A Bayesian approach to modeling lost person behaviors based on terrain features in Wilderness Search and Rescue

Abstract

In Wilderness Search and Rescue (WiSAR), the incident commander (IC) creates a probability distribution map of the likely location of the missing person. This map is important because it guides the IC in allocating search resources and coordinating efforts, but it often depends almost exclusively on the missing person profile, prior experience, and subjective judgment. We propose a Bayesian model that uses publicly available terrain features data to help model lost-person behaviors. This approach enables domain experts to encode uncertainty in their prior estimations and also makes it possible to incorporate human behavior data collected in the form of posterior distributions, which are used to build a first-order Markov transition matrix for generating a temporal, posterior predictive probability distribution map. The map can work as a base to be augmented by search and rescue workers to incorporate additional information. Using a Bayesian ? 2 test for goodness-of-fit, we show that the model fits a synthetic dataset well. This model also serves as a foundation for a larger framework that allows for easy expansion to incorporate additional factors such as season and weather conditions that affect the lost-person's behaviors.