A Data-Driven System for Probabilistic Lost Person Location Prediction

Today, when a report of a lost person occurs, both the Search And Rescue (SAR) team and Lost Person (LP) have limited access to assistive technologies, leaving manual or ad-hoc search planning as an all too common solution. Geospatial data exists, however, that when coupled with appropriate models and algorithms can enable decision support systems to help predict the location of lost persons and provide guidance for optimal search execution given the available search resources. The environments and context for application of these technologies, however, introduce several key complexities. The data required for accurate analysis and prediction (e.g., elevation, land cover, exclusion zones, known markers) can be large and the exact subset needed for any particular incident may not be known until the lost person event occurs. The algorithms required to generate location probability distributions are compute intensive in comparison to the limited compute resources available on the devices located closest to the incident or carried by a search team. That search team is by design, distributed, conducting operations with multiple independent operators, often in areas with limited, degraded access to network infrastructure. This paper describes the design, algorithms, models, and evaluation of software entitled LandSAR that employs geospatial datasets and tooling in a distributed context to address these challenges and enable such capabilities at