Predictive Probability Density Mapping for Search and Rescue Using An Agent-Based Approach with Sparse Data

TL;DR

This work tackles the challenge of predicting lost-person locations under sparse data by introducing J2, an agent-based predictive density mapper that simulates four land-cover–driven LP behaviors to navigate real landscapes without location-specific training. The method combines Monte Carlo path generation with mobility-time sampling and validates the PDMs using Gaussian Processes trained on real SAR data, enabling robust starting-point sampling from limited data. Results on Isle of Arran show that J2 achieves substantially closer alignment to historical location data than prior approaches, with a symmetric KL score of $61.56$ versus $306.02$ for the earlier version and far better than a random baseline. The approach demonstrates data-efficient generalization and has practical implications for UAV-assisted SAR by providing reliable PDMs that adapt across locations and terrain types.

Abstract

Predicting the location where a lost person could be found is crucial for search and rescue operations with limited resources. To improve the precision and efficiency of these predictions, simulated agents can be created to emulate the behavior of the lost person. Within this study, we introduce an innovative agent-based model designed to replicate diverse psychological profiles of lost persons, allowing these agents to navigate real-world landscapes while making decisions autonomously without the need for location-specific training. The probability distribution map depicting the potential location of the lost person emerges through a combination of Monte Carlo simulations and mobility-time-based sampling. Validation of the model is achieved using real-world Search and Rescue data to train a Gaussian Process model. This allows generalization of the data to sample initial starting points for the agents during validation. Comparative analysis with historical data showcases promising outcomes relative to alternative methods. This work introduces a flexible agent that can be employed in search and rescue operations, offering adaptability across various geographical locations.

Figures (12)

• Figure 1: The process of extracting the weighted average angle $\bar{\theta}$ from $E^{k-\text{nearest}}$ for a given position $\mathbf{m}$ on the grid.
• Figure 2: Example path network. If the agent moves from $0$ to $2$, the edge $(0,2)$ will have a score of $0.1$ assigned to it whilst both $(2,3)$ and $(2,1)$ will have a score of $1$. After normalizing the scores, this gives a probability of $4.76\%$ to backtrack to $0$.
• Figure 3: Discrete mobility time (hours) from both koester_lost_2008 (n=232) and perkins_missing_2011 (n=132) along with their respective continuous Log-Normal curve. perkins_missing_2011 provides distance from PLS data in kilometers, and was scaled by $3.87km\per h$ from gast_preferred_2019
• Figure 4: Normalized ($x$-,$y$- and $z$-axis) heatmap derived from the raw SMR PLS data
• Figure 5: The resultant heatmaps after applying conventional methods to Fig. \ref{['fig:pls_raw_data_heatmap']}