PNNL

Abstract

Aerial application of fire retardant is a critical tool for the management of wildland fire spread. Retardant applications are carefully planned to maximize fire line effectiveness, improve firefighter safety, protect high-value resources and assets, and limit environmental impact. However, to-pography, wind, visibility, and aircraft orientation can lead to differences between planned drop locations and the actual placement of the retardant. Information on the precise placement and areal extent of the dropped retardant can provide wildland fire managers with key information to 1) adaptively manage event resources, 2) assess the effectiveness of retardant slowing or stopping fire spread, and 3) document location in relation to ecologically sensitive areas. In this study, we use Sentinel-2 satellite data and machine learning classifiers to test an automated approach for map-ping retardant application. We show that a multiclass model (retardant, burned, unburned and cloud artifact classes) outperforms a single class retardant model and that image differencing (post-application-pre-application) outperforms single image models. Compared to random forest and support vector machine, the gradient boosting model performed the best with an overall ac-curacy of 0.88 and an F1 Score of 0.76 for fire retardant, though results were comparable for all three. Our model maps the full areal extent of the dropped retardant rather than simple lines currently mapped by aerial survey. Spaceborne remote sensing coupled with machine learning models can provide the capability to rapidly map the footprint of recently dropped fire retardant over large regions.