Development of a methodology to predict the sites with the highest accumulation of wildlife roadkill (hot spots) on roads administered by the La Pintada Concession based on artificial intelligence algorithms and geographic information systems.

Wildlife roadkill is a phenomenon that arises from the construction of road infrastructures, this is characterized by collisions between vehicles and wildlife. Unfortunately, Colombia is no stranger to the problem of wildlife roadkill, reaching millions of dead animals per year on the country’s roads. This phenomenon is characterized by following spatial patterns that can be analyzed through geostatistical techniques. In recent years, artificial intelligence has been used to predict spatial phenomena such as forest fires, surprise floods, among others. However, despite being a spacial phenomenon, currently there are no investigations that apply artificial intelligence techniques to the prediction of roadkill hot spots. Therefore, this project aims to develop a methodology to predict roadkill hot spots on roads administered by the La Pintada Concession (Antioquia – Colombia) based on artificial intelligence algorithms and geographic information systems, this will be achieved through algorithm training based on roadkill data, this training must be validated through cross-validation. Additionally, a transfer learning experiment will be carried out on roads administered by the La Pintada Concession, which will be validated based on the information obtained in the field. The methodology developed will allow reducing times and costs associated with the previous studies for the mitigation of wildlife roadkill. Additionally, it would allow the prediction of roadkill hotspots during the planning stage of road infrastructures. With this information more wildlife friendly designs could be created which would save millions of wildlife lives per year, protecting the ecosystem services associated with each of the species, thus allowing sustainable development.