Modeling dunnian runoff dynamics during flood events in the Loing watershed (France)

Flood events are significant hydrological phenomena that can lead to severe human and economic damages. In the Seine River basin (France), the floods of May-June 2016 resulted in four fatalities and economic losses estimated between 0.8 and 1.25 billion euros, according to the French reinsurance fund. This event followed an unusually wet month of May, marked by intense rainfall concentrated in the southern part of the basin, primarily within the Loing River watershed. At the outlet of this watershed, an unprecedented peak discharge of nearly 500 m³.s^-1 was recorded, with a return period now estimated to be between 400 and 1,000 years. The latest IPCC report emphasizes the increasing frequency and severity of extreme weather events driven by climate change, highlighting the need for a better understanding of the hydrological processes leading to major floods to improve forecasting and mitigation efforts.

Given that flood dynamics typically result from a combination of processes including river overflow, subsurface runoff and groundwater discharge, a coupled surface and groundwater hydrological application of the Loing River watershed is currently being developed. The CaWaQS modeling platform is used to (i) simulate the key hydrological processes within each component of the Loing hydrosystem (soil, river system, vadose zone, and aquifer system) and (ii) generate daily key variables of interest, such as distributed discharge and hydraulic heads. The simulation of surface flow processes relies on a reservoir-based conceptual approach, utilizing sets of seven calibration parameters (or production-functions), distributed according to the intersection of soil and land use types. An initial simulation solely based on production-function parameters inherited from the CaWaQS-Seine basin regional application (15 production-functions, 105 parameters) led to largely underestimated flows. As a result, local specificities such as the extensive artificial drainage of arable lands was incorporated through four additional functions, bringing the total number of parameters to 133. Bayesian inversion and frequency analysis of observed discharge data were used to calibrate AET fluxes and effective rainfall partionning into runoff and infiltration flows, although this approach was still insufficient to accurately represent flood dynamics. To address this, the CaWaQS source code was enhanced to explicitly incorporate Dunne-type runoff processes related to soil saturation. A new calibration was based on analogies between physical and conceptual parameters, reducing the number of parameters requiring adjustment from 57 to just 2: drainage efficiency and kinematic porosity. These two parameters, initially not spatially discretized, are calibrated using an automated screening procedure.

This revised conceptualization and its associated fitting methodology enable the simulation of runoff processes triggered by soil saturation and drainage in agricultural areas, providing a differentiated assessment of their impact on the hydrosystem. It also allows a more accurate representation of flow dynamics during flood events.