Hybrid modeling of karstic springs: Error correction of conceptual reservoir models with machine learning

Abstract

The simulation of spring discharge is of high interest for water supply management. Accurate modeling and prediction of spring discharge can help water authorities and managers optimize water use, face seasonal variability, and prepare for possible drought conditions. Therefore, developing reliable simulation models and forecasting tools for spring discharge is essential for the effective management of groundwater resources relying on karstic springs. While the use of complementary modeling approaches has been explored in various hydrological contexts, its application to spring discharge modeling remains relatively unexplored. Previous studies on spring discharge modeling have predominantly focused on either conceptual/distributed models or data-driven models separately, without fully investigating the potential advantages of combining these two approaches. Consequently, there is a research gap in understanding the potential benefits and effectiveness of integrating these models for spring discharge modeling. This study developed and evaluated a hybrid modeling approach integrating a conceptual hydrological model with a machine learning technique to simulate spring discharge from the largest karst aquifer supplying the city of Bordeaux, France. Specifically, the study combined the five-parameter GR5J reservoir model with a random forests data-driven model in a hybrid framework. Model performance was assessed through comparison with….