Évaluation de l’utilisation des données des réanalyses ERA5 et ERA5-Land pour la modélisation de l’évapotranspiration

Abstract

This study examines the added value of using data from ERA5 and ERA5-Land reanalyses for the estimation of evapotranspiration in hydrological modeling. The modeling tools (models and internal formulas) must be adjusted locally and often do not allow the extraction of intermediate results from the final response due to a lack of physical representativeness within the models themselves. Unfortunately, the use of these tools is also due to the lack of abundant availability of hydrological and climatic data. In an effort to explore the path to more representative tools, the study uses new data that are much more comprehensive than observations, allowing the testing of physically based formulas to characterize the underlying processes of a hydrological model. More precisely the latest generation of ERA5 and ERA5-Land reanalysis data. Coupled with this data is the testing of 14 different estimation formulas of evapotranspiration. This arrangement makes it possible to test the potential of the use of more physical potential evapotranspiration (PET) formulations on the hydrological modeling results and thus to open the door to these changes. In order to ensure the greatest generalization of this study, 1169 watersheds are analyzed. They are spread all over North America, so in Canada, the United States and Mexico. Added to this is the use of two hydrological models. The analysis of the results is done through the quality of the modeling which is judged by the value of the Kling-Gupta efficiency criterion (KGE). Observations remain the leader if only the KGE values obtained in calibration are considered. However, there are exceptions for mountainous regions such as the Rockies in Canada, more northern regions of Canada and for Mexico. In these cases, better calibration results with reanalysis data (ERA5 and ERA5-Land) are observable. All in all, the superiority in calibration of the observations does not imply poor results for the reanalysis data. The results in hydrological modeling for the combinations using the more physically based PET show that it is possible to obtain models as well calibrated as by the through simpler and more empirical PET formulations. In general, some simpler PET formulations generate better results and there is a noticeable difficulty in modeling in the central region of the United States. Overall, the study shows the favorable hydrological modeling potential of reanalyses as well as the potential of reanalysis data combined with more physical-based methods to estimate PET.

Date	18 Aug 2022
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Annie Poulin (Supervisor) & Richard Arsenault (Co-supervisor)