A spatially explicit suspended-sediment load model for western Oregon

Abstract

We calibrated the watershed model SPARROW (Spatially Referenced Regressions on Watershed attributes) to give estimates of suspended-sediment loads for western Oregon and parts of northwestern California. Estimates of suspended-sediment loads were derived from a nonlinear least squares regression that related explanatory variables representing landscape and transport conditions to measured suspended-sediment loads at 68 measurement stations. The model gives estimates of model coefficients and their uncertainty within a spatial framework defined by the National Hydrography Dataset Plus hydrologic network. The resulting model explained 64 percent of the variability in suspended-sediment yield and had a root mean squared error value of 0.737. The predictor variables selected for the final model were (1) generalized lithologic province, (2) mean annual precipitation, and (3) burned area (by recent wildfire). Other landscape characteristics also were considered, but they were not significant predictors of sediment transport, were strongly correlated with another predictor variable, or were not as significant as the predictors selected for the final model. The northern Oregon coastal drainages had the highest predicted suspended sediment yields (median yield 475 kilograms per hectare per year) and the Klamath River Basin had the lowest (median yield 53 kilograms per hectare per year). Quaternary deposits were, on average, the largest contributor to incremental suspended-sediment yield even though this lithologic province only makes up 17 percent of the modeling domain. Coast Range sedimentary rocks and Coast Range volcanic rocks had high suspended-sediment yields whereas, in addition to the Klamath terrane, the Western Cascade and High Cascade lithologic provinces had low suspended-sediment yields. Precipitation and the area affected by recent wildfire both positively correlated with suspended-sediment load. Suspended-sediment transport rates predicted by this SPARROW model are less than historical (1956–73) and long‑term (thousands of years) geological rates. This difference likely results, in part, from biases in the data underlying the SPARROW model, probably resulting in predicted suspended-sediment estimates that underestimate actual transport rates. However, the differences also likely owe to natural and human-caused variation in suspended-sediment yields as they respond to changes in climate, vegetation, fire frequency, and land use. In particular, decreases in mean annual suspended-sediment yields within the Umpqua River Basin since 1956–73 may owe to less intense forest harvest, passage of the Oregon Forest Practices Act of 1971, and increased emphasis in habitat protection in recent decades. Such sensitivity may have implications for the spatial and temporal distributions of aquatic and riparian habitats. Knowledge of the regionally important patterns and factors in suspended-sediment sources and transport could support broad-scale, water-quality management objectives and priorities. Because of biases and limitations of this model, however, these results are most applicable for general comparisons and for broad areas such as large watersheds. For example, despite having similar area, precipitation, and land-use, the Umpqua River Basin generates 68 percent more suspended sediment than the Rogue River Basin, chiefly because of the large area of Coast Range sedimentary province in the Umpqua River Basin. By contrast, the Rogue River Basin contains a much larger area of Klamath terrane rocks, which produce significantly less suspended load, although recent fire disturbance (in 2002) has apparently elevated suspended sediment yields in the tributary Illinois River watershed. Fine-scaled analysis, however, will require more intensive, locally focused measurements.