Abstract: To evaluate site quality and provide scientific evidence, the mixed effect model of polymorphic site index of Juglans mandshurica was accurately simulated in the three provinces of Northeast China. A total of 197 plots in the 23 typical regions of Liaoning, Jilin and Heilongjiang provinces were set by the sample circle method, from which we got 1537 height-age data of J. mandshurica. The site factors were divided, assigned and calculated using variance analysis and model fitting. The results showed that slope position was the dominant factor affecting the growth of dominant wood of J. mandshurica, followed by soil depth, slope and aspect. After fitting and analyzing eight common basic models, we found that the logistic model H=a/[1+exp(b+cA)] was the optimal one (R²=0.70), with a mean absolute error (MAE) of 2.52. When the four main influen-cing factors were randomly combined, the optimal site index model of the stochastic combination (M_8.15) was obtained with R² of 0.90, which improved the fitting accuracy of the base model. The K-means clustering method was used to further divide the initial groups of site types into six groups. We established the nonlinear mixed effect model M_final, H=(20.1837+u_i)/[1+exp (1.7352-0.0961A)]+ε_ij, with R² and Akaike's information criterion (AIC) being 0.92 and 912.65, respectively, which could significantly improve the fitting and accuracy of the model. The equation could be used for the accurate evaluation of site quality of J. mandshurica under the complex site types in the three provinces of Northeast China.

Key words: Juglans mandshurica, random effect model, site index