Laserfiche WebLink
declined from over 3,300 fish in 2001 to about 2,142 individuals in 2003, an apparent reduction <br />of 35%. Non-overlapping confidence limits for the years 2001 and 2003 for basin-wide Green <br />River abundance estimates also suggested a significant decline in abundance. An even greater <br />reduction of 48% was observed if one adds lower Green and Desolation Gray Canyon estimates <br />from 2001 (as a measure of their year 2000 abundance), to the middle Green River estimate from <br />2000. Among individual reach estimates, confidence limits among Green River Basin abundance <br />estimates from 2001 to 2003 generally overlapped with three exceptions. Those were for the <br />largest populations, the middle Green River, when confidence limits for the relatively low 2002 <br />and 2003 estimates did not overlap with confidence limits for the higher 2000 abundance <br />estimate, and for the White River when the 2000 estimate was higher and the confidence limits <br />did not overlap with the 2003 estimate. Regression of loge abundance estimates of Colorado <br />pikeminnow adults as a function of time suggested strong negative relationships over time for the <br />middle Green River and White River, and marginally significant and negative relationships over <br />time for the Yampa River and the lower Green River. The adult Colorado pikeminnow <br />population in the Desolation-Gray Canyon reach of the Green River showed only a slight and <br />non-significant reduction in abundance over time. Below we discuss the abundance estimation <br />model, model assumptions, implications of parameter estimates, and potential reasons for the <br />apparent decline of Colorado pikeminnow abundance in the Green River Basin during the study <br />period. <br />Model selection and assumptions.--We explored a series of increasingly complex models <br />that led us to select the Huggins robust-design multi-strata model as the most realistic one for <br />estimation. The sampling design (three or four primary annual sampling periods with three <br />closely-spaced secondary sampling occasions in each year) led logically to a robust-design <br />39