
statistics. This error increases with the decreasing class 2113), it was assumed that these are of
476
 Analyzing building stock using topographic maps and GIS

Figure 4 Result of SEMENTA,17 GIS layers in an automatically generated ArcMAP project

commercial/industrial use because such great depths intuitive user interface. In the menu, the input data
cannot be found in residential buildings due to the (DTK25 and object classes 2111 –2114 of the ATKIS
demand for natural lighting. Dealing only with two- basic DLM) can be selected and the programme par-
dimensional morphological information, there are ameters residents per footprint area density (1/m2),
limitations in detecting non-domestic or mixed-use in dwellings per footprint area density (1/m2), and the
buildings (e.g. stores or offices in ground floor). median number of floors and heights (m) of the
Another problem concerns the unknown vacancy rate defined building types can be modified. The complex
in domestic buildings, which is equally distributed to processing not only produces output files in ESRI
all residential buildings. Additionally, the assumption Shape1 format, but also generates a Microsoft Excel1-
of regionalized standard uniform heights for all build- compatible statistical file (CSV file) and automatically
ings of one type can be grossly inaccurate in some cases opens an ArcMap project that visualizes the planning-
(e.g. central city blocks). This can lead to over- or related result layers (urban block type, dwelling
underestimations of block-based floor area ratio, resi- number, etc.). Figure 4 shows an example: classified
dent number, dwelling units or building volume. buildings together with the derived urban block types.

On the basis of the data, a three-dimensional represen-

Method implementation tation of the building and block classification can be
The fully automated processes described for pre-proces- generated by combining the results with a three-dimen-
sing, image processing (building extraction), object sional modelling software (e.g. ESRI ArcScene).
surveying (parameter calculation), classification, and
deriving settlement structure parameters have been
implemented in the form of a toolbar for ArcMap Building-based settlement development
under ArcGIS (ESRI). The ArcGIS software has the For ex-post analysis, spatial information concerning
greatest distribution and is used extensively in planning building construction and wrecking over the last
and regional administration in Germany. This decades is required. In the framework of the research
SEMENTA1 (SettlementAnalyzer) toolbar has an project (REFINA, 2008), a method was developed
477
Meinel et al.

for a multitemporal analysis of topographic maps at a Table 2 Typical examples of building stock development
scale of 1 : 25 000 to detect changes in the building
stock and the settlement area. Firstly, the method Change in building stock /time slice T1 T2 T3 T4 T5
described above was used for maps of various time
slices. Comparison of the settlement structure par- No change, building in existence before T1 1 1 1 1 1
ameters calculated showed that the influence of the Demolition between T3 and T4 1 1 1 0 0
quality of the map sheets representing various time
New construction between T3 and T4 0 0 0 1 1
slices was too great to attain robust results across
time. For example, the size of buildings in the maps Demolition between T2 and T3 and new 1 1 0 0 1
construction between T4 and T5
was altered by up to 50% through new cartographic
representation or later scanning of originals with a
Note: 0, Not existent in time slice; 1, existent in time slice.
lower binarization threshold. For this reason, the pro-
cedure was supplemented for a multitemporal appli-
cation by a building linkage process described below. raster maps of historical time slices can thus automati-
cally be rectified linearly with reference to the latest
Older editions of topographic maps are available only DTK25-V using about 1000 identical points.
in analogue form and for the purpose of comparative
analysis have to be homogenized with respect to map Even after map rectification, the extracted buildings are
representation, data structure, and geometric location. not entirely congruent due to differing map represen-
The latest cartographic material served as reference, tations, scanning conditions, and residual geometric
since it can be assumed that this will provide the best filling. Therefore, a vertical building linkage method
quality of representation and position preservation. was developed in order to describe the life span of the
Initially, a black layer has to be extracted from a building. The iterative building linkage follows on the
colour scan of older maps and binarized. This is fol- basis of the latest time slice Ti. All buildings of the
lowed by high precision position adjustment, because next oldest time slice Ti – 1 are entered into the most
identical buildings have to be congruent in maps recent layer where no building exists in this location
showing successive time periods. The IMAGINE Auto- (e.g. after demolition). A structure is assumed to exist
Sync1 programme was used for this purpose under the where buildings overlap. The next step, on the basis of
image-processing software ERDAS IMAGINE, which this layer, is to enter the buildings of the next oldest
permits the automatic rectification of image data time slice Ti – 2, including attributes up to the oldest
using matching algorithms (AutoSync, 2006). Binary time slice T1. The relevant attributes are defined by

Figure 5 Settlement area in 1984 compared with 2005 ^ example: Xanten

478
 Analyzing building stock using topographic maps and GIS

means of a geometric 1 : 1 combination of the buildings information about the life span of buildings, including
in the largest overlapping area. Where a building in the building size, building type, etc (Table 2). The building
older time slice is represented by several building poly- type determined by SEMENTA was corrected by
gons, only the attributes of the polygon with the greatest means of a plausibility test through assignment of the
overlapping area are transferred. It should be noted here most frequent type of building over time.
that tracking of the evolution of the built form over time
was not the focus of this work, but it only can be ana- For older time slices (before 2000), for which there are
lysed for buildings with a 1 : 1 linkage by measuring generally no ATKIS data, urban blocks are projected.
the changes in building size. This allows ‘initial development’ and ‘infill develop-
ment’ to be distinguished on a block-and-building
The result is an analytic layer in which all buildings basis. Figure 5 shows the projected settlement areas
that have ever existed are described with the latest of 1984 (block with at least one building) compared
geometry. The attribute table of the layer provides with 2005.

Figure 6 Work£ow of automated settlement evolution analysis

479
Meinel et al.

‘Initial development’ is identified where a building is maps of radiation density, which show the extent of
constructed on a block still undeveloped at the time, danger zones in the vicinity of transmission installa-
while a new building assigned to an already developed tions, are becoming increasingly established. By over-
block is classified as ‘infill development’. This differen- laying the hazardous areas with SEMENTA results,
tiation offers a good basis for estimating inner and supplementary information for disaster planning and
outer development. The retrospective projection of hazard prevention can be obtained.
blocks will no longer be necessary in the future with
reliably updated digital spatial base data available for
each recent time slice, since each new block can then Infrastructure planning
be classified as outer development. After building Under conditions of shrinkage (Deilmann et al., 2009),
assignment and correction of building type, robust funding for the maintenance, expansion or downsizing
data are available for a differentiated appraisal of of services of general interest will demand much more
new buildings, demolitions, and significant conver- precise and sustainable planning in the future. This
sions. The procedure for building-based analysis of includes all spending on the technical infrastructure –
settlement development is described in detail by telecommunications, gas, water, electricity, and
Hecht et al. (2008). Figure 6 shows the workflow of waste – and the social infrastructure – education,
the proposed approach. health, recreation, and supply. For concrete planning,
fine-grained information on population, housing, and
building densities is indispensable. The fine-grained
Applications settlement structure parameters generated by the
procedure can be used in planning essential services,
For medium-scale, supra-local planning, various appli- school networks, and in calculating the catchment
cations are conceivable. Some are briefly discussed in areas of health facilities and pharmacies. The results
this section. can also be used in public transportation planning to
optimize stops and station networks or to model the
interaction of land use and transportation (Lautso
Tra⁄c route planning et al., 2004). In this case disaggregated raster-based
A settlement structure analysis for major planning representation of land was used for the modelling
projects and traffic route planning is possible with (Spiekermann, 2003). Furthermore, the classified build-
comparatively little effort. Land-use intensity can ings can be used for material flow modelling (e.g. Schil-
now also be estimated automatically with great pre- ler, 2007). Finally, small-scale maps of residential
cision. This considerably increases efficiency, replacing building, dwelling and resident density are of growing
estimates of parameters with real data. The calculation interest in elaborate energy concepts.
of morphological parameters, for example, the classifi-
cation of a city in density zones, could be standardized
and automated in the future as well. This is the case State, regional, and local planning
both for functionally structural parameters such as In these three areas the approach is particularly
type of building development, type and intensity of useful, since very fine-grained base data can be gener-
use, building spacing, distance between frontage and ated with ease and rapidly for very substantial areas.
street axis, and the number of persons affected, and On the basis of the parameters calculated, compre-
for morphological parameters such as community, hensible, comparable, and objective spatial demar-
village, and location typifications. The method would cation and typification of settlement areas could be
enable the calculation of other parameters: minimum achieved. One example includes the spatial classifi-
and median building distance from the street axis, cation of urban regions into urban core, central
urban canyon width ratio (for calculating noise and location, peripheral location, and the outer zone
pollution), proportion of infill sites (share of undeve- with the aid of density parameters. The precise
loped frontage/street section length), and the number spatial demarcation and typification of village devel-
of frontage residents (important for traffic noise opment structures in terms of a typology would also
calculations). be possible.

Hazard cadastre/hazard maps Geomarketing

Hazard maps are becoming increasingly important for Geoinformatics has long since become part the
risk and disaster management and to inform the public. approach for optimizing economic and sales concepts.
Their preparation is to some extent already required by The basis includes very detailed data on residents and
state and federal legislation. For instance, flood risk dwellings from a range of private data providers (e.g.
areas are required to be designated by the Saxon Microm, GFK Macon). The high price of such data
Water Act (SächsWG). A European Union flood Direc- reflects both the considerable effort required for cap-
tive is being prepared. Apart from flood hazard maps, turing them and the appreciation of users. Some of
480
 Analyzing building stock using topographic maps and GIS

the available data can be generated more precisely by ATKIS data is about six times higher than that for
the method developed and, owing to the high degree raster-based DTK25 data.
of automation, are much more affordable.
The combination by means of SEMENTA building and
Overall, the method offers a multitude of applications. block-related parameters with other data such as build-
For the first time, the settlement structure can be pre- ing addresses, vacancy information, street data, and
cisely described by means of a system of quantitative industrial information can be achieved by means of
parameters through fully automatic surveying and GIS, and is extremely useful, extending the field of
typification of the building stock. This fundamentally application for the programme. The results of a
improves the information situation for planning and multi-temporal application of the method for different
decision-making processes. Spatially and substantively time slices could enable the implementation of plan-
imprecise assumptions and generalizations, as well as ning goals through comparison with older planning.
costly field surveys, which in effect are possible only For the first time in Germany, a foundation has been
for small areas, can be replaced by the method. created for taking stock of the ratio of inner develop-
ment to outer development that does not operate
with labour-intensive aerial photographic interpret-
ation. Thus, technical support can be provided for
Prospects the national strategy on sustainability with its territor-
Under conditions of shrinkage, sustainable infrastruc- ial policy goals.
ture planning requires detailed information on settle-
ment structure. The spatial resolution of local Currently, the programme is only applicable to
authorities and the statistical districts of cities, cur- German topographic maps. Applications on topo-
rently the smallest statistical territorial units in graphic maps of other countries with a similar scale
Germany, are not sufficiently high for this purpose. are possible, yet require special adjustments to the
Since raster-based regional statistics are still a thing characteristics of the respective topographic maps
of the distant future in Germany, a method for and the other underlying building typology. In the
spatial disaggregation through the combination of future, the adaptation of the SEMENTA programme
spatial base data and statistical data is presented. As to various national spatial base data in Europe is
a result, block and raster-based, high-resolution settle- planned. In addition, the methodology can also be
ment structure parameters are available. implemented in other GIS software packages.

Against the background of European application,

administrative areas are not suitable for spatial analy-
sis. With the GEOSTAT project there is a proposal for References
a Eurogrid system that provides a hierarchy of a stable, ADV (2007) Richtlinie über Gebühren für die Bereitstellung und
neutrally harmonized system of grids that serve as a Nutzung von Geobasisdaten der Vermessungsverwaltungen
foundation for spatial analysis across all disciplines der Länder der Bundesrepublik Deutschland vom 11.
(Backer, 2008). Therefore, the aim is to get an Dezember 2007 (available at: http://www.adv-online.de/
extdeu/binarywriterservlet?imgUid¼95e7590f-9498-b11a-
additional calculation, on a regular raster cell grid, 3b21-718a438ad1b2&uBasVariant¼22222222-2222-2222-
with INSPIRE-compliant data specifications defined 2222-222222222222) (accessed on 9 March 2009).
in INSPIRE (2007). ADV (2008a) ATKISw-Objektartenkatalog für das Basis-DLM.
Teil D1 der ATKISw-Gesamtdokumentation (ATKISw-Basis-
OK) (available at: http://www.adv-online.de/) (accessed on
The method has great potential, since it is based solely
9 March 2009).
on generally accessible, reliably updated data. By using ADV (2008b) Produktblatt Digitales Basis-Landschaftsmodell
spatial base data that are prepared and kept up to date (Basis-DLM) 2008. Produktblatt ATKIS – Digitales
at great expense and effort by surveying authorities, Geländemodell (available at: http://www.adv-online.de)
the value of this data can also be exploited for planning (accessed on 9 March 2009).
ADV (2008c) Produktblatt DTK25 (Digitale Topographische
processes. In the future, sectoral planning should exert Karte 1 : 25 000). Produktblatt DGM 2008 (available at:
greater influence on the further development of spatial http://www.adv-online.de) (accessed on 9 March 2009).
base data; these data are amenable to the inclusion of AutoSync (2006) IMAGINE AutoSyncTM White Paper (available at:
relevant sectoral data by nature. The increasing avail- http://gi.leica-geosystems.com/documents/pdf/IMAGINE
ability of map sheets in the new design (derived directly AutoSyncWhitePapeFeb06.pdf) (accessed on 25 May 2008).
Backer, H. (2008) The Geostat Project Proposal (Version
from ATKIS base DLM) considerably enhances the 20081103), in L.H. Backer (ed.) GEOSTAT Initiative Euro-
precision of results, buildings are mapped more accu- pean Forum for GeoStatistics, Stockholm (available at:
rately, and, above all, stored in a separate layer, http://www.stat.si/forumGs/Project%20proposal_partners
which avoids complex and problematic building %2020080929.pdf).
Bin, D. and Cheong, W.C. (1998) A system for automatic extrac-
extraction. The exclusive use of the ATKIS base tion of road network maps, in Proceedings. IEEE Inter-
DLM will be possible once it covers the entire building national Joint Symposia on Intelligence and Systems, (Cat
stock. However, the price of the necessary vector-bases No. 98EX174), IEEE, Rockville, MD, pp. 359–366.

481
Meinel et al.

Bundesamt für Bauwesen und Raumordnung (BBR) (2007) Lau- Meinel, G., Hecht, R., Herold, H. and Schiller, G. (2008a) Auto-
fende Raumbeobachtung -Raumabgrenzungen (available at: matische Ableitung von stadtstrukturellen Grundlagendaten
http://www.bbr.bund.de/nn_103086/DE/Raumbeobachtung/ und Integration in einem Geographischen Informationssys-
Werkzeuge/Raumabgrenzungen/Siedlungsstrukturelle tem, Forschungen Vol. 134, 2008, Bundesamt für Bauwesen
Gebietstypen/Gemeindetypen/gemeindetypen.html) (accessed und Raumordnung (BBR), Bonn.
on March 2009). Meinel, G. Herold, H. and Hecht, R. (2007). Gebäudebasierte,
Chiang, Y., Knoblock, C.A., Shahabi, C. and Chen, C. (2009) vollautomatische Erhebung und Analyse der Siedlungsstruk-
Automatic and accurate extraction of road intersections tur – Grundlage für ein Monitoring und die Bewertung der
from raster maps. Geoinformatica, 13(2), 121–157. Siedlungsentwicklung, in Schrenk, M., Popovich, V.V. and
Deilmann, C., Effenberger, K.H. and Banse, J. (2009) Housing Benedikt, J. (eds) Proceedings of REAL CORP 2007,
stock shrinkage: vacancy and demolition trends in Germany. Vienna, Austria, 20 –23 May 2007, pp. 815– 824.
Building Research & Information, 37(5/6), 658–666. Meinel, G., Knop, M. and Hecht, R. (2008b) Qualitätsaspekte
European Commission (2008) EUROSTAT – Conference on und Verfügbarkeit digitaler Geobasisdaten in Deutschland
Modern Statistics for Modern Society, Luxemburg, 6– 7 unter besonderer Berücksichtigung des ATKISw Basis-DLM
December 2007, Office for Official Publications of the und der DTK25(-V). Photogrammetrie – Fernerkundung –
European Communities, Luxemburg. Geoinformation, 1/2008, 29–40.
Federal Government of Germany (2003) Perspectives for Meinel, G. and Siedentop, S. (2007). Erhebung und indikatoren-
Germany – our strategy for sustainable development. gestützte Bewertung der Siedlungsstruktur und ihrer
Paper presented at the 2nd International Conference on Entwicklung – Konzept ‘Deutschlandmonitor Siedlungs-
Early Warning, Bonn, Germany (available at: http://www. und Freiraumentwicklung’, in Strobl, J., Blaschke, Th. and
ewc2.org/upload/downloads/national_strategy_germany.pdf). Griesebner, G. (eds) Angewandte Geoinformatik 2007.
Frischknecht, S. and Kanani, E. (1998) Automatic Interpretation 19th AGIT-Symposium, Salzburg, Austria, pp. 473 –481,
of Scanned Topographic Maps: A Raster-Based Approach. Wichmann, Heidelberg.
Lecture Notes in Computer Science No. 1389/1998, Nippel, T. and Klingl, T. (1998) Swiss Land Use in the European
Springer, Heidelberg. Context – Integration of Swiss Land Use Statistics with
Hecht, R., Herold, H. and Meinel, G. (2008) Gebäudescharfe CORINE Land Cover, Swiss Federal Statistical Office &
Analyse der Siedlungsentwicklung auf Grundlage Swiss Agency for the Environment, Neuchâtel.
mittelmaßstäbiger Karten, in J. Strobl, Th. Blaschke and G. REFINA (2008) BMBF Research for the Reduction of Land Con-
Griesebner (eds) Angewandte Geoinformatik 2008. Beiträge sumption and for Sustainable Land Management, REFINA
zum 20. AGIT-Symposium Salzburg 2008 Wichmann, (available at: http://www.refina-info.de/en/) (accessed on
Heidelberg, pp. 11–17. 25 June 2009).
INSPIRE (2007) Drafting Team ‘Data Specifications’ 2007. Defi- Salat, S. (2007) Energy and bioclimatic efficiency of urban
nition of Annex Themes and Scope, Eurostat, Luxembourg morphologies – towards a comparative analysis of Asian
(available at: http://www.epsiplus.net/content/download/ and European cities, in Proceedings of the International
3952/49703/file/inspireDataspecD2_3v2.0.pdf). Conference on Sustainable Building Asia (SB2007 Seoul),
Koç San, D. and Turker, M. (2007) Automatic building detection Seoul, Republic of Korea, 27– 29 June 2007.
from high resolution stereo satellite images. Paper presented Schiller, G. (2007) Urban infrastructure: challenges for resource
at the Conference on Information Extraction from SAR and efficiency in the building stock. Building Research & Infor-
Optical Data, with Emphasis on Developing Countries, mation, 35(4), 399–411.
Istanbul, Turkey, 16– 18 May 2007 (available at: http:// Shan, J. and Lee, D.S. (2005) Quality of building extraction from
www.isprs2007ist.itu.edu.tr/39.pdf). IKONOS imagery. Journal of Surveying Engineering, Amer-
Lari, Z. and Ebadi, H. (2007) Automated Building Extraction ican Society of Civil Engineers, 31(1), 27–32.
from High-Resolution Satellite Imagery Using Spectral and Spiekermann, K. (2003) The PROPOLIS Raster Module.
Structural Information Based on Artificial Neural Deliverable D4 of PROPOLIS, Spiekermann & Wegener
Network, ISPRS Commission III, WGIII/4, Hannover, Urban and Regional Research, Dortmund.
(available at: http://www.ipi.uni-hannover.de/fileadmin/ Steiniger, S., Lange, T., Burghart, D. and Weibel, R. (2008) An
institut/pdf/Lari_Ebadi.pdf). approach for the classification of urban building structures
Lautso, K., Spiekermann, K., Wegener, M., Sheppard, I., Stead- based on discriminant analysis techniques. Transactions in
man, P., Martino, A., Domingo, R. and Gayda, S. (2004) GIS, 12(1), 31–59.
PROPOLIS: Planning and Research of Policies for Land Steinnocher, K. (2006) Linking remote sensing and demographic
Use and Transport for Increasing Urban Sustainability. analysis in urbanised areas. Paper presented at the 1st Work-
PROPOLIS Final Report, LT Consultants, Helsinki. shop of the EARSeL Special Interest Group on Urban Remote
Lee, D.S., Shan, J. and Bethel, J.S. (2003) Class-guided building Sensing ‘Challenges and Solutions’, Berlin, Germany.
extraction from Ikonos imagery. Photogrammetric Engin- Wonka, E. (2006) Regionalstatistik in Österreich – Von der
eering and Remote Sensing, 69(2), 143–150. Tabelle zu räumlichen Analysen und Visualisierungen.
Lefèvre, S., Weber, J. and Sheeren, D. (2007) Automatic building Salzburger Geographische Arbeiten, Band 39, Hrsg. J. Strobl.
extraction in VHR images using advanced morphological
operators. Paper presented at the 2007 Urban Remote
Sensing Joint Event, Paris, France, pp. 1–5 (available at:
http://dpt-info.u-strasbg.fr/lefevre/publis/urban2007.
pdf) (accessed on 25 June 2009). Endnote
Maderlechner, G. and Mayer, H. (1994) Automated acquisition
1
of geographic information from scanned maps for GIS IMAGINE AutoSync, ESRI ArcGIS, ESRI Shape, ERDAS
using frames and semantic networks, in Proceedings of the IMAGINE, HALCON and Microsoft Excel, SEMENTA and
12th IAPR International Conference on Pattern Recognition ATKIS are trade names, trademarks and/or registered trade-
(Cat. No. 94CH3440-5), pp. 361–363. marks of the respective owners.

482