Wildfires damage assessment Via LiDAR

Abstract

This paper examines the phenomenon of wildfires in California and investigates the buildings affected by the Woolsey Fire in Central Malibu in 2018. We focus empirically on machine learning to identify damaged objects from point-cloud data. This project includes a literature review with references to methods used for wildfire research and LiDAR data processing. In this study, researchers trained an existing deep learning model to determine if it offers an effective solution for extracting damaged objects. Data sources for this study include point-cloud data retrieved via the LidarExplorer tool and Kaggle’s 2013–2020 California wildfire data. Using two layers of building footprints in the Malibu “T-Zone” revealed 907 structures, of which 435 were damaged or destroyed based on map observations. This analysis of structure identification supports the literature that deep learning can successfully classify objects damaged by wildfires.