Artificial intelligence techniques in hydrology and water resources management and their applicability to Sri Lankan river basins

Abstract

Artificial Intelligence techniques are increasingly being used in hydrology for tasks such as groundwater modelling, streamflow prediction, and rainfall time series generation. In Sri Lanka, traditional water resource management methods have limitations and are less accurate in predicting rainfall-runoff, flood events, and drought conditions due to complex parameters and seasonal rainfall patterns. AI methodologies were integrated into hydrological modelling to enhance water resource management practices in Sri Lankan River basins. The study evaluated the applicability of AI techniques in hydrology and water resources management by using data-driven models like RNN-LSTM and RNN-GRU, and physical-based models like HEC-HMS. The study focused on the Kalu River basin and Kirindi Oya basin from October 01, 2000, to September 30, 2011. The evaluation criteria included NASH, MRAE, and R2, as determined based on existing literature. LSTM and GRU models performed well simulating Kalu River basin streamflow. However, all three models failed to simulate streamflow accurately in the Kirindi Oya basin due to inconsistency of input features. While AI models offer efficient simulation of flash flood scenarios, limited and unreliable rainfall data can impact accuracy. Dry zone simulations require further model development to improve reliability as current models perform well only in wet zones.