Machine Learning Models Using SAR Data Map Flood Susceptibility in Kenya’s Nyando Watershed

Global: Machine Learning Models Using SAR Data Map Flood Susceptibility in Kenya’s Nyando Watershed

A research team has leveraged Synthetic Aperture Radar (SAR) imagery together with environmental and hydrological variables to model flood susceptibility in the River Nyando watershed of western Kenya. The study focused on the May 2024 flood event, aiming to improve disaster risk reduction, land‑use planning, and early warning capabilities.

Data Acquisition and Processing

Sentinel‑1 dual‑polarization SAR data captured during the May 2024 event were processed to generate a binary flood inventory. This inventory served as the training dataset for subsequent machine‑learning analyses.

Conditioning Factors

Six ancillary variables—slope, elevation, aspect, land‑use/land‑cover, soil type, and distance from streams—were integrated with the SAR‑derived flood map to provide contextual information for the classifiers.

Machine‑Learning Approach

Four supervised classifiers were evaluated: Logistic Regression (LR), Classification and Regression Trees (CART), Support Vector Machines (SVM), and Random Forest (RF). Model performance was quantified using overall accuracy, Cohen’s Kappa, and Receiver Operating Characteristic (ROC) analysis.

Results

The Random Forest model achieved the highest predictive performance, recording an accuracy of 0.762 and a Kappa coefficient of 0.480. These metrics surpassed those of LR, CART, and SVM.

Implications

The resulting susceptibility map highlights the low‑lying Kano Plains near Lake Victoria as the area of greatest flood vulnerability, aligning with historical flood records and the observed impacts of the May 2024 event. The authors suggest that such maps can inform targeted mitigation strategies and support the development of early warning systems in data‑scarce regions.

This report is based on information from arXiv, licensed under Academic Preprint / Open Access. Based on the abstract of the research paper. Full text available via ArXiv.