Discussion of Results


The calibrations of the PNW NHD SPARROW models resulted in a better overall fit with the calibration load data compared to the RF1 models (table 8). These improvements in model fit were likely due to a combination of factors: (1) the improved resolution of the NHD network compared to the RF1 network, (2) differences in the set of nutrient loads used to calibrate the models, and (3) refinements in the estimates of anthropogenic nutrient sources. The better fit of the NHD models compared to the RF1 models resulted in less uncertainty in the predictions of nutrient loads, incremental nutrient yields, and nutrient source shares. Model uncertainty associated with the PNW RF1 models, which also applies to the PNW NHD models, is discussed in Wise and Johnson (2011). 


The coefficients for the nutrient source terms (tables 3 and 4) provide some insight into why certain results were obtained. The coefficients for the nutrient source terms representing land cover types (expressed as kilograms per square kilometers per year [(kg/km²)-yr]) were analogous (but not directly comparable) to yield values or export coefficients, whereas the coefficients for the nutrient source terms representing mass loading (expressed as kilograms per year [kg/yr]) were an indication of the availability for delivery to streams. In both the TN and TP models, developed land had the largest coefficient of any land cover source term, which reflected the large number of nonpoint nutrient sources in urban areas and the relative ease with which they are transported to streams. The highest incremental yields were predicted for catchments dominated by urban nutrient sources because of the large coefficients for developed land and the fact that nutrients from point sources were discharged directly to streams. In both models the land cover terms representing natural nutrient sources had much smaller coefficients than developed land, which explained why catchments where the largest share of the local load was from these land cover types had relatively low incremental yields.


The coefficients for the farm fertilizer source terms indicated that a greater proportion of nitrogen was delivered to streams compared to phosphorus and this explained why catchments where the largest share of the local load was from farm fertilizer had relatively high incremental TN yields but the lowest incremental TP yields. These results are consistent with the physical and chemical processes affecting the mobility of nitrogen and phosphorus. Although the coefficients for the livestock manure source terms in both models were greater than those for the farm fertilizer source terms, catchments where the largest share of the local load was from livestock manure had relatively low incremental yields. This was because most of the manure was from grazing livestock, which represents a more diffuse source of nutrients compared to farm fertilizer. The coefficients for the source term representing point sources and the source term representing springs and power returns in the TN model were about 1, but the coefficients for these nutrient source terms in the TP model were 1.20 and 1.67, respectively. Because these coefficients acted as scaling factors for the estimated load, the model results indicated that our estimates of TP load from point sources and from springs and power returns were likely too low.


There were potential advantages to using a geologic phosphorus source term that was based on a phosphorus index instead of the land cover source terms that were used in the NHD TP model. This was the approach that was used in RF1 SPARROW TP model (Wieczorek and Lamotte, 2013; Wise and Johnson, 2011). One potential advantage was that the phosphorus index provided a spatially continuous distribution of geologic phosphorus. The source terms for land cover did not account for geologic phosphorus released from land cover types other than forestland, scrubland, and grassland. Geologic phosphorus released from agricultural and developed land was not a significant nutrient source in the TP model because it was negligible compared to the contributions from anthropogenic activities (farm fertilizer, confined cattle, and developed land). Another potential advantage was that the phosphorus index accounted for regional differences in the phosphorus content of geologic material whereas the land cover source terms did not. There was no evidence, however, that using the phosphorus index provided a better model fit in areas with relatively high concentrations of geologic phosphorus. The under predictions at the calibration stations located in the Snake River headwaters (which are in the Western Phosphate Field), for example, were greater when the phosphorus index was used in place of the land cover source terms.


The better fit for the TP model using the source terms for land cover compared to the phosphorus index indicated that weathering processes were a more important control on the availability of natural phosphorus than the phosphorus content of the rocks themselves. Additionally, the coefficient for the west-side forest source term being the largest of the three geologic source terms indicated that the geologic phosphorus in this land cover type was more readily delivered to streams compared to the geologic phosphorus in the other land cover types (east side forests and scrubland/grassland). These results were most likely due to differences between the amount of precipitation that falls on the west and east sides of the Cascade Range. The mean annual precipitation in predominantly (greater than 50 percent) forestland catchments on the west side of the Cascade Range (194 cm) was greater than the mean annual precipitation in predominantly forestland catchments on the east side (84 cm) and predominantly scrubland and grassland catchments (43 cm; almost all located on the east side). Soil erosion rates in forested catchments on the west side of the Cascade Range are greater than soil erosion rates in forested catchments on the east side (Elliot, 2006) and, presumably, also greater than soil erosion rates in scrubland and grassland. As a result, the export of geologic phosphorus from west-side forestland was expected to be greater than the export of geologic phosphorus from east-side forestland, scrubland, and grassland. The relatively small coefficient for the scrubland and grassland source term (compared to the forestland terms) might have reflected slower weathering processes in arid and generally low slope environments.


The calibration results for the delivery terms in the TN and TP models were consistent with our assumptions about the processes that occur within watersheds. The delivery of nitrogen and phosphorus to streams in catchments receiving high precipitation was larger than in similar catchments receiving less precipitation. Similarly, in arid areas agricultural catchments having a large percentage of irrigated land contributed more nitrogen to nearby streams than similar catchments having less irrigated land. Catchments having a large baseflow index tended to contribute less nitrogen than catchments having a small baseflow index, presumably because of denitrification in soil and groundwater. The contribution of phosphorus also was less from catchments having a large baseflow index compared to catchments having a small baseflow index. This most likely resulted from less overland flow in catchments having a large baseflow index, meaning that less sediment-bound phosphorus was available for transport. The use of solar radiation as a delivery term is unique to the PNW NHD SPARROW model for TN. The positive relation between solar radiation and forest productivity could explain the significance of this delivery term. Forested catchments that receive more solar radiation have greater productivity rates and, presumably greater nitrogen uptake from the soil than forested watersheds that receive less solar radiation. This could leave less nitrogen available for permanent removal through soil denitrification.


The results obtained for nutrient loss in free-flowing streams and impoundments in the NHD models were not typical of recent SPARROW nutrient applications for large hydrologic regions (see Hoos and McMahon, 2009; Brown and others, 2011; Garcia and others, 2011; Moore and others, 2011; Rebich and others, 2011; and Robertson and Saad, 2011) but were consistent with the PNW RF1 models. All the other regional SPARROW applications identified TN loss in free-flowing streams as a significant removal process, whereas the PNW NHD TN model did not. The result could have reflected the importance of fixation in the cycling of nitrogen in PNW streams, meaning that a substantial amount of the nitrogen lost through denitrification was replaced with nitrogen fixed from the atmosphere. All of the other regional SPARROW applications also identified TP loss in impoundments as a significant removal process whereas the PNW NHD TP did not. This result indicates that settling within impoundments was not an important mechanism for phosphorus removal in PNW surface waters. This could have been because a large proportion of the phosphorus was in the dissolved form (and was not settling out before flowing out of the impoundments) or because the settling of phosphorus in impoundments was balanced by phosphorus dissolution and resuspension.


First posted July 17, 2013