Search and Rescue

In Wilderness Search and Rescue, finding the lost target as quickly as possible is the best way to ensure the target's survival. One way to facilitate a quick rescue is knowing where to look. My research focuses on developing probability based models predicting lost person behavior. This project is affiliated and supported by SARBayes.

Current Work:

Using gas simulations inspired by Burgen and Darken, I am developing a diffusion based model for human movement. Using elevation and land cover data, I estimate the relative walking speed in each cell. This is used as the maximum speed of diffusion across the cell. This model, run out to infinite time/equilibrium, has been tested on MapScore for roughly 200 cases in Yosemite National Park, and is showing promising results thus far.

DiffusionElev_YOSAR101
DiffusionElev_YOSAR090
DiffusionElev_YOSAR058
DiffusionElev_YOSAR026
DiffusionElev_YOSAR008
DiffusionElev_YOSAR005
DiffusionElev_YOSAR001
DiffusionElev_Arizona13
DiffusionElev_Arizona12
DiffusionElev_YOSAR085

Sample maps generated with an elevation based diffusion model. The lighter pixels denote higher probabilities of the target being found in the corresponding cell.

Past Work:

Using the International Search and Rescue Incident Database (ISRID)and Bob Koester's Lost Person Behavior, I have created different models in ArcGIS and Python. These models are based off of features such as distance from the last known point, elevation, and offset from roads/trails/rivers/etc. These models were also scored with Mapscore.

 

One specific model fitted distance from the last known point to a log-normal distribution. The distributions were calibrated to data from ISRID for each target category. I then ran statistical cross validation on the results, and a complete rightup can be found here.

Distance2
Linear Features
Land Classification
Elevation
DELL