Laserfiche WebLink
<br />STATUS OF GREEN RIVER COLORADO PIKEMINNOW <br /> <br />1361 <br /> <br />was used to estimate survival in year t (S,), the <br />probability of transition between reaches i and j (\/1i)' <br />capture-recapture probapilities within reach i for each <br />year t and sampling pas~ k (p. k)' and the abundance of <br />, It <br />adult and recruit Coloraqo pikeminnow in each reach i <br />for each year t (N ). Abtfudance was estimated with the <br />It ! <br />Huggins estimator (San~thanan 1972; Huggins 1989, <br />1991; Alho 1990). A~undance estimates from the <br />Huggins model were detived by the equation <br />I <br />I <br />h1t+1 <br />N = t(l/p7), <br />!i=l <br />I <br />where M'+l is the num~r of unique animals captured <br />over all short-term samp~ing passes and <br />i <br />I , <br />p7 = 1 t D (1- Pj;), <br /> <br />where Pji is the probabili~y of initial capture within the <br />sampling season. I <br />Animals in the populJtion that were never captured <br />have capture probability 1(1 - p) but are removed from <br />the likelihood function, The new multinomial distribu- <br />tion still sums to one, add because only fish that were <br />captured are included, iJdividual covariates (here, TL <br />or polynomials for such~were incorporated to estimate <br />p, "', and S, where appr priate. Information for the p* <br />estimates was taken fr I m both the closed-captures <br />portion of the likelihoodl (used for abundance estima- <br />tion) and the Cormack-Jolly-Seber (CJS) component <br />of the model (used to ~stimate annual survival rates <br />across all reaches.) With the information about p* from <br />the CJS portion of the lik~lihood, the individual p' s per <br />pass within the annual s4nPling period are identifiable <br />based on the numbers o~ fish initially captured during <br />each sampling pass within a year. Thus, we could <br />estimate abundance for \ river reaches and years in <br />which no fish were recrptured between passes in a <br />given year. In additiop to the across-years CJS <br />contribution to p*, redaptures of fish in reaches <br />between passes within ~ single year provided more <br />efficient estimates of abldance. Riverwide abundance <br />estimates were obtained y summing the separate reach <br />estimates by year. Sta dard errors for riverwide <br />I <br />estimates were obtained from the variance-covariance <br />matrix of the likelihood from program MARK. <br />The covariate TL (l~ngth at first capture) was <br />standardized with a z trJsformation, namely, <br />TLNi = (d - Lmean)/SD, <br /> <br />where TLN. is the transfobed length of fish i, L is the <br />TL of fish i, L is thd mean TL for all fish \n the <br />mean I <br />J>OP"btinn, ~d SD;,." rp""tinn """""d d~;,tinn <br /> <br />of TL. The z transformation was used because the <br />numerically smaller covariates avoid the numerical <br />optimization difficulties encountered when covariates <br />have large values or a wide range of values. Because <br />adult Colorado pikeminnow grow very slowly (Os- <br />mundson et a1. 1997), use of length at first capture as <br />the covariate was deemed appropriate for fish that may <br />be captured several times over the study period. <br />Inclusion of the covariate TL in the modeling was <br />important because of the potential effects of fish size <br />on the probability of capture (Reynolds 1983). <br />Abundance estimators such as those in program <br />CAPTURE (White et a1. 1982) do not have the <br />capability to use individual covariates because the <br />estimating likelihood includes probabilities for animals <br />that are never captured, so the covariates are unknown. <br />Selection between models was performed with information- <br />theoretic procedures (i.e., Akaike's information crite- <br />rion adjusted for small sample size [AlCc] , after <br />Bumham and Anderson 1998). Differences among <br />pairs of point estimates were deemed significant (P ::; <br />0.05) if the confidence intervals did not overlap, <br />although we recognize that this method sometimes <br />provides conservative results (i.e., fewer significant <br />tests; Schenker and Gentleman 2001). <br />Survival and finite population rate-of-change mod- <br />els.-Jolly-Seber-type models (recaptures only and <br />Pradel's survival and population rate-of-change mod- <br />els) were used in program MARK (Cormack 1964; <br />Jolly 1965; Seber 1965; Pradel 1996; White and <br />Burnham 1999) to estimate apparent survival (1991- <br />1999) and population rates of change (A; 1991-2003), <br />respectively, for Colorado pikeminnow captured in the <br />Green River basin. The goal of the survival analyses <br />was to determine whether a composite survival rate <br />from 1991 to 1999 was different from survival from <br />2000 to 2003. <br />We used Pradel's model (Pradel 1996) to estimate A, <br />that is, <br /> <br />Ai = Ni+dNi = <Pi + Ii, <br /> <br />where N is population size at time i or i + 1, <Pi is the <br />survival rate, and!; is the number of fish recruited to <br />the population at time i per adult in the population at <br />time i. A value of A less than 1 indicates a declining <br />population, a value greater than 1 an increasing <br />population, and a value of exactly 1 a stable <br />population. Functionally, the Pradel model is similar <br />to the Jolly-Seber model used to predict survival rates <br />but uses the capture history in reverse order to predict <br />the probability of entering the population (Nichols et <br />a1. 2000). A main assumption of Pradel's model is that <br />the size of the study area is constant over the sampling <br />period, which precluded the use of capture data from <br />