Application of Artificial Neural Networks for Accuracy Enhancements of Real-Time Flood Forecasting in the Imjin Basin

Abstract

Hydrometeorological forecasts provide future flooding estimates to reduce damages. Despite the advances and progresses in Numerical Weather Prediction (NWP) models, they are still subject to many uncertainties, which cause significant errors forecasting precipitation. Statistical postprocessing techniques can improve forecast skills by reducing the systematic biases in NWP models. Artificial Neural Networks (ANNs) can model complex relationships between input and output data. The application of ANN in water-related research is widely studied; however, there is a lack of studies quantifying the improvement of coupled hydrometeorological model accuracy that use ANN for bias correction of real-time rainfall forecasts. The aim of this study is to evaluate the real-time bias correction of precipitation data, and from a hydrometeorological point of view, an assessment of hydrological model improvements in real-time flood forecasting for the Imjin River (South and North Korea) is performed. The comparison of the forecasted rainfall before and after the bias correction indicated a significant improvement in the statistical error measurement and a decrease in the underestimation of WRF model. The error was reduced remarkably over the Imjin catchment for the accumulated Mean Areal Precipitation (MAP). The performance of the real-time flood forecast improved using the ANN bias correction method.