Extreme low flow estimation: added value of piezometry to constrain the asymptotic behavior of a lumped rainfall-runoff model

The adaptation to climate change of thermal power plants necessitates the identification and characterization of high impact hazards. Extremely low river flow is one of these situations. The estimation methods traditionally used today still rely on extreme value theory (i.e., statistical adjustment on few observations and/or simulations), but these methods suffer from numerous limitations. Recent developments now make it possible to consider another approach, based on hydro-climatic simulations: extreme low flow quantiles are estimated by coupling a climate generator and a hydrological model. A first proof of concept was recently tested on a single basin and showed significant potential (Parey et al. 2022). Areas for improvement were also identified, both on the climate generator and the hydrological model.

The purpose of this work was then (i) to extend this first proof of concept to a larger number of basins and (ii) to quantify the sensitivity of the simulation chain (i.e., extreme low flow quantiles estimation) to the parameters of the hydrological model (in our case, the MORDOR-SD daily lumped rainfall-runoff model, Garavaglia et al. 2017).

A dataset of 33 catchments, each of them being associated with at least one piezometer, was selected to investigate whether the MORDOR-SD model could be constrained by piezometric time series to improve low flow simulations. By performing calibrations using only streamflow information we first confirmed that a particular state of the model was well correlated with piezometry in most studied catchments (the level of the so-called « deep » store, dedicated to the baseflow component).

A multi-objective calibration approach was then implemented, optimizing both (i) flow simulation with 4 criterions focusing on different streamflow signatures and (ii) eventually one supplementary criterion base on the affine correspondence between the deep storage level of the model and piezometry (i.e., calibration with or without piezometric information).

The results led us to propose a classification of the 33 basins based on two indices. The first index characterizes the importance of the baseflow in the streamflow (BFI = baseflow index). The second index characterizes the a priori representativity of the piezometric time series during low flows (Cor QMNA/ZMNA, index also used in Andreassian & Pelletier 2023).

For 14 out of the 33 basins (BFI > 0.7), piezometric information was almost neutral and did not lead to a significant improvement: the streamflow information was sufficient to constrain the low flow simulations. For 11 out of the 33 basins (Cor QMNA/ZMNA < 0.6 and BFI < 0.7), piezometric information was misleading and degraded the results: we assume that the piezometric information was not sufficiently relevant. Ultimately, only 8 out of the 33 basins (Cor QMNA/ZMNA > 0.6 and BFI < 0.7) emerged as interesting case studies. For these 8 watersheds, the piezometric information appears relevant to be included in the calibration process to derive a physics-based extrapolation of extremely low flow quantiles.