School of Botany - Theses

Permanent URI for this collection

http://hdl.handle.net/11343/307

Filters

2012 1
Reset filters

Settings
Statistics
Citations
Author: Morris, William Keith × End date: 2015 × Start date: 2010 × Type: Masters Research thesis ×

Search Results

Now showing 1 - 1 of 1
  • Item
    Thumbnail Image
    Bayesian modelling of tree mortality
    Morris, William Keith ( 2012)
    In this work I aimed to relate two topics of importance in applied ecology and the statistical analysis of ecological data: the challenge of analysing tree mortality data and the use of informative priors in Bayesian ecological models. To achieve these aims I conducted analyses on two separate case studies of tree mortality. The first study concerned the mortality of transplanted eucalypt tree seedlings during an experiment carried out in Southeastern Australia. For the second case study I used data collected over 20 years in a 50 ha tropical forest dynamics plot located in Northern Thailand. A key challenge when modelling death data is how to account for a hazard that varies through time. Ignoring a variable hazard leads to biased estimates of mortality rate. I identified evidence of a declining hazard rate in the eucalypt seedling data. To account for a potential source of bias in estimating the mortality rate of the seedlings I introduced a novel method whereby the expected mortality is a function of the sum of mortality experienced in the past. Bayesian methods have become common in the ecological research literature. Yet ecologists rarely use informative priors in their Bayesian models—a feature which has been shown to increase the precision of model predictions. This reluctance may in part stem from a perception that informative priors could introduce bias and lead to less accurate models, compared with those that use uninformative priors or fit with non-Bayesian methods. To date there have been no comprehensive tests of the effect of informative priors on the accuracy of Bayesian ecological models. Using the eucalypt seedling dataset I demonstrate that incorporating an informative prior based on the results of a pilot study does not negatively affect the accuracy of the model, while increasing the precision and making cost-effective use of the available information. With the tropical forest dataset I tested the effect on model accuracy of informative priors. In the first comprehensive test of this kind, I demonstrate that a properly specified informative prior, in this case based on the correlation between tree growth and mortality, will on average have no effect on the accuracy of models of mortality rate.