Near future flash flood prediction in an arid region under climate change

Abstract

Abstract Flash floods represent a significant threat, triggering severe natural disasters and leading to extensive damage to properties and infrastructure, which in turn results in the loss of lives and significant economic damages. In this study, a comprehensive statistical approach was applied to future flood predictions in the coastal basin of North Al-Abatinah, Oman. In this context, the initial step involves analyzing eighteen General Circulation Models (GCMs) to identify the most suitable one. Subsequently, we assessed four CMIP6 scenarios for future rainfall analysis. Next, different Machine Learning (ML) algorithms were employed through H2O-AutoML to identify the best model for downscaling future rainfall predictions. Forty distribution functions were then fitted to the future daily rainfall, and the best-fit model was selected to project future Intensity-Duration-Frequency (IDF) curves. Finally, the Soil and Water Assessment Tool (SWAT) model was utilized with sub-daily time steps to make accurate flash flood predictions in the study area. The findings reveal that IITM-ESM is the most effective among GCM models. Additionally, the application of stacked ensemble ML model proved to be the most reliable in downscaling future rainfall. Furthermore, this study highlighted that floods entering urbanized areas could reach 20.33 and 20.70 m³/s under pessimistic scenarios during rainfall events with 100 and 200-year return periods, respectively. This hierarchical comprehensive approach provides reliable results by utilizing the most effective model at each step, offering in-depth insight into future flash flood prediction.

Keywords

• Future flood prediction
• Climate change
• SWAT model
• Machine learning
• CMIP6 scenarios