Hybrid watershed models based on nonlinear regression are useful tools for estimating the magnitude of loading rates (i.e., export coefficients) for various pollutant sources within large‐scale river basins. Few such models, however, have incorporated temporal variability in either source distributions or climate, despite evidence that precipitation is the primary driver in interannual variability in loading rates. The model developed here includes changes in precipitation, land use, point source discharge, and livestock operations to capture temporal variability in nitrogen loads. Precipitation is incorporated directly in the formulation of export rates using coefficients that vary by source type. Instream and reservoir retention of nitrogen is included to account for nitrogen sinks within the watershed. A Bayesian hierarchical approach is employed to integrate uncertainty in loading estimates, include prior knowledge of parameters, address intrawatershed correlation, and estimate export coefficients probabilistically. We apply this method to three North Carolina river basins that have experienced substantial growth in urban development and livestock operations in the past few decades, and where eutrophication‐related water quality problems are common. Accounting for temporal variability constrains uncertainties in nonpoint source export coefficients by nearly 50%, relative to a spatial‐only model. Results indicate that livestock operations are a significant contributor of nitrogen throughout much of the study area. Precipitation is shown to have a larger influence on export rates for agricultural than for developed lands, creating a system dominated by agricultural total nitrogen during high precipitation years and by developed (urban) regions during low precipitation years.

No abstract