Integrating riparian landscape metrics into Bayesian mixed-effects model to explore river nitrogen concentration dynamics

Accurately modeling river nitrogen (N) concentration dynamics is crucial for pollution mitigation, yet statistical models often lack explanatory power while process-based models demand extensive data. This predictive gap often stems from a focus on landscape composition, neglecting the critical role of landscape configuration—the spatial arrangement of land uses—in governing nutrient transport. To address this gap, we developed a Bayesian mixed-effects model to capture spatiotemporal dependencies while integrating key landscape configuration metrics. In the Qinhuai River watershed, our new model (BMECONFI) achieved high predictive performance (R2=66%−69%; RMSE=0. 207–0. 218), significantly outperforming models based on composition alone. Posterior distributions revealed that human activity intensity (NLI) and impervious surface aggregation (PLADJImpervious) positively influenced river N concentrations, while water landscape connectivity (IJIWater) exerted a negative effect. This study provides a robust and interpretable framework that bridges the gap between oversimplified statistical approaches and complex process models, offering a practical tool for designing targeted, landscape-based N pollution mitigation strategies.