Capture-recapture modelling of population dynamics

点击翻译按钮获取中文摘要

This research involves the use of statistics in studying the dynamics of animal populations and other ecological problems. The research is motivated by real problems encountered by ecologists. Much of the research involves capture-recapture experiments. In these experiments, animals are captured, tagged, released and then subsequently recaptured or resighted. The pattern of captures and recaptures/resightings enables the researcher to estimate survival rates, the number of animals present in the population and other population parameters.The research aims to develop new statistical methodology to help answer questions such as "Is there evidence of compensatory mortality from predation of smolts in the Columbia River Basin." In such cases, many thousands of smolts are tagged with PIT (passive integrated transponder) tags with detectors on many dams that the smolts encounter during their downstream migration and their return as adults. Avian predation has been steadily increasing over the last 20 years with the expansion of tern and cormorant colonies in the basin. Predation pressure also varies within a year. The tags from some predated smolts can be detected on the avian colonies as a measure of avian predation (after adjusting for deposition and detection efficiency on the colonies). This makes a natural experiment to investigate the relationship between mortality due to avian predators and all other sources of mortality. If there is no relationship, then predator controls may have little impact on subsequent returns of adult fish and other methods may be needed to help restore endangered stocks. Standard statistical methods applied to this type of data are inadequate to to distinguish compensatory mortality from heterogeneity in survival among subgroups, artefacts that look like compensatory mortality such as sampling correlation, and biases introduced from competing risks. Complex Bayesian models are needed to disentangle these effects. Furthermore, the large dataset (millions of smolts have been marked over the last 20 years) introduces computational challenges in using fitting a Bayesian model to the data. Opportunities are also present to modify where detector arrays are installed which may lead to dramatic improvements in efficiency to detect compensatory effects -- just how big of an improvement can be expected and is it cost effective compared to simply increasing tagging efforts?The statistical methods methods developed are broadly applicable in many wildlife management situations where there are different sources of mortality and a choice of management actions to increase survival at different parts of the life cycle. Is there evidence that the management action would have any impact? As what stage in the lifecycle is the best place to implement such actions?