mobalichris/Analysis-of-flood-susceptibility-in-the-city-of-Kinshasa

Kinshasa Flood Analysis (2015-2025): A reproducible ML framework (MaxEnt, RF, BRT) mapping flood susceptibility. Using Sentinel-2 & SRTM data, it identifies urban density as a key driver. Includes a policy matrix for SDGs 11 & 13, adaptable to other data-scarce African megacities. # Flood Susceptibility Mapping in Kinshasa This repository presents a geospatial workflow for flood susceptibility mapping in Kinshasa, Democratic Republic of the Congo. The main analysis is implemented in `Flood_vf.ipynb` and combines environmental raster data, flood occurrence points, machine learning models, spatial validation, map comparison, and zonal prioritization. ## Objectives - map flood susceptibility at 30 m resolution - compare the predictive performance of multiple models - evaluate spatial agreement and divergence between susceptibility maps - identify high-priority flood-risk zones - communicate results through maps, tables, and dashboards ## Main Models The final workflow focuses on three models: - MaxEnt - Random Forest - Boosted Regression Trees (BRT) ## Data ### Inputs - `points.shp`: flood occurrence points - raster predictors stored in `resultats/` ### Main predictor variables - slope - NDVI - built-up area - river proximity - BSI - population - TRI - TWI - SPI log - HAND - flow accumulation - elevation ## Workflow The notebook follows these main steps: 1. load flood occurrence points and raster predictors 2. reproject all raster layers to a common grid (`EPSG:32733`, 30 m) 3. extract raster values at flood points 4. generate pseudo-absence points 5. build the final modelling dataset 6. assess predictor quality using `InGR`, `VIF`, and tolerance 7. perform spatial block cross-validation 8. train MaxEnt, RF, and BRT 9. produce flood susceptibility maps 10. evaluate model performance at optimal thresholds 11. compare maps using correlation, divergence, and spatial agreement 12. compute zonal flood proportions and rank priority areas ## Validation and Performance Assessment The project includes: - spatial block cross-validation - ROC curves and AUC - accuracy - precision and recall - specificity - F1-score - PPV and NPV - TSS-based optimal threshold selection ## Main Outputs The `outputs/` folder contains: - susceptibility maps …