It is a pleasure to write that I am now, nearly, degreed.
My PhD was recently accepted into the Library at The University of Melbourne. I thought I would mark the occasion with short summary of the beast.
Plant height and growth are fundamental to the understanding of species’ ecological strategies, to the description and prediction of ecosystem dynamics and to vegetation management. I explored how plant functional traits can be used to predict woody plant growth for many species. I demonstrated internal and out-of-sample prediction of species growth trajectories from traits, I dissected methods to evaluate the predictive capacity of growth models and I outlined a virtual ecologist approach to designing robust field studies for complex analysis.
Chapter one incorporates plant functional traits into multi-species hierarchical non-linear models of plant growth. This approach increases our understanding of trait-growth relationships but also aids our ability to draw predictive inferences from them. I built and parameterized models with a case-study of time since fire in semi-arid mallee woodland. I demonstrated inference by predicting species height-growth trajectories from traits to species with few data, to species with no growth data, only trait information, and for hypothetical species with defined trait combinations.
Chapter two contributed to the growth modeling literature by focusing on evaluating the predictive capacity of non-linear growth models using cross-validation. I demonstrated why cross-validation is important compared to naïve performance metrics and demonstrated the value of using multiple metrics to capture different aspects of model performance.
Gaining greater predictive capacity in trait-based ecology may also require stronger quantitative tests of model transferability, which is a severe test of how general a model actually is. In chapter three I tested the out-of-sample predictive ability or transferability of my trait-growth models by using traits to predict the growth trajectories between species in three different ecosystems.
The worth of predictions and inference from data analysis is intimately linked to the statistical and ecological assumptions of fitted models and fundamentally to the data underlying the analysis. My final chapter demonstrated a virtual ecologist approach to aid in the design of studies that use complex analysis techniques. I used a simulation based on realistic fieldwork constraints such as species detection and occupancy rates, as well as travel times and unpredictable field conditions. This assists planning how much data is needed and how long that data will take to collect for hierarchical multi-species nonlinear models.
In the unlikely event someone would like to read my thesis, get in touch and I will send you the open access link.