Automatic Multi-Scale Calibration Procedure for Nested Hydrological-Hydrogeological Regional Models

Large hydrosystem modelling and understanding is a complex process depending on regional and local processes. A nested interface concept has been implemented in the hydrosystem modelling platform for a large alluvial plain model (300 km²) part of a 11000 km² multi-layer aquifer system, included in the Seine basin (65000 km², France). The platform couples hydrological and hydrogeological processes through four spatially distributed modules (Mass balance, Unsaturated Zone, River and Groundwater). An automatic multi-scale calibration procedure is proposed. Using different data sets from regional scale (117 gauging stations and 183 piezometers over the 65000 km²) to the intermediate scale(dense past piezometric snapshot), it permits the calibration and homogenization of model parameters over scales.

The stepwise procedure starts with the optimisation of the water mass balance parameters at regional scale using a conceptual 7 parameters bucket model coupled with the inverse modelling tool PEST. The multi-objective function is derived from river discharges and their de-composition by hydrograph separation. The separation is performed at each gauging station using an automatic procedure based one Chapman filter. Then, the model is run at the regional scale to provide recharge estimate and regional fluxes to the groundwater local model. Another inversion method is then used to determine the local hydrodynamic parameters. This procedure used an initial kriged transmissivity field which is successively updated until the simulated hydraulic head distribution equals a reference one obtained by krigging. Then, the local parameters are upscaled to the regional model by renormalisation procedure.
This multi-scale automatic calibration procedure enhances both the local and regional processes representation. Indeed, it permits a better description of local heterogeneities and of the associated processes which are transposed into the regional model, improving the overall performances of the two nested models.