An Open-Source Tool for Mapping War Destruction at Scale in Ukraine using Sentinel-1 Time Series

TL;DR

A scalable method for estimating building damage resulting from armed conflicts is presented, by training a machine learning model on Synthetic Aperture Radar image time series to generate probabilistic damage estimates at the building level, leveraging existing damage assessments and open building footprints.

Abstract

Access to detailed war impact assessments is crucial for humanitarian organizations to assist affected populations effectively. However, maintaining a comprehensive understanding of the situation on the ground is challenging, especially in widespread and prolonged conflicts. Here we present a scalable method for estimating building damage resulting from armed conflicts. By training a machine learning model on Synthetic Aperture Radar image time series, we generate probabilistic damage estimates at the building level, leveraging existing damage assessments and open building footprints. To allow large-scale inference and ensure accessibility, we tie our method to run on Google Earth Engine. Users can adjust confidence intervals to suit their needs, enabling rapid and flexible assessments of war-related damage across large areas. We provide two publicly accessible dashboards: a Ukraine Damage Explorer to dynamically view our precomputed estimates, and a Rapid Damage Mapping Tool to run our method and generate custom maps.

Figures (11)

• Figure 1: Spatial distribution of building damage in Ukraine after two years of conflict. Percentage of buildings likely damaged within the first two years of the war, aggregated by hromadas. The predictions were thresholded at 0.655, and only buildings larger than 50$\,$m2 were considered.
• Figure 2: Damage estimates and UNOSAT labels for Chernihiv. Building damage estimate for Chernihiv thresholded with the standard confidence threshold of 0.655, aggregated over the first two years of conflict. For clarity, we only present the damage heatmap in the main figure, while three zoomed insets show building-level predictions. Red dots represent UNOSAT annotations indicating buildings marked as either destroyed or severely damaged. The VHR satellite layer is displayed solely for visualization, all results are derived from 10$\,$m-resolution Sentinel-1 images. Readers are encouraged to explore the maps on the interactive https://olidietrich.users.earthengine.app/view/ukraine-damage-explorer. Sources: Google Earth/Maxar Technologies, Overture Maps building footprints, UNITAR/UNOSAT damage annotations.
• Figure 3: The Rapid Damage Mapping Tool. Screenshot from the https://olidietrich.users.earthengine.app/view/rapid-damage-assessment-sentinel1, displaying damage estimates for a region in Mariupol, alongside the Sentinel-1 time series visualizer.
• Figure 4: Regions analyzed by UNOSAT. Overview of the 18 regions assessed by UNOSAT in Ukraine. Blue and orange dots indicate cities used for training and testing, respectively. The three insets highlight the variation in damage density across different regions. Background map: OpenStreetMap.
• Figure 5: Overview of the machine learning framework. For training (left), we use per-pixel Sentinel-1 time series extracted at the location of UNOSAT point annotations. The model is fed with a pair of time series from the same location. The first one spans a fixed 12-month time interval $T_0$ from 2020, and the second one spans one of the 3-month time intervals $T_n$ between 2021 and 2023. Both time series are encoded with a custom features extractor, and damage labels are dynamically assigned according to $T_n$. At inference time (center), the model generates a damage probability heatmap valid at $T_n$ and spanning the entire country, aggregating the predictions of different Sentinel-1 orbits. The raw damage probabilities are intersected with building footprints. For the final map the estimates for different time intervals $T_n$ are thresholded and aggregated.