Interactive, GPU-Based Urban Growth Simulation For Agile Urban Policy Modelling

In this paper we present a novel approach of simulating urban growth by utilising the computation power of modern GPUs. The simulation results can be used in urban policy modelling to reduce turnaround times in the policy cycle. We use a state-of-the-art agent-based simulation model that consists of rules to describe human behaviour. The simulation incorporates geospatial information such as land-use, current population density and road network data. In order to simulate the phenomena of urbanisation, in our model citizens more likely settle near roads or existing settlements/cities. In this paper we present our implementation that is based on the FLAME GPU framework. Each agent on the GPU represents a group of citizens at a specific location. In order to evaluate our approach we present a practical use case. We measure the performance of our implementation and compare it with a Java-based solution. Finally, we discuss our approach and show opportunities for agile and interactive urban policy modelling. INTRODUCTION AND MOTIVATION The term “urban sprawl” describes the problem of modern cities growing quickly resulting in wide-spread developments with a very low density. This often has negative effects on environment and therefore on people’s health: more land is covered with buildings or streets; public transport in suburbs is often not sufficiently developed and so citizens have to use the car to get to their job or to the city centre which effectively leads to a higher air pollution. Besides, urban sprawl may also affect the cultural life and family life. People living in suburbs sometimes participate less in cultural events than people living near the city centre. Long travels to work and back to home reduce the time an employee can spend with his or her family. Urban planning and policy modelling therefore aim for creating more compact but at the same time sustainable and healthy cities. This development requires infrastructure changes that have to be well thought out. So, urban planners more and more involve citizens in the discussion about urban development plans in order to create a city that is well received by everyone. They make use of simulations based on geospatial information. Innovative techniques such as 3D visualisation help urban planners to present the simulation results to decision makers and to the public. Modern urban policy modelling deploys a so-called policy cycle (see Krämer et al., 2013). Simulations and 3D visualisations are used to gain feedback from decision makers and citizens. This feedback can then be incorporated in new simulations which are presented to the public again. This loop repeats until a general agreement on the planning has been found. The shorter the feedback cycle is, the faster a final decision can be made. Creating such simulations is currently a timeconsuming task that may take several hours or even days with existing solutions (see section “Performance” below). Urban planners often make use of modern satellite imagery to improve their calculations. For example, satellite images or LIDAR data spanning several years allow urban planners to calculate urban growth and hence to estimate future trends. The ongoing development of sensor technology leads to more accurate data sets which may be exploited to achieve better simulation results. However, at the same time the volume of data to process becomes larger and larger which makes them harder to process with standard geospatial information systems (GIS). Nonetheless, quickly creating simulations based on such data sets is a crucial part for the urban policy feedback cycle. Modern computer architectures with multi-core CPUs and GPUs allow for creating high performance applications (cf. Owens et al., 2007). However, current GIS solutions do not fully take advantage of this yet. In practice, urban planners process raster data or point clouds such as satellite images or LIDAR data respectively with software tailored to simple workstations. In recent years these workstations have evolved and already include sophisticated graphics hardware. With this hardware it now becomes possible to not only create high performance 3D visualisations but also to make use of the thousands Proceedings 27th European Conference on Modelling and Simulation ©ECMS Webjørn Rekdalsbakken, Robin T. Bye, Houxiang Zhang (Editors) ISBN: 978-0-9564944-6-7 / ISBN: 978-0-9564944-7-4 (CD) and millions of cores offered by a modern GPU to create geospatial simulations. Modelling the behaviour of citizens in an urban environment can be rather complex in that it is non-linear and possibly chaotic. A lot of individual factors have to be taken into account that make the model large and hard to comprehend. Agent-based modelling (ABM) attempts to simplify such problems. Agents are autonomous units that act on their own, just like citizens. Modelling urban life becomes a lot easier by considering only one citizen or a group of similar citizens and by representing them as individual agents. Modern graphics hardware allows agent-based simulations to run on the GPU. So, it is possible to create high-performance simulations modelling urban life on the graphics card. To summarise, in order to create sustainable, compact cities, urban planners deploy a feedback cycle that is based on geospatial simulations. The shorter this cycle is, the faster decisions can be made. However, geospatial data—which provides the basis for such simulations— becomes larger and larger and so simulations take more and more time with current GIS technology. In this paper we therefore present a new approach of interactively simulating urban development with modern GPU hardware. We use agents to model real urban life. We describe our implementation and evaluate its performance compared to a pure Java application. We conclude with a final discussion on the applicability of our approach to a practical use case, and we show opportunities for agile urban policy modelling.