Water resource management in Québec is essential due to its strong reliance on hydroelectricity and the impacts of climate variability, whether from natural fluctuations or anthropogenic climate change. Currently, Hydro-Québec uses ensemble streamflow predictions (ESP) to produce probabilistic volume scenarios, but their biases limit operational efficiency during extreme events such as spring floods. This thesis primarily aims to develop the SynthEau method, which enhances the accuracy of inflow volume forecasts and the temporal consistency of probabilistic scenarios derived from ESP. The goal is to better manage volume discrepancies during extreme events and to compare SynthEau with the currently used method (HMAPPORT) to assess its effectiveness.
The research focuses on the analysis of the La Grande and Saint-Maurice River basins in Québec. Historical inflow data and archived ensemble hydrological forecasts were used. Two transformation models were compared: HMAPPORT, which uses the temporal median as a basis for redistributing the predicted volume, and SynthEau, a newly developed method that ensures the temporal consistency of probabilistic inflow volume quantiles through a cumulative-decumulative approach. The evaluation is based on mean relative error, biases against real observations, and scenario reliability.
The SynthEau model shows a reduction in average inflow volume forecast errors compared to HMAPPORT, particularly over a 30-day medium-term forecast horizon for high (15% exceedance) and low (85% exceedance) scenarios. The results indicate that SynthEau provides a better representation of forecasted volumes under extreme hydrological conditions, with biases generally below 0.5%. However, the comparison of median scenarios with actual volumes shows similar performance between SynthEau and HMAPPORT, highlighting the dominant influence of the ensemble streamflow forecasts themselves. Reliability analysis reveals a slight underestimation of extreme events at the 15% quantile and an overestimation of frequent scenarios at the 85% quantile for both models, as they rely on the same input data.
SynthEau improves the estimation of forecasted volumes in probabilistic scenarios derived from ESP, providing tangible operational benefits for Hydro-Québec’s water resource management. The observed biases relative to actual inflows are primarily caused by the ESP forecasts, underscoring the importance of improving these forecasts upstream. Future research could integrate advanced data assimilation methods or bias correction directly into ensemble forecasts to further enhance the precision and reliability of operational scenarios.
Marinelli, J. (Author),
Arsenault (Supervisor) &
Brissette (Co-supervisor),
22 Apr 2025Student thesis: Master's thesis › Master in Engineering: Construction Engineering