Advancing hydropower with AI: Probabilistic inflow forecasting at Cahora Bassa reservoir

Inflow forecasts are essential for the optimal management of reservoirs. Small run-of-the-river plants can benefit from forecasts to make the most of spot energy markets on a short-term scale. For large storage infrastructure, long-term information can be provided to help reap the most benefits of seasonal water transfers. Regardless of the desired time scale, forecasts and uncertainty go hand-in-hand. This paper discusses an artificial intelligence-based inflow forecasting system at Cahora Bassa, in Mozambique, used to provide reliable probabilistic forecasts on which reservoir management decisions can be taken. Despite advances in hydrological modelling capabilities, what can be inferred about streamflow with lead times of several months is limited by the complexity of meteorological patterns which, for the time being, meteorological models struggle to capture. In such a setting, the best decisions are those that acknowledge unknowns.