Partitioning multi-source uncertainties in simulating nitrogen loading in stream water using a coherent, stochastic framework: Application to a rice agricultural watershed in subtropical China

Highlights: • BayeANOVA was used to partition uncertainty sources in modeling stream N loadings. • Relative contribution of 5-source uncertainties to total uncertainty was compared. • Average total uncertainty was greater in rice-growing season than in fallow season. • NO₃⁻-N prediction uncertainty contributed to 75.48% of TN prediction uncertainty. Uncertainty is recognized as a critical consideration for accurately predicting stream water nitrogen (N) loading, but identifying the relative contribution of individual uncertainty sources within the total uncertainty remains unclear. In this study, a powerful method, referred to as the Bayesian inference combined with analysis of variance (BayeANOVA) was adopted to detect the timing and magnitude of multiple uncertainty sources and their relative contributions to total uncertainty in simulating daily loadings of three stream water N species (ammonium-N: NH₄⁺-N, nitrate-N: NO₃⁻-N and total N: TN) in a rice agricultural watershed (the Tuojia watershed) as influenced by non-point source N pollution. Five sources of uncertainty have been analyzed in this study, which arise from model structure, parameters, inputs, interaction effects between parameters and inputs, and internal variability (induced by random errors of model or environment). The results show that uncertainty in parameters relating to the processes of both N and hydrologic cycles contributed the largest fractions of total uncertainty in N loading simulations (58.83%, 63.48% and 61.64% for NH₄⁺-N, NO₃⁻-N and TN loading, respectively). Additionally, three of the largest uncertainties (i.e. parameters, inputs and interaction effects) in all three simulated N loadings were on average significantly greater in the rice-growing season relative to the fallow season, primarily due to the excess fertilization application during the rice-growing season. The predicted TN uncertainty was mainly attributed to the inaccuracy of NO₃⁻-N simulation, which contributed to 75.48% of predicted TN uncertainty. It is concluded that reducing the parameter uncertainty of NO₃⁻-N loading simulation during the rice-growing season is the key factor to improving stream water N modeling precision in rice agricultural watersheds.