A reproducible geospatial machine learning pipeline to detect earthquake impact in Myanmar using Sentinel-2 satellite data and OpenStreetMap features.

Earthquake Date: March 28, 2025
AOI: 50km radius around epicenter (Mandalay) in Myanmar
Goal: Identify damaged zones, infrastructure vulnerability, and prioritize emergency response using ML and geospatial analysis.

This project performs change detection using pre- and post-earthquake Sentinel-2 imagery. It computes spectral indices (NDVI, NDBI, NDWI), generates difference maps, performs patch-level statistical analysis, and visualizes the impact with overlays from OpenStreetMap (OSM). The outputs include static difference maps, patch-wise heatmaps, and an interactive web map.

The pipeline is:

• Fully reproducible
• Modular (scripted in distinct steps)
• Impact-focused (vegetation, water, built-up changes)
• Integrated with OSM for infrastructure awareness
• Visualization-ready (static plots and Folium maps)

• Download Sentinel-2 pre/post-event data using 01_download_sentinel.py
• Product IDs are set inside the script; change as needed
• Filtered by <10% cloud cover

• Run 02_generate_aoi_bbox.py to define the AOI for Mandalay + 50km buffer

• Use 03_download_osm_data.py to download roads, hospitals, landuse, and other key infrastructure layers

• Use 04_clip_and_stack.py to clip all bands to AOI and stack them
• Resamples 20m bands to 10m resolution

• 05_compute_indices.py computes indices and generates difference maps (Post - Pre)
• Difference rasters are reprojected to EPSG:4326

• 06_visualize_indices.py saves side-by-side static maps for pre, post, and diff

• 07_patch_stats.py divides diff rasters into 128x128 pixel patches and computes average change
• Outputs saved as GeoJSON and CSV

• 08_plot_patch_stats.py renders patch maps with diverging color maps for each index

• 09_interactive_map.py builds a layer-controlled Folium map grouping patch data by index
• Output: outputs/maps_interactive/map_grouped.html

git clone https://github.com/nithingm/Myanmar_Earthquake_Change_Detection.git
cd Myanmar_Earthquake_Change_Detection

Using uv:

uv venv
source .venv/bin/activate
uv pip install -r requirements.txt

Or using conda:

conda create -n myanmar_eq python=3.10 -y
conda activate myanmar_eq
pip install -r requirements.txt

export EODAG__COP_DATASPACE__AUTH__CREDENTIALS__USERNAME="your_copernicus_username"
export EODAG__COP_DATASPACE__AUTH__CREDENTIALS__PASSWORD="your_copernicus_password"

Use the master runner script:

python scripts/00_run_pipeline.py

This sequentially runs:

1. Sentinel download
2. AOI generation
3. OSM data fetch
4. Clipping + stacking
5. Index computation
6. Static visualizations
7. Patch Stats
8. Plot Patch Stats
9. Interactive Map

Bounding box used:

North: 27.5
South: 17.05
East: 98.4
West: 95.5

File: data/aoi/myanmar_aoi_bbox.geojson

• Integrate Maxar/UNOSAT post-disaster building data
• Estimate infrastructure damage severity from index scores
• Add classification model (CNN) for binary damage prediction
• Build emergency response prioritization tool

• Copernicus Open Access Hub
• EODAG
• OSM Overpass API
• UNOSAT Damage Assessments

MIT License

Developed by Nithin George
ML for Impact | Geospatial AI | Disaster Response