Kernel density estimations for visual analysis of emergency response data

The purpose of this chapter is to investigate the calculation and representation of geocoded fire & rescue service missions. The study of relationships between the incident distribution and the identification of high (or low) incident density areas supports the general emergency preparedness planning and resource allocation. Point density information can be included into broad risk analysis procedures, which consider the spatial distribution of the phenomena and its relevance to other geographical and socio-economical data (e.g., age distribution, workspace distribution). The service mission reports include individual points representing the x/y coordinates of the incident locations. These points can be represented as a continuous function to result in an effective and accurate impression of the incident distribution. The continuity is recognized by kernel density calculations, which replaces each point with a three-dimensional moving function. This method allows to control the degree of density smoothing by the search radius (also referred to as bandwidth) of the kernels. The choice of the kernel bandwidth strongly influences the resulting density surface. If the bandwidth is too large the estimated densities will be similar everywhere and close to the average point density of the entire study area. When the bandwidth is too small, the surface pattern will be focused on the individual point records. Experimentation is necessary to derive the optimal bandwidth setting to acquire a satisfactory case-specific density surface. The kernel density tools provided in standard GIS (like ArcGIS) software suggest a default calculation of the search radius based on the linear units based on the projection of the output spatial reference, which seems to be inadequate to display incident density. Within this chapter we provide a flexible approach to explore the point patterns by displaying the changes in density representation with changing search radius. We will investigate how the parameters can be optimized for displaying incident distribution in relation to the service areas of Fire and Rescue stations.