Sentinel-2 Urban Change Analysis for the Venezuela Earthquake June 2026

This dataset provides a rapid experimental Sentinel-1-based damage assessment product for the broader impacted area from the 24 June 2026 Venezuelan earthquakes. It includes both the entire analyzed area of the SAR analysis and the Overture Maps building footprints identified as likely damaged. This SAR product estimates 81,437 likely damaged buildings across the analyzed impacted area. This preliminary damage proxy product is intended to support rapid screening and prioritization, targeting of follow-up assessment, triangulation with VHR optical and radar images, and other damage assessments and products.

Following the June 2026 earthquakes that struck the central coast of Venezuela, the Microsoft AI for Good Lab ran their building damage assessment AI models to map and classify the buildings affected by the earthquakes. The models were applied to satellite imagery captured on 25 and 26 June 2026, producing building footprints with predicted damage across the affected areas of Catia La Mar, La Guaira, and Caraballeda. For each assessed area, the collection provides a reference area map, a valid-area mask describing the extent of the analysis, and a GeoPackage of predicted building damage footprints. This collection mirrors the dataset published on the Humanitarian Data Exchange (HDX) by the Microsoft AI for Good Lab and made available under the Creative Commons Attribution 4.0 International (CC-BY-4.0) license. ## Methodology The Microsoft AI for Good Lab ran their damage assessment AI models on satellite images taken 25 and 26 June 2026 and have mapped out the affected buildings. ## Disclaimer This analysis represents a portion of the damage from the earthquake; other damage in the region is not included in this assessment. While these insights offer a valuable initial overview, they remain preliminary and require on-the-ground verification for a comprehensive understanding of the damage. ## Caveats Please note that the source imagery has orthorectification quality issues, which may cause building footprints to be misaligned with existing basemaps.

# Venezuela Earthquake Caracas: Damage Proxy Maps, Jun 2026 The **Earth Observatory of Singapore – Remote Sensing Lab (EOS-RS)** produced these damage proxy maps for Venezuela Earthquake Caracas (Jun 2026) from synthetic aperture radar (SAR) satellite images. Each map is a separate ZIP below with its GeoTIFF, a KMZ for Google Earth, the PNG figures and a text file describing how it was made and how to read it. ## Maps in this dataset - **EOS-RS Damage Proxy Map: La Guaira, Venezuela, Earthquakes, 25 June 2026, v0.8 ** - **EOS-RS Damage Proxy Map: Yuracuy Region, Venezuela, Earthquakes, 25 June 2026, v0.8 ** **More map details and files:** **Legal Disclaimer:** **Credits:** Earth Observatory of Singapore - Remote Sensing Lab (EOS-RS), Contains modified Copernicus Sentinel data (2026)

Dates of Images: Post-Event: 06/25/2026 Pre-Event: 04/07/2026-06/13/2026 (8 images total) Summary: The NASA GSFC landslides team produced this landslide proxy heatmap from Sentinel-1 imagery. Potential landslides were detected by calculating the change in the backscatter coefficient before and after the triggering event using the log ratio approach (Iratio). The Iratio is generated by subtracting median pre- and post-event Radiometrically Terrain Corrected (RTC) gamma0 backscatter in decibel scale at 10 m resolution obtained from the Alaska Sar Facility’s Hybrid Pluggable Processing Pipeline (HyP3). This map should be used as a guide to identify areas likely affected by landslides. This is a rapid response product. We have not done any form of manual verification or corrections to remove false positives. The likely landslide pixels are identified using 99th percentile Iratio threshold considering areas greater than 10-degree in slope and not affected by radar shadow and layover effects. A heatmap is generated using identified landslide pixels for visualization ease. Pre-event imagery = median imagery generated using images from 04/07/2026-06/13/2026 (8 images total) Post-event imagery = 06/25/2026 Suggested Use: The red and yellow areas indicate potential zones of dense landsliding. False positives related to natural and human-made ground surface change, including but not limited to deformation from mining, construction, deforestation, agriculture, flooding, snow cover, and changes in reservoir water levels can also be within the red and yellow areas. Satellite/Sensor/Resolution: Synthetic Aperture Radar on European Space Agency's (ESA) Copernicus Sentinel-1 satellite, 10 meters Credits: NASA GSFC landslides team, Copernicus Sentinel-1 data Citation: Handwerger AL, Huang M-H, Jones SY, Amatya P, Kerner HR, Kirschbaum DB. 2022. Generating landslide density heatmaps for rapid detection using open-access satellite radar data in Google Earth Engine. Nat. Hazards Earth Syst. Sci. Copernicus Publications, 22(3): 753–773. https://doi.org/10.5194/nhess-22-753-2022. Esri REST Endpoint: See URL to the right. WMS Endpoint: https://gis.earthdata.nasa.gov/portal/services/DISASTERS_202606_EARTHQUAKE_VENEZUELA/202611_sentinel1_landslide_heatmap/ImageServer/WMSServer

Date of Image(s) Sentinel-1 Synthetic Aperture Radar (SAR): 12/28/2024 (Pre-fire) 01/09/2025 (Called post-fire, but fires may still burning) Sentinel-2 MultiSpectral Instrument (MSI): 01/02/2025 (Pre-fire) 01/12/2025 (Called post-fire, but fires may still burning) Ancillary data ESA Global Land Cover in 2021 01/23/2025 Summary The burned areas and burn severity levels for vegetation and urban areas were generated separately based on an integration of Sentinel-2 and Sentinel-1 data. Specifically, the Sentinel-2 MSI data provide large changes in vegetation reflectance after burning due to changes in near-infrared (NIR) and shortwave-infrared (SWIR) bands, but changes in these spectral ranges were less obvious or nonexistent in affected urban environments. In a complementary sense, Sentinel-1 synthetic aperture radar (SAR) is sensitive to surface roughness changes, a signal which is more apparent (usually rougher or higher backscatter) due to the fire destruction of built infrastructure. Thus, we leveraged these two key physical characteristics together to distinguish the urban-burned areas and vegetation-burned areas. Note that while the magnitude of calculated changes in NIR, SWIR, and SAR backscatter in pre-and post-event imagery imply a range of burn severity, these metrics are still considered proxies for actual burn severity and should be interpreted as such. We first calculated the differenced Normalized Burn Ratio (dNBR = NBR_pre – NBR_post) from Sentinel-2 data in which the NBR = (NIR – SWIR)/(NIR+SWIR). This provides critical changes in vegetation due to burning. Second, we calculated the differenced vertical/horizontal polarization (VH) backscatter (dVH = VH_post – VH_pre) from Sentinel-1 SAR data. The ESA Land Cover classification map and the relationship between dNBR and dVH were used to distinguish urban and vegetation-burned areas. The generated dNBR was used to divide the vegetation-burned areas into 4 groups: Low (values ≤0.25), Moderate (0.250.45). The generated dVH was used to divide the urban-burned areas into 3 groups: Low (values ≤1), Moderate (14). Notes that these thresholds were defined for contrast visualization of severity levels across four southern California fires and may not reflect other regions. All analysis was performed in Google Earth Engine. From these layers, we can clearly see the vegetation and urban-burned areas and their burn levels. The structure damage could be overlayed on top of burn maps generated by the integration of Sentinel-1 and -2. The maps were preliminary results only and no ground-validation has been considered at the time of posting. To separate vegetation and urban-burned areas, we applied the empirical thresholds of Sentinel-2 dNBR and Sentinel-1 dVH, which could be imperfect in some locations on a large scale. Sentinel-1 SAR data was subject to signal noise in some areas likely due to terrain and speckle noise, which could lead to incorrect severity levels in urban-burned areas to some extents. Refinement of the results is in progress. Suggested Usage Despite the rawer form of this quick-look product, it is still useful for distinguishing vegetation and urban-burned areas and severity assessment. Resolution Sentinel-1A SAR C-band, 10m spatial resolution Sentinel-2 MSI, 10-60m spatial resolution Credits Khuong Tran (ARC), Taejin Park (ARC), Aakash Chhabra (ARC), Weile Wang (ARC), Kyle Kabasares (ARC). Service URL https://gis.earthdata.nasa.gov/portal/home/item.html?id=88fe789c74f9459392d293931ac7aac5 WMS Endpoint https://gis.earthdata.nasa.gov/portal/services/DISASTERS_202501_FIRE_CA/2501_s1_s2_burnseverity/ImageServer/WMSServer

This dataset provides a rapid damage screening assessment for the 26 August 2026 Nepal flash flood and debris-flow event using 10-m Sentinel-1 Synthetic Aperture Radar (SAR) data. A post-event radar acquisition acquired on 28 August 2026 was compared with a one-year pre-event Sentinel-1 baseline to identify unusually strong changes in radar backscatter. The resulting damage proxy was spatially averaged using a 50 m neighbourhood and restricted to a built-environment search domain (residential and non-residential) defined from pre-event Overture building footprints and a 30 m surrounding buffer. A conservative operational threshold (K50 absolute z-score ≥ 1.60) was used to identify high-confidence likely-damage/change areas. The resulting layer was visually reviewed at post-processing level to remove identified artefacts. Across the assessed area, approximately 245.13 ha of high-confidence change were identified within approximately 4755 ha of assessed built-environment domain across 58,557 ha of assessed AOI. Admin-3 summary statistics are provided for 16 municipalities intersecting the analysis area. The product is a likely-damage / unusual-change proxy and not proof of structural or building-level damage. It is intended to support rapid geographic prioritisation, complementary imagery review and field assessment.