Abstract Process‐based watershed models require robust, spatially distributed data for accurate total nitrogen (TN) simulation. However, traditional calibrations are often constrained by sparse in situ monitoring networks, resulting in substantial estimation errors and biases, particularly in ungauged sub‐basins. While newly available spatial inversion data offer high‐resolution observations of TN, their complex, heterogeneous, and non‐stationary error structures pose challenges for rigorous integration into process‐based models. This study presents a Hybrid Inversion‐Observation Bayesian Data Fusion (HIOBF) framework to address this issue. The framework incorporates a geography‐driven spatial extrapolation model to characterize inversion data errors in unmonitored areas, systematically merging high‐resolution spatial inversion data with conventional point‐scale measurements. We implemented and tested this framework using the Soil and Water Assessment Tool model in the Dongjiang River Basin, China. Results from independent validation demonstrated that by synergizing the spatial coverage of inversion data with the temporal precision of gauge measurements, the HIOBF framework significantly reduced parametric uncertainty and improved the Nash‐Sutcliffe Efficiency by an average margin of 55% compared to the traditional gauge measurement‐only baseline. This work provides a transferable methodology, offering a robust approach to enhance simulation accuracy and provide defensible uncertainty estimates for water resource management in data‐scarce contexts.
Zhou et al. (Wed,) studied this question.