Semi-Conceptual Spatialized Modeling of Surface – Underground Water Flows in a Karst Environment

Soutenance de thèse de doctorat de Ibrahim Al Khoury.

The Jury members will be:

  *   Ms. Hélène CELLE, reviewer, Université Franche Comté
  *   Mr. Nicolas Massei, reviewer, Université de Rouen Normandie
  *   Mr. Pieter VAN BEEK, examiner, Université Toulouse III – Paul Sabatier
  *   Mr. Stéphane BINET, examiner, Agence de l’eau Adour-Garonne
  *   Mr. Ryan BAILEY, examiner, Colorado State University
  *   Mr. David LABAT, thesis director, Université Toulouse III – Paul Sabatier
  *   Ms. Laurie BOITHIAS, thesis co-director, Université Toulouse III – Paul Sabatier

Summary:
Karst aquifers, which supply freshwater to nearly 25% of the global population, are facing depletion due to climate change and anthropogenic pressure. Hydrological models have been advocated for effective karst water resources planning and management, but studies integrating the recharge-discharge characteristics of karst watersheds and evaluating their response to changes in the flow dominant controls remain limited due to their inherent heterogeneity, anisotropy, flow duality, and non-linearity. This thesis developed the semi-conceptual spatialized numerical model ISPEEKH (Integration of Surface ProcEssEs in Karst Hydrology) by coupling SWAT+, the restructured version of the semi-distributed eco-hydrological model SWAT (Soil and Water Assessment Tool), with the non-linear epikarst-matrix-conduit reservoir module of the rainfall-runoff model KarstMod to simulate the surface-underground water flows in karst watersheds. ISPEEKH was applied to simulate the daily water balance of the Baget catchment (13.25 km2), located in a poorly gauged region of the French Pyrénées and characterized by conduit-dominated non-linear flow. The model simulated the catchment streamflow satisfactorily (NSE = 0.67, R² = 0.68, and PBIAS = 0.7% for the 2008−2013 calibration period, and NSE = 0.65, R² = 0.69, and PBIAS = -13.83% for the 2014−2018 validation period), allowing the estimation of the epikarst, matrix, and conduit fluxes, including the bidirectional matrix-conduit exchange flow rate, the contribution of the matrix and conduit outflows to spring flow, and their seasonal variability. The Baget catchment’s hydrological response to synthetic land-use change scenarios of afforestation and deforestation was then assessed using ISPEEKH. Results showed that afforestation over the entire catchment did not significantly affect its water balance, while deforestation for wood production increased the mean annual discharge by 6−9%, notably in the low-flow periods, and deforestation for pastureland development reduced the mean annual discharge by 5−7%, mainly in the high-flow period. Various precipitation datasets were then evaluated for the simulation of daily streamflow in the catchment from 2006−2018, including the gauge-based (CPC and E-OBS), reanalysis (SAFRAN, COMEPHORE and ERA5-Land), and satellite-based (PERSIANN-CDR, IMERG-LR, SM2RAIN-ASCAT and CHIRPS) products. ISPEEKH was integrated with a PEST framework for automated calibration, sensitivity analysis, and uncertainty quantification. Results showed that streamflow was significantly underestimated under the ensemble of the precipitation products. The gauge- and satellite-based precipitation products had the worst performance, with a flow underestimation bias ranging from 48 to 74%, while the reanalysis products yielded better streamflow simulation results with a flow underestimation bias of 30−44%. The CPC, E-OBS, ERA5-Land, IMERG-LR, and merged CPC-IMERG-LR datasets downscaled to 1-km spatial resolution did not improve the model predictive performance compared to the coarse datasets. The downscaled datasets along with COMEPHORE were bias corrected to reduce the water balance discrepancy and re-applied for hydrological modeling. Significant improvement in the streamflow simulation were observed under the corrected COMEPHORE and downscaled E-OBS, CPC, and merged CPC-IMERG-LR precipitation datasets, with COMEPHORE yielding the best model predictive performance (NSE = 0.719, R² = 0.736, and PBIAS = 3.2% for the calibration period, and NSE = 0.637, R2 = 0.732, and PBIAS = -10.65% for the validation period), suggesting that fine-resolution native reanalysis precipitation could be used as a base dataset for the hydrological modeling of remote meso-scale karst catchments.