Multistate mark-recapture models provide a flexible framework for estimating life history and movement parameters of many types of marked organisms. For organisms marked with battery powered transmitters, battery life can impact the probability of observation independent of the behavior or life history of the organism. We present a general framework for multistate mark-recapture models that explicitly incorporates information on anticipated transmitter battery life. The model was developed for understanding the movement patterns of invasive bighead carp (Hypophthalmichthys nobilis) and silver carp (H. molitrix) on the Illinois River. This understanding can aid in developing and deploying efficient control measures to prevent further upstream advances of established populations. We first constructed a simplified model with states for live versus deceased fish, each with or without active batteries. We considered live fish with active batteries to be observable and batteries past the expected expiration were treated as a known ‘expired’ state. After testing this simplified model, we expanded it to include monthly movement transitions between six reaches on the Illinois River. We ran the models in a Bayesian framework using the NIMBLE package in R which provides several features for efficient Markov chain Monte Carlo (MCMC) sampling of latent state models.
Results/Conclusions
Fitting a battery life parameter in the model allowed us to include data from fish whose tags were past their expired dates in the model without censoring or making assumptions about whether those fish were still living or not. The simplified model fit to a known (i.e., simulated) dataset was able to successfully recover the originating parameters. The full multistate model fit to a large multi-agency dataset of bighead and silver carp tagged in the Illinois River showed survival estimates similar to previously constructed frequentist models for this system. Reach to reach movement parameter estimates were consistent with observations with most tagged fish only detected within a single reach (~72% for both species). Constructing this model in a Bayesian framework allowed great flexibility to customize the model to the transmitter-receiver system that is currently in place to monitor these species, including the ability to incorporate a battery life parameter. In addition, posterior samples from the model can be used directly in population simulation models developed for assisting management decisions. We plan to continue to refine and test various model configurations to distinguish ones that have the most explanatory power for our data.