Application de techniques de post-traitement des prévisions météorologiques pour améliorer les prévisions hydrologiques opérationnelles: étude de cas sur plusieurs bassins versants canadiens

Abstract

Hydrological forecasts allow estimating streamflow in rivers runoff. This is vital for various activities such as drought and flood prevention, water supply management and hydroelectric plant operations. Medium-range hydrological forecasts (3-15 days) use hydrological models driven by dynamical weather prediction forecasts. However, hydrological forecasts often contain biases due to various factors such as uncertainties in the hydrological model's parameters, deficiencies in model structure and meteorological input, unreliable measurement of input data, and mis specified initial and boundary conditions. Generally, meteorological input is the main source of uncertainty in hydrological forecasts, uncertainty that make them less useful to water resources system managers. However, they can be further improved using post-processing methods. Post-processing has the capability to reduce overall bias and improve the uncertainty quantification (spread), in order to enhance the usefulness of the forecasts in decision-making. In this study, the Quantile Mapping (QM) post-processing technique was implemented on meteorological forecasts, more specifically on the precipitation input. The QM method is designed to be a simple and effective post-processing method for streamflow forecasting, intended for operational use by end-users. As a result, QM was assessed in three different configurations: A monthly, a seasonal, and an annual quantile mapping application scheme. The evaluation was carried out over 20 watersheds with different surface areas from 30 km2 to over 30 000 km2 in Canada. Post-processing methods were trained on the precipitation of the 2015-2019 ECMWF Overall Operational Forecast and then applied in the 2020 forecast, each forecast an ensemble of 50 members. Eight global hydrological models were used: CEQUEAU, GR5dt, HBV, HYMOD, IHACRES, MOHYSE, SIMHYD and TOPMODEL. Hydrological models were calibrated and assimilated using the Covariance Matrix Adaptation Evolution Strategy (CMAES) and the Ensemble Kalman Filters algorithm (EnKF). The postprocessed forecasts were then introduced into the previous eight models, generating the hydrological forecasts for the year 2020 with a forecast period of 10 days and a time step of 6 hours. The robustness of the methodology and the results were evaluated using the Continuous Ranked Probability Score (CRPS) metric and a new metric we called the Runoff-Specific CRPS (RSCRPS), which is calculated by dividing the mean daily CRPS by the mean runoff of the basin. Noticeable enhancements are observed when employing QM in precipitation forecasts; however, these improvements do not translate into enhanced hydrological forecasts. Findings indicate that the three QM configurations do not enhance the accuracy of streamflow forecasts in most catchments, and at times, the processed forecasts exhibit poorer performance compared to the raw forecasts, highlighting the need to ensure precipitation post-processing methods positively impact hydrological forecasts.

Date	14 Dec 2023
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Annie Poulin (Supervisor) & Richard Arsenault (Co-supervisor)