| Abstract | Sub-seasonal to seasonal (S2S) streamflow forecasts play a critical role in the planning and management of water resources for various purposes, such as optimization of hydropower production, ensuring sufficient water supplies for various usages, mitigating flood and drought risks, and management of nutrients from industrial and agricultural sources. Contrary to day-to-day operational activities, such forecasts can provide an extended operational window to various levels of the government for taking appropriate actions and issuing timely directives. Compared to the vast amount of hydrologic literature on short-term streamflow forecasting, S2S forecasting area is still not well-developed. This paper reviews state-of-the-art in S2S streamflow forecasting, considering conventional process-based and statistical modeling approaches, emerging machine learning (ML) techniques, and hybrid options. The generated knowledge and insights are intended to guide the development of operational tools for S2S forecasting for Alberta, Saskatchewan, and Manitoba provinces of Canada, and can also be used for developing similar tools for other regions of the world. Apart from discussing various modeling challenges, data availability constraints, and quantification of uncertainties, the paper also presents a systematic framework for developing ML-based S2S streamflow forecasting tools. Various limitations of the reviewed approaches and potential avenues of future research are also discussed to advance research and applications in S2S forecasting area. It is found that the potential of ML in addressing scaling issues in hydrology, through S2S forecasting, and investigating relevant hydrologic mechanisms at coarse spatial and temporal resolutions are not adequately explored. This is a significant path forward for ML in hydrology. |
|---|
