Spatial distribution simulation of recruitment trees of natural secondary forest based on geographically weighted regression

  Objective  By studying the amount and spatial distribution pattern of recruitment trees in natural secondary forest, the response of recruitment trees to various variables was analyzed, the reasonable method of processing spatial non-stationary data was explored, and the optimal model of the amount of recruitment trees was constructed. It is expected to provide more accurate technical means for the study of growth dynamics of natural secondary forests, and to provide a reference for the accurate improvement of forest quality of natural secondary forests.

  Method  Based on the data collected from 106 bureau level permanent sample plots in Tazigou Forest Farm of Wangqing Forestry Bureau in Jilin Province of northeastern China during 1997 and 2007, we taken stand factor, topography factor and soil factor as the influencing factors and established conventional Poisson regression (PR), geographically weighted Poisson regression (GWPR) and semiparametric geographically weighted Poisson regression (SGWPR), respectively to simulate the status of amount and distribution of recruitment trees of natural secondary forest in the area. Coefficient of determination (R²), mean square error (MSE) and Akaike’s information criterion (AIC) were used to evaluate the fitting effects of three models. The spatial autocorrelation and local spatial aggregation of residuals of the three models were analyzed by global and local Moran’s I. The spatial distribution of recruitment trees in the research area was drawn with the fitting results of SGWPR, and the distribution pattern of recruitment trees in the research area was analyzed.

  Result  (1) In the three models, both stand factor and topographic factor had a great influence on the amount of recruitment trees, among which the average DBH of stand was the variable with the greatest influence, and there was a significantly negative correlation between them; (2) GWPR was obviously better than PR in the fitting effect, among which SGWPR had the best fitting effect. For the fitting of the strong influencing points which deviated far from the expected value, it showed excellent effect; (3) GWPR had a better stability and can significantly reduce the spatial autocorrelation of model residual. By contrast, SGWPR can minimize the spatial distribution of residual with similar aggregation; (4) ten years later, in more than 83% area of Tazigou Forest Farm, the number of recruitment trees was 0−683 per hectare. The overall condition of northern area was better than southern area, and the maximum value of local area was mainly located in the marginal hillside of the northeast of forest farm.

  Conclusion  The optimal model for the amount of recruitment trees can be obtained by SGWPR. When constructing the model of amount of recruitment trees, not all variables need to consider the geographically weighting, which should be determined according to the specific research content and data.