Hybrid fitting of a hydrosystem model using dense spatio-temporally distributed data: the Beauce aquifer functioning over 40 yr (France)

This study focuses on the Beauce aquifer (8 000 km2, unconfined) over a 40-year period. The mono-layer aquifer system is part of the hydrosystem Loire (surface basin of 117 000 km2) which is composed of a multi-layer aquifer system. This area is documented with various types of structural (land use, geology) and hydrological data (precipitation, potential evapotranspiration, water volume withdrawn at pumping wells and their location) from which a distributed process-based model has been implemented to model the surface, the unsaturated zone and the aquifer system. The surface model contains 37 620 cells ranging from 1 to 16 km2, 16 141 among them are river cells. Beauce aquifer unit is simulated with 4 489 groundwater cells. To understand the Beauce hydrological functioning and quantify exchanged fluxes, a pragmatic hybrid fitting method has been developed. First the parameters of the water mass balance module are calibrated based on in-river gauging stations selected from a morphological analysis. Then the surface and river routing modules are calibrated based on the analysis of flood discharge peaks at 157 gauging stations. After a pre-calibration of the surface modules for the whole Loire basin, the hybrid fitting methodology focuses on the Beauce aquifer system. It couples manual and automatic iterative calibration. Roughly, the automatic calibration aims at inversing a low water piezometric head map for a steady state using the successive flux estimation. Then the transient manual calibration aims at calibrating others parameters in transient state. The model performances are assessed with a multicriteria approach using global RMSE and bias, and criteria computed for 78 piezometers and for 157 gauging stations. Inspired from soft computing techniques, the hybrid fitting methodology involves three data subsets: a calibration one (10 yr), a validation one (10 yr) and a test one (35 yr). The global RMSE on piezometric head is around 2.5 m for the three subsets and is rather uniformly spatially distributed over 78 piezometers. The sensitivity of the simulation to the different steps of the calibration process is investigated. The transmissivity field fitting is based on the long term evolution (low frequency) of the recharge signal whereas the storage coefficient is fitted taking into account the short term variations of the piezometric head (higher frequency). For long term insight into aquifer system functioning, the priority is thus to first fit the transmissivity field and to assess accurately the distributed aquifer recharge. The fitted model has then been used to quantify the hydrosystem mass balance over a 35-year period of time leading to the conclusion that there is no trend of water resources decrease leading to scarcity neither due to climate nor to human activities.