Individual Tree Mortality Models For Mixed Spruce-fir-broadleaf Forests In Jingouling Region

The spruce-fir mixed broadleaved-conifer forest is one of the main forest types in northeast China,trees mortality is an important part of forest growth models,understand the pattern of this forest types,it is helpful to grasp the dynamic change of its stand and provides reference for local forest managers to make the forest management plans.This study used the trees mortality of Jingouling forest farm control method plot first block and second block’s data form 1987 to 2012,based on the ecological characteristics of the local major tree species,it was divided into four species groups: slow-growing coniferous,moderate-growing coniferous,slow-growing broad-leaved,slow-growing broad-leaved,establish binary logistic model(GLM),stepwise regression was used to select the optimal combination of variables that affect the mortality,then Maximum likelihood estimation was used to calculate parameter values.The random effect of the sample land level was added to the basic model to establish the binary logistic mixed model(GLMM).At the same time,the optimal threshold was selected by comparing various threshold selection methods,draw the ROC curve and calculated the AUC value,Pearson chi-square test was performed by diameter class to examine the performance of model.The result shows as follows: the variables related to DBH were selected into four species group models,the DBH was positively correlated in the groups of slow-growing conifers species group.In the moderate-growing conifers species group,the rate of mortality was u-shaped with the increase of DBH,and it decreased when the DBH was 5-20 cm,and then increased again.Because there are not enough big trees,DBH was positively correlated in slow-growing broad-leaved species group and moderate-growing broad-leaved species group.In the stand factor,the rate of mortality of slow-growing conifers species group,slow-growing broad-leaved species group and moderate-growing broad-leaved species group increases with the rise of stand density.Because young fir develops better when the light is low,number per hectare was negatively correlated moderate-growing conifers species group.The competitive factors were also added to all tree group models,the greater competitive advantage of trees in stand,the lower the mortality rate.the sum of larger than object trees basal area was positively correlated in moderate-growing coniferous and moderate-growing broad-leaved species groups,number of trees that DBH larger than the object trees was positively correlated in moderate-growing coniferous and slow-growing broad-leaved,Ratio of DBH squared to basal area was correlated negatively correlated in slow-growing coniferous.in the site factor,only the elevation correlation coefficient was positively correlated in the middle broad-leaved tree species group.The mixed effect model(GLMM)of four tree species groups compared with the base model(GLM),The AUC values were improved by 15.4%,18.2%,25.3% and 13.6%,respectively.In the threshold methods,only the threshold selected according to the maximum principle of the sum of specificity and sensitivity(MST principle)passed the Pearson chisquare test of dividing diameter class.These results suggest that because the ecological difference among species groups.The types and degrees of factors affecting tree species were different,the binary logistic mixed model(GLMM)has higher stability and prediction accuracy than the basic model,which can solve the problem that data with autocorrelation and heteroscedasticity;MST principle is the threshold value selection method most applicable to this study,which can improve the accuracy of prediction.