Application of a Bayesian Watershed Model Linking Multivariate Statistical Analysis to Support Watershed-Scale Nitrogen Management in China

Abstract

Excessive nitrogen loads and subsequent eutrophication risk have led to a series of critical water quality problems in Chinese watersheds. To address this issue, a modeling approach is useful for quantifying nitrogen sources, assessing source apportionment, and guiding management responses. In this study, we modeled the main hydrochemical processes of the Lian River watershed located in the south of China using the Regional Nutrient Management (ReNuMa) model, a model derived from the Generalized Watershed Loading Function (GWLF) model and incorporating Net Anthropogenic Nitrogen Inputs (NANI) to estimate runoff nitrogen concentrations. An informal Bayesian method, the Generalized Likelihood Uncertainty Estimation (GLUE) procedure, was applied for model calibration and uncertainty analysis. The resulting modeled monthly total nitrogen fluxes have high Nash-Sutcliff coefficients (>0.85) for the calibration (2005–2009) and verification (2003, 2004 and 2010) periods, representing an acceptable goodness-of-fit. The model outputs were further processed using multivariate statistical analysis to determine latent rules of nitrogen source apportionment under different circumstances, including different water regimes, seasonal patterns, and loading levels. The main nitrogen contributions in different natural and management-driven conditions have been identified, and appear to be significant for supporting decision-making priorities. We find that the ReNuMa model, with its Bayesian procedure and the linkage of subsequent multivariate statistical analysis, represents a useful approach with applicability within China and a great potential to be extended elsewhere.