Estimation of nitrogen inputs to catchments: comparison of methods and consequences for riverine export prediction

Catchment nitrogen (N) budgets are a valuable tool to assess relative magnitude of N inputs and predict losses via riverine export. However, a range of computational approaches may be chosen, potentially affecting the modeled relationship between inputs and exports. To determine the influence of various assumptions and computational details on the effectiveness of N input estimates in predicting riverine N export, we compared eight separate net anthropogenic N input budgets and one soils compartment budget for each of 18 Lake Michigan catchments. N input estimation methods that took into account seasonal fluctuations in livestock numbers and estimated crop N-fixation by legume yield rather than area harvested best predicted river N export. The average annual river export of N from the 18 catchments ranged from less than 300 kg N km(-2) year(-1) in forested areas to more than 800 kg-N km(-2) year(-1) in agricultural catchments and 1,580 kg-N km(-2) year(-1) in small urban catchments. Using the most effective model (R (2) = 0.95, median prediction error = 1.8%) riverine N exports were found to account for 21% of N inputs. Other methods predicted riverine N exports less well (R (2) = 0.61-0.73), bias was greater, and the fractional export of N inputs by rivers decreased to similar to 13%. The soil N budget also was a less effective predictor of river export. This comparison demonstrates that N budgeting that incorporates more detailed description of agricultural N sources can substantially improve prediction of riverine N exports from catchments with a wide range of landscape characteristics.