Artificial Neural Network Modeling of Rainfall-Runoff Extreme Value Distributions: A Focus on the Shape Parameter

In this study, we explore the variables impacting the shape parameter changes in a Generalized Extreme Value (GEV) distribution and model the parameter to predict flood risk areas. Our dataset comprises satellite, meteorological, and hydrological data from the Mun River Basin collected over 13 years (2010–2022) from 17 stations of the Thai Meteorological Department and agricultural stations. The shape parameter is estimated via an Artificial Neural Network (ANN) approach for a non-stationary process and by Maximum Likelihood Estimation (MLE) for a stationary process. Using the Nash–Sutcliffe Coefficient (NSE) for comparison, we found that the non-stationary model was more suitable for the monthly maximum rainfall data from 10 stations, while the stationary model was applicable for the remaining 7 stations. Overall, all models had predictive accuracy with an average NSE greater than 0.80, indicating their suitability for monthly maximum rainfall data. The study culminates with the presentation of the return level of monthly maximum rainfall data for various return periods via a 2D map.