keepcalculator

06 Numbers and Powers

Disaster Damage Mapping: Satellite + Engineering Data

Card No. 06·73268 · Updated · Standard formula, shown below

This guide explains how to combine satellite imagery with engineering data to create a damage map after a disaster. You will learn the workflow for aligning imagery with structural and infrastructure records, classifying damage levels, and producing a map that emergency responders and engineers can use for triage and resource allocation. The result is a practical, evidence-based damage assessment that leverages the strengths of both data types.

What You’ll Need

  • Satellite imagery: Pre-disaster and post-disaster images. Free sources include Sentinel-2 (10m resolution, optical), Landsat 8/9 (30m), and NASA FIRMS for active fire data. For higher resolution, consider commercial sources like Maxar or Planet (often made available during disasters).
  • Engineering data: Building footprints, infrastructure networks (roads, bridges, power lines), structural characteristics (material, height, age), and construction codes. These may come from municipal GIS departments, OpenStreetMap, or national databases.
  • GIS software: QGIS (free, open-source) or ArcGIS Pro. You will use these for layering, analysis, and map production.
  • Image processing tools: For change detection, you can use the Semi-Automatic Classification Plugin (QGIS), Google Earth Engine (code editor), or even basic raster math in your GIS.
  • Coordinate reference system (CRS): Ensure all datasets are in the same CRS (e.g., WGS 84 / UTM zone). This is critical for accurate overlay.
  • Field validation data (optional but recommended): A small set of ground-truth points (e.g., from a rapid field survey or drone) to calibrate and validate your damage classification.

Step-by-Step

  1. Acquire and preprocess satellite imagery. Download pre- and post-event images from your chosen source. Ensure they are cloud-free over your area of interest (AOI). If using Sentinel-2, apply atmospheric correction (e.g., using the Sen2Cor plugin or the Semi-Automatic Classification Plugin in QGIS). Clip both images to your AOI.
  2. Load and align engineering data. Import building footprints, road networks, and other infrastructure layers into your GIS. Check that they align with the satellite imagery. If there is a spatial offset (common with OpenStreetMap data), use a georeferencing tool to shift the engineering layer to match the imagery. This step is non-negotiable; misaligned data will produce a useless map.
  3. Perform change detection on the imagery. Use a change detection method to highlight areas of change between pre- and post-event images. A simple and effective method is the Normalized Difference Water Index (NDWI) for flood mapping, or the Normalized Burn Ratio (NBR) for fire. For general structural damage, calculate the difference in the Normalized Difference Vegetation Index (NDVI) or use a simple band difference (e.g., the red band). In QGIS, use the Raster Calculator: change = post_event_band - pre_event_band. High absolute values indicate significant change.
  4. Extract candidate damage zones. From the change raster, create a binary mask of pixels that exceed a threshold (e.g., change > 2 standard deviations from the mean). This mask represents areas of likely damage. Convert this raster to a polygon layer for further analysis.
  5. Intersect damage zones with engineering data. Use the “Intersect” or “Spatial Join” tool to combine the damage polygons with your building footprints or infrastructure lines. Each building or road segment will now have a “change value” attribute from the satellite analysis.
  6. Classify damage levels. Based on the change value and the engineering attributes, assign a damage class. A common scale is the European Macroseismic Scale (EMS-98): Grade 1 (negligible) to Grade 5 (total collapse). For example, a building with a very high change value and a known weak material (e.g., unreinforced masonry) might be classified as Grade 4 or 5. Buildings with low change values and reinforced concrete frames might be Grade 1 or 2. You can do this in the attribute table with a conditional expression (e.g., CASE WHEN change_value > 0.5 AND material = 'masonry' THEN 'Grade 4' ...).
  7. Validate with ground truth (if available). Compare your classification against any field data or drone imagery. Calculate a confusion matrix to see where your map is over- or under-estimating damage. Adjust your thresholds or classification rules accordingly.
  8. Produce the final damage map. Symbolize the classified layer with a color ramp (e.g., green to red). Add contextual layers (roads, place names, administrative boundaries). Create a layout with a legend, scale bar, north arrow, and a clear title. Export as a PDF or high-resolution PNG for sharing.

Common Mistakes

  • Ignoring CRS alignment: Overlaying data in different coordinate systems will cause severe spatial errors. Always check and reproject all layers to a common CRS.
  • Using only post-disaster imagery: Without a pre-disaster baseline, you cannot distinguish damage from pre-existing conditions (e.g., a building that was already a ruin).
  • Applying the same threshold to all areas: A change threshold that works in a dense urban area may produce many false positives in a rural area with vegetation changes. Use a local threshold or manually review high-change zones.
  • Forgetting to account for shadows and clouds: Shadows from tall buildings or clouds can create false change signals. Mask out these areas or use imagery from similar sun angles.
  • Over-relying on satellite data: Satellite imagery has limits; it cannot see through roofs. A building with a collapsed roof but intact walls may show little change from above. Engineering data helps infer this, but field verification is essential for high-stakes decisions.

Tips

  • Use a pre-event footprint layer to mask out non-building areas. This reduces false positives from vegetation or bare soil changes.
  • Consider using radar (SAR) data. Sentinel-1 SAR is excellent for damage mapping because it works day/night and through clouds, which is common after disasters. Look for a decrease in backscatter coherence.
  • Automate the workflow with a model. In QGIS, you can use the Graphical Modeler to chain the steps (raster calc, threshold, intersection) so you can re-run the analysis quickly if new imagery arrives.
  • Share your map in a web format. Use QGIS Cloud or export to GeoJSON and view in a tool like Kepler.gl to make it interactive for field teams.
  • Document every step. Keep a log of data sources, dates, thresholds, and classification rules. This is critical for the map to be defensible in post-disaster reviews or insurance claims.

Troubleshooting

  • My change detection shows too much noise. This usually means your threshold is too low. Recalculate the standard deviation of the change raster and set the threshold to 2.5 or 3 standard deviations. Also, apply a median filter (3×3 or 5×5) to the change raster to smooth out speckle.
  • The engineering data does not line up with the imagery. If the offset is consistent across the whole area, use the “Translate” tool to shift the layer by a fixed distance. If it varies, you may need to use a more complex georeferencing approach or find a better data source.
  • My classification is too aggressive (everything is Grade 5). Your change threshold is likely too low, or you are not weighting the engineering data enough. Revisit your classification rules; a high change value alone does not mean total collapse. Incorporate structural type and age more heavily.
  • I have no pre-disaster imagery. Check the USGS EarthExplorer or Copernicus Open Access Hub for archived scenes. If truly none exist, you can use a building’s age and construction type as a proxy, and use the post-event imagery only to look for visible debris fields or missing structures, but this is far less reliable.
  • The map is too slow to produce. If your AOI is very large, use a tiling approach. Process the area in smaller tiles and merge the results. Or, move the analysis to Google Earth Engine, which is designed for large-scale raster processing.

Frequently Asked Questions

Q: Can I use free satellite data for a reliable damage map?
A: Yes, for large-scale damage (e.g., flooded neighborhoods, burnt areas), Sentinel-2 (10m) is often sufficient. For individual building-level damage, you will need higher resolution commercial imagery or drone data.

Q: What is the best change detection index for earthquakes?
A: For earthquakes, there is no single best index. A combination of NDVI difference (to detect vegetation loss/debris) and a textural change (e.g., using the GLCM texture tool) often works. SAR coherence loss is also very effective.

Q: How do I handle buildings that are partially damaged?
A: This is the hardest case. You will need to look at the shape of the change polygon. If it covers only a portion of the building footprint, classify it as partial damage. Field validation is crucial here.

Q: Is it better to use machine learning for classification?
A: For large areas, yes. A random forest classifier using spectral bands, change indices, and engineering attributes as features can be more accurate than manual thresholds. However, it requires a good training dataset, which is often unavailable immediately after a disaster.

Q: How accurate is this method compared to field surveys?
A: Typically, satellite-based methods achieve 70-90% accuracy for distinguishing damaged vs. undamaged, but accuracy drops significantly when trying to differentiate between damage grades (e.g., Grade 3 vs. Grade 4). Field surveys remain the gold standard.

Q: What if my engineering data is incomplete?
A: You can still produce a useful map. Use the satellite change detection as the primary layer and only add engineering data where it exists. Clearly mark areas with no data as “unassessed” rather than “undamaged.”

Related

Method: the standard formula for this calculation, rounded as each result states. See how the calculators work or report an error.

More cards in this drawer

The whole Numbers and Powers drawer →