AI-based techniques for multi-step streamflow forecasts: application for multi-objective reservoir operation optimization and performance assessment

Abstract. Streamflow forecasts are traditionally effective in mitigating water scarcity and flood defense. This study developed an artificial intelligence (AI)-based management methodology that integrated multi-step streamflow multi-objective reservoir operation optimization for resource allocation. Following the methodology, we aimed to assess forecast quality forecast-informed performance together due influence of inflow uncertainty. Varying combinations climate hydrological variables were input into three AI-based models, namely a long short-term memory (LSTM), gated recurrent unit (GRU), least-squares support vector machine (LSSVM), streamflow. Based on deterministic forecasts, stochastic scenarios further using Bayesian model averaging (BMA) quantifying The forecasting scheme was coupled with multi-reservoir model, programming solved parameterized robust decision-making (MORDM) approach. framework applied demonstrated over system (25 reservoirs) Zhoushan Islands, China. Three main conclusions drawn from this study: (1) GRU LSTM performed equally well might be preferred method LSTM, given it had simpler structures less modeling time; (2) higher could lead improved operation, while uncertain more valuable than regarding two metrics, i.e., supply reliability operating costs; (3) relationship between horizon complex depended configurations (forecast uncertainty) measures. reinforces potential seek strategies under