Back-propagation (BP) algorithm of artificial neural network (ANN) was applied to tree height prediction of Larch plantation in northeast China by taking logsigmoid function of logsig and linear function of purelin in Matlab as the neural functions. One input variable of tree diameter and one output variable of tree height was used in the model with one hidden layer of 5 hidden neurons. Model developed was evaluated graphically and statistically. Results showed that model performs well with mean square error (MSE) of 0.130901 and model precision of 97.6%. The graphical comparisons between the actual measured data and the network predicted output clearly demonstrate very good agreement between the actual and predicted performance.

Water content of Korean pine samples was studied using near infrared spectrascopy (NIR) combinded with parial least squares regression (PLS) analysis. Seven typical wave lengths were selceted to characterize the wood water content of Korean pine. Prediction models of wood water content were developed based on the original NIR spectra at seven varied wave lengths. The best model was showed between wavelengths from 1000 to 2100 nm with correlation coefficient (R) of 0.9871, standard error of calibration (SEC) and root mean square error of calibration (RMSEC) of 0.0337 and 0.0334, respectively. The correlation coefficient（R） for model validation was 0.9817 with standard error of prediction (SEP) and root mean square error of prediction (RMSEP) of 0.0470 and 0.0465, respectively. This study indicated that NIR is a useful tool for fast prediction of wood water content of Korean pine. The modeling proficiency could be improved significantly for some certain wave length. It could be a research focus for future related studies.

Model for predicting wood density of Larch was established using near-infrared spectroscopy (NIR) combined with support vector machine (SVM). A hundred and seventeen Larch samples were used in the study. Wood density of samples was measured according to standard test methods for physical and mechanical properties of wood. Support vector machines for regression (SVR) was used for model building. Radial basis function (RBF) was used as kernel function to establish a model for predicting wood density. For the train set, the coefficient of determination (R2) and the mean square error (MSE) were 0.8504 and 0.6460×10-3, while the R2 and MSE was 0.8520 and 0.4451×10-3, respectively, for the test set. Results showed that using SVM in near-infrared spectroscopy calibration could significantly improve the model performance in order to rapidly and accurately predict wood density.

In this paper, an integration of BP neural network and PCA for modeling wood water content of larch combined with NIRS was investigated. The original spectra were collected and pretreated with 9 point smoothing and multiplicative scatter correction (MSC). Five typical principal components were extracted from PCA with the application of establishing prediction model. Full cross-validation approach was applied to achieve desirable modeling performance. The prediction correlation coefficient (R) was 0.952 while the mean square error of prediction (MSEP) was 38.27. This study indicated that NIR is a useful tool for rapid and accurate prediction of wood water content.

In this study, the sample data was based on 2190 branch length samples of 30 trees from dahurian larch (Larix gmeliniiRupr.) plantations located in Wuying forest bureau in Heilongjiang Province. A second order polynomial equation with linear mixed-effects was used for modeling branch length of larch tree. The LME procedure in S-Plus is used to fit the mixed-effects models for the branch length data. The results showed that the polynomial model with three random parameters could significantly improve the model performance. The fitted mixed effects model was also evaluated using mean error, mean absolute error, mean percent error, and mean absolute percent error. The mixed model was found to predict branch length better than the original model fitted using ordinary least squares based on all errors. The application of mixed branch length model not only showed the mean trends of branch length, but also showed the individual difference based on variance-covariance structure.