| Forest pests and diseases have seriously harmed China’s forestry resources and have become one of the important factors that seriously restrict the sustainable development of China’s forestry industry.As a unique harmful insect in China,Dendroctonus armandi has seriously harmed the ecological environment of the Shennongjia forest district.At present,the monitoring and prediction research of remote sensing technology in Huashan pine moths is relatively small,and the application of remote sensing data source is relatively single.It is impossible to realize the spatial prediction of the spread of Dendroctonus armandi pests.To investigate the relationship between remote sensing indicators and the spatial dispersal pattern of Dendroctonus armandi,this paper taked Shennongjia National Forest Park as the study area and extracted the dynamic changes of deadwood points of Dendroctonus armandi during2018-2019 for analyzed the dispersal pattern of Dendroctonus armandi pests.The spatial pattern distribution of pests was analyzed based on hotspot analysis and descriptive statistics to determine the pest distribution patterns of different scenarios;the correlation between multi-source remote sensing feature variables and pest spread was analyzed based on ANOVA and mutual information method;the prediction of pest spread of Dendroctonus armandi was realized based on Extreme Boosting Gradient Regression Tree algorithm,and the feature importance analysis and partial dependence analysis were used to determine the influence of multi-source remote sensing feature variables on the pest spread of Dendroctonus armandi.The feature importance analysis and partial dependence analysis were used to determine the influence of multi-source remote sensing variables on the prediction of pest spread of Dendroctonus armandi.The analysis of the importance of features and the partial dependence analysis was used to determine the influence of multi-source remote sensing variables on the prediction of the spread of the pest.The main findings of this paper are as follows.(1)During 2018-2019,the number of deadwood spots of Pinus armandii infested with Dendroctonus armandi in the study area increased sharply,with diffusion to the south and east,and more cold hot spots distributed in the healthy area diffusion(a)area in the south and east;however,the diffusion was more serious in the orioriginallyfected area diffusion(b)area,and there were hot spots with a larger range.(2)Among these variables,the number of points of the grid with new dead wood from the nearest diseased wood grid,the maximum value of annual soil conditioning vegetation index,and the distance of the grid with new dead wood from the nearest diseased wood grid had significant linear or non-linear relationships with the diffusion of Dendroctonus armandi;the slope direction,10% canopy return point cloud density,70% canopy return point cloud density,and 100% canopy return point cloud density.There were no linear or non-linear relationships between the characteristic variables and the diffusion of Dendroctonus armandi.(3)Modeling and prediction analysis of the degree of pest diffusion of Dendroctonus armandi we obtained a prediction accuracy R2 value of 69.09%.Based on this modeling analysis,we found that the prediction model for healthy areas had a lower accuracy of 21.38%,while the prediction model for diseased areas had a higher accuracy of 75.85%.(4)According to the feature importance method,the feature variables with the highest feature importance are pressure variables,followed by height variables and spectral variables.The feature correlations of the forest stand variables,topographic variables,and density variables were low.The variables that have a significant impact on the prediction of the diffusion of Dendroctonus armandi were the number of new deadwood grids from the nearest diseased wood grids,the variance of vegetation height,the maximum value of annual soil-conditioned vegetation index,and the maximum value of annual normalized red edge vegetation index. |
PDF Full Text Request