Academic · Geospatial Analytics · Sustainability
Flood Risk Vulnerability Mapping: Trincomalee & Galle
NUS MSc Data Science for Sustainability · Geospatial Analytics for Disaster Risk Reduction · Team of five · AY 2025/26
How does coastal flood risk differ between two Sri Lankan districts hit by the 2004 Indian Ocean tsunami? A five-person team project comparing Trincomalee and Galle with one shared geospatial method.
The question
How do coastal flood-risk and infrastructure-exposure patterns differ between Trincomalee and Galle?
Both districts were analysed inside the same 10 km coastal buffer, at the same resolution and with the same indicators, so differences in risk zones and exposure come from the districts themselves.
My part
A team of five. In our shared repository I wrote the first Trincomalee notebook in Python with Google Earth Engine and the final two-district analysis notebook, whose outputs match the risk shares, exposure figures and Moran’s I values in the report.
Where the districts sit
Data & geospatial workflow
EPSG:32644 · 30 m analytical resolution · 10 km coastal buffer
All raster analysis was standardized around a common projected coordinate system and analytical resolution. Data was ingested, reprojected and resampled, then turned into hazard and vulnerability indicators, a weighted risk score, risk classes, exposure overlays and spatial-autocorrelation tests.
Built-up exposure uses ESA WorldCover 2021 as a proxy. Its gaps are listed under Limitations.
Processing
- Google Earth Engine
- Python
- Jupyter
GIS / Cartography
- QGIS
Data
- Landsat 5
- SRTM DEM
- HydroSHEDS
- JRC Global Surface Water
- ESA WorldCover 2021
- WorldPop 2020
- OpenStreetMap
- LSIB coastline/boundaries
Two-stage risk framework
Stage 1Weighted index
A weighted index combined hazard and vulnerability indicators.
Hazard, 60%: elevation 30, coast distance 25, historical inundation proxy 20, TWI 15, slope 10.
Vulnerability, 40%: population density 40, settlement type 25, road access 20, hospital access 15.
Stage 2Rule-based classification
Low, Medium and High classes were assigned using rule-based physical thresholds informed by coastal-risk literature.
Flood-risk patterns
- High
- Medium
- Low
Interactive district comparison
Both districts share the same 10 km coastal buffer, resolution and indicator framework, so their reported risk, built-up exposure, vulnerability and clustering can be read side by side.
Infrastructure exposure
Using the ESA WorldCover 2021 built-up class as a proxy, Galle has substantially more of its built-up proxy in the combined Medium/High classes. Trincomalee has the larger share specifically in the High-risk class.
Population exposure
Spatial clustering
The project used Global Moran's I for overall risk clustering, Bivariate Moran's I for co-location between risk and built-up exposure, LISA hotspot analysis and factor decomposition. The bivariate results indicate statistically significant spatial co-location in both districts.
Methods
Global Moran's I
Measures spatial clustering of the risk index.
Bivariate Moran's I
Tests co-location between risk and the built-up exposure proxy.
LISA & decomposition
Locates hotspots and examines indicator contributions.
The Galle railway corridor
Key findings
- Galle has the larger Medium-risk share and the larger combined Medium/High built-up proxy share.
- Trincomalee has the larger share specifically in the High-risk class.
- Galle's mean Vulnerability Index is higher, with population-density contribution particularly associated with that difference.
- Both districts show clustered risk and statistically significant spatial co-location between risk and the built-up exposure proxy.
Limitations
- The index is not validated against observed flood outcomes. Its weights were assigned from the literature and expert judgement, not fitted to data.
- The Landsat-derived historical inundation proxy substantially under-detected the actual 2004 tsunami extent and therefore contributed nothing to the weighted score in either district, despite its 20% weight.
- ESA WorldCover 2021 is only a proxy for later built-up exposure and may undercount informal coastal structures.
- Shared indicator weights across both districts may miss region-specific risk drivers.
Next steps
- Validate the risk classes against recorded flood events before trusting the weights.
- Test region-specific indicator weights against the shared-weight baseline.
- Improve historical inundation reconstruction with a better-validated event proxy.
- Evaluate higher-resolution built-up exposure data, including informal coastal structures where available.