A conceptual approach for correcting global snow water equivalent data and evaluating its impact on streamflow forecast

This thesis addresses the challenges of using satellite-derived snow water equivalent (SWE) products from passive microwave (PMW) observations in hydrological forecasts, such as flood forecasting and hydropower optimization. The study focuses on reducing the bias of GlobSnow, a global SWE product, which suffers from low spatial resolution and underestimation of deep snow due to volume scattering effects. A novel bias correction method, the Watershed Scale Correction (WSC), is introduced to adjust GlobSnow SWE based on direct runoff measurements during the spring melt season. The first part of the thesis develops and applies the WSC approach to eight watersheds in Quebec, resulting in an average bias reduction from 33.5% to 18%. The corrected SWE product is then assimilated into GR4J, a lumped conceptual hydrological model, to improve streamflow forecasts for the spring flood season. The second part explores the assimilation of both the original and corrected SWE products, as well as point data measurements, into the GR4J model through deterministic and probabilistic experiments. Deterministic experiments showed that assimilating uncorrected SWE often degraded performance, notably in smaller basins, while assimilating the corrected SWE product consistently improved forecasts and in some cases (e.g., Manic-5 and St. Francis River watersheds, 2015–2016) matched or outperformed point-based snow course data. Probabilistic experiments confirmed that the assimilation of the uncorrected SWE product degraded skill as melt advanced, whereas the corrected product reduced Normalized Root Mean Square Error NRMSE and improved Continuous Ranked Probability Skill Score CRPSS on most forecast dates, with the largest gains in high-SWE or high-bias years (e.g., 2015). Improvements were less pronounced in low-variability years for SWE (e.g., 2012) and were sometimes reduced by inconsistencies between the precipitation and temperature dataset (ERA5-Land) used for hydrological model calibration and the forecast forcing dataset (EM-Earth), which introduced systematic precipitation biases. Finally, uncertainty analysis using a range of WSC-derived correction factors highlighted that SWE-related uncertainty can be as large as, or larger than, meteorological uncertainty in some watersheds. Results showed that SWE uncertainty dominated in Manic-5 across all years, in St. Francis in most years (except 2013), while Batiscan was more strongly influenced by meteorological forcing. This research contributes to advancing SWE-based hydrological forecasting by demonstrating how watershed-scale correction of PMW-derived SWE can reduce systematic biases in satellite data and improve streamflow forecasting model performance. It also provides probabilistic bounds that explicitly account for both meteorological and snow-related uncertainties, which are critical for applications such as flood prediction and hydropower production.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.