Robust models for forest growth and yield predictions under climate change

Climate change poses an enormous challenge to those modelling forest ecosystems: the unprecedented nature of projected future climatic conditions challenges the fundamental assumption in empirical modelling that the past is a good predictor of the future. Instead, models must be selected and fit using existing data, while predictions are required in the context of an uncertain, but different, future. Although there have been many calls for climate sensitive models in forestry literature, relatively little examination of the statistical design and performance of these models has occurred in the context of the challenges imposed by climate change. This dissertation first presents a comprehensive examination of the requirements for effective climate sensitive models, especially noting the lack of appropriate statistical techniques for model selection and fitting in forestry and ecology literature. Two potential solutions are proposed: the use of statistical techniques more suited to prediction in a setting of extrapolation, and the use of a wider selection of data sources. The example of height increment in spruce (Picea glauca, Picea englemannii and hybrids thereof) is used to examine the efficacy of a number of potential model fitting and selection methods, with cross validation and regularization identified as key components of effective methods. A novel combination of forest inventory and provenance trial data is used for this analysis, facilitated through the use of methods new to forestry literature such as the inverse hyperbolic sine transformation. The effect of this novel dataset is further examined in a sensitivity analysis, using both spruce height increment and mortality models. The combination of data types is found to result in more effective predictions by allowing models to account for both climatic and genetic effects. Overall, a collection of methods for model fitting, selection, and integration of disparate data sources provides guidance to forest modellers facing the challenges of climate change.