WSP12246 - Laserfiche WebLink

Home Browse Search

~4GJ The null hypothesis is that there is no trend, or the value of coefficient b equals zero. If there is a sig- nificant trend, the value of b is the magnitude of the trend. Use of equation 3 does not adjust the annual or seasonal loads for the effects of streamflow. To remove the variation in annual or seasonal loads caused by streamflow, the same flow-adjustment method that was used for the seasonal Kendall test also was used for the regression method. Flow-adjusted linear regression was done in a two-step method. First, the loads were regressed against streamflow using hyperbolic func- tions, which had the highest coefficients of determina- tion when compared to other flow-adjustment models. Residuals from the flow-adjustment models were tested for normality and constant variance to verify that assumptions for using the parametric method were not violated. Second, the residuals from the flow- adjustment models, or the flow-adjusted loads, were regressed against time using equation 3. The null hypothesis is that there is no trend in the flow-adjusted loads or that the value of coefficient b in equation 3 is not significantly different from zero. Regression trend tests were done using procedures in SAS (SAS Institute, 1982). Step-Trend Analysis Step-trend analysis can be used to determine if there is a difference in population means or medians between two or more sets of data. Parametric or non- parametric methods can be used. The parametric test for step trends is the two-sample t-test (lman and Conover, ] 983). When using the !-lest, it is assumed that the data sets are normally distributed about their mean values. The t-test determines if there is a signif- icant difference between the means of two data sets. Parametric tests have diminished power to detect true differences in mean values when applied to data that are not normally distributed. A commonly used nonparametric test for step trends is the Wilcoxon rank-sum test. That test is computed using a two- sample I-test applied to the ranks of the data instead of using the original data. The Wilcoxon rank-sum test has no assumptions concerning data distributions. The rank-sum test is used to test for the difference in medi- ans between two data sets. The step-trend tests were done using procedures in SAS (SAS Institute, 1982). For tests involving com- parisons among three or more data sets, multiple t-tests were used on the original data (parametric method) and on the ranks of the data (non parametric method). Repeated t-tests were done because they are more applicable to unequal sample sizes than are other tests, such as the Duncan multiple range test (SAS Institute, 1982). TREND ANALYSIS FOR THE COLORADO AND GUNNISON RIVERS Monotonic trends in dissolved-solids and major- ion data at three sites were investigated to examine possible effects on salinity in the Colorado River from salinity-control projects in the Grand Valley (the Grand Valley Unit) and in the lower Gunnison River Basin (the Lower Gunnison Basin Unit). Gaging station 09163500 (fig. I) on the Colorado River near the Colorado-Utah State line is the outflow site and is downstream from the Grand Valley and the Gunnison River. It was not sufficient to evaluate trends only at station 09163500 because trends at that station might have been induced by salinity trends in the Colorado River upstream from the Grand Valley or by trends in the Gunnison River. Based on data in table I, about 83 percent of the annual mean dissolved-solids load at station 09163500 for water years 1970-93 was accounted for at inflow gaging stations 09095500 Colorado River near Cameo and 09152500 Gunnison River near Grand Junction (fig. I). Therefore, stations 09095500 and 09152500 were included in the trend analysis. Plateau Creek (station 09105000) accounted for only about 2 percent of the annual mean dissolved-solids load at station 09163500 during 1970-93 (table I). Trends in Plateau Creek were unlikely to have an effect on trends in the Colorado River; therefore, trend analysis was not done for station 09105000. The period of record for the trend analysis was water years 1970-93 (October I 969-September 1993). Most major water-storage projects in the Colorado River Basin upstream from the Grand Valley were completed by ] 965. Reservoir construction during a time period being analyzed for water-quality trends would complicate the flow-adjustment trend analysis or render it unfeasible. Although there are dissolved- solids data for ]966-93, the trend tests were done on data from water years 1970-93 so as to have concurrent records with periodic major-ion data for the three stations. Two shorter periods also were examined for trends. Water years 1980-93 were TREND ANALYSIS FOR THE COLORADO AND GUNNISON RIVERS 13