Forensic engineering framework for evaluating meteorological forecasting models applied to flood control

Accurate precipitation forecasting is fundamental to flood prediction, reservoir operation, and disaster risk reduction. However, systematic biases in numerical weather prediction (NWP) models often reduce the reliability of hydrological simulations. This study proposes a forensic engineering framework to evaluate the performance of two widely used NWP models, which are the Global Forecast System (GFS) and the Weather Research and Forecasting (WRF) model, for flood forecasting in the Karkheh watershed, southwestern Iran. Support Vector Regression (SVR) was employed to improve precipitation forecasts as a post-processing bias-correction technique, and the corrected forecasts were subsequently used as inputs to a rainfall–runoff model for reconstructing the 2019 flood event. The forecasting performance of the original and bias-corrected precipitation datasets was evaluated using statistical performance indicators, and their impacts on runoff simulation were assessed through a forensic engineering analysis of the 2019 flood. This work’s results demonstrate that SVR-based bias correction substantially enhanced the forecasting accuracy of both the GFS and WRF models and significantly improved the agreement between predicted and observed runoff volumes. Bias correction applied to the first and second events reduced runoff prediction errors by 73.4, 136.6, and 78.9 million m 3 for the WRF model and by 74.6, 162.0, and 50.4 million m 3 for the GFS model with respect to the 1-, 2-, and 3-day lead forecasting times, respectively. These improvements resulted in more reliable runoff estimates and closer agreement with the observed runoff volumes. Overall, the proposed forensic engineering framework demonstrates that integrating SVR-based bias correction with numerical weather prediction models considerably improves precipitation and runoff forecasting accuracy. The proposed methodology provides a practical decision-support framework for flood forecasting, reservoir operation, and flood risk management in watersheds exposed to extreme hydrological events.