Combining Weibull distribution and k-nearest neighbor imputation method to predict wall-to-wall tree lists for the entire forest region of Northeast China

Abstract

Abstract Key message We propose a coupled framework to combine the strengths of the Weibull function in modeling diameter distributions and the ability of the k-nearest neighbor (kNN) method to impute spatially continuous forest stand attributes for the prediction of wall-to-wall tree lists (lists of stems per hectare by species and diameter at breast height (DBH)) at regional scales. The tree lists of entire Northeast China’s forests predicted by the above framework reasonably reflect the species-specific tree density and diameter distributions. Context Detailed tree lists provide information about forest stocks disaggregated by species and size classes, which are crucial for forest managers to accurately characterize the current forest stand state to formulate targeted forest management strategies. However, regional tree list information is still lacking due to limited forest inventory. Aims We aimed to develop a coupled framework to enable the prediction of wall-to-wall tree lists for the entire forest region of Northeast China, then analyze the species-specific diameter distributions and reveal the spatial patterns of tree density by species. Methods A two-parameter Weibull function was used to model the species-specific diameter distributions in the sample plots, and a maximum likelihood estimation (MLE) was used to predict the parameters of the Weibull distributions. The goodness-of-fit of the predicted species-specific Weibull diameter distributions in each plot was evaluated by Kolmogorov-Smirnov (KS) test and an error index. The kNN model was used to impute the pixel-level stand mean DBH. Results Weibull distribution accurately described the species-specific diameter distributions. The imputed stand mean DBH from the kNN model showed comparable accuracy with earlier studies. No difference was detected between predicted and observed tree lists, with a small error index (0.24–0.58) of diameter distributions by species. The fitted species-specific diameter distributions generally showed a right-skewed unimodal or reverse J-shaped pattern. Conclusion Overall, the coupled framework developed in this study was well-suited for predicting the tree lists of large forested areas. Our results evidenced the spatial patterns and abundance of tree species in Northeast China and captured the forest regions affected by disturbances such as fire.