Objective Health assessment is one of the important prerequisites for implementing sustainable forest management, however most of the previous studies were carried out only on a single scale, without considering the hierarchical structures of forest ecosystems. Therefore, the present study focused on the canopy characteristics, and studied the method of scale transformation for the forest health assessment by the remote sensing and statistical method, which can provide theoretical support and guidance for the forest health management in China.
Method Based on the datasets of individual-tree health survey from 50 sample plots in Pangu Forest Farm, the health assessment model of individual-tree was constructed using the entropy-AHP comprehensive index method. Five commonly used statistical indicators, namely mean value (Hm), standard deviation (Hstd), coefficient of variation (Hcv), skewness (Hpd) and kurtosis (Hfd), were summarized for each sample plot based on the health assessment results from tree-level. Then, a comprehensive forest health assessment model of regional-level was developed by combining the Landsat TM and topographic data using the nonlinear error-in-variable simultaneous equations model. Finally, the forest health status and their spatial distribution characteristics of Pangu Forest Farm were quantitatively analyzed.
Result The sample plot survey datasets indicated that the average health score of individual-tree in Pangu Forest Farm was 0.663 8 ± 0.091 2, belonging to the sub-health level, among which the proportion of sub-healthy trees was the highest (79.43%); the differences of the health grades among different tree species were significant, namely Picea asperata > Betula platyphylla > Larix gmelinii > Populus davidiana > Pinus sylvestris; the statistical values of Hm, Hstd, Hcv, Hpd and Hfd, for the health scores at stand-level were 0.663 3, 0.084 1, 12.84, −0.607 6 and 0.846 0, respectively, indicating that approximately 78.43% of the total forests had a significant left-pointed normal distribution; the remote sensing inversion results showed that the regional-level health score Hm was about 0.619 4 ± 0.054 3, in which topographic (DEM), vegetation index (RVI, DVI, EVI and Green) and original bands (B1, B3) were the key driving factors. The estimated accuracy of the constructed NESEM model was all larger than 75%, which could meet the needs of forest health assessment; in addition, a significant pattern that gradually decreased from north to south was observed for the mean forest health scores, in which the higher scores of Hm were usually concentrated in the convenient transportation areas, such as the areas of residential and forest roads.
Conclusion The forests in study area were mainly sub-health, which may be urgent to carry out scientific health management. Meanwhile, the multi-scale transformation method presented in the study, namely combining the canopy characteristics with the results of forest health assessments by remote sensing and statistical methods, could achieve the scale conversions of forest health assessments among different levels very well.