Coupling Multiagent Geosimulation and Spatial OLAP for Better Geosimulation Data Analysis

INTRODUCTION AND MOTIVATIONS Simulation output analysis is a very relevant step in a simulation study. This step is necessary to test different ideas and to learn about the simulation model and the corresponding simulation system. A user needs to better understand the simulation model's output and, consequently, needs appropriate techniques to analyze the simulation output. Some researchers (Seila 1992, Alexopoulos et al. 1998, Alexopolous 2000, Sanchez 2001) propose different analysis techniques for simulation outputs. Decision makers usually find the statistical and mathematical aspects of these analysis techniques difficult to use. To overcome this difficulty, Grier (1992) proposed a graphical statistical analysis technique that can be used to analyze simulation outputs. The visual display of the results quickly conveys information about the simulation model. Users who rely on simulations to support their decisions prefer graphical analyses because they are easy to understand. Blaisdell (Blaisdell et al. 1992) proposed SIMSTAT, a tool used to analyze simulations based on a statistical graphical analysis technique that is combined with several discrete-event simulation tools. Using statistical graphical analysis is efficient for event-discrete simulations or other kinds of simulations that do not deal with spatial or geographic data. However, for certain geosimulation (GS) fields such as urban simulation (US) or multiagent geosimulation (MAGS), spatial data become an important issue for decision makers. In such simulation fields, a classical analysis technique based on tables and statistical graphs is too limited for spatial analysis: no spatial analysis (e.g., correlation between spatial variables), no spatial visualization, no map-based or cartographical exploration of spatial data (which is relevant to reveal clusters, the proximity between two phenomena or the spatial area of a phenomenon), etc. Multiagent geosimulation (MAGS) has a great potential when explaining the subtle interactions of heterogeneous actors in complex social systems taking into account the geographic aspect of the simulation environment. The characteristics of the agents (autonomy, social ability, proactiveness, advanced spatial behavior such as perception, navigation, and memorization, etc.) and the spatial features of the simulation environment make MAGS an attractive approach to develop simulations of complex systems. In addition, the complexity of the simulation models and their visualization capabilities (cartographical visualization, 2-D-3-D displays) make them more realistic and, therefore, closer to users' mental models. Generally, simulation applications generate outputs, which need to be interpreted by the users. The huge volume and the complexity of these outputs, a big part of which is spatial, make them difficult to be interpreted and analyzed by the users. To analyze these outputs of simulations that are performed in spatial environments, we need more sophisticated analysis techniques, which can be used to analyze and explore complex simulation models and outputs involving geographic data. These analysis techniques must also be compatible with users' mental models and must generate analysis results that can be easily exploited by users. Consequently, the traditional statistical or mathematical analysis techniques are less suited to analyze geosimulation outputs because they are less efficient to analyze and explore the spatial aspects of the simulation outputs, which are very important for a geosimulation. After an in-depth comparison of several analysis techniques and tools (Seila 1992, Blaisdell et al. 1992, Grier 1992, Kelton 1997, Alexopolous et al. 1998, Sanchez 2001, Alexopolous 2002), we found that the most appropriate one for simulation outputs is the online analytical process (OLAP) because: (1) OLAP allows users and analysts to explore data in the way they think, across multiple variables, called dimensions, at the same time (Codd et al. …