Abstract: [Objective] The aim of this study was to develop taper functions of dahurian larch(Larix gmelinii)based on widely used variable-exponent taper models developed by Kozak(1988),Kozak(1994),Kozak(2001)and Kozak(2002),and the significances of different regions were also tested.[Method] Stem taper data of dahurian larch was collected from three regions of Daxing'anling. SUR method in SAS software was used to fit the four taper equations. Coefficient determination(R_adj²), mean absolute bias(MAB), root mean square error(RMSE), mean percentage of bias(MPB)and prediction accuracy(P%)were employed to evaluate the precision of different models combining multicollinearity (CN)and graph of the residuals. The non-linear extra sum of squares method was used for regional comparison. The method required the fitting of full model and reduced model.[Result] 1)Variable-exponent taper models developed by Kozak(1988)and Kozak(1994)were better based on goodness of fit statistics and prediction precision, but they had high multicollinearity. The other models decreased multicollinearity and improved prediction precision of upper stem diameters. 2)F-test of regional comparison showed that stem taper in region 3 had large differences with region 1 and region 2. Stem taper in region 1 had small difference with region 2. Any two regional test showed significant difference(P<0.0001). It was indicated that any two regions could not use the same parameter estimates and separated parameter estimates were necessary. 3)Through the stem taper curve simulations in three regions, it could be seen that the same model showed the different taper curve simulations in three regions, especially the taper curves of model 3 showed significant difference in three regions. The taper curve simulations of other models showed small differences with region 1 and region 2, and significant difference was found for region 3. The results of simulations were consistent with the results from F-test. Different region had a significant effect on tree stem curve. Incorrectly application of taper models in different regions would result in larger prediction error.[Conclusion] Taper model developed by Kozak(2002)showed the consistent performance on fitting statistics, residual distribution, and multicollinearity. The prediction precision was more than 99%, and could be selected as the best taper equation for the three regions of Daxing'anling.

Key words: variable-exponent taper models, multicollinearity, dummy variables, nonlinear regression