A Monte Carlo approach to modeling lost person behavior in wilderness areas using a geographic information system

Abstract

Monte Carlo modeling techniques using mean information fields (MIF), developed by Torsten Hagerstrand in the 1950s, were integrated with a geographic information system (GIS) to simulate lost person behavior in wilderness areas. Big Bend Ranch State Natural Area, a 265,000 acre park acquired by Texas Parks and Wildlife, was used as the study area. A comprehensive GIS data base for the park was developed, including relevant layers of contours, hydrography, vegetation, roads/trails, and other man-made features. Travel behaviors were collected from 1 18 lost person mission reports provided by the National Association of Search and Rescue (NASAR). Two models for travel rate and time mobile were developed from these data using linear regression; the first model providing probabilities for input into the MIF based on four significant variables: age, age,-4, sex, and equipment. Only age was significant in determining the total time mobile for the second model. In a simulation program written in ARC/INFO, these initial probabilities in the MIF were modified according to attributes of the terrain underlying the MIF and according to the lost person's characteristics. Visibility was calculated for each cell in the MIF, and the probabilities of the visible cells were converted to ranges. A random number was generated, and the cell encompassing that random number became the new center of the MIF. Several Monte Carlo simulation runs were conducted, and model behavior was observed by outputting the simulation waypoints and the topographic data on 1:24,000 color maps. The adult lost person profile model migrated an average of 4 kilometers in a predominantly northwest direction from the test point and chose locations which contained drainage features, roads, or gentle slopes. The northwest direction of travel appeared to be partly a result of the southeastern trend of the underlying topography. These choices reflect the lost person behavior data model well, although distance traveled was overestimated and underestimated for minimum and maximum values, respectively. Five simulations for a child profile were performed as a comparison to the adult profile. These simulation runs yielded shorter distances and a preference for gentler slopes.