Abstract: [Objective] Through the analysis of the spatial correlation and spatial distribution pattern of forest regeneration in different locations of natural secondary forests in Daxing'an Mountains, this study was implemented to explore the impacts of scale effect on the spatial autocorrelation, to understand the potential regularity in regeneration dynamics from a deeper level by analyzing the influencing factors of forest regeneration, and finally to provide theoretical basis and technical support for the operation and decision of natural secondary forest in this area.[Method] We took Cuigang forestry station of Xinlin Forestry Bureau of Daxing'an Mountains in Heilongjiang Province as the research area. Based on the data of 45 permanent sample plots established in the research area from July to August 2018, we selected 9 factors in 5 aspects including stand factors, topographic factors, forest stand spatial structure, soil thickness and species diversity as the independent variables, and established the global Poisson regression model and geographically weighted Poisson regression(GWPR)model under 4 scales(5 km, 10 km, 15 km and 20 km)on the basis of geographically weighted regression model to simulate the regeneration status of natural secondary forest in this area. Global Moran I and local Moran I were used to respectively describe the global spatial autocorrelation and spatial distribution of model residuals, to evaluate the fitting effects of global model and of local models under different scales, and to explain the differences among the each local model under different scales. Finally, the local model under 5 km was adopted to draw the spatial distribution plan of forest regeneration in the research area so as to evaluate and analyze the forest regeneration in the research area.[Result] The local model under 5 km made the best local spatial distribution of model residuals, formed the ideal distribution state of aggregated distribution of different model residuals, the parameter estimates of the model variables produced the largest range of variation, and had the best stability. With the gradual increase of the spatial scale, the stability of the model declined gradually, but still generally better than that of the global model. Meanwhile, the local model under this scale showed the lowest spatial autocorrelation of model residuals. The fitting effects of local models were better than those of global models, where, the local model under 5 km had the minimum MSE value and AIC value. In the research area, the number of regeneration individuals in the south part was larger than the one in the north part, while the differences between the east and the west were not obvious.[Conclusion] The influences of spatial scale shall be taken into consideration when the local model is established. The local model under 5 km adopted in this study could well simulate the spatial distribution of natural secondary forest regeneration in the research area, and could effectively reduce or even remove the spatial autocorrelation. In the research area, the number of stand regeneration individuals is mainly between 1 000-2 000 hm^-2, that is, the overall regeneration is at a bad level, and the natural forest regeneration ability is generally weak, therefore, management measures such as artificial promoting of natural regeneration shall be taken for forest management.

Key words: forest regeneration, spatial autocorrelation, geographically weighted Poisson regression model, spatial scale, Daxing'an Mountains