7930 - Laserfiche WebLink

Home Browse Search

best estimate of actual sediment yield from the watersheds. RESULTS AND DISCUSSION Analysis of the data from the Boise and Clear- water National Forests resulted in 87 suspended, 83 bedload, and 81 total sediment rating equations (Qs = aQb). Thirty-eight equations of each sediment type were derived from the Silver Creek data. Rating Equation Success The number of equations that were statistically significant at the 95 percent level of confidence var- ied by sediment type (table 1). An arbitrary criterion was used to establish "useful" equations. Equations were used to predict sediment yield if the explana- tory (independent) variable, Q, explained at least 60 percent of the variation in the response (dependent) variable, Q. (that is, R2--0.60). The better controlled sampling conditions in Silver Creek should have provided data that define the sediment-discharge relation more accurately. However, the variability of sediment discharge was no better defined by the regressions on streamflow in Silver Creek than at the natural cross-sections on the two Forests (table 1). Significant relationships were much more evident for bedload sediment than for suspended sediment. This supported the observations of Shen (1972), Holeman (1975), and Rannie (1977) that suspended sediment discharge on large rivers was more depen- dent on watershed properties and perturbations than on streamflow, and that bedload should be more closely correlated with streamflow. Only 11 percent of the suspended sediment equations but half of the bedload equations had R2>0.60. Total sediment load rating equations fared much better on the National Forests than in Silver Creek; 78 percent versus 47 percent had R2>0.60. The research watersheds were typically smaller than the monitored watersheds on the Forests, but no correlation was found between watershed size and rating equation success. All suspended sediment regressions in this paper used sediment concentration in milligrams per liter (mg/L) and streamflow in cubic feet per second (ft3/s). Sometimes suspended sediment is reported as a rate, such as kilograms per hour (kg/h). If this rate is regressed against streamflow, the coefficient of determination will be considerably higher than if the same data in milligrams per liter were used. The result, however, is a spurious self-correlation because the conversion of concentration, milligrams per liter, to a rate in kilograms per hour involves streamflow (Kenney 1982). The regression using kilograms per hour is really aQ vs Q. The self-correlation of Q vs Q does not alter the sediment concentration predicted using streamflow. But a cause-and-effect relationship between the two, with properties of the resulting equation, cannot be claimed. The rating equations for total sediment were also plagued by spurious self-correlation because sus- pended sediment concentration must be converted to a rate before being added to the bedload component. This problem is unavoidable with current sampling devices that measure suspended sediment as a con- centration and bedload as a rate. Several factors may contribute to the low success rate for the rating equations. First, the sample size was small, typically 15 or less. With small samples, the table t value becomes large due to the small degrees of freedom. This makes it more difficult to show statistical significance. A related problem may be that the 10 to 15 samples did not adequately sample the hydrograph to define a good sediment- discharge relationship. The hysteresis effect may be inherent in the data, even though there are not enough data points to analyze for these effects. The presence of hysteresis merely adds to the variance due to concentration. Various levels of management activities existed prior to and during the period of monitoring on most of the Forest watersheds. This could continu- ally vary the sediment supply to the streams such that a good sediment-discharge relationship for a given year may not exist. In the research area, how- ever, management activities were controlled. Most of the drainages were undisturbed for 2 or 3 years and then impacted with specified levels of management. The introduction of disturbances in these small watersheds did not have an adverse effect on the rating equation success. The rating equation success increased slightly after management activities occurred in the drainages. Because these drainages Table 1-Statistical significance of sediment rating equations in the Idaho batholith Sediment Location type' No. rating equations No. significant rating equations2 (%) No. significant rating equations w/R2>0.60 (%) National Forests3 SS 87 38 (44) 10 (11) BL 83 59 (71) 47 (57) Tot 81 71 (88) 63 (78) Silver Creek SS 38 18 (47) 4 (11) BL 38 27 (71) 16 (42) Tot 38 30 (79) 18 (47) 'Type of sediment: SS = suspended; BL = bedload; Tot = total sediment. 2Significant rating equations are those with coefficients that are significant at the 95 percent level of confidence 3Data from granitic watersheds on the Clearwater and Boise National Forests, Idaho.