Global Flood Monitoring

CHD Dsci Exploration

Author

OCHA Centre for Humanitarian Data

Published

September 2, 2026

Preface

This book documents the development of a near-real-time flood monitoring pipeline built on the Copernicus Global Flood Monitoring (GFM) system. The work is led by the OCHA Centre for Humanitarian Data with the goal of providing timely, actionable flood extent and population exposure information for humanitarian response.

What is GFM?

The Global Flood Monitoring system produces 20-metre resolution flood extent maps derived from Sentinel-1 SAR (Synthetic Aperture Radar) imagery. Three independent algorithms (TU Wien, DLR, LIST) each classify pixels as flooded or not, and an ensemble consensus (2-of-3 agreement) produces the final binary flood extent. Because SAR penetrates cloud cover, GFM can observe flooding even during active storm events — a critical advantage over optical sensors.

GFM data is distributed through a STAC (SpatioTemporal Asset Catalog) API hosted by EODC, making it accessible for cloud-native geospatial workflows.

What we built

The ds_flood_gfm Python library and its accompanying CLI scripts form a pipeline that:

Queries GFM satellite observations for any area of interest via the STAC API
Composites multiple observations over time into gap-free flood extent maps, tracking which satellite pass each pixel came from (provenance)
Overlays flood extent with GHSL population density to estimate affected population by administrative division
Exports Cloud-Optimized GeoTIFFs, flood polygons with provenance metadata, choropleths, and CSV summaries to Azure Blob Storage

The production pipeline is driven by four CLI scripts that can be run for any configured country with a single command.

What this book covers

1 Production Pipeline showcases the complete CLI pipeline with real command examples and the maps they produce — this is the core of the book
2 GFDS vs FloodScan evaluates the GDACS Global Flood Detection System (GFDS) as a complementary flood source, validated against FloodScan over the Nigeria 2022 floods

The appendices hold the supporting material:

GFM Methodology (Appendix A — Querying the Global Flood Monitoring System, Appendix B — Temporal Compositing and Provenance, Appendix C — Population Exposure Analysis) walks through the concepts behind each pipeline stage: querying the STAC API, temporal compositing with provenance tracking, and the population overlay
Design Decisions documents why we built a custom pipeline instead of using the GFM product layers directly
Early Project Development preserves the proof-of-concept notebooks and event case studies that shaped the pipeline

# Preface {.unnumbered} This book documents the development of a near-real-time flood monitoring pipeline built on the [Copernicus Global Flood Monitoring (GFM)](https://global-flood.emergency.copernicus.eu/) system. The work is led by the OCHA Centre for Humanitarian Data with the goal of providing timely, actionable flood extent and population exposure information for humanitarian response. ## What is GFM? The Global Flood Monitoring system produces 20-metre resolution flood extent maps derived from Sentinel-1 SAR (Synthetic Aperture Radar) imagery. Three independent algorithms (TU Wien, DLR, LIST) each classify pixels as flooded or not, and an ensemble consensus (2-of-3 agreement) produces the final binary flood extent. Because SAR penetrates cloud cover, GFM can observe flooding even during active storm events --- a critical advantage over optical sensors. GFM data is distributed through a STAC (SpatioTemporal Asset Catalog) API hosted by EODC, making it accessible for cloud-native geospatial workflows. ## What we built The `ds_flood_gfm` Python library and its accompanying CLI scripts form a pipeline that: 1. **Queries** GFM satellite observations for any area of interest via the STAC API 2. **Composites** multiple observations over time into gap-free flood extent maps, tracking which satellite pass each pixel came from (provenance) 3. **Overlays** flood extent with GHSL population density to estimate affected population by administrative division 4. **Exports** Cloud-Optimized GeoTIFFs, flood polygons with provenance metadata, choropleths, and CSV summaries to Azure Blob Storage The production pipeline is driven by four CLI scripts that can be run for any configured country with a single command. ## What this book covers - @sec-production-pipeline showcases the complete CLI pipeline with real command examples and the maps they produce --- this is the core of the book - @sec-alternative-sources evaluates the GDACS Global Flood Detection System (GFDS) as a complementary flood source, validated against FloodScan over the Nigeria 2022 floods The appendices hold the supporting material: - **GFM Methodology** (@sec-querying-gfm, @sec-temporal-compositing, @sec-population-exposure) walks through the concepts behind each pipeline stage: querying the STAC API, temporal compositing with provenance tracking, and the population overlay - **Design Decisions** documents why we built a custom pipeline instead of using the GFM product layers directly - **Early Project Development** preserves the proof-of-concept notebooks and event case studies that shaped the pipeline