Author(s): Rishma Chengot; Helen Baron; Nathan Rickards
Linked Author(s):
Keywords: Machine learning; Random forest; Reservoir forecasting; Water resources
Abstract: Reservoirs play a critical role in global water resources by storing and regulating flow. With climate change and population growth, reliable reservoir predictions are essential for early warning systems and sustainable water management. While Machine Learning techniques are widely used in water resource management, few studies focus on reservoir storage, especially long-term forecasting. This study examines the performance of various Random Forest (RF) algorithms in predicting long-term reservoir behavior using UK reservoirs as a case study, comparing their performance with that of a static ARIMA model. Predictions were made for 1,3, and 6-month periods, using precipitation, temperature, and historical reservoir storage as inputs. The performance of the RF models is acceptable. Among the selected RF algorithms, Extra Trees performed best, showing strong results for both monthly and seasonal forecasts, although performance decreases over time; specifically, the model performs well for a 1-month lead time but less effectively for a 6-month lead time. The findings suggest that incorporating more catchment parameters could enhance model performance and support broader applications in data-scarce regions.
Year: 2024