Construction of tree growth model based on spatial autocorrelation

We established a permanent 100 m×100 m plot located at Tano secondary forest in Miyazaki prefecture, southwestern Japan between 2004 and 2008. Within this plot, we recorded all stems with the girth at breast height (GBH) no less than 15 cm, mapped the positions of corresponding stem bases, and identified all tree species with recorded GBH. We focused on studying one dominant tree species, Distylium racemosum. A Bayesian statistical approach was applied to quantify a spatially autocorrelated random effect. The effect of competition on individual growth was also considered. We compared the model including the spatially autocorrelated random effect with that did not. The results showed that competition, especially symmetrical competition, played a very important role in affecting individual growth. When spatially autocorrelated random effect was included, compared with the model that did not include the random effect (R2=0.74), the model accounted for significant proportions of the variation (R2=0.83). We also analyzed other tree species in the plot, and high correlation was consistently found when spatial autocorrelation was included. The Bayesian approach used in this study, including the intrinsic CAR model, is a powerful tool that can obtain important ecological information from forest census data, and provide theories and methods for future researches on simulation of individual tree growth.