Grant: Alliance Grants

Forecasting inflows in rivers and reservoirs is an important aspect of keeping populations safe from potential flood events and managing water resources sustainably. However, our ability to forecast inflows is limited by numerous factors, including the capacity of hydrological models to convert weather information into precise flows, as well as the uncertainty related to future weather in the first place. Hydrologists therefore use probabilistic methods to evaluate not only the expected inflow volumes, but also the risks and probabilities of flood event occurrences. This is a difficult task, especially on longer time scales, due to the chaotic nature of the weather system. Nonetheless, some methods exist to extract as much information as possible from our existing systems. These processing steps allow hydrologists to better predict flows and their uncertainty, but they are still far from perfect due to limitations in the current operational forecasting settings. In this project, we aim to improve probabilistic inflow forecasting by applying statistical postprocessing methods and deep learning techniques to develop new predictive tools that will allow improving the accuracy as well as the uncertainty estimations of the inflows. This will be performed on 200 catchments across Canada using operational weather forecasting data that the research team has been collecting over the past 8 years, which will allow large-scale learning and improvements compared to the current state-of-the-art methods.