7075 - Laserfiche WebLink

Home Browse Search

Bias Correction of Deerlodge Park and Mathers Hole Rating Curves Both O'Brien and Elliott developed curves relating mean daily discharge to sediment load for various fractions of sediment, including suspended, bedload, and total load. Andrews (1986, 1979, 1978) also developed suspended sediment curves and other curves for the Green and Yampa River stations using the same methods. All of these original rating curves used least squares linear regression of log transformed values resulting in power curves of the form: load - aqb. Least squares regressions using log transformations typically underestimate sediment load and the degree of underestimation increases with the degree of scatter about the rating curve (Ferguson, 1986). Both Ferguson (1986) and Miller (1984) provide bias correction factors to improve rating curves developed in this manner. Ferguson's correction is based on transformations using base 10 logarithms, while Miller's is for natural logarithm transformations. Both methods require the actual data in order to determine the variance needed to calculate the correction factor. Elliott's report (1984) included as supplemental data the individual measurements for all of his rating curves. These were utilized to calculate the correction factor which is applied to the intercept ('a' coefficient) of the original power curve. Using each method, the same power curve for total sediment load at Deerlodge Park was obtained. Tables 3 and 4 show the corrections using base 10 (Ferguson) and base a logarithms (Miller), respectively. Koch and Smillie (1986) commented on Ferguson's paper (1986) and compared observed sample means to uncorrected and corrected mean annual sediment loads for the Little Snake near Lily and Yampa River near Maybell stations. They showed the uncorrected curve underestimated the annual loads while the corrected curve, using the method discussed by Ferguson, overestimated the annual load. Application of a nonparametric smearing estimate resulted in even greater overestimation. Their main conclusion was that application of a bias correction based on the assumption of normally distributed residuals may not be valid in all situations using rating curves to predict loads. Ferguson (1986) responded to Koch and Smillie's comments by stating that the inability of the nonparametric correction factor to produce better estimates suggests that the problem may not be a violation of the assumption of normally distributed residuals, but perhaps the violation of two other assumptions. The two possibilities mentioned were the true log-log rating curve may not be linear and the scatter may not have the same variance at all discharges. In discussing this matter with Robert Milhous (pers. comm. 3/7/88) an alternate form of the power curve may provide a better fit of a gravel bed river where the sediment load is strongly influenced by a critical discharge associated with the size of the material on the surface of the armour layer. The form is: load = a(Q - Qc)b, where Qc is the critical discharge.