Significance of rainfall distribution in predicting eucalypt plantation growth, management options, and risk assessment using the process-based model CABALA

In the past, criteria for site selection for plantations of Eucalyptus globulus have focused on the expected yield predicted using long-term average climate data. When selecting sites using predictive models of plantation growth, scant consideration has been given to climatic variation and how this affects the probability that droughts will reduce yield or cause death of trees. This is often because either the time step of the models does not provide the required resolution, or because climatic data of the required resolution is not available. In particular, rainfall patterns are often represented simplistically. In this work we generate daily rainfall patterns with different methods (observed sequences, naive daily distribution of monthly rainfall totals, and simulation of daily rainfall using a Markov rainfall intensity and dry spell simulator) and use these as input sequences into the process-based tree-growth model CABALA [For. Ecol. Manage., this volume]. We examine how rainfall representation influences prediction of growth and the development of water stress in modelled plantations and compare predictions of growth, pre-dawn soil-water potential and aspects of nitrogen uptake and utilisation with observations made over a rainfall gradient in southwest Western Australia. The naive rainfall representation led to biased predictions, failed to represent the soil-water recharge benefit of high intensity rainfall events, under predicted tree drought stress, and failed to adequately represent nitrogen leaching and conditions for nitrogen mineralisation in plantation soils. Furthermore, it was clear from the distribution of plantation yields that were predicted using observed weather sequences that appropriate evaluation of plantation risk requires the representation of rainfall in a way that captures inter-annual variability in rainfall amount. We concluded that if we are to obtain maximum benefit from the application of process-based models it is necessary that input variables such as rainfall distribution are adequately represented.