Study Of GIS And ANN Based Individual Tree Growth Model

A set of spatial-temporal correlation individual tree growth model for Masson pine in plantation, was developed by ANN and integrated with GIS. The study adopted the theory of biogeographically statistics, and biological significance of these modes was considered. Data for the study was analysis stem data and management density data, and the data was from the fixed and temporary sample plots. The first step of the study was to reconstruct the information of deceased tree by servitor tree in near mature plots. The deceased tree information was about diameter, height, volume, location, density and growth (increment) process.The second step is spatial correlation analysis of these reconstructed dada. Through the validation statistics and related function analysis of deceased tree and survival tree, to compute the autocovariance distance between objects tree and competitors for its neighboring trees. The definition of autocorariance distance, independent distance and significantly correlated distance between object-tree and its competitors were given in chapter 4 of this thesis.The third step (chapter 5) developed three kinds of new competition indices of individual tree: (1) Crown Area Index, a kind of distance-independent competition index, could be obtained from RS image;(2) TGP and TVP, a kind of indirect distance-dependent competition index, based on Thiessen Polygon Method;(3) BDR, BBR, BDDR and BDBR, distance-dependent competition indices, were constructed by the correlated distance that computed in chapter 4 between object tree and its competitor. The parameters of second and third competition indices were estimated by the spatial function of GIS. The correlation coefficient of logarithm regression showed a significant correlation between these indices and diameter, height and volume value. It was concluded that those indices are suitable for diameter, height and volume individual tree growth competitor indices. After comparative study, we found TGP, BBR and BDBR had better fitting effect, and their computed parameters are easier to capture.The last step of the study was to develop individual tree growth models based on the competitor indices of CA, TGP, BBR and BDBR, which were constructed in chapter 5. The method for modeling was BP neural network modeling technology, and the model structure was 5:3:1. A full discussion of model development and model structure was presented in chapter 6. The results of training and simulation test showed these models had good performance in prediction of growth, and the best predicting period is 6 year.