Empirical Models Based on the Universal Soil Loss Equation Fail to Predict Sediment Discharges from Chesapeake Bay Catchments

• Published in: Journal of environmental quality 2008-01, Vol.37 (1), p.79-89
• Main Authors: Boomer, K.B, Weller, D.E, Jordan, T.E
• Format: Article
• Language: English
• Subjects: algorithms; Basins; Catchments; Estimates; estimation; Flow velocity; forests; Geographic information systems; Geologic Sediments; hydrologic models; Mid-Atlantic Region; Models, Theoretical; prediction; Rain; regression analysis; sediment deposition; Sediment discharge; Sediment transport; Sediment yield; Soil; Soil erosion; soil erosion models; Stream degradation; Stream discharge; Stream flow; Stream water; Suspended sediments; Universal Soil Loss Equation; Water Movements; Water quality; Water shortages; Water Supply; watershed hydrology; Watersheds
• ISSN: 0047-2425; 1537-2537; 1537-2537
• DOI: 10.2134/jeq2007.0094

The Universal Soil Loss Equation (USLE) and its derivatives are widely used for identifying watersheds with a high potential for degrading stream water quality. We compared sediment yields estimated from regional application of the USLE, the automated revised RUSLE2, and five sediment delivery ratio algorithms to measured annual average sediment delivery in 78 catchments of the Chesapeake Bay watershed. We did the same comparisons for another 23 catchments monitored by the USGS. Predictions exceeded observed sediment yields by more than 100% and were highly correlated with USLE erosion predictions (Pearson r range, 0.73-0.92; p < 0.001). RUSLE2-erosion estimates were highly correlated with USLE estimates (r = 0.87; p < 001), so the method of implementing the USLE model did not change the results. In ranked comparisons between observed and predicted sediment yields, the models failed to identify catchments with higher yields (r range, -0.28-0.00; p > 0.14). In a multiple regression analysis, soil erodibility, log (stream flow), basin shape (topographic relief ratio), the square-root transformed proportion of forest, and occurrence in the Appalachian Plateau province explained 55% of the observed variance in measured suspended sediment loads, but the model performed poorly (r2 = 0.06) at predicting loads in the 23 USGS watersheds not used in fitting the model. The use of USLE or multiple regression models to predict sediment yields is not advisable despite their present widespread application. Integrated watershed models based on the USLE may also be unsuitable for making management decisions.