Professional Documents
Culture Documents
doi:10.1068/b36009
Jana Hoymann
Institute for Landscape Architecture and Environmental Planning, Technical University of Berlin,
EB 4-2, StraÞe des 17. Juni 145, 10623 Berlin, Germany; e-mail: jana.hoymann@tu-berlin.de
Received 15 January 2009; in revised form 1 December 2009; published online 18 August 2010
Abstract. In this paper a scenario study of the residential land-use development in the Elbe River
Basin is presented. The study uses an approach that empirically determines suitability maps for the
application within a land-use-change model. Recent urbanisation processes are explained in the first
step and are used in simulations of future land use in the second step. Binomial logistic regression
analysis is applied in an analysis of the location characteristics influencing residential land-use
change. Estimation results are adapted and used as weights in the calculation of suitability maps,
which consist of the location characteristics of residential land-use change. Including policy maps
in suitability calculations allows important spatial restrictions to be accounted for and enables the
impact of spatial planning on the allocation of residential developments to be analysed. The suitability
maps are further applied to the Land Use Scanner model to simulate spatially explicit residential
land-use developments in the Elbe River Basin. Results of this study show that empirically determined
suitability maps used in models of land-use change can contribute to the operational use of scenario
studies in political discussion support. Considerable differences in applied policy maps in terms of
their contribution to the sustainable development of residential and use are discussed.
1 Introduction
Sustainable development measures with respect to urban land consumption are purely
quantitative in Germany. Reducing urban land consumption from 120 ha to 30 ha per
day by 2020 is the objective of the German strategy for sustainable development
(RNE, 2004). The supplementary consideration of the spatial configuration of urban
land-use changes (for example, extensive suburbanisation processes), which influence
economic as well as ecological processes, should also be taken into account (Lambin
et al, 2001). For example, the impacts and costs of suburbanisation and sprawl are
reviewed by Gordon and Wong (1985). Christaller's central place theory provides one
way of considering spatial configuration and has been implemented in the national
planning policy to guide the development of urban areas (ROG, 2004). This leads to
the following research questions. Where will residential areas be developed in the future
and what are the determinants of location choice? What are the potential impacts of
the determined central places on the future patterns of observed settlement structure in
contrast to those observed when no spatial policy is implemented? To answer these
questions, a scenario study was conducted for the Elbe River Basin.
Land-use-change models can be applied to the allocation of residential developments.
Extensive reviews with respect to the purpose, scale, and theory of land-use-change models
can be found in Agarwal et al (2002), Briassoulis (2000), or Verburg et al (2004b). In this
study, the Land Use Scanner (LUS) (http://www.lumos.info/landusescanner.htm) is applied
to allocate future developments on grid cells (Hilferink and Rietveld, 1999; Schotten
et al, 2001). The model's ability to simulate urban and nonurban land-use types simul-
taneously and to integrate maps that are considered in the spatial allocation process
makes it appropriate for use in this study. Previous applications of the model include
the impact assessment of land-use changes for water management and flood risk or the
derivation of land-use-change scenarios within the framework of the Fifth National
912 J Hoymann
Physical Planning Report in the Netherlands (Dekkers and Koomen, 2007; Schotten
et al, 2001; Van der Hoeven et al, 2008).
The major challenge of this type of land-use-change model is the calculation of
suitability maps for each land-use type. Weighting and combining the different
determinants of location choice by expert knowledge is a straightforward means of
calculating suitability maps. This approach was applied several times for simulations
with the LUS (Borsboom-van Beurden et al, 2007; Hilferink and Rietveld, 1999).
However, the influence of location factors on the spatial allocation is not easily
determined, which leads to the consideration of alternative approaches. The approaches
reviewed by Verburg et al (2004b) include the recent application of logistic regression
analysis, which offers the opportunity to determine empirically the influence of differ-
ent location factors. Approaches can be categorised into binomial and multinomial
models, as well as into models explaining existing land-use patterns or past land-use
changes (eg Aspinall, 2004; Dendoncker et al, 2007; Loonen and Koomen, 2009;
Pontius and Batchu, 2003; Turner et al, 1996). While many studies use logistic regres-
sion analysis to explain landscape patterns or landscape changes, only a few have
applied the regression results to a future projection (Millington et al, 2007; Pontius
and Batchu, 2003; Wear and Bolstad, 1998). None of these projections takes into
account change in location preferences through time. This is the first study to apply
a logistic regression analysis to determine suitability maps empirically using the esti-
mated coefficients as weights for location factors that are summed up to one suitability
map. In addition, this study is the first that modifies the coefficients with respect to
scenario storylines to account for changes in location preferences and applies these
changes into the land-use-change model LUS. Accounting for these changes through
time allows the feasibility of regional planning instruments to be tested to determine
their influence on the allocation of residential areas.
Figure 1. Study area. The white box indicates the subset presented in figure 3.
and the higher the change in the probability for residential change when the independent
variable changes by one unit (Backhaus et al, 2006).
The regression analysis included three estimations. The first estimation considered
only the land use in neighbouring cells. The second estimation applied only location
factors that account for accessibility and physical characteristics. Finally, a full model
that considered both land use in neighbouring cells and location factors is estimated.
The first estimation is called the autoregressive model and is applied because spatial
dependence of the data is expected. Surrounding land-use types are assumed to have
a substantial influence on land use at a certain location (Verburg et al, 2004a). This
influence was assigned explicitly and therefore accounts for spatial autocorrelation
(Anselin, 1988; Tobler, 1979). The application of a separate autoregressive model
explicitly identified the influence of neighbourhood on residential land-use change.
The second estimation identified the influence of the location factors. All estimation
approaches were applied to residential changes between 1990 and 2000 as in Verburg
et al (2004a). A full set of variables is presented in table 1.
A sample of the database was chosen because the database is very large (containing
2.3 million grid cells) and the proportion of residential land in comparison to all other
land-use types was very disproportionate. Choice-based sample was applied where
sampling rates for categories of the dependent variable were unequal. King and Zeng
(2001) and Prentice and Pyke (1979) showed that estimated coefficients are not biased
by this sampling technique. The total sample size was 8500, consisting of 30% new
residential land and 70% of cells showing no change.
The results were validated by applying an ROC curve (relative operating character-
istic; Pontius and Schneider, 2001). The ROC is a statistic that was used in this present
study to measure the extent to which grid cells with larger probability values were
concentrated at locations that truly became residential areas between 1990 and 2000.
The ROC statistic measures the area under the curve with values from 0.5 (completely
random) and 1 (perfect fit).
3.2 The land-use-change model ö Land Use Scanner
The allocation of residential land-use change was simulated with the LUS model. The
LUS is a GIS-based operational, spatially explicit simulation model that allocates land-
use changes on grid cells. Therefore, exogenously calculated regional claim sets and
suitability maps for each land-use type should be provided. See Hilferink and Rietveld
(1999) and Dekkers and Koomen (2007) for a more detailed description. With respect
to the applied logistic regression analysis, the suitability, sc , for residential land use
corresponds with the logit of Pc :
Pc
y 1
sc ln bX c . (2)
1 ÿ Pc
y 1
As mentioned previously in section 3.1, in the basic version of the LUS model, a logit-
type approach was applied to determine the probability that a cell is converted into a
certain land-use type depending on the suitability of that cell. The model in this form
does not guarantee agreement with the exogenously provided regional claims (Rietveld
et al, 2001). Therefore, allocation is conducted with a doubly constrained logit model
that assigns land uses according to both their suitability and the pressure on land
induced by the regional claim sets according to the following equation:
Mcj aj bc exp
ascj , (3)
Table 1. List of implemented suitability maps.
915
N:/psfiles/epb3705w/
916 J Hoymann
where
Mcj is the amount of land in cell c of land-use type j,
aj is a balancing factor to ensure that regional claims are kept,
bc is a balancing factor ensuring the total amount of land per cell to equal the size
of that cell,
a is a factor determining the extent to which mixed land use occurs in cells (a 0
random allocation, a > 1 tendency for homogeneity of land use per cell).
The constraints, implemented as balancing factors aj and bc , ensure that the
regional claims are kept and that the total amounts of land use expected per grid cell
are equal to the size of the cell (Hilferink and Rietveld, 1999). The solution of this
model is found through iteration to find values to balance the factors until convergence
is achieved between the demand for land and the allocated land.
Thus, the LUS is an equilibrium model that acts by balancing the demand and
supply of different land-use types using a bidding process (Koomen and Buurman,
2002). The LUS is based on the classical models of von Thu«nen, Losch, and Alonso,
since transport and accessibility play an important role as determinants of land
uses (Rietveld et al, 2001). The algorithm itself uses discrete choice theory to allocate
land-use changes to grid cells.
The LUS emphasises two very important characteristics. Firstly, it integrates urban
and nonurban land-use types simultaneously (Hilferink and Rietveld, 1999). Second,
land is not always allocated on the basis of the highest bidding land-use types as there
are strong public interventions in the spatial planning policy. The LUS is able to
integrate regional spatial planning documents to account for these restrictions.
3.3 Variables and data sources
For the complete study area of the Elbe River Basin, approximately 2.3 million cells
with a spatial resolution of 250 m were calculated. Regional claims were based on
administrative regions (Hoymann, 2008). The calculation of suitability maps did not
incorporate statistical indicators such as population or economic development, because
these factors were considered in the regional claim sets. These factors did not differ in
this study, so only allocation differences due to differences in location preferences
could be compared. These location preferences (suitability maps) differed only in the
applied weights with respect to scenario storylines, while the underlying datasets
remain constant over time. This was done so that the allocation changes can be
analysed independently when the weights of location factors change. A complete list
of variables applied is presented in table 1.
The CORINE land-cover database of the year 1990 was applied to calibrate the
suitability maps in the logistic regression analysis (UBA, 2004). The corresponding
dataset from the year 2000 was used for quasivalidation of the regression analysis
and was implemented into the LUS. The land-cover types were aggregated into sixteen
classes, whereas only residential, arable land, and pasture were assumed to change over
time.
The autoregressive and the full model measures need to be defined to include the
land-use types of neighbouring cells. For these surroundings, multiple rings with differ-
ent radii and widths were defined [figure 2(a)]. The first ring with a diameter of three
cells contained the eight cells directly surrounding a given location (Fc3). The second
ring contained the next sixteen cells outside the first ring (Fc5); the third ring the
subsequent fifty-six cells (Fc9) and so on. In total, eleven rings were defined with an
outer radius of 5000 m (Fc41). This radius was chosen because residential development
took place at distances of up to 5000 m from existing residential areas between 1990
and 2000 [figure 2(b)]. The number of cells of a certain land-use type was counted for
Spatial allocation of future residential land use 917
12
Fc3 6
Fc5
Fc9 4
Fc13
2
Fc17
0
10 20 30 40
Diameter of outer ring (number of cells)
(a) (b)
Figure 2. (a) Definition of the autoregressive variables; (b) distribution of new residential cells
between 1990 and 2000 amongst the defined rings.
every ring and divided by the total number of cells in that ring. This resulted in the
proportion of each land-use type (residential, arable land, pasture) within the rings
surrounding a given location. The underlying assumption was that there was a positive
correlation in a given ring between the proportion of residential areas and the number
of new residential areas. Sine residential development took place nearly exclusively on
agricultural land, it was also assumed that there was a positive correlation between
these variables.
Slope, which is calculated from Shuttle Radar Topography Mission (http://www2.jpl.
nasa.gov/srtm/) data, is assumed to be the most important physical characteristic
for the allocation of residential areas; slope restricts urbanisation in a considerable
part of the study area that is characterised by low mountain ranges (CIAT, 2004).
Accessibility measures were calculated as Euclidean distances to existing residential
areas, metropolitan cities, and middle-order and high-order centres. The classification
of cities into the hierarchies of central-place theory is defined in the regional spatial-
planning documents. For the Czech part of the study area, cities were classified
manually using the criteria for central places. In addition, Euclidean distance to rail-
way stations, highway exits, railroads, roads, and airports was calculated using digital
terrain models of the German Federal Agency for Cartography and Geodesy (http://
www.bkg.bund.de/) and ESRI datasets for the Czech part of the study area. Proximity
to infrastructure, market places, and leisure facilities was assumed to be important for
residential development.
Seven types of nature conservation areas were included in the LUS. On the basis
of their regulatory framework, nature conservation areas are expected to have a
lasting effect on the choice of location for residential land use (Heiland et al, 2006).
The types of conservation area in the analysis were grouped into strictly protected
areas (strcprot), including nature conservation areas and national parks, and into areas
that have the potential to be strictly protected (potstrcprot) if the regulatory framework
is adjusted to include all other types of nature conservation areas.
Regional spatial planning datasets originate either from state development plans
or from regional planning documents. The applied spatial planning instruments aim
to preserve land use, land-use functions, and landscapes worthy of being protected.
918 J Hoymann
These instruments were divided into priority areas and reserve areas. While the first
category excludes land-use types that impact the designated priority functions, the
latter reflects the importance of those land-use types in comparison with competing
uses. The instrument `potential residential areas' attracts the allocation of residential
land use. In the Czech part of the study area, an indicator approach was applied at
the municipal level to define development axes for urban development (MMR and
UèUèR, 2006).
Nature conservation maps and maps of regional spatial planning were not used in
the logistic regression analysis but were used in the calculation of the suitability maps
when adapting the scenario storylines within the LUS for two reasons. First, a refer-
ence image of historic trends was used to evaluate the impact of policy and no-policy
scenarios. Secondly, the underlying planning documents were first enacted at the end
of the studied decade or later and many of the nature protection areas were also
established during that time. This is especially true for the East German parts of the
study area. For the Czech Republic, the documents were not developed until the year
2000. Therefore, these documents did not influence the development of the land-use
changes between 1990 and 2000. This second factor was particularly relevant, because
a large part of the study area is located in East Germany and the Czech Republic.
3.4 Scenario description
This study is part of a larger research project being conducted until the year 2020
examining the impacts of global change on the water cycle in the Elbe River Basin
[GLOWA-Elbe (http://www.glowa-elbe.de/)]. Thus, scenarios A1 and B2 (see below),
which were based on the IPCC-SRES (Intergovernmental Panel on Climate Change.
Special Report on Emissions Scenarios) (Nakicenovic and Swart, 2000), were analysed
with two trajectories of regional economic and demographic development and were
combined with two orientations of land-use policy (0, ) (Hartje et al, 2008). The 0
scenarios represent weak spatial planning policies while the scenarios indicate very
restrictive policies.
Scenario A10 assumed higher growth rates in economic development and spatial
spillover effects from metropolitan regions to peripheral regions (Blazejczak et al, 2008).
This development was supported by a weak spatial planning policy. Only the strictly
protected nature conservation areas influenced the allocation of residential areas.
On the basis of the same socioeconomic development as in scenario A10, scenario
A1 was characterised by a spatial planning policy that implements sustainable devel-
opment by strictly following the spatial planning documents to realise a development
in the sense of central place theory. Due to the strict nature of this policy, residential
developments were not permitted in nature conservation areas.
Scenario B20 assumed lower economic growth rates are concentrated in the metro-
politan regions (Blazejczak et al, 2008). A weak regional spatial planning policy should
lead to a decrease in urban sprawl at the periphery of metropolitan regions. Only the
strictly protected nature conservation areas influenced the allocation of residential areas.
Scenario B2 was based on the same socioeconomic development as scenario B20.
However, the spatial planning policy that was consequently implemented promotes
the central place theory. Again, spatial policy restricts residential developments in any
nature conservation area.
3.5 Adaptation of regression results to the scenarios studies
The logistic regression approach was inherently similar to the allocation algorithm of
the LUS, and thus the regression estimation coefficients were directly implemented as
weights in calculating suitability values in the LUS with the following scenario-specific
adaptations.
Spatial allocation of future residential land use 919
First, the accessibility grid for the middle-order and high-order centres was used
only in the A1 scenarios while the distance to metropolitan regions was used in the B2
scenarios. Additionally, the weights were set to higher values to promote the assumed
development of the scenario storylines.
Second, the inner rings of the neighbourhood measures of residential development
were given higher values for the residential development of the B2 scenarios to restrict
the developments to the immediate neighbourhood of existing residential areas. It was
assumed that this restriction avoids sprawl.
Finally, weights for these policy maps were applied only in the scenarios with
stronger spatial policy () to reduce urban sprawl and to demonstrate their influence
on residential development.
To allow for a comparison of the different adaptations in the simulation of the
different scenarios a baseline simulation was added, consisting of the unchanged
coefficients estimated by the regression analysis.
4 Results
4.1 Location preferences for residential developments
This section presents the results of the three binomial logistic regression analyses for
residential land-use changes between 1990 and 2000. An initial version of the regres-
sion models found two difficulties. Due to the inherent proximity of airports and
large cities, the distance to airports shows a strong correlation with the distances to
metropolitan regions and high-order central places. Furthermore, the neighbourhood
measures reveal uncertainty about their influence with increasing distance to a cell.
This occurs because the signs of the coefficients change between rings of larger
distance and are not significant although residential developments occur within
certain distances of existing residential areas. Therefore, all neighbourhood measures
with an outer diameter of seventeen or more cells and the distance to airports were
excluded from the regression analysis. The final regression analyses were calculated
with the reduced set of variables. The results are presented in table 2.
The signs of the coefficients of the autoregressive model were as expected. The
probability of residential development increases with increasing quantities of residen-
tial grid cells in the neighbourhood and then decreases. This relationship is associated
with the mean patch size of residential areas. Repeated increased probabilities for
residential areas indicate their development in proximity to, though not necessarily in
direct connection with, existing residential areas. The positive influence of agricul-
tural areas in residential neighbourhoods confirms current developments in which
residential areas are developed on agricultural land.
The second estimation, which included only the accessibility measures and physical
characteristics as explanatory variables, also showed a significant influence on the
locations of residential areas. The probability of developing new residential areas
decreases strongly with increasing distance to railway stations, roads, freeway exits,
residential areas, and slope. The distance to railroads had the opposite influence
because they cannot be accessed anywhere but at railway stations. Noise exposure is
another possible explanation for the positive correlation observed between distance to
railroads and residential development. The latter is also true for distance to roads and
stations, but access to roads and stations is essential to residential development and the
mobility of the residents. In contrast to railroads, access to roads is not only through
freeway exits. This also leads to the question of what comes first: settlements or roads?
Indeed, both settlements and roads influence each other.
The most important factor was distance to existing residential areas, but regional
differences were observed in distance maps to central places. These maps indicated a
920 J Hoymann
Table 2. Estimated coefficients of the binomial logit models for residential land-use changes
between 1990 and 2000. Method is Enter. (See table 1 for explanation of variables.)
that cells were close to larger settlements became disproportionately high. Therefore,
the sign of FC13 GG1 90 changed to account for new residential areas around smaller
settlements. Furthermore, in the full model, most of the accessibility measures, physical
characteristics, and neighbourhood measures showed smaller absolute values of the
coefficients to balance their impact.
For validation purposes, the corresponding ROC value for residential development
in all three regression models are presented in table 2. Both models (the autoregressive
model as well as the model using the accessibility measures and physical characteristics
as explanatory variables) showed high ROC values. This result indicates the impor-
tance of both sets of variables in explaining the allocation of residential land-use
change. The ROC value could be marginally increased due to the combination of
both sets of variables in the full model, and therefore has the best predictive power
of the three regression models. Therefore, the full factor model is further used for the
simulation of residential land-use change.
4.2 Simulation of residential development
The estimated regression coefficients of the full binomial logistic model were adapted
and implemented into the LUS and future residential land-use change was simulated for
all five scenarios. Table 3 summarises indicators that quantify the degree of urbanisa-
tion and urban sprawl in the five simulations of total residential area (Ritsema van Eck
and Koomen, 2008). Compared with the year 2000, the number of patches decreased in
B2 scenarios and increased in A1 scenarios. Considerable increases in average patch
size accompanied this development. Changes in connectivity were only marginal in
comparison with 2000 in all five simulations. Patches of residential areas merged,
which resulted in increased patch sizes, a decreased number of patches, and decreasing
connectivity between patches. The latter was due to the merging of patches. Thus, small
distances between patches were not reflected in the statistics and the distance between
patches increased. This result was based on the development of landscape metrics
between 1990, 2000, and the baseline scenario. When applying scenario-specific weights
for suitability maps, the merging process was not pronounced. Instead, smaller patches
were observed. In all scenarios and in all parts of the study area, the connectivity index
decreased with an increase in the influence of policy ( scenarios).
Figure 3 shows an increase in residential land in grid cells. All five scenarios showed
hotspots of residential development in and around the metropolitan regions. This was a
continuum of the current trend. Additionally, the high-order and middle-order central
places showed significant increases in residential development, particularly in the A1
scenarios. The maps clearly show that the residential developments of the B2 scenarios
were not limited to the immediate neighbourhood of the metropolitan city of Berlin,
Table 3. Urbanisation indicators.
Current Scenarios
land use
baseline A10 A1 B20 B2
Total residential 941 812.5 992 325 994 237.5 994 162.5 991 787.5 992 018.75
area (ha)
Share of residential 6.4420 6.7902 6.8027 6.8033 6.7881 6.7865
area within
study area (%)
Number of patches 8 859 8 831 8 947 8 946 8 783 8 769
Mean residential 106.3114 112.3684 111.1252 111.11293 112.9213 113.1279
patch size (ha)
Connectivity 0.0013 0.0012 0.0013 0.0014 0.0012 0.0012
922 J Hoymann
A10 A1
B20 B2
0 5 10 20 km
Baseline
Figure 3. Absolute growth of residential area on grid cells for the five scenarios. See figure 1 for
an overview of the study area. [Source: FGG Elbe, 2005 (www.fgg-elbe.de/) and own simulations.]
Spatial allocation of future residential land use 923
but were also found in smaller settlements of the hinterland. This was an unexpected
result. This development in the hinterland was expected for the A1 scenario. Instead,
there was an overall concentration on the metropolitan city. This caused polarisation
between the metropolitan cities and the rural areas in the study area. When considering
different socioeconomic developments within the regional claims, this development
might again be different when a larger growth is expected for the A1 scenarios.
To represent the pattern observed for all planning documents, the effect of the
nature conservation areas on allocation is presented in figure 3. Without accentuat-
ing the importance of metropolitan cities such as Berlin, the distribution of new
residential areas was throughout the hinterland. This can be observed in a comparison
of the B20 and B2 scenarios. If nature conservation areas were strictly protected as
in the scenarios, development was relocated outside the nature conservation areas.
In scenario B2, the new residential area occurred in the strip that was not protected,
particularly to the west of Falkensee and in northeast Berlin.
The vast majority of municipalities showed an increase in residential areas that
were less than 5 ha in size (figure 4). Increases in residential land of more than
50 ha were found in 2.5% and 2% of all municipalities in scenarios A1 and B2,
respectively, which represented 70% and 50%, respectively, of the newly allocated
residential land. This was associated with a strong correlation between the absolute
size of current residential areas and the growth up to 2020.
100
80
Cumulative share (%)
60
40 Scenarios
A10
A1
20 B20
B2
Baseline
0
4000
Number of cells
3000
2000
Scenarios
A10
A1
1000 B20
B2
Baseline
1900 ^ 2000
0
1000 2000 3000 4000 5000 1000 2000 3000 4000 5000
Distance (ha) Distance (ha)
(a) (b)
Figure 5. Distribution of residential development for two types of settlement structure within
different distances from current residential areas and scenarios (cumulative). (a) Central places of
high and medium order; (b) rural municipalities.
Table 4. Number of cells allocated as residential by at least 33% of the cells in nature conservation
and planning areas (percentage change shown in parentheses).
Current Scenarios
land use
baseline A10 A1 B20 B2
Nature conservation
strcprot a 346 367 (6.07) 367 (6.07) 376 (8.67) 371 (7.23) 377 (8.96)
potstrcprot a 27 037 28 732 (6.27) 28 524 (5.50) 27 385 (1.29) 28 734 (6.28) 27 455 (1.55)
Total in nature 22 103 23 461 (6.14) 23 349 (5.64) 22 406 (1.37) 23 490 (6.28) 22 460 (1.62)
conservation area
Percent change from 6.14 5.64 1.37 6.28 1.62
current residential
land in nature
conservation areas
Share of residential areas 2.15 2.29 2.27 2.18 2.29 2.19
in nature conservation
areas (%)
Spatial planning
Instruments attracting 3 203 3 961 (23.67) 4 146 (29.44) 4 222 (31.81) 3 956 (23.51) 4 232 (32.13)
residential
developments
Instruments restricting 15 832 16 350 (3.34) 16 405 (3.62) 15 842 (0.06) 16 308 (3.01) 15 847 (0.09)
residential
developments
Total in planning areas 21 857 23 355 (6.85) 23 569 (7.83) 23 153 (5.93) 23 315 (6.67) 23 158 (5.95)
Percent change from 6.85 7.83 5.93 6.67 5.95
current residential
land in planning
areas
Share of residential areas 3.08 3.29 3.32 3.26 3.28 3.26
in planning areas (%)
a See table 1.
Spatial allocation of future residential land use 925
Dosch F, Beckmann G, 1999, ``Trends und Szenarien der Siedlungfla«chenentwicklung bis 2010'',
in Perspektiven der kunëftigen Raum- und Siedlungsentwicklung Eds W Strubelt, H-P Gatzweiler,
R Kaltenbrunner (Bundesamt fu«r Bauwesen und Raumordnung, Bonn) pp 827 ^ 842
Eurostat, 2007, ``Bodennutzung nach Hauptkategorien'', Statistical Office of the European
Communities, http://www.epp.eurostat.ec.europa.eu/portal/page? pageid=1996,45323734
& dad=portal& schema=PORTAL&screen=welcomeref&open=/env/env land&language
=de&product=EU MASTER environment&root=EU MASTER environment&scrollto=0
Gordon P, Wong H L, 1985, ``The cost of urban sprawl: some new evidence'' Environment and
Planning A 17 661 ^ 666
Hartje V, Ansmann T, Blazejczak J, Go«mann H, Gornig M, Grossman M, Hillenbrand T,
Hoymann J, Kreins P, Markewitz P, Mutafoglu K, Richmann A, Sartorius C, Schulz E,
Vo«gele S, Walz R, 2008, ``Regionalisierung der Szenarioanalyse (der Antriebskra«fte und des
Nutzungsdruckes) des globalen Wandels fu«r die Wasserwirtschaft'' Wirkungen des globalen
Wandels auf den Wasserkreislauf im Elbegebietö Risiken und Optionen. Schlussbericht zum
BMBF-Vorhaben GLOWA-Elbe II (Potsdam), pp 7 ^ 39, http://www.glowa-elbe.de/pdf/
schlussbericht glowaii/kapitel/kapitel2 part1.pdf
Heiland S, Reinke M, Siedentop S, Tra«ger T, Knigge M, Meyer-Ohlendorf N, Blobel D, 2006
Beitrag naturschutz politischer Instrumente zur Steuerung der Fla«cheninanspruchnahme Bonn,
http://www.uni-stuttgart.de/ireus/institut/siedentop.html
Hilferink M, Rietveld P, 1999, ``LAND USE SCANNER: an integrated GIS based model for long
term projections of land use in urban and rural areas'' Journal of Geographic Information
Systems 1 155 ^ 177
Hoymann J, 2008, ``Szenarien der Siedlungsfla«chenentwicklung im Elbeeinzugsgebiet'' Wirkungen
des globalenWandels auf denWasserkreislauf im ElbegebietöRisiken und Optionen. Schlussbericht
zum BMBF-Vorhaben GLOWA-Elbe II (Potsdam), pp 140 ^ 170, http://www.glowa-elbe.de/pdf/
schlussbericht glowaii/kapitel/kapitel2 part1.pdf
King G, Zen L, 2001, ``Logistic regression in rare events data'' Political Analysis 9 137 ^ 163
Koomen E, Buurman J, 2002, ``Economic theory and land prices in land use modeling'', in
Proceedings of the 5th AGILE Conference on Geographic Information Science, Palma, Balearic
Islands Eds M Ruiz, M Gould, J Rumson, http://www.agile-online.org
Lambin E F, Turner B L, Geist H J, Agbola S B, Angelsen A, Bruce J W, Coomes O T, Dirzo R,
Fischer G, Folke C, George P S, Homewood K, Imbernon J, Leemans R, Li X, Moran E F,
Mortimer M, Ramakrishnan P S, Richards J F, Skanes H, Steffen W, Stone G D, Svedin U,
Velkamp A, Vogel C, Xu J, 2001, ``The causes of land-use and land-cover change: moving
beyond the myths'' Global Environmental Change 11 261 ^ 269
Loonen W, Koomen E, 2009, ``Calibration and validation of the Land Use Scanner allocation
algorithms'', in Background Studies (Milieu- en Natuurplanbureau, Bilthoven)
Millington J D A, Perry G L W, Romero-Calcerrada R, 2007, ``Regression techniques for
examining land use/cover change: a case study of a Mediterranean landscape'' Ecosystems
10 562 ^ 578
MMR, UèUèR, 2006, ``Spatial development policy of the Czech Republic'', Institute for Spatial
Development, Ministry for Regional Development of the Czech Republic, http://www.uur.cz
Nakicenovic N, Swart R, 2000 Special Report Emissions Scenarios (Cambridge University Press,
Cambridge)
Penn-Bressel G, 2003, ``Reduzierung der Fla«cheninanspruchnahme durch Siedlung und Verkehr.
Materialienband'', Federal Environment Agency, Berlin, http://www.umweltdaten.de/
publikationen/fpdf-1/2587.pdf
Pontius R G Jr, Batchu K, 2003, ``Using the relative operating characteristic to quantify certainty
in prediction of location of land cover change in India'' Transactions in GIS 7 467 ^ 484
Pontius R G, Schneider L C, 2001, ``Land-cover change model validation by an ROC method
for the Ipswich watershed, Massachusetts, USA'' Agriculture Ecosystems and Environment
85 239 ^ 248
Prentice R L, Pyke R, 1979,``Logistic disease incidence models and case-control studies'' Biometrika
66 403 ^ 411
Rietveld P, Scholten H J, Stillwell J, 2001, ``EuroScanner: a simulation model for land use change in
Europe'', in Land Use Simulation for Europe Eds J Stillwell, H J Scholten (Kluwer, Dordrecht)
pp 235 ^ 244
Ritsema van Eck J, Koomen E, 2008, ``Characterising urban concentration and land-use diversity
in simulations of future land use'' Annals of Regional Science 42 123 ^ 140
928 J Hoymann
RNE, 2004, ``Mehr Wert fu«r die Fla«che: Das `Ziel-30-ha' fu«r die Nachhaltigkeit in Stadt und Land'',
Rat fu«r Nachhaltige Entwicklung, German Council for Sustainable Development,
http://www.nachhaltigkeitrat.de/uploads/media/Broschuere Flaechenempfehlung 01.pdf
ROG, 18 August 1997 (BGB1. I S 2080), zuletzt gea«ndert am 24. Juni 2004 (BGB1. S. 1359),
Raumordnungsgesetz [Regional Planning Act]
Schotten C G J, Goetgeluk R, Hilferink M, Rietveld P, Scholten H J, 2001,``Residential construction,
land use and the environment: simulations for the Netherlands using a GIS-based land use
model'' Environmental Modeling and Assessment 6 133 ^ 143
STABU, 2006, ``Zunahme der Siedlungs- und Verkehrsfla«che: 114 ha/Tag'', Statistics Bundesamt,
http://www.destatis.de/presse/deutsch/pm2006/p4920112.htm
STABU, 2007, ``Siedlungs- und Verkehrsfla«che nach Art der tatsa«chlichen Nutzung. Erla«uterungen
und Eckzahlen 2006'' Bodenfla«che nach Art der tatsa«chlichen Nutzung Fachserie 3/Reihe 5.1,
Statistiches Bundesamt
Tobler W, 1979, ``Cellular geography'', in Philosophy in Geography Eds S Gale, G Ollson (Reidel,
Dordrecht) pp 379 ^ 386
Turner M G,Wear D N, Flamm R O, 1996, ``Land ownership and land-cover change in the Southern
Appalachian Highlands and the Olympic Peninsula'' Ecological Applications 6 1150 ^ 1172
UBA, 2004, ``CORINE Land Cover'', Federal Environment Agency, DLR ^ DFD, German
Aerospace Centre ^ German Remote Datasensing Centre, http://www.corine.dfd.dlr.de/
workshop2004 en.html
Van der Hoeven N, Aerts J, Van der Klis H, Koomen E, 2008, ``An integrated discussion support
system for new Dutch flood risk management strategies'', in Planning Support Systems: Best
Practices and New Methods Eds S Geertman, J Stillwell (Springer, Berlin) pp 159 ^ 174
Verburg P H, Ritsema van Eck J R, de Nijs T C M, Dijst M J, Schot P, 2004a, ``Determinants
of land-use change patterns in the Netherlands'' Environment and Planning B: Planning and
Design 31 125 ^ 150
Verburg P H, Schot P, Dijst M J,Veldkamp A, 2004b, ``Land use change modelling: current practice
and research priorities'' GeoJournal 61 309 ^ 324
Wear D N, Bolstad P, 1998,``Land-use changes in Southern Appalachian landscapes: spatial analysis
and forecast evaluation'' Ecosystems 1 575 ^ 594