Sensitivity of riverine nitrate modelling to territorial estimations of diffuse sources in agricultural catchments

Coastal eutrophication has been linked to excessive nitrogen inputs from intensive agricultural practices on contributing watersheds and is associated with multiple ecological alterations such as green tides and toxic algal blooms, as observed in the Bay of Brest (Brittany, France). Mechanistic modelling is a powerful tool for improving our understanding of nitrogen transfers from the land to the sea through river networks. However, its effectiveness strongly depends on the accurate estimation of territory-specific boundary conditions, which remains challenging due to the scarcity of existing observational measurements, especially in small watersheds.

The sensitivity of simulated riverine nitrate concentrations to different boundary condition datasets was assessed using the pyNuts-Riverstrahler modelling platform applied to the two main watersheds draining into the Bay of Brest (~320 and 1700 km²). This model explicitly represents water fluxes (including discharges from wastewater treatment plants, water withdrawals, and dams) and associated concentrations of carbon and nutrients (N, P, and Si) across the entire river network at a kilometre-scale spatial resolution. Simulated riverine nitrate concentrations were compared to observational data to evaluate the performance of different input datasets for each watershed. First, two alternative baseflow estimation approaches were tested, namely a statistical recursive digital filter (BFLOW) and the conductivity mass balance (CMB) method. Second, diffuse agricultural nitrogen inputs through surface runoff were estimated from the GRAFS methodology at two spatial scales: regional and municipal.

Results indicate that estimates derived from the CMB method predict lower baseflow contribution to total streamflow and show greater spatial variability across the watersheds than those obtained with BFLOW. Nitrate simulations driven by municipality-scale GRAFS inputs better reproduce observed nitrate concentrations and their spatial heterogeneities along the river network, despite data gaps due to the partial confidentiality of agricultural statistics at this scale. The simulation combining CMB methodology with municipality-scale GRAFS inputs appears to be the most representative, both in terms of nitrate concentration levels and seasonal dynamics simulation, particularly in areas with complex hydrogeological functioning. 

Overall, this work highlights the critical role of boundary condition estimation in mechanistic hydro-biogeochemical modelling, with direct implications for understanding and managing coastal eutrophication in agricultural watersheds.