Integrating machine learning and physically based hydrodynamic modeling for flood hazard mapping: a case study of the Takkalasi watershed, Indonesia

Abstract

Floods are among the most frequent and damaging natural disasters in Indonesia, with increasing intensity and frequency driven by climate and land cover changes. Physically based hydrodynamic models such as HEC-RAS 2D Rain-on-Grid can simulate flood processes accurately but require extensive data and computational resources, limiting their application in data-scarce regions. This study develops an integrated framework that combines HEC-RAS 2D Rain-on-Grid simulations with the Random Forest (RF) algorithm for flood hazard mapping in the Takkalasi watershed, South Sulawesi, Indonesia. Flood depth and inundation maps from the 21 December 2024 flood event simulated by HEC-RAS were used as training data for RF classification and regression models. The models achieved high predictive accuracy and low flood-depth error, producing flood probability, susceptibility, and depth maps. The results show that upstream–downstream rainfall synchronization, flat downstream topography, limited drainage capacity, and land cover changes strongly influence flood extent and depth. This integrated approach enables rapid flood hazard mapping using readily available spatial data and can be applied to other watersheds. Its implementation supports adaptive spatial planning and disaster risk reduction, particularly in regions with limited hydrological observations.

Access to Document

Other files and links

• Link to publication in Scopus
• Open Access Version Available

Fingerprint