Deep-learning-based streamflow prediction in monsoon rainfall-dominated catchment under projected climate scenarios for water resource management

The monsoonal rivers exhibit a notable discrepancy in discharge levels between high flow and low flow periods due to rainfall variance and extreme events of precipitation within their catchments. Conventional machine learning-based hydrological models show poor accuracy in rainfall-runoff prediction for catchments with a skewed distribution of extremes. Given the unpredictability of extreme events and the anticipated increase in their frequency and magnitude, it is crucial to comprehend the hydrological dynamics of monsoonal rivers affected by climate change. In this study, a hybrid DL model was developed by combining 1D-CNN and Bi-LSTM for daily streamflow modelling and evaluating its proficiency to extrapolate beyond the bounds of historical climate data in the Brahmani River Basin using mean ensembles of the CMIP6 GCMs scenarios data. A rise in seasonal streamflow is predicted for the monsoon season, while a decline in mean annual discharge is predicted for pre-monsoon, post-monsoon and winter seasons in future (2031-2100) under SSP245 and SSP585 scenarios. Additionally, there is a higher likelihood of pluvial flooding in the upper part of the catchment for the future periods. Therefore, an integrated water management strategy needs to be adopted in the BRB for enhancing the resilience of water resource systems in the context of climate change.