Wildfire Burn Volume and Severity Estimation Using Drone Remote Sensing

Abstract

Wildfires are becoming a major environmental and social challenge worldwide. Accurate estimation of burn volume and burn severity is essential for post-fire management. This study used Unmanned Aerial Vehicles (UAVs) remote sensing as an efficient way to collect pre-fire and post-fire data and proposed a machine-learning algorithm to predict burn volume and burn severity using multispectral drone images. Researchers conducted one-year sequential drone mapping of a 4.1 km2 canyon site before and after a controlled burn that was conducted in Salinas, California, in October 2022. Based on in-situ ecological research, the high-resolution visible pre-fire mapping was processed by a supervised, object-based classification method to produce a pre-fire vegetation cover map. The burn volume estimation was based on the difference between the pre-fire and post-fire 3D models, using the highest temperature during the fire and the species classification as analysis thresholds. The Random Forest algorithm was applied to multispectral drone images to predict fire severity and vegetation resprouting. Post-fire ecological indices in 40 sampling points were collected in situ and used as labels. The fire severity index was predicted for the whole canyon site and was compared to the Normalized Burn Ratio index from multispectral drone images. Predicted resprouting patterns were evaluated by in-situ ecological research. This study composed an effective way to estimate the burn volume and to predict burn severity in wildfire events using drone remote sensing. The UAV-based approach can provide valuable information for fire management and ecological restoration in California.