One project that Darryl was working on this month was tidying up an analysis of the survival and growth rates of southern beech trees near Arthur's Pass in New Zealand's South Island, and in particular to estimate what impact a major earthquake in 1994 had on those rates.
Trees in study plots from across the region had been measured periodically since 1974, enabling a suitable period of time to establish baseline, or pre-earthquake, survival and growth rates, and what effect a range of factors (e.g., exposure, various soil measurements, etc) had on those factors. The intent of the analysis was to examine how the effect of these factors changed in the 5-year period following the earthquake, and then again 5+ years post-earthquake. The basic hypothesis was that we would see changes in that initial post-earthquake interval, and then things would be more 'normal' later on. Of specific interest was what effect distance from the epicentre had on survival and growth.
The data set provided a number of challenging aspects for the analysis, including the clustered sampling (i.e., plots were sampled from the area, not individual trees), unequal time intervals between surveys of different plots, missing data, and the correlation between tree survival and growth. To overcome this, Darryl developed a joint, hierarchical, tree survival and growth model, analysed within a Bayesian framework. Below is a screenshot of the system resources mid-analysis, with no other software running!
In a nutshell, the results were largely as expected with evidence of a reduction in survival and growth immediately post-earthquake, particularly nearer the epicenter. A paper will be forthcoming in the Journal of Ecology on this study, hopefully in the next few months. We'll be sure to share it when it's published.
