Validation of water transit traced by a distributed hydrological model using a geochemical end-member mixing approach

Knowing not only quantities but also pathways of water through a catchment is crucial for an in-depth understanding of hydrological processes. This is particularly true when aiming to understand the pathways of pollutants in the environment. We use a geochemical end-member mixing approach to validate the water transfer predicted by a process-based distributed hydrological model (J2000) developed over the Claduègne rural mesoscale catchment (42 km²) under Mediterranean climate. This hydrological model represents explicitly the heterogeneity of the catchment through hydrologic response units, including 6 land cover and 4 lithology classes. It also includes a detailed representation of human activity (drinking water extraction, wastewater effluent discharge, urban overland flow, irrigation, livestock breeding). It was calibrated on streamflow discharge at three hydrometric stations throughout the catchment. It tracks water volumes through the catchment by spatial origin and four different flow processes (overland flow, subsurface storm flow; and slow and rapid groundwater flow). The mixing model distinguishes water originating from six end-members, including subsurface water from two types of lithology (sedimentary [sed] vs. basaltic [bas]) and two classes of land cover (crop and pasture [open] vs. shrubland and forest [closed]) as well as overland flow [OF] and urban sources. Each end-member was characterized by dissolved concentration of 47 elements in water samples collected at different locations and under various hydrological conditions (97 samples in total), quantified via Inductively Coupled Plasma - Mass Spectrometry (ICP-MS).

End-member signatures were repeatedly drawn from Weibull distributions fitted to samples for each end-member. Non-negative mixing contributions were optimized in order to represent measured concentrations at the outlet. Only draws resulting in a sum of contributions of 100 % (±5 %) were kept; the average of the 100 best drawn combinations (least residuals) was used as final contribution. The end-member mixing model was applied to 256 samples taken at high frequency (up to 2 h^-1) during flood events (14 and 4 events at the Claduègne and Gazel outlets, respectively), in addition to low-flow samples collected at different seasons.

The contribution of overland flow was zero most of the time and peaked during flood events, with proportions up to 80-90 %. The urban contribution was mostly below 10 %, with some higher values during low flow periods.

Results were compared to tracking results from the hydrological model, run at an hourly time step. Direct per-sample correlations between the two models had the following pearson R values: urban: 0.60, OF: 0.52, sed closed: 0.54, sed open: 0.41, bas closed: 0.23, bas open: 0.31. All were significant with p<0.05, except bas closed (p<0.1). In a more qualitative way, the two models agreed on patterns over the course of flood events and over the seasons, as well as contribution-discharge relationships. We demonstrated that these two independent approaches produce coherent results, validating the hydrological model’s representation of water transit through the catchment.