GeoTechLabs

Floods are devastating natural disasters causing significant economic losses in the region around Lake Baringo. This study leverages Synthetic Aperture Radar (SAR) data from Sentinel-1 and machine learning models, focusing on the Baringo watershed in Kenya. The motivation arises from recurrent flooding in Baringo, leading to inadequate flood inventory maps and hindering effective flood management. The objectives include developing flood inventory maps for the years 2019, 2020, and 2021, identifying key factors influencing flood occurrence, generating flood susceptibility maps, and assessing the accuracy and reliability of the RF and SVM models.

The study found that elevation, slope, rainfall, soil type, and land use were the most influential factors in predicting flood susceptibility. The flood inventory was randomly divided into training which is 70% and 30% used for testing. This research compares Random Forest (RF) and Support Vector Machine (SVM) algorithms to develop a comprehensive flood susceptibility map. Accuracy assessment using ROC curves indicates high performance for both RF with 96.49% and SVM with 92.27% in overall accuracy.

This research developed flood susceptibility maps and identified high-susceptible zones. This was classified anywhere from very low to very high classifications for its potential to flood. Both RF and SVM models demonstrated excellent discrimination ability, but RF outperformed SVM in terms of accuracy and reliability. This contributes significantly to flood management strategies, providing a comprehensive approach to mapping susceptibility, especially in regions prone to recurrent flooding like the Baringo watershed.

KEY WORDS: Flood susceptibility, SAR, RF model, SVM model, Flood Conditioning Factors