Workshop A Workshop on KDE+: A method for identification of animal–vehicle–collision hotspots

This workshop will cover:

  • Methodological issues: KDE+ basics
  • Introduction to KDE+ software
  • Practical application to test your own data
  • Discussion

Presenters: Michal Bíl, Richard Andrášik, Jiří Sedoník

Affiliation: CDV – Transport Research Centre, Brno, Czech Rep.

Keywords:road kill; traffic accidents; hotspots; kernel density estimation; Monte Carlo simulations

Workshop summary:

The following workshop is a hands-on session run on computers during which participants will test the KDE+ software and apply the spatial clustering method to their own data.

The KDE+ method is based on kernel density estimation and Monte Carlo simulations and objectively selects significant clusters. The strength of a cluster is important to drivers as it represents the individual danger. Road administrators are, however, more interested in mitigating locations with a high density of animal-vehicle collisions. We investigated the relationship between the strength of a cluster and the density of animal-vehicle collisions within the cluster. We found that these two attributes of a cluster can be used to express the collective danger. The clusters should therefore be ranked according to this alternative measure in order to select those locations that should receive a higher priority for mitigation. If our aim is to alert drivers regarding the most dangerous locations, we should use the strength of a cluster for ranking. Both measures can be used to rank the clusters across the entire road network.

The original KDE+ method is designed for animal-vehicle collisions that are localized by GPS. However, a number of databases in European countries do not contain GPS locations or the GPS locations contain significant errors. As a result, we cannot be sure about the exact position of an animal-vehicle collision. We consequently extended the applicability of the KDE+ method to also address inaccurate position data by deriving a new kernel function that considers this uncertainty. Although the accuracy of results depends on the precision of input data, the localization of clusters is also satisfactory in the case of inaccurate position data.

KDE+ is a unique method because it not only determines the significance of the clusters objectively but is also capable of processing inaccurate position data. In addition, it is not influenced by under-reporting of animal-vehicle collisions as most of the methods for crash-frequency data are. Based on our results, only half or even less of the data is sufficient for precise localization and ranking of the clusters.

We have developed standalone KDE+ software which can be downloaded free of charge from We have also recently programmed ArcGIS ToolBox based on the KDE+ approach. Both software will be available for testing.

Information for participants:

The participants are encouraged to bring notebooks with their own data. There is no need to have any GIS software on your PC if you only plan to test the standalone KDE+. ArcGIS 10.2 or higher is necessary if you wish to test ArcGIS Toolbox. The presenters also provide the participant with test data sets from actual AVC data.

road kill; traffic accidents; hotspots; kernel density estimation; Monte Carlo simulations