Modelling nutrient loads in data-scarce large catchments using spatially regularized ensemble calibration

Nutrient pollution in freshwater systems poses major ecological challenges, requiring robust modelling tools to support effective water management. However, catchment-scale water quality modelling is often constrained by sparse monitoring networks and high parameter uncertainty associated with complex biogeochemical systems. This study presents an integrated modelling framework combining the MOBIDIC–ADR hydrological model with a newly developed BIO–ALGAE reactive component to simulate nutrient dynamics across large catchments. The framework employs spatially regularized ensemble calibration using PEST++ iterative ensemble smoother to estimate distributed pollutant loads while quantifying parameter and predictive uncertainty. The model was applied to the Arno River catchment (7990 km 2 ) in Italy, simulating 8 water quality constituents including dissolved oxygen, nitrogen species, phosphorus compounds, and algal biomass over a ten-year period (2011–2020). Using 8151 observations from sparse sampling locations, the calibration demonstrated the model's ability to reproduce observed patterns across multiple constituents. The model proved effective in identifying pollution hotspots, highlighting strong associations between urban areas and elevated carbonaceous biochemical oxygen demand and ammonium loads, whereas phosphorus displayed a more heterogeneous spatial distribution indicative of multiple source contributions. Despite limitations under low dissolved oxygen conditions, the approach effectively captured first-order reactive processes and provided spatially explicit load estimates with uncertainty bounds. This framework offers a practical decision-support tool for targeted water quality management in data-scarce environments.