Influence of Climate and Human Activities on the Relationship between Watershed Nitrogen Input and River Export

Abstract

River export of nitrogen (N) is influenced strongly by spatial variation in anthropogenic N inputs and climatic variation. We developed a model of riverine N export for 18 Lake Michigan Basin watersheds based on N budgets at 5-year intervals from 1974 to 1992. N inputs explained a high proportion of the spatial variation in river export but virtually none of the temporal variation, whereas between year N export was related to variation in discharge for over one-half of the rivers. A regression model of riverine N exports as an exponential function of N inputs and a powerfunction of annual water discharge accounted for 87% of the variation in annual total nitrogen fluxes over space and time. Application of this model to three scenarios of future land use, including business as usual, greater reliance on organic farming methods, and expanded corn-based ethanol production, and two climate scenarios, including increases in water discharge by 5% and 10%, suggests that riverine N export is likely to increase by as much as 24% in response to heavierfertilizer use for expanded corn production and a 10% increase in annual discharge. However, N export by rivers could decrease below present-day export through reduced reliance on commercial fertilizer use.

A lagged variable model for characterizing temporally dynamic export of legacy anthropogenic nitrogen from watersheds to rivers

“…3) suggest that the calibrated results are highly consistent, especially considering the complexities of N delivery across watershed landscapes to rivers. These calibrated results equal or exceed those obtained with other watershed N simulations using mechanistic models such as SWAT, AGNPS, and HSPF and lumped watershed models such as the export coefficient model, SPARROW and PolFlow model, as well as statistical models developed between NANI and riverine TN flux (Nash-Sutcliffe coefficient varied between 0.65 and 0.90 (>0.65 is considered very good as reviewed by Moriasi et al (2007)) and R 2 varied between 0.70 and 0.96 (McIsaac et al 2001;De wit et al 2003;Li et al 2014;Han et al 2009;Chen et al 2014). Although this LVM lacks the ability to predict seasonal/monthly/daily riverine N export (it is difficult to evaluate watershed N budgets at finer temporal resolutions) compared with mechanistic models, it has the advantage of simplicity and importantly can quantify the contribution of legacy N sources (the lag effect of NANI) to riverine N exports.…”

“…NANI has been widely recognized as an effective predictor of riverine N fluxes and their relationship is commonly described using an exponential function due to the effects of progressive N saturation in landscapes (McIsaac et al 2001;Han et al 2009;Hong et al 2012;Chen et al 2014). In addition, fractional export of NANI by rivers is strongly related as a power function to various watershed attributes, including hydroclimate (e.g., water yield, precipitation, and temperature) (McIsaac et al 2001;De wit et al 2003;Han et al 2009;Howarth et al 2012;Li et al 2014), land use (Groffman et al 2004;Han et al 2009), and agricultural management (e.g., drainage systems, fertilizer application, and tillage) (Sobota et al 2009;Kopáček et al 2013;Chen et al 2014).…”

“…In addition, fractional export of NANI by rivers is strongly related as a power function to various watershed attributes, including hydroclimate (e.g., water yield, precipitation, and temperature) (McIsaac et al 2001;De wit et al 2003;Han et al 2009;Howarth et al 2012;Li et al 2014), land use (Groffman et al 2004;Han et al 2009), and agricultural management (e.g., drainage systems, fertilizer application, and tillage) (Sobota et al 2009;Kopáček et al 2013;Chen et al 2014). Current models are generally established using a multi-year average temporal resolution to reduce the uncertainty derived from the N leaching lag effect (Howarth et al 2006;Hong et al 2012;Swaney et al 2012;Li et al 2014).…”

“…) that incorporates both NANI from the current year and from any previous year (lag terms) based on the observed relationships between NANI and riverine N flux in previous studies (McIsaac et al 2001;Han et al 2009;Swaney et al 2012;Chen et al 2014):…”

“…A detailed description of the Bayesian calibration approach for relevant water quality model parameters is available in Shen and Zhao (2010) and Chen et al (2012). According to parameter values from previous relevant studies (McIsaac et al 2001;Howarth et al 2006;Han et al 2009;Chen et al 2014), the prior distribution of the four unknown parameters was assumed to follow a normal distribution, i.e., a, 5.0±5.0; b j , 1.0±1.0; β 0 , 0.018±0.018; and λ, 0.5±0.5. The WinBUGS code for Bayesian calibration is available in Electronic supplementary material E. To obtain the best-fit posterior model for parameters a, b j , β 0 , and λ, two Markov chains were initiated at different arbitrary initial values as well as being independent of each other using WinBUGS 1.4.…”