The effect of hydrological model structure on spring flow forecasts when assimilating a distributed snow product

Abstract

Listen

Translate article

Accurate forecasting of spring flow is essential for mitigating flood damage and optimizing hydroelectric power production. In northern countries such as Canada, this flow is mainly driven by snowmelt processes. By integrating snowpack data from diverse sources (in situ, remote sensing, and reanalysis) with modelled snow-related state variables through data assimilation (DA), it is possible to leverage both modeling and observations for more accurate spring flow estimates. Accurate estimates of snow water equivalent (SWE) within a heterogeneous snowpack are crucial for optimizing the advantages of snow DA. Here we assess the potential effect of distributed SNOw Data Assimilation System (SNODAS) SWE data on improving spring flow in two hydrological models having distinct inner structures: HSAMI, a lumped model, and HYDROTEL, a distributed model. DA analyses used an ensemble Kalman filter scheme, which was run over a three-year period. The results were then compared with those from an open-loop experiment. We used Nash–Sutcliffe efficiency (NSE) and bias to assess the results and found that SNODAS SWE DA did not improve 1-day spring flow forecasts for the lumped model. In contrast, HYDROTEL produced more accurate 1-day spring flow estimates for all 3 years, improving NSE of the spring flow forecasts from 0.52 to 0.70, 0.32 to 0.68, and 0.39 to 0.67 for the 2014–2017 period. This study demonstrates that incorporating distributed snow data into a distributed hydrological model can improve spring flow forecasts. Given the availability of SNODAS dataset over most Canadian watersheds, this DA framework is aimed to serve as an asset to enhance operational flood forecasting systems across Canada.