Potentiel des données de précipitation et température des réanalyses atmosphériques en modélisation hydrologique

Abstract

The sparse coverage of weather stations over several regions of the world limits the ability of hydrological models to adequately simulate river flows. The objective of this thesis is to evaluate the potential of atmospheric reanalyses as an alternative to weather stations to overcome the lack of information in areas where these stations are sparse or nonexistent. To do this, precipitation and temperature data from three recent global atmospheric reanalyses (ERA-Interim, CFSR and MERRA) were used as meteorological inputs to a hydrological model to simulate daily discharges on 800 watersheds located in different climatic regions of the USA and Canada. First, a pre-validation of precipitation and temperature datasets from the global reanalyses was performed by comparing them to observational datasets over the USA where the spatial coverage of weather stations is high. Each dataset was then used to calibrate a hydrological model and to simulate daily river flows of 370 US watersheds. Results showed that temperatures from reanalyses were similar to that of observational data over most of the USA. On the other hand, precipitation from all three global reanalyses was biased, especially in summer and winter in south-eastern USA. Despite these biases, the simulated flows forced by the reanalysis datasets were similar to those forced by observations, except in the humid continental and subtropical climatic regions, where the poor precipitation seasonality of reanalyses degraded river flow simulations. In Canada where the spatial coverage of weather stations is lower, the accuracy of the simulated streamflows of 316 watersheds using reanalysis data was compared to that of the flows simulated using observational data, according to the density of weather stations. Results showed that the simulated streamflows using precipitation and temperature data from CFSR were generally similar to those simulated using gridded observations, regardless of the weather station density. On the other hand, ERA-Interim and MERRA performed significantly better than the gridded observations in the Mountain region, especially when the density of weather stations is less than 1 station per 1000 km2. Finally, the impacts of the combination of the three global atmospheric reanalyses and observational data on the accuracy of the simulated streamflows was evaluated. Two combination approaches were considered. The first consists of using a weighted average of meteorological inputs (precipitation and temperature) from all the databases, to calibrate the hydrological model and to simulate streamflow. The second approach consists of using all meteorological inputs separately to simulate hydrographs and to compute a weighted average of the simulated hydrographs. Results showed significant improvements of the accuracy of simulated streamflows in both combination cases over most watersheds. Moreover, in 100% of the cases where the accuracy of the simulated streamflows using only observational data was low (corresponding to a Nash-Sutcliffe value < 0.5), taking into account reanalyses data greatly improved the accuracy of the simulated streamflows (Nash-Sutcliffe values increased by at least 0.3). Overall, the results of this thesis suggest that precipitation and temperature from global atmospheric reanalyses can be used for hydrological modeling studies in regions where there are few weather stations. However, since the potential for precipitation and temperature data from reanalyses varies spatially, they should be used with caution in hydrological studies.

Date	1 Aug 2016
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	François Brissette (Supervisor)