International Journal of
2019, Vol. 17(4) 336 –356
© The Author(s) 2019
Article reuse guidelines:
Planning support systems (PSS) are simply defined as computer-based tools to support planners in undertak-
ing planning specific activities.1,2 PSS are spatial data-driven tools designed for measuring, mapping or evalu-
ating impacts arising from likely urban development scenarios.3,4 When PSS is developed using best practice
evidence, they can be applied to meaningfully derive information and analyse data to support spatial planning
practices.5 PSS are designed to support collaboration throughout the urban planning process with simple user
interfaces that allow testing and evaluation of various urban design scenarios4–6 and are commonly built using
geographical information systems (GIS), designed to collect, analyse and visualise spatial data.7
At the same time, automation in planning processes helps end-users to go through data within multiple lay-
ers to detect potential issues, arrive at conclusions and reinforce iteration in the process. Manual tasks lead to
inevitable human errors and inconsistency in the technique. Automation eliminates these chances of mistakes
and helps the city planning process to rely more strongly on the accurate nature of the data. By reducing the
error-prone manual process, decision makers can focus on outcomes which can lead to a practical resolution.
Planning support systems
for organisational teams
Pradeep Alva , Patrick Janssen and Rudi Stouffs
In practice, most planners do not make significant use of planning support systems. Although extensive research has
been conducted, the focus tends to be on supporting individual tasks, and the outcomes are often the development of
new stand-alone tools that are difficult to integrate into existing workflows. The knowledge contribution in this article
focuses on developing a novel spatial decision support framework focusing on the workflows and tool-chains that span
across different teams with varying skill sets and objectives, within an organisation. In the proposed framework, the
core decision-making process uses a set of decision parameters that are combined using a weighted decision tree. The
framework is evaluated by developing and testing a workflow and GIS tool-chain for a real-world case study of land
suitability and mixed-use potentiality analysis.
GIS, SDSS, PSS, planning automation, TOD, raster geoprocessing, data analytics, geoinformatics
National University of Singapore, Singapore
Pradeep Alva, National University of Singapore, 4 Architecture Drive, Singapore 117566.
891569JAC0010.1177/1478077119891569International Journal of Architectural ComputingAlva et al.
Special Issue Article
Alva et al. 337
While many spatial decision support systems (SDSS) and PSS exist, most planners do not make signifi-
cant use of such systems in practice.8–10 Pelzer et al.11 discuss the concept of task-technology fit and PSS.
They believe this concept can lead to a better understanding of the utility that is reachable by different plan-
ning support capabilities of a PSS for different tasks. However, PSS research typically focuses on supporting
individual tasks, and the outcomes are often the development of new stand-alone tools that are difficult to
integrate into existing workflows. In order to improve the support function of PSS, a better conceptual and
empirical understanding of the relation between planning tasks and PSS is required.
We believe that in order to improve the probabilities of tool usability and generation of good results, agil-
ity has to be introduced as an ancillary to the framework. A human-centred development method can support
flexible iteration and gradual development which can respond to the rapidly changing end-user require-
ments.12 The adoption of advanced automation processes in practice requires focusing on improving the
experience for users as much as it requires focusing on improving the performance of algorithms. Before
developing the tool, one needs to observe a user’s workflow, identify tasks that should be automated and
propose methods for ensuring tools integrate into – rather than interrupt – existing processes.13
In our proposed framework, an agile approach is carried out with inputs from organisational teams and
automation working in tandem to achieve better outcomes. Most importantly, the spatial decision support
framework is designed in collaboration with the organisational team and not in isolation. This resulted in
tools which can support active intervention from the organisational team. This approach seeks higher effi-
ciency and speed throughout the process and as a result, giving more time for planners to spend on assess-
ment and arriving at accurate decision making. The aim of this research is to explore an alternative approach
to the effective use of planning support tools, evaluated through two case studies. The focus is on developing
systems for large organisations with teams of diverse professionals with different expertise.
In this article, we present a framework which focuses on workflows rather than individual tasks. Rather
than developing stand-alone tools, we aim to support specific workflows by developing tool-chains on top
of existing GIS frameworks. These workflows and tool-chains will typically span across different teams
within the organisation, with varying skill sets and objectives. For instance, a key challenge is a typical
workflow in which planning decisions need to be made based on geoprocessed data.14 In such a workflow,
the task for the GIS team involves analysing large amounts of geospatial data, which is typically slow and
requires a high level of technical knowledge. The task for planners focuses on decision making using that
processed data, which must be interactive and requires domain knowledge. Finally, the results of the deci-
sions need to be communicated by the managerial team. This article discusses how such planning tasks
within an organisation can be linked into seamless workflows, supported by geospatial tool-chains.
The automated workflows of planning tasks using geospatial tool-chains were discussed with the organi-
sational teams in practice on a regular basis. This continuous iterative process of redesigning the tool based
on the feedback from the users led to the development of subsequent versions of the tool. We also believe
that this approach could improve practicality and usability of the tool developed.
To demonstrate the proposed framework and tool-chains, two case studies with different data types – ras-
ter and vector – are presented. A case study of mixed-use potentiality uses raster datasets, and a case study
on land suitability assessment uses vector data. Both case studies have been applied in a Singapore context.
In this research, a decision support framework is developed to support long-term planning of regional mixed-
use nodes and short-term land suitability assessment. This framework helps planners analyse the potentiality
of possible spatial locations where such developments can materialise.
In the proposed framework, the core decision-making process uses a set of spatial decision parameters
that are combined into a single metric. The parameters are first calculated based on the latest geospatial data.
Once each parameter is calculated, the overall fitness of each spatial location is then derived using a weighted
338 International Journal of Architectural Computing 17(4)
decision tree.15 The weights in the decision tree may vary depending on any kind of location choice
approach.16 As a result, the weights must be user-defined.
Each spatial location is then given a percentage score, 100% being the maximum score for the specific
query. This score can be visualised spatially as a heat map so that areas of high potentiality or suitability can
be easily identified. Geospatial tool-chains are used to connect these workflows of acquiring, operating,
assessing and visualising data.
The three stages
The research proposes a workflow consisting of three main stages, with customised tools targeting different
teams with varying expertise: (1) geoprocessing performed by the GIS team, (2) interactive decision mak-
ing performed by the planning team and (3) communications performed by the managerial team (Figure 1).
A series of tools are developed for performing specific tasks in the workflow. In addition, an important
issue is also the flow of data between specific tasks performed by different teams:
Stage 1. The geoprocessing task is performed by the GIS team, as this stage deals with compiling the
latest data for the decision tree. The decision tree will need to be defined before this stage can be car-
Stage 2. The decision-making task is performed by planners with specific domain skills in an organisa-
tion. This process may include extensive customisation and exploration of the decision-tree weights.
Stage 3. The communications task is performed by the managerial team, creating reports to be presented
and deployed to planning agencies. If the requirements cannot be met, the process will usually go back to
The geoprocessing task handled by the GIS team is the slowest task to execute but is executed infre-
quently and does not need to be interactive. This task results in datasets that are then consumed by the
decision-making task. This task must be fast and interactive so that the planners are able to iteratively
Figure 1. Three stages of the tool-chain workflow.
Alva et al. 339
explore many alternative scenarios. Finally, the communications task will consume the datasets created by
the decision-making task and produce a set of final reports for presentation to planning agencies.
In this division of labour, the geoprocessing task is only executed when new input data are added to the
GIS database, such as changes to the master plans and new land-use information. The decision-making task,
however, can be executed multiple times without having to re-execute the time-consuming geoprocessing
task. If the planning requirements cannot be met, then the communications task can loop back to the deci-
sion-making task to consider alternative results. The formalisation of these three stages makes the overall
decision-making logic more transparent and evidence-based.17
Case study of mixed-use potentiality analysis
Transit-oriented development (TOD) is an urban development model that considers various principles for
designing sustainable cities. It advocates multifunctional centralities in a multimodal mobility network that
employs high density, mixed uses, walkability, and a variety of transportation options so that the basic urban
needs are easily accessible.18–20
This case study focuses on analysing spatial potentiality for regional mixed-use nodes. We adopt a location
choice approach,16 considering four key factors: (1) land availability, (2) accessibility to transit, (3) presence
of parks and (4) exposure to traffic noise.
Although other relevant factors can be identified or considered, our selection has been mainly driven by
data availability. Nevertheless, we acknowledge the limitations of our selection and we have been investigat-
ing other factors for future inclusion. However, more important than being complete in our analysis is to
support planners in developing acceptable scenarios or providing arguments to have these scenarios accepted
by higher management. As such, we need to balance research and practice, informing our approach from
research but adhering to a practical aim.
For the mixed-use potentiality, a grid-based method is used to analyse the four factors. The results are
then aggregated in a hierarchical manner, specifying weights at each level where these weights can be identi-
fied as parameters to be adjusted by the planners (Figure 2). We used an analysis grid rather than plots, as
plots may be combined or divided, local and primary access roads may be added, retraced or even removed,
Figure 2. Decision tree for mixed-use potentiality analysis.
340 International Journal of Architectural Computing 17(4)
canals may be dug, land set aside for future green or land may be reclaimed from the sea or other water bod-
ies. A geospatial tool-chain for mixed-use potentiality analysis with three stages is shown in Figure 3.
Land availability. This parameter is affected by land leases and other land-based temporal data. In principle,
land availability is measured at a point in time. However, in order to be able to combine the results with the
other analysis results, we are using a sigmoid function to transform a point in time into a value between 0
and 1, where 0 means no availability at all and 1 means immediate availability. Because 2050 is considered
as the target date, we consider 2050 as the inflection point and 2010–2090 as the primary range for the
sigmoid curve using equation (1). When applied to lease ends, any lease ending before 2050 will result in a
value greater than 0.5, while any lease ending after 2050 will result in a value less than 0.5 (see Figure 4)
Figure 3. Mixed-use potentiality analysis tool-chain.
Figure 4. Sigmoid equation (1) transforming points in time (x-axis) into the range of 0–1 (y-axis).
Alva et al. 341
where e is Euler’s number and x is the year of land availability (e.g. end of lease).
Overall, land availability is affected by lease ends and other land-based temporal data. So, the final value
for land availability is determined as the minimum value within land-based temporal data, as altogether it
should be read as restricting availability.
Accessibility to transit. Transit planners generally consider a walkable distance of 400 m. However, studies on
average walking distances in different cities have resulted in varying averages and correlations between
walking distances (to bus stops) and the percentage of people (transit users) walking at least this distance. A
survey study conducted in Singapore revealed an average walking distance of 187 m to bus stops, 226 m to
light rail transit (LRT) stations and 608 m to mass rapid transit (MRT) stations.21
While these distances are measured over actual walkways, in the context of future urban developments,
we must instead use straight-line distances and reduce these averages accordingly. We adopt the assumption
that in a porous area or over a well-connected grid, the straight-line distance may be about 80% of the actual
walking distance. Furthermore, we use curve-fitting to deduce a function that can transform distances into
accessibility values between 0 and 1. Simplifying the exponent values in the functions, we arrived at a gen-
eral function of probability equation (2) with an average walking distance for bus stops of 150 m and for
MRT stations of 450 m (Figure 5)
where distance is the Euclidian distance to the nearest transit node and average is the average walking dis-
tance based on transit node type.
Figure 5. Comparison of the adopted equation (2) (highlighted in orange) transforming distances to bus stops
(x-axis) into probabilities within the range of 0–1 (y-axis), with fitted curves derived from two Singapore studies:
an NTU-LTA study21 and an FCL-URA study.22 Note that the latter is a general walkability study conducted in the
central business district.
NTU-LTA: Nanyang Technological University-Land Transport Authority; FCL-URA: Future Cities Laboratory-Urban
342 International Journal of Architectural Computing 17(4)
Overall, accessibility to transit is a combination of accessibility to pick-up/drop-off points for different
modes of public transportation. Here, we distinguish between bus stops and MRT stations and acknowledge
a preference of MRT over buses, beyond the distinction in average walking distance. Therefore, we are
adopting weights of 1 and 0.75 for MRT stations and bus stops, respectively, with the final value for acces-
sibility to transit as the maximum of two values when multiplied by the respective weights. We acknowledge
that these weights may seem rather arbitrary; however, we consider them as mere parameters that could be
adjusted by the planner in an exploratory manner.
Presence of green spaces. As is common in a location choice approach,16 we consider both the accessibility and the
area of green spaces in the analysis. Walkable distance to parks generally depends on quality and amenities. We
consider a conservative measure of accessibility by adopting the same 150 m radius for (projected) green spaces
as we have done for bus stops. For areal measurement, we consider a well-connected neighbourhood as the basis
with an 800-m walking distance or a 640-m radius and measure the total green area within this neighbourhood.
Once again, we adopt a sigmoid function to normalise the area within the range of 0–1. With a few exceptions,
most large recreational parks in Singapore have an area of around 0.5–0.6 km2. As such, we have selected 0.25 km2
as the inflection point of the sigmoid curve. Note that the total area of a neighbourhood with 640 m radius is almost
1.29 km2. Projected green spaces may include both recreational parks and stretches of greenery, the latter lining up
selected coastlines, river banks, existing or projected canals, and minor arterial roads.
For this reason, we take both parks and greenery into account with respect to accessibility, while only the
former contributes to the areal analysis. As such, when combining both analyses under a weighted sum, the
greenery will play a minor role with respect to the parks. As a default, we consider a 50/50 weighing of
accessibility and area of green spaces, although these weights can be easily adapted by the planners.
Exposure to traffic noise. Another impact is exposure to traffic noise. We consider the measured value of 70
dBA as guiding. Using a predictive tool,23 we approximated the expected distance for a reduced noise level
of 50 dBA at about 340 m.24 Considering that further reductions in the noise level require ever larger dis-
tances, we selected 150 m as the half-way point (y = 0.5) using equation (3) (see Figure 6)
where e is Euler’s number and x is the Euclidian distance to road causing traffic noise.
Figure 6. Adapted equation (3) transforming distances to the expressway (x-axis) within the range of 0–1 (y-axis),
guided by measured and predicted noise levels.
Alva et al. 343
To determine the mixed-use potentiality of each grid location, we first consider the aggregation of the four
aspects mentioned above: land availability, accessibility to transit, presence of green spaces and exposure
to traffic noise. The aggregation is achieved by summing the weighted outcomes of these four analyses
(Figure 2). The actual weights should be considered as parameters for the planner to play with.
Mixed-use nodes essentially need a critical mass. For this reason, we additionally compute a normalised cumu-
lative mixed-use potentiality over a ‘neighbourhood’ with a 640-m radius about the grid cell. This accumulation
will favour areas with consistently high values, rather than peak values surrounded by much lower urban potential.
To normalise the result, we can divide the cumulative urban potential value by the summed area of all grid cells
within the neighbourhood. This cumulative potential denotes the growth area for mixed-use development.
Automated potentiality analysis tool-chain
Our initial attempts to develop a methodology were based on vector data grid (fishnet). Grids of 30 × 30 m2
and 100 × 100 m2 size fishnet polygons were tested. However, the geoprocessing using vector data grid was
very slow to execute considering a vast set of polygons. An alternative approach using a raster grid was,
therefore, explored. A raster grid of 30 × 30 m2 cell sizes was created with floating-point pixel type to repre-
sent continuous data. These cells carry normalised values of each geoprocessing step in a sequence. The
extent of the raster grid is based on the input extent of plot data. Geoprocessing using the raster grid was
much faster when compared to the vector approach. However, as expected, the processing duration increases
with the extent of the raster grid (Figure 7).
The 30 × 30 m2 grid size is small enough for detailed analyses, at the overall Singapore scale or, approxi-
mately, even at the level of a single plot or plots, but not too small to make the computation process too
heavy. It is also the same resolution as, among others, Landsat 7 satellite images and the United States
Geological Survey (USGS) ground cover map.
Figure 7. Raster grid geoprocessing workflow.
344 International Journal of Architectural Computing 17(4)
A geospatial tool-chain for mixed-use potentiality analysis has been developed. A python toolbox (.pyt)
was implemented using the raster grid in the ArcGIS platform.25 This platform was chosen since it is com-
monly used within planning organisations, making tool adaptation and utilisation easier.
Most source input data are in vector format (.shp, polygon feature class), and these discrete datasets need
to be converted to lossless raster grid format. The spatial analyst module, an extension of ArcPy Python
package, is used for geoprocessing raster objects in ArcGIS. These raster objects created are temporary and
can be permanent when necessary by calling out the save method on the raster object. The remaining raster
objects analysis steps are done on the fly and execute fast. This also means that python scripting for raster
objects is more advanced in comparison with automating vector data.
The four normalised values for (1) land availability, (2) accessibility to transit, (3) presence of green
spaces and (4) exposure to traffic noise were calculated and weighted as per user priority. The weighted
values were then aggregated, and a cumulative mixed-used potential value was calculated for each grid cell
for 640 m radius using the focal statistics tool. Individual raster grids for each parameter along with the
aggregated and cumulative potential value outputs were produced. These final lossless raster grid outputs
can be converted to either vector formats (.shp, point feature) for further analysis or other visual output for-
mats (PDF, JPEG) for mapping and documentation.
This geospatial tool-chain includes a graphical user interface (GUI) to allow users to enter the parameters,
with built-in validation and error checking. It automatically filters out invalid data types and field names and
will flag errors for the user, giving feedback and warnings. This tool-chain can be easily shared and deployed,
becoming an integral part of the everyday working environment. All the output data of the geoprocessing
step are saved inside geodatabase files (.gdb) located in a shared workspace in a network drive. This allows
teams in the organisation to easily work together, linking and sharing data in a seamless way.17
This also makes room for iteration without repeating the geoprocessing step. The user can insert the out-
put geodatabase file (.gdb) from the previous session and just add different weightages. This step requires
very little time and executes in a matter of a few seconds. This helps cater to different skill sets of an organi-
sational team. A GIS expert can run the geoprocessing step, and a planner/decision maker can directly use
the derived data from geodatabase file shared on a local network and iterate various outputs rapidly and
intuitively (Figure 8).
Figure 8. Output iteration using the geodatabase file.
Alva et al. 345
This tool-chain has been tested with open data collected for existing land and transit nodes in Singapore. For
example, a mixed-use potentiality analysis at the scale of Singapore is presented using the Land Use Master
Plan 201426 and public transit access data27 from open data websites ‘data.gov.sg’28 and ‘mytransport.sg’,
respectively. The extent of the land-use polygons is used to create a raster extent with 30 × 30 m2 cells. This
grid is then used as a base for all geoprocessing steps ahead.
In order to test the framework, a hypothetical scenario was created to investigate potentiality under the
assumption that all existing urban areas are available for development, but at the same time maintaining all
existing parks. Land availability was set to have a lease end year of 2019 while existing parks were separated
from other land uses and assigned 2090 as lease end year.
These park polygons also serve as input for analysing park proximity, calculating Euclidian distances to
every cell of base raster grid. Polygons defining expressway road outer edges are extracted from the land-use
data and added as a source input for assessing traffic noise exposure. Euclidian distances from the road to
each cell of the base raster grid are calculated. Similarly, distances from public transit nodes (MRT stations
and bus stops) to cells are measured. This Stage-1 geoprocessing task took about 3 h for analysing an island
extent raster grid. Stage-1 of a regional scale extent of around 100 km2 executes in 15–20 min.
Equations (1)–(3) are, respectively, used for transforming values between 0 and 1, to create the key four
factors of mixed-use potentiality analysis. An aggregated mixed-use potentiality is calculated with assigned
weightages of 30% for land availability, 40% for accessibility to transit, 20% for the presence of parks and
10% for exposure to traffic noise factor (Figure 9).
A neighbourhood cumulative mixed-use potentiality is calculated for the 640-m radius (Figure 10). This
Stage-2 decision making takes about 30 s altogether together for assigning weightages and visualising the data
for the island extent raster grid. For a 100-km2 extent raster grid, Stage-2 executes in 1–3 s. Stage-3 report or
documentation export to visual (PDF, JPEG) formats is optional and takes a similar duration as Stage-2.
Figure 9. Aggregate potentiality result.
346 International Journal of Architectural Computing 17(4)
Figure 10. Cumulative potentiality result.
From the cumulative mixed-use potentiality, mixed-use nodes could be selected automatically as centred
on cells with the highest cumulative mixed-use potentiality values. The size of the mixed-use nodes or
growth area could subsequently be deduced from the aggregate mixed-use potentiality through sampling.
However, mixed-use nodes should be some minimum distance apart. As a result, selecting nodes would need
to be an iterative process of selecting the highest cumulative value, determining the radius of the mixed-use
nodes from the landscape of aggregate values, then excluding a buffer area (with a minimum radius of twice
the radius of the mixed-use nodes), before selecting another node, and so on.
We argue that this selection process should not be automated and should, instead, be left to the urban
planners. This decision can be based on the aggregate, cumulative mixed-use potentiality, any other informa-
tion generated from the methodology described here and, of course, other information the urban planners
may deem relevant or appropriate.
In this light, the automated quantitative approach can be used as an exploratory means to identify mixed-
use potentiality and understand the impact of the different aspects with respect to this potential. This approach
provides an objective underpinning that can serve as arguments to support the decision-making process on
the development location of mixed-use node. Once the selection of mixed-use nodes has been made, this
selection, in turn, serves to guide the land use and density planning that could serve to influence the future
master plan. The computational support to this planning process will follow a similar approach of location
choice and aggregate analysis, although the analyses will be time-dependent and differ both in the analyses,
including urban catchment, and in the aggregation.
Although the data considered for the sample here in the article focus on existing conditions, we argue this
method can be an approach to the decision-making process for future scenarios. Any development planned
for the future can be compared with the results from the existing data, thereby helping planners in the deci-
Alva et al. 347
Case study of land suitability analysis
The land suitability process is initiated by land developers looking for real estate developments. These land
developers will approach organisations and will submit a request for available land parcels that meet speci-
fied requirements. Through the land suitability assessment process, the organisation will select appropriate
land parcels to present to the land developers. The land parcels are selected based on the requirements, taking
into account planning regulations and other long-term strategic planning goals. The process iterates back and
forth filtering land parcels for different requirements and constraints in parallel.
A tool-chain has been developed and tested to semi-automate the land suitability assessment process. The
tool-chain combines various requirements and constraints into a single integrated query and helps planners
to select land parcels based on geospatial data analysis.
The land suitability tool-chain allows the planners to interactively query and interrogate an integrated
database of land parcel information. The decision-making tool also provides the ability to customise the land
query based on specific land developer requests. The tool-chains operate on GIS datasets called feature
classes. Feature classes are homogeneous collections of common features, each having the same spatial
representation, such as points, lines or polygons, and a common set of attribute columns. The geoprocessing
is continued in the vector format, as the output data type required is also plot-based and discrete. The output
feature classes are stored in geodatabase files (.gdb) due to the compact and efficient nature of the format.
The case study focuses on a land suitability assessment process for different land-use developments using
open-data available. We are considering four key factors for land suitability assessment: (1) land area, (2)
plot ratio, (3) location and (4) shape (see Figure 11).
Figure 11. Decision tree for land suitability analysis.
348 International Journal of Architectural Computing 17(4)
Land area. It is one of the key factors to be considered for land suitability. Along with the input plots, a new
set of plots are created by combining adjacent plots. A combination of second order is carried out using the
merge tool option in GIS (see Figure 12). Size of the land required is used as the threshold in this land-filter-
ing process. All input plots and combined plots with less than the required area are filtered out. By including
larger plots, a planner can eventually decide to subdivide provided it satisfies other requirements. A normal-
ised value based on the land area is calculated using equation (4)
where x is the area of the plot polygon and area is the required land area.
Plot ratio. The plot ratio values are assigned to each plot by the local authorities and are firmly depicted on a
land-use master plan. The required gross plot ratio (GPR) by the developer is taken as a threshold in the
query process to find the most suitable plot. Plots with less than the required GPR are filtered out. A normal-
ised value based on the plot ratio is calculated using equation (5) (see Figure 13)
Figure 12. Adapted equation (4) transforming land areas into the range of 0–1 (y-axis). For example, the required
area considered here is 2 ha (x-axis).
Figure 13. Adapted equation (5) transforming gross plot ratios into the range of 0–1 (y-axis). For example, the
required plot ratio considered here is 2 (x-axis).
Alva et al. 349
where x is the GPR of the plot polygon and PR is the required GPR.
Location. The need for good public transportation access, like plot proximities to MRT stations, public bus
interchanges and bus stops, is optional and user specific. This kind of location preference can also extend to
zones and specific points. Using the average distance values to the preferred location in equation (2) gives
the normalised values for every location preference. A weighted sum of these different normalised values
gives a location aggregate (see Figure 14).
Shape. A preference for the shape of a plot is optional and depends on a case-by-case requirement. A specific
plot requirement of a regular or non-elongated plot is normalised to a range of 0–1 using equation (6). Using
this equation, all the square plots will be most fit and have a normalised value of 1. The mid-range is consid-
ered as a rectangle with a length-to-width ratio of 2:1 (see Figure 15)
Figure 14. Adapted equation (2) transforming distances to MRT stations into probabilities within the range of 0–1
(y-axis). The average distance to MRT stations considered here is 450 m (x-axis).
Figure 15. Adapted equation (6) transforming plot shapes into probabilities within the range of 0–1 (y-axis).
350 International Journal of Architectural Computing 17(4)
where ratio is the length-to-width ratio of the plot.
To determine the land suitability of each plot, we consider the aggregation of the four aspects mentioned above:
land area, plot ratio, location and shape. The aggregation is achieved by summing the weighted outcomes of these
four analyses (Figure 11). The actual weights should be considered as parameters for the planner to play with.
Automated land suitability tool-chain
A geospatial tool-chain has been implemented within the ArcGIS platform. ArcGIS is commonly used within
planning organisations, making tool adaptation and utilisation easier. Our initial attempts to create tool-
chains were developed using the Model Builder visual programming environment within ArcGIS. However,
the programming paradigm lacked scalability and as a result, models quickly became very complex and
difficult to understand. In the final version, all tools were implemented as Python tool scripts. The tools can
be easily shared and deployed, becoming an integral part of the everyday working environment.
Documentation for each tool has also been developed, which can be accessed in the same way as the docu-
mentation for system tools.25 All the three tools include GUIs to allow users to enter the parameters. The
tools automatically filter out invalid data types and field names and will flag errors for the user, giving feed-
back and warnings. In addition, all the output data of the three tools are saved inside geodatabase files (.gdb)
located in a shared workspace in a network drive. This allows teams in the organisation to easily work
together, linking and sharing data in a seamless way (see Figure 16).
Geoprocessing tool. It calculates spatial data required for the analysis performed with the decision-making
tool. The main input to the geoprocessing tool consists of a feature class containing a set of land parcel
polygons, referred to as plots. The results of the geoprocessing tool are added to this feature class as addi-
tional attributes. Other feature classes required by the tool include the master plan with GPR information
and public transit information (for buses and MRT). The types of analysis include shape analysis, regional
or zone analysis, next to a specific customer/supplier, and additionally customised location preferences.
Additional feature classes may also be required depending on the types of analysis being performed.
Figure 16. Land suitability analysis tool-chain.
Alva et al. 351
Decision-making tool. It allows planners to interactively and iteratively assess available plots by setting the
requirements and defining parameter weightages (see Figure 17). The planners use the tool to enter the
requirements from the land request and the weightages for the decision tree parameters. The tool then assesses
the plot and assigns an overall fitness value to each plot, normalised to the range of 0%–100%. The fitness
value is added as an additional attribute to the feature class. Plots are then ranked, with the plots with higher
fitness being more suitable. In order to visualise this data, a heatmap of the plot fitness values is automatically
generated. The planner can then review the plots with high fitness values and either select a preferred plot or
adjust the parameter weightages and run a new query.
Communications tool. It helps the managerial team to check various additional requirements and constraints
based on the local authority code of practices for a selected plot. The tool checks two key constraints. The
first constraint to be checked is the effective buildable area of the selected plot. The buildable area is calcu-
lated taking into account various no-build zones. These include no-build zones imposed by adjacent roads,
with different road categories having different offsets. The second constraint to be checked is the permissible
building height of the selected plot. The building height is constrained by limits imposed by various govern-
ment agencies. The maximum floor area for the development is first calculated based on the plot area and
GPR, which is derived from the Master Plan 2014. The building height is then calculated by applying a set
of predefined rules for the selected building typology for the development of the buildable area of the plot.
This building height is then checked against the height limits for the plot.
The tool also automatically generates a formatted report (in .docx format) that summarises all relevant
information about the selected plot (see Figure 18). The information in the report includes text and tables
Figure 17. User interface of decision-making tool for land suitability in ArcGIS.
352 International Journal of Architectural Computing 17(4)
describing the key data for the selected plot, as well as all the regulations and codes of practice that are
applicable. The report also includes a formatted plan diagram of the selected plot, showing graphical repre-
sentations of some of the key constraints (see Figure 19). This report can be used for effective communica-
tion to the land developers or to officially register the chosen plot for tenure.
Figure 18. Workflow of communications tool for land suitability.
Figure 19. Result of communications tool for land suitability. A plot plan showing the calculated buildable area.
Alva et al. 353
The tool-chain has been tested with a wide variety of queries using open data collected for plots in Singapore.
As an example, a query for an industrial plot is presented. The query specifies the required plot area, GPR,
plot shape, location, together with a set of decision tree weights. The results of the query are described below
(see Figures 20 and 21). The Stage-1 geoprocessing tool was tested with a feature class containing 111,525
plot polygons. The time taken to execute the tool was around 30 min. Stage-1 found 7484 industrial plots and
created 11,555 second-order plot combinations. This resulted in a total of 19,039 plots available for query.
The stage-2 decision-making tool then filters these plots based on a set of specified weights and other
requirements, resulting in a total of 413 available plots. The decision maker can then select the most suitable
plot. Finally, the Stage-3 communications tool can then be used to check additional requirements and
Figure 20. Result of land suitability assessment query using tool-chains.
Figure 21. Input parameters and query output of the land suitability assessment.
354 International Journal of Architectural Computing 17(4)
constraints and to generate a report. At the end of Stage-3, the registration of the selected plot is fed back into
the database. This becomes a part of an organisational process, as the managerial team informs the GIS team
of these updates, thereby closing the geospatial data loop.
Within the framework, the system, organisation and its members go hand-in-hand in maintaining and
manipulating geospatial data. The tool-chain is currently limited to a specific number of input parameters. If
additional parameters need to be added, then these can be added. This would require both the geospatial data
schema and the Python script to be updated. However, such modifications may be difficult for users who are
not familiar with Python programming and who do not have knowledge of the various tool templates defined
within ArcGIS. Approaches are being investigated that are more flexible and modular, allowing additional
parameters to be defined without requiring programming.
The article has documented a spatial decision support framework and implementation focusing on the work-
flows that span across different teams within the organisation, with varying skill sets and objectives.
Geospatial tool-chains were implemented on top of ArcGIS, supporting decision making for planners. These
tool-chains help planners to visualise land potentials based on geospatial data analysis. The demonstration
shows how planning tasks within an organisation can be linked to seamless workflows.
The broader aim of this research is to develop a practice-orientated approach to creating PSS. A key prob-
lem with the existing SPSS and PSS systems is that they are developed in isolation of the workflows which
exist within organisations. This results in a lack of synergy between the tool and the workflows in practice.
Usability of the tool is also dependent on the existing workflows.
This research started by first mapping out end-to-end workflows that existed within the organisation.
Based on these workflows, modular and flexible tools were then developed on top of the existing GIS
infrastructure already being used within the organisation. This approach ensures that tools and systems are
well aligned with the workflows, resulting in effective usage. It enables effective intervention by the plan-
ners, improving the efficacy of the outcomes. Future research will apply the same overall approach to other
planning workflows, including long-term planning projections considering various dynamic urban and
Nevertheless, the automation procedures need to acknowledge planners’ points of view and the interests of
the planning community within their development. This requires further exploration of potential methods to
provide this inclusivity in automation. At the same time, ways of visualising data have to be evolved towards
engaging dialogues among decision makers who can think and plan for complexities of future cities.
The authors would like to thank the JTC Corporation, under the Deputy Directorship of Aloysius Iwan Handono, for
their support in providing information and technical data to facilitate the development of the workflows and
Declaration of conflicting interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship and/or publication of this
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this
article: This research received funding support from NUS-JTC Industrial Infrastructure Innovation Centre.
Alva et al. 355
Pradeep Alva https://orcid.org/0000-0001-7997-1894
Rudi Stouffs https://orcid.org/0000-0002-4200-5833
1. Batty M. Planning support systems and the new logic of computation. Reg Develop Dial 1995; 16(1): 1–17.
2. Klosterman RE. Planning support systems: a new perspective on computer-aided planning. J Plan Educ Res 1997;
3. Geertman S. Participatory planning and GIS: a PSS to bridge the gap. Environ Plann B 2002; 29(1): 21–35.
4. Geertman S and Stillwell J. Planning support systems: an inventory of current practice. Comput Environ Urban
2004; 28(4): 291–310.
5. Geertman S, Stillwell J and Toppen F. Introduction to ‘planning support systems for sustainable urban develop-
ment’. In: Geertman S, Toppen F and Stillwell J (eds) Planning support systems for sustainable urban development
(Lecture Notes in Geoinformation and Cartography), vol. 195. Berlin: Springer, 2013, pp. 1–17.
6. Arciniegas G and Janssen R. Spatial decision support for collaborative land use planning workshops. Landscape
Urban Plan 2012; 107(3): 332–342.
7. Boulange C, Pettit C, Gunn LD, et al. Improving planning analysis and decision making: the development and
application of a walkability planning support System. J Transp Geogr 2018; 69: 129–137.
8. Geertman S and Stillwell J. Interactive support systems for participatory planning. In: Geertman S and Stillwell J
(eds) Planning support systems in practice (Advances in Spatial Science). Berlin: Springer, 2003, pp. 25–44.
9. Vonk G. Improving planning support: the use of planning support systems for spatial planning. PhD Thesis,
Nederlandse Geograﬁsche Studies, Utrecht, 2006.
10. Schilder MC. Planning support systems in urban development in the Netherlands. MSc Thesis, Delft University of
Technology, Delft, 2016.
11. Pelzer P, Arciniegas G, Geertman S, et al. Planning support systems and task-technology fit: a comparative case
study. Appl Spat Anal Polic 2015; 8(2): 155–175.
12. Marshall S, Hudson-Smith A and Farndon D. Digital participation – taking ‘planning’ into the third dimension.
Town Country Plann 2019; 88(1): 11–14.
13. Heumann A and Davis D. Humanizing architectural automation: a case study in office layouts. In: Gengnagel
C, Baverel O, Burry J, et al. (eds) Design modeling symposium: design with all senses. Cham: Springer, 2019,
14. Zhu Y and Ferreira J. Data integration to create large-scale spatially detailed synthetic populations. In: Geertman S,
Ferreira J, Goodspeed R, et al. (eds) Planning support systems and smart cities (Lecture Notes in Geoinformation
and Cartography). Cham: Springer, 2015, pp. 121–141.
15. Klosterman RE, Brooks K, Drucker J, et al. Planning support methods: urban and regional analysis and projection.
Lanham, MD: Rowman & Littlefield, 2018.
16. Schirmer P, van Eggermond M and Axhausen K. The role of location in residential location choice models: a
review of literature. J Transp Land Use 2014; 7(2): 3–21.
17. Alva P, Janssen P and Stouffs R. A spatial decision support framework for planning: creating tool-chains for
organisational teams. In: Proceedings of the 24th CAADRIA conference on intelligent & informed (eds M Haeusler,
MA Schnabel and T Fukuda), vol. 2, Wellington, 15–18 April 2019, pp. 11–20, http://papers.cumincad.org/data/
18. Ogra A and Ndbele R. The role of 6Ds in transit oriented development (TOD). In: Proceedings of the neo-
international conference on habitable environments (eds S Bahga and A Singla), 2014, pp. 535–542. Jalandhar,
India: CreateSpace, https://core.ac.uk/download/pdf/54194442.pdf
19. Farr D. Sustainable urbanism: urban design with nature. Hoboken, NJ: John Wiley & Sons, 2008.
20. Suzuki H, Cervero R and Iuchi K. Transforming cities with transit: transit and land-use integration for sustainable
urban development. Washington, DC: World Bank, 2013.
21. Olszewski P and Wibowo SS. Using equivalent walking distance to assess pedestrian accessibility to transit stations
in Singapore. Transp Res Record 2005; 1927(1): 38–45.
356 International Journal of Architectural Computing 17(4)
22. Erath A, van Eggermond M, Ordonez S, et al. Modelling for walkability: understanding pedestrians’ preferences
in Singapore. In: Proceedings of the 14th international conference on travel behavior research, Windsor, 19–23
23. Calculation of road traffic noise according to the 2002 version of the Dutch calculation method, 1998. https://
rigolett.home.xs4all.nl/ENGELS/vlgcalc.htm (accessed 10 June 2019).
24. Lau SSY, Zhang J, Lau ESK, et al. A comparative study of road traffic noise based on measurement, prediction
and simulation according to the CRTN method in Singapore. In: Proceedings of the 9th cross-strait acoustic
conference, 12 August 2016, pp. 11–12. Macau, China: Macau University of Science and Technology.
25. Zandbergen PA (ed.). Python scripting for ArcGIS. 2nd ed. Redlands, CA: Esri Press, 2015.
26. Urban Redevelopment Authority (URA). Master Plan 2014 – statutory land use plan, https://www.ura.gov.sg/
Corporate/Planning/Master-Plan (accessed 18 October 2019).
27. The contents of the website – ‘MyTransportSG’ (https://www.mytransport.sg/) is owned and operated by the Land
Transport Authority of Singapore (LTA). The data for Mass rapid Transit station and Bus Stops are from the
following links from ‘MyTransportSG’ website. https://www.mytransport.sg/content/mytransport/home/dataMall/
static-data.html (accessed 18 October 2019).
28. Data.gov.sg – Singapore’s open data portal. Government of Singapore, 2017, https://data.gov.sg/ (accessed 18