Laserfiche WebLink
<br />~ <br />~ <br />~. <br />~ <br /> <br /> <br />The accuracy of regression estimates from base-station data was tested <br />for each of the five streamflow record-extension sites. The tests were made <br />using split-sample analysis, in which the daily data for a site Were divided <br />into two approximately equal periods based 'on the number of water years in the <br />overall period of analysis. The later years were used to calibrate a regres- <br />sion model for each site. The calibrated models were then used with data from <br />the earlier years to verify the accuracy of the model predictions. Accuracy <br />was measured by the standard error of predicted streamflow, defined as the <br />square root of the mean squared deviation between the observed and predicted <br />values. The period of analysis, mean streamflows, standard deviations, and <br />standard errors are listed in table 3. The overall period of analysis used <br />for a site generally was the concurrent period of record for the record- <br />extension and base-station sites. However, for sites that were affected by <br />completion of upstream reservoirs during the period of record, the period of <br />analysis was restricted to the time prior to regulation of streamflow. This <br />restriction was applied for record-extension site 12, Duchesne River near <br />Randlett, Utah, which was affected by Starvation Reservoir beginning in water <br />year 1970, and for record-extension site 16, San Rafael River near Green <br />River, Utah, which was affected by Joes Valley Reservoir beginning in water <br />year 1966. <br /> <br />. <br /> <br />Standard errors of prediction ranged from 9 to 116 percent of the <br />observed mean daily streamflow. When the daily estimates were added to yield <br />monthly valuea, the errors decreased to 5 to 74 percent. The site with the <br />largest percentage error also had the amallest mean streamflow, and its <br />absolute error was small in comparison with the error at the other sites. <br /> <br />EXTENSION OF DISSOLVED-SOLIDS RECORDS <br /> <br />The data used to extend records of dissolved-solids concentration <br />primarily were the monthly values of streamflow and flow-weighted dissolved- <br />solids concentration computed for the 12th biennial report (U.S. Department of <br />the Interior, 1985). The procedure used to compute flow-weighted monthly <br />dissolved-solids concentration is described by Liebermann and others (1987). <br />Total monthly streamflow volume and dissolved-solids discharge (mass) were <br />determined for each month of the period of record. Then the flow-weighted <br />dissolved-solids concentration was computed as the ratio of dissolved-solids <br />discharge to streamflow. <br /> <br />Three methods were used to estimate dissolved-solids concentration during <br />the extension period. Weighted regression on streamflow at the record- <br />extension site was used for sites that had a complete streamflow record during <br />the extension period. If streamflow data were not available for the record- <br />extension site, dissolved-solids concentrations were estimated by weighted <br />regression on streamflow at a base-station site. Weighted-regression was used <br />to decrease monthly differences in the variability of the residuals. <br />(Weighted regression should not be confused with the flow-weighting procedure <br />used to compute the historical dissolved-solids concentrations.) For several <br />record-extension sites, water-quality data were available at a suitable base- <br />station site. Dissolved-solids concentrations at these record-extension sites <br />were estimated by linear regression on dissolved-solids concentration at the <br />base-station site. In all three methods, flow-weighted concentration was the <br /> <br />6 <br />