Reference architecture
Discipline: GeoAI
Industry: Disaster management
Technology: GeoAI
Cloud: Google Cloud
Region: Illustrative
Risk maps refreshed in hours instead of weeks
Reference architecture
National flood-risk intelligence platform
A civil-protection agency mapped flood exposure manually, basin by basin, so each refresh took weeks and could not be repeated consistently between analysts.
Conceptual visualization only
Engagement at a glance
Objectives
- Produce comparable flood exposure maps across all monitored basins
- Reduce refresh time from weeks to hours after new imagery lands
- Keep every published layer traceable back to source imagery
Data sources
- Sentinel-1 SAR and Sentinel-2 optical scenes
- National digital elevation model
- Building footprints and critical-infrastructure registers
- Historical flood extent records for validation
Architecture and approach
Earth Engine imagery ingestion, Python segmentation models, PostGIS storage, Web GIS delivery.
Step 1
Scene ingestion
Scheduled pulls with cloud and orbit filtering, plus terrain correction.
Step 2
Water extraction
SAR thresholding combined with a trained segmentation model for mixed land cover.
Step 3
Exposure overlay
Intersection with buildings, roads and population grids to quantify who is affected.
Step 4
Validation
Comparison against historical flood extents with per-basin accuracy reporting.
Step 5
Publication
Versioned layers published to the Web GIS with change log and download.
Technologies used
- Google Earth Engine for imagery access and preprocessing
- Python segmentation models (rasterio, xarray, PyTorch)
- PostGIS for versioned vector storage
- Web GIS delivery with tiled services and export
Security and governance
- Role-scoped access per basin and agency unit
- No personal data processed; exposure reported at grid level
- Full audit trail from scene ID to published layer version
Output
Reported against the objectives set out for this reference architecture.
Refresh cycle
Weeks → hours
Automated pipeline triggered by new scene availability.
Coverage
All monitored basins
Consistent method replaces analyst-specific workflows.
Traceability
100% of layers
Every output records inputs, model version and run time.
Lessons learned
- SAR alone misjudges urban water; a hybrid model was needed for built-up areas.
- Validation data quality set the ceiling on reported accuracy, so it was curated first.
Reference architecture illustrating method and architecture only. It is not a verified client project, and no client commissioned it.
Ready to scope your AI initiative?
Share your objective and constraints. A specialist reviews every request before any recommendation is issued.