Near Real Time Damage Detection


June 23, 2023


The purpose of this project is to build a platform capable of delivering accurate and timely damage assessments in the aftermath of natural disasters and armed conflict. The backbone of this platform is a new algorithm that uses open-access radar satellite imagery to detect building damage over large areas, twice per week regardless of weather conditions. Damage assessments using this methodology have been featured in the Economist twice:

  • For all of Ukraine at several intervals throughout the war

  • Following the devastating earthquake in Turkey


The Pixel-Wise T-Test (PWTT) is new Building Damage Assessment algorithm that uses Synthetic Aperture Radar imagery. The algorithm has been deployed and validated in several cities damaged during the war in Ukraine. Despite being simple and lightweight, the algorithm produces results with accuracy statistics rivalling State of the Art methods that use deep learning and expensive high resolution imagery. Furthermore, the workflow is deployed entirely within the Google Earth Engine environment, allowing for the generation of near-real time damage maps that allow humanitarian practitioners to immediately get the count of damaged buildings in a user-specified area of interest.

The next two sections showcase the PWTT algorithm applied to two different case studies: Mariupol in Ukraine, and Kahramanmaras in Turkey. On the “How it Works” page is a poster presented at the 2023 International Conference on Learning Representations which describes the PWTT algorithm. Further technical details can be found in the peer reviewed conference paper on the same page. Finally, an annotated walkthrough of an earlier version of the source code applied to the 2020 Beirut port explosion can be found in the Beirut Blast Assessment tab.

Ukraine Damage Assessment

The interactive application below shows the PWTT algorithm applied to Mariupol, Ukraine. Red and purple areas indicate a high likelihood of building damage. The application allows users to zoom in and out, and to pan around the map. Users can draw a polygon around a specific area of interest, which will display the footprints of buildings that are likely to be damaged, and report the number and proportion of damaged buildings in the area.

To assess the accuracy of this workflow, manually annotated building damage labels from UNOSAT are used. A dropdown menu under the “Accuracy Assessment” heading in the tool will visualize these labeled building footprints for either Mariupol or Irpin, and generate an ROC Curve.

For areas in which such labeled data are not available, the “Show Geolocated Footage” button will display conflict events from the Centre for Information Resilience (CIR) Ukraine Monitoring Map. These events can be used to visually verify the accuracy of the PWTT algorithm.

Turkey Earthquake Damage Assessment

A key strength of the PWTT is that it can be applied to any area of the world, regardless of weather conditions, differences in building types, and different disaster types. Below, the PWTT algorithm applied to Kahramanmaras, Turkey, following the 2023 earthquake. Userts can pan/zoom to other affected cities, such as Hatay in the south. Predicted damage can be visually verified using the optical basemap, which displays post-disaster imagery.

For details on how the PWTT algorithm works, see How it Works.