Near real-time wildfire building damage assessment with ICEYE SAR data

Wildfires are becoming larger and more frequent as a result of global warming, since hotter temperatures help create the conditions for increased fire activity. Wildfires thus constitute a serious threat for human life, and can cause catastrophic damages and property losses. When such disasters occur, it is critical to quickly assess their impact, so that authorities can make informed decisions about the safety of the population, and insurance practices can be initiated for the damaged properties. 

Typically, such assessment was done by manually inspecting aerial optical imagery, once available, to determine damages to the individual properties. However, safely flying over affected areas requires sufficient visibility and favorable wind conditions. Moreover, both the data acquisition and the   successive manual inspection are costly and time consuming processes. 

Satellite remote sensing can avoid the risk and costs associated with on-site surveys. In particular, Synthetic Aperture Radar (SAR) satellite sensors do not rely on daylight and are insensitive to smoke and clouds, therefore they can be used at all times of day and night during and after the event, in all weather conditions. The limiting factor for satellite imagery is normally the timeliness, given that the frequency with which a certain area is overpassed by most  satellites is in the range of several days or even longer. This is why having access to a constellation of SAR satellites that deliver near-real time imagery, on a  global scale, is a game changer in disaster assessment. 

ICEYE developed a specific solution for building damage evaluation, relying on prompt tasking SAR images  from their  large constellation of NewSpace satellites, and using machine learning models to quickly provide situational awareness on the whole fire perimeter, at a building level. The method is based on a post-event image only, without requiring any pre-event imagery.  In this presentation,  we present several case studies  of wildfire events that occurred in different regions of the US in 2023 and 2024, showing the damage maps obtained, and the performance achieved in classifying each building as damaged or undamaged. The results were compared with the ground truth information such as provided by official governmental entities or retrieved from aerial photographs. The automatic assessment performance metrics were then derived.