Results


Model Calibration


The TN model included 10 nutrient source terms and 4 delivery terms and the TP model included 9 nutrient source terms and 2 delivery terms (tables 3 and 4). The signs of the delivery terms, rather than the magnitudes, provide information on how they influenced the models. Three delivery terms in the TN model (mean annual precipitation, mean annual solar radiation, and arid land irrigation) had positive coefficients and acted to enhance the delivery of nitrogen from land to water. One delivery term (base flow index) had a negative coefficient and acted to attenuate the delivery of nitrogen from land to water. One delivery term in the TP model (mean annual precipitation) had a positive coefficient and one delivery term (base flow index) had a negative coefficient. Attenuation was not a significant removal mechanism in free-flowing streams or impoundments in either the TN or the TP model.


Based on the R² of yield and the RMSE values, the TN model showed better fit with the calibration data set than the TP model (tables 3 and 4), and the NLLS coefficient estimates for both the TN and TP models were generally close in value to the nonparametric bootstrap estimates (meaning that the uncertainty of the NLLS coefficients was low). The exceptions were the coefficients for scrubland and grassland in the TP model and atmospheric deposition and arid land irrigation in the TN model. The negative value for scrubland and grassland in the TP model indicated that there was large uncertainty associated with the coefficient for this source term. Based on an evaluation of the model residuals, the best model fit for TN was in the lower Columbia River basin (LCOL) and the best model fit for TP was in the northern Oregon coastal drainages (NOCR) (appendix B; figs. B1 and B2). The poorest model fit for TN was in the Yakima River basin (with no substantial bias toward over or under prediction) and the poorest model fit for TP was in the Snake River headwaters (due exclusively to under prediction). There were two clusters of under prediction that might have resulted from underestimation of natural nutrient sources. The cluster of under prediction in the Middle Snake-Boise River basin was in an area with documented deposits of nitrate salts (Mansfield and Boardman, 1932), which were not accounted for in the TN model, and the Snake River headwaters lie in a predominantly forested watershed in a region of extensive phosphate deposits called the Western Phosphate Field (U.S Geological Survey, 2002).


The nitrogen and phosphorus content of livestock manure was represented by two nutrient source terms: confined cattle and grazing livestock. The confined cattle source term represented the manure from cattle associated with a registered dairy or feedlot, and the grazing livestock source term primarily represented the manure from cattle that were not associated with a registered dairy or feedlot, but also included manure from a relatively small number of other non-poultry animals. These two types of manure were included as distinct nutrient sources in the TP model. In contrast, these two nutrient sources were combined and modeled as one source in the TN model because of difficulties experienced during model calibration. Specifically, there was a conflict between the nutrient source terms representing grazing livestock manure and atmospheric deposition; neither source term was significant when both were included in the model along with confined cattle manure. However, when confined cattle and grazing livestock manure were combined and represented by one nutrient source term, the resulting TN model included significant coefficients for both livestock manure (confined and grazing) and atmospheric deposition. Combining the two types of manure in the TN model was justified because the coefficient values of the confined cattle and grazing livestock source terms were similar when modeled separately, which indicated no substantial difference in availability.


The decision to include source terms representing nutrients from nonpoint urban sources and springs and power returns was not based solely on the statistical results from the calibrations. The nitrogen and phosphorus loading from nonpoint urban sources was represented by the area of developed land rather than by an alternative surrogate equal to non-farm fertilizer use (appendix A). Either one of these nutrient source terms was significant in the TN and TP models, but developed land was selected because it accounted for most nonpoint sources of urban¹ nutrients, including fertilizer use, leaking sewer lines, animal manure, as well as other sources, whereas non-farm fertilizer represented only one source. The load from springs and power returns was a significant nutrient source term in the TN model but not in the TP model (p-value = 0.1408). This nutrient source term was retained in the TP model, however, because it was an important local source of phosphorus for some stream reaches. This determination was based on the large under predictions that were observed at calibration stations located downstream of springs and power returns when they were not included in the TP model calibration.


Model Predictions


The incremental yields of TN and TP were much greater on the western side of the Cascade Range compared to the eastern side (table 5; figs. B3 and B4). The largest median incremental yields of TN and TP were predicted for the northern Oregon coast (NOCR) and the Washington coast (WACR), respectively. The high yields of TN and TP in these watersheds were directly related to the large amount of precipitation. The incremental yields of TN and TP in other areas west of the Cascade Range (PUGT, LCOL, WILL, and SOCR) were less than in WACR and NOCR, but still greater than areas east of the Cascade Range. In the east side watersheds the greatest median incremental TN yields were predicted in the Clearwater (CLRW) and Middle Columbia River (MCOL) basins and the greatest median incremental TP yields were predicted in the Clearwater and Spokane River (SPOK) basins. The large median TN and TP yields in the Clearwater River basin were due to high precipitation compared to the other east side HUC6 watersheds. The large median TN yield in the Middle Columbia River basin was due to a combination of high average input from farm fertilizer and high average precipitation compared to the other east side HUC6 watersheds and the large median TP yield in the Spokane River basin was due to high levels of urbanization in this watershed compared to other east side HUC6 watersheds.


The largest local sources of TN and TP load are shown in figures 2 and 3 and summarized in table 6. The figures do not show catchments where the largest local source was the load at a boundary reach or the load from springs and power returns because these represented less than 1 percent of the total number of catchments. The largest local source of TN in more than 50 percent of the catchments was from forestland or alder trees and the largest local source of TP in almost 90 percent of the catchments was from livestock manure (primarily grazing livestock) and geologic sources. The highest median TN and TP yields were predicted for catchments where urban sources were the largest local source, the lowest median TN yields were predicted for catchments where atmospheric deposition was the largest local source, and the lowest median TP yields were predicted for catchments where farm fertilizer was the largest local source (table 6). There were significant (α < 0.05) differences between each of the median incremental TN yields and each of the median incremental TP yields shown in table 6.


On average, the largest contributor to the total TN load in PNW streams was forestland, which was responsible for at least 63 percent of the TN load in one-half of the reaches (table 7). On average, the largest contributors to total TP load in PNW streams were geologic phosphorus (which was represented by forestland, scrubland, and grassland) and livestock manure (primarily grazing livestock). In one-half of the reaches these two nutrient sources were responsible for at least 58 and 30 percent, respectively, of the TP load. Although on average diffuse nutrient sources (both natural and anthropogenic) were responsible for most of the total TN and TP load in PNW streams, concentrated anthropogenic nutrient sources contributed much of the total load in some of the large rivers. Two examples were the Willamette River in western Oregon (fig. 4) and the Snake River, which flows through southern Idaho, northeastern Oregon, and southeastern Washington (fig. 5). Both rivers drain watersheds containing a mix of agricultural, urban, and undeveloped land. Forestland and farm fertilizer were the largest contributors to TN load in the Willamette River upstream of Eugene and Springfield and farm fertilizer was generally the largest contributor to TN load downstream of this point. The exception was the 45 km of river immediately downstream of Eugene and Springfield, where the inputs of nitrogen from urban sources (primarily wastewater treatment plants) resulted in point sources being the largest contributor to TN load. Geologic phosphorus was the largest contributor to TP load in the Snake River between its headwaters and Idaho Falls, whereas point sources and agricultural nutrient sources (farm fertilizer and livestock manure) were generally the largest contributors downstream of this point. The livestock manure generated along this part of the Snake River was mostly from cattle in dairies and feedlots. Almost all of the contribution from urban nutrient sources upstream of the Boise River was from fish farms and hatcheries², whereas downstream, a large percentage was from wastewater treatments plants, especially those that discharged to the Boise River. The three-fold increase in TP load between Twin Falls and King Hill was mostly due to phosphorus input from springs and the large number of fish farms and hatcheries located along this reach. The contribution from urban nutrient sources (almost exclusively fish farms and hatcheries) was as high as 50 percent of the total TP load in this segment of the Snake River. The three-fold increase in TP load downstream of the Boise River was mostly due to phosphorus input from this large tributary.


¹ The exception was nitrogen leaching from septic tanks, which was modeled as a separate source in areas that were not served by municipal sewer lines in 2002.


² Although fish farms and hatcheries are not typically in urban areas they were grouped with urban sources in this study because their effects could not be distinguished from urban point sources during model calibration.


First posted July 17, 2013