School of Botany - Theses
Permanent URI for this collection
Search Results
1 - 1 of 1
-
ItemNovel methods to account for individual heterogeneity in capture-recapture studies( 2015)A major challenge in ecological analyses is estimating population-level parameters such as survival, births and population size when most individuals are not observed by sampling. Capture-mark-recapture (CMR) models provide the ability to understand what proportion of the population is missing in sampling and to account for this missingness in inference. Yet one of the standard assumptions behind CMR analyses is that the probability of sampling an individual is either constant across all individuals or can be modelled as a function of fully-observed covariates. However, in many cases this is not true, a factor that can seriously bias estimates of population-level demographic processes. This thesis develops novel methods to account for individual heterogeneity in capture probabilities. I first provide a general review of CMR studies and history of their development, as well as the application of Bayesian state-space models to CMR studies. I discuss the motivating sampling scenario in which a number of common sources of bias are present, then describe how CMR sampling approaches were deployed to try and account for these biases. In chapter two, I propose a solution to the problem of bias due to temporary migration. I describe a capture-recapture sampling scenario on a population that is closed to births and deaths but in which capture probabilities are confounded by migration, then develop a model to account for migration using auxiliary radio telemetry data. The telemetry data provide a means to estimate migration rates, which can then be used to account for the bias in estimated capture probabilities. Simulation studies show that this approach allows for unbiased estimates of population sizes and should be applicable to a range of situations. Chapter three then considers the problem of state misclassification in data where mortality of animals can be inferred via remote observations of movement patterns. I show that in some cases, misclassification of individuals as dead can result in biased estimates of survival rates. I employ a state misclassification model in order to correct for such errors and show that in wild populations of native fish such bias can result in a reduction in estimated survival rates of up to 50%. Next I describe a means of estimating age-specific capture and survival rates in CMR scenarios involving animals with well-de ned growth patterns. The approach involves estimating growth parameters from length interval data in a capture-recapture context or from other ageing methods, then using these growth as prior information in a CMR model in order to estimate age at first capture. We show how otolith and CMR-based estimates of age correspond well, and simulation studies show how estimate survival rates closely match true values. Finally, I consider a large-scale sampling scenario in which multiple sources of bias are possible and multiple sources of data are available to help account for these errors. I develop a state-space modelling approach to incorporating multiple sources of data in this context. I show how it is possible to evaluate the relative fit of models with differing types of data on the same population through a combination of root-mean-square error evaluation, simulation modelling and comparison of the magnitude of various model parameters. I show how single sources of CMR data in this context are likely influenced by various kinds of individual heterogeneity, but show through simulations how it is possible to help correct for this source of bias by incorporating multiple data sources. I also show how incorporating latent state data can help correct parameter estimates that would otherwise be heavily biased in a model that otherwise appears to have a relatively good fit.