Estimating population trend by integrating point counts, removal sampling, and mark-recapture observations in a single hierarchical model

Background/Question/Methods

To estimate population abundance and trend ecologists must often confront two interrelated challenges, (1) imperfect detection of individuals limiting inference from counts and (2) inconsistent survey protocols over time-spans when population trends may be of interest and detectable. I demonstrate that a conditional multinomial N-mixture models can estimate abundance while accounting for imperfect detection by integrating observations provided by point-counts, removal sampling, and mark recapture surveys that also varied in effort and the numbers of replicates or capture sessions. I demonstrate the application of such a model using captures of Sonoyta pupfish (Cyprinodon eremus) a highly endangered species whose monitoring has spanned five decades with sampling protocols changing multiple times.

Results/Conclusions

Preliminary results show detection probability of pupfish varied from 0.026 to 0.477 across surveys while total capture probability varied from 0.012 to 0.886 and was highest in removal sampling consisting of 5 or more removal trapping sessions with 60 traps. Total capture probability was lowest in point-count replicates with 20 or fewer traps. Estimated pupfish abundance varied from a low of 1222 to a high of 8651 a log-linear trend of r = 0.019 suggested pupfish abundance has been stable over the previous 50 years (Bayesian p-value = 0.105). Integrating data from multiple sampling protocols into hierarchical models can leverage monitoring data spanning time periods when trends may become detectable and may be applicable to many monitoring programs, providing new opportunities to describe and explain population dynamics.