Improvement of high and low flow simulation in the hydrological model chain SASER.

SASER (SAfran-Surfex-Eaudysee-Rapid) is a distributed and physically-based modeling chain. Currently, SASER has been implemented for different spatial domains and resolutions. The Pyrenean application of the model at 2.5 km of spatial resolution has a good performance, but it can be improved. We have evaluated the simulated streamflows using the KGE score, which is above 0.5 over 57% of the near-natural catchments. We have seen that SASER simulates reasonably well high, but not extreme, and median daily streamflows, but low flows and peak flows are underestimated. Our hypothesis is that low flows are underestimated due to the lack of a groundwater model and that peak flows are underestimated due to low intensities hourly precipitation in SAFRAN, among other issues. 

The objective of this study is to improve the streamflow simulated by SASER by improving the intensity of the hourly precipitation produced by SAFRAN and by introducing a simple conceptual model to simulate groundwater effects.

SAFRAN ingests daily precipitation observations, which are distributed to an hourly scale using relative humidity, which generates low precipitation intensities. This is not realistic at all in a Mediterranean climate. Our hypothesis is that we can improve hourly intensities by using the outputs of an RCM simulation, forced by a reanalysis, to distribute the hourly precipitation. In this experiment we have used the CNRM-ALADIN model, forced by ERA-Interim, from the EURO-CORDEX database. We keep the daily amounts from SAFRAN (over windows that span from 1 to 14 days) and we redistribute hourly precipitation according to the RCM simulation. We evaluated the results by comparing the hourly precipitation distribution of a set of 13 precipitation stations from the Ebro Basin real-time observation system (SAIH), using the Perkin Skill Score (PSS), which improved from an average of 0.70 in the standard SAFRAN product to 0.88 in our best configuration. Consequently, we now have a new precipitation dataset with improved precipitation intensity patterns.

To improve SASER low flows, we followed the steps of Getirana et al. (2014) and Artinyan et al. (2008), we introduced a linear reservoir at grid point resolution between the LSM and the routing scheme. We calibrated the reservoir parameters catchment-by-catchment in near-natural sub-catchments. The KGE score of the square root of the streamflow shows on average an improvement of 21% with respect to default simulation (without reservoir)

The regionalization approach was chosen to set the reservoir parameters in human-influenced catchments, where calibration is unfeasible. This approach allows us to link physical characteristics with the reservoir parameters through a linear equation, as did Beck et al. (2020).  In this process, an evolutionary algorithm was implemented, which optimizes the equation coefficients, thereby we were able to produce maps (at 2.5 km resolution) of the model parameters based on physiographic data. Preliminary results show using this approach we obtain performances close to those obtained by a classical calibration procedure.

This work was funded by the HUMID project (CGL2017-85687-R, AEI/FEDER, UE), EFA210/16-PIRAGUA and IDEWA (PCI2020-112043) projects, and the predoctoral grant PRE2018-085027 (AEI/FSE).