Author(s): Rishma Chengot; Helen Baron; Nathan Rickards
Linked Author(s): Helen Baron
Keywords: Decision tree forecasting hydrology random forest
Abstract: The regional modeling problem, characterized by the challenge of using a single model to simulate hydrological parameters across large areas, is a long-standing issue in hydrological sciences. While this problem has been explored to some extent in the context of rainfall-runoff modeling, the specific challenges associated with reservoir prediction remain underexplored. Numerous studies have applied machine learning (ML) algorithms for reservoir forecasting; however, a significant gap exists in research focused specifically on reservoir storage predictions. In this study, we examine the performance of different machine learning algorithms in predicting long-term reservoir behavior, using catchment characteristics such as precipitation, temperature, and historical reservoir storage as inputs. Among various ML models, Extra Trees performed the best, and this ET model was run by leveraging the Caravan dataset (part of the CAMELS series) as input for the global model. We compared its predictive capabilities with those of a localized ML model. Our findings aim to enhance the understanding of the effectiveness of local versus global approaches in reservoir storage prediction, contributing valuable insights to the field of hydrological modeling.
Year: 2025