Sirmj-1986/Sentinel1-Flood-Mapping-LOEO

Reproducibility code for multi-event Sentinel-1 flood mapping using adaptive Otsu thresholding, Leave-One-Event-Out validation, and machine learning in the Niger-Benue confluence, Nigeria. # Multi-Event Sentinel-1 Flood Mapping Reproducibility Repository This repository contains the reproducibility workflow for the manuscript: **Multi-Event Sentinel-1 Flood Mapping Using Adaptive Otsu Thresholding and Machine Learning in the Lokoja–Bassa–Kogi Sector of the Niger–Benue Confluence** The public code is intentionally limited to the analytical pathway reported in the manuscript. Superseded experiments, deep-learning development, temporary runtime-recovery cells, author-specific Google Drive paths, and exploratory cartographic code are excluded. The manuscript study-area map was prepared separately in **ArcGIS Pro** and is not generated by this repository. ## Repository scope The notebook reproduces the reported analytical workflow: 1. Environment and configuration 2. Sentinel-1 preprocessing and acquisition matching 3. Predictor construction and grid harmonisation 4. Reference-sample preparation 5. Leave-One-Event-Out (LOEO) partitioning and spatial blocking 6. Event-specific adaptive Otsu benchmark 7. Random Forest, XGBoost, and LightGBM development comparison 8. Independent held-out evaluation and full-area LightGBM inference 9. Paired Otsu-LightGBM statistical comparison 10. Full-area flood mapping and spatial agreement 11. Held-out TreeSHAP interpretation 12. Post-selection sensitivity analyses 13. External NEMA flooded-location corroboration 14. Reproducibility audit and manuscript outputs The repository does **not** contain a `02 Study data` notebook section and does not reproduce the study-area map. ## Main code `reproduce_flood_mapping_study.ipynb` This is the single public analytical notebook. It uses repository-relative paths and does not require access to the authors' Google Drive. ## Reproducibility levels ### Level A: Core numerical reproduction The smallest reproducibility package can contain: - `data/processed/loeo_12predictor_table.csv` - `data/processed/otsu_heldout_predictions.csv` With these files, the notebook can reprod …