Machine learning-based flash flood susceptibility mapping and risk assessment in the Mae Chan watershed, Chiang Rai, Thailand | |
| Author | Rajaratnam, Ponnarashi |
| Call Number | AIT Thesis no.WM-25-06 |
| Subject(s) | Floods--Risk assessment--Thailand Floods--Data processing Machine learning |
| Note | A thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering in Water Engineering and Management |
| Publisher | Asian Institute of Technology |
| Abstract | Flash floods in the Mae Chan basin, characterized by complex terrain, pose significant threats, often occurring without warning. This study aimed to identify key features influencing flash floods and to generate susceptibility, vulnerability, and risk maps for mitigation. Sixteen conditioning factors—including slope, Topographic Wetness Index (TWI), Stream Power Index (SPI), Drainage Density (DD), Distance to Stream (DtS), rainfall, NDVI, lithology, LULC, and soil type—were used. A flash flood inventory for the 2023 events was generated using SAR Sentinel-1A and Landsat 8/9 imagery, complemented by historical flood data (2004–2014) for validation. Feature selection via multicollinearity and RFECV (Random Forest and XGBoost) identified optimal predictors. Both RFECV methods highlighted TWI, DtS, and SPI; RF also selected Elevation, DD, NDVI, Lithology, and Plan Curvature, while XGBoost chose LULC. RF and XGBoost models were evaluated for flash flood susceptibility mapping through confusion matrices and statistical metrics. The models achieved strong predictive capability, with AUC-ROC scores exceeding 0.97. XGBoost showed slightly superior recall (0.9688) and AUC-ROC (0.9866), indicating better detection of actual flash flood occurrences. RF, conversely, achieved higher precision (0.9091) and specificity (0.8929), proving more effective in minimizing false alarms. Validation with 50 historical events confirmed robust spatial correspondence: 64% for RF and 68% for XGBoost events fell into high/very high susceptibility zones, validating their predictive capability.A vulnerability map was developed using population density, LULC, and road density, identifying Mae Chan District and Chiang Saen as highly vulnerable. Integrating the XGBoost susceptibility with this vulnerability, the risk map identified high-risk zones (8.0% or 95.23 km²) in Mae Chan and Chiang Saen districts where high susceptibility overlaps with significant exposure. |
| Year | 2025 |
| Type | Thesis |
| School | School of Engineering and Technology |
| Department | Department of Civil and Infrastucture Engineering (DCIE) |
| Academic Program/FoS | Water Engineering and Management (WM) |
| Chairperson(s) | Shrestha, Sangam; |
| Examination Committee(s) | Shanmugam, Mohana Sundaram;Sarawut Ninsawat; |
| Scholarship Donor(s) | Global Water and Sanitation Center (GWSC);AIT Scholarship; |
| Degree | Thesis (M. Eng.) - Asian Institute of Technology, 2025 |