Severity Monitoring Of Erannis Jacobsoni Djak. Based On UAV Remote Sensing And Characterization Of Changes In Biochemical Components Of Forest Canopy

As the largest terrestrial ecosystem,forests play an important role in the Earth’s biosphere.As one of the three major forest disasters,pests and diseases damage the external structure and internal physiological and biochemical components of forest trees and pose a great threat to forest ecosystems.Based on this,this study takes Erannis jacobsoni Djak.pest attacking forest trees as the research object,and constructs the pest severity monitoring model and forest canopy biochemical component estimation model by using feature extraction method and machine learning through the affected forest tree canopy appearance characteristics and internal biochemical component data combined with UAV multispectral features,and analyzes the intra-and inter-month variation characteristics of different pest severity.The results of the study are as follows:（1）The spectral reflectance of UAV multispectral images in different pest severity cases has significant differences.The spectral reflectance in the green,red-edge and near-infrared bands decreased with increasing damage;while the spectral reflectance in the blue and red bands gradually increased.ANOVA found that only EVI and EVIreg were lower than the F-test value,indicating that the selected features showed significant sensitivity.A total of9 VIs and 12 VIs+TF were obtained for monitoring model construction,and14,8 and 16 VIs were used for CHLC,LWCF and LWCD estimation model construction,respectively,by extracting sensitive features through SPA.SPA can effectively reduce the number of sensitive features and prevent model overfitting.（2）The overall accuracy,Kappa,R_macroand F1_macrocoefficients of the six pest severity models constructed by VIs and VIs+TF spectral feature sets were higher than 0.8,indicating that the models could effectively monitor pest severity.The comparison revealed that the 1D-CNN model achieved the best accuracy in both feature sets.The overall accuracies of 1D-CNN based on VIs and VIs+TF were 0.8950 and 8800,respectively.meanwhile,the accuracy of the model constructed based on VIs was higher than that of the model constructed based on VIs+TF.It can be seen that the VIs and 1D-CNN can effectively identify the severity of Erannis jacobsoni Djak.pest by VIs and 1D-CNN,and the method has important practical value for forest pest monitoring.（3）In the estimation of biochemical components of the forest canopy,the estimation models constructed by 1D-CNN,RF and SVM all achieved an R²of 0.9 or more,which were effective in estimating the CHLC,LWCF and LWCD of the forest canopy.（R²=0.9406）;for LWCF,the 1D-CNN model is the best（R²=0.9477）,RF is the second best（R²=0.9318）,and SVM is the worst（R²=0.9156）;for LWCD,the 1D-CNN model is the best（R²=0.9595）;RF is the second best（R²=0.9513）;SVM is relatively poor（R²=0.9511）.It can be seen that the R²of 1D-CNN is higher than that of both RF and SVM,and its estimation effect is significantly better than other models.（4）By analyzing the intra-monthly and inter-monthly variation of canopy biochemical fractions with different pest severity,we found that the canopy biochemical fractions under pest stress were able to recover to some extent after the short-term outbreak of pests.For the intra-month period,CHLC,LWCF and LWCD showed a decreasing trend with increasing damage severity;for the inter-month period,CHLC and LWCD showed an overall increase and then a flat trend,while LWCF showed a continuous increase.Specifically,we found that CHLC and LWCD of healthy trees throughout the month showed a continuous decline;LWCF showed a continuous increase.Meanwhile,only for healthy stands in June,CHLC and LWCF were characterized by decreasing and then increasing,and LWCD was continuously decreasing;for light stands in June,CHLC showed decreasing and then increasing characteristics,and LWCF and LWCD showed continuous increasing trends;for moderate stands in June,CHLC and LWCD showed increasing and then flat trends,while LWCF showed continuous increasing.In June,the CHLC and LWCF showed a continuous increase,while the LWCD showed an increase and then a flat trend for heavy stands.The above inter-month variation characteristics were complexly attributed to the fact that the Yarrow larch looper still infested the forest trees in early July,which caused the change of forest tree pest severity in July compared with June.The results can provide an important reference for forest ecological restoration projects affected by the pest.In conclusion,the spectral vegetation index can effectively construct the insect pest severity monitoring model and biochemical component estimation model,and the insect-stressed forest can achieve a certain degree of recovery in the short term.In addition,due to the lack of data before the pest outbreak and July,the analysis of the whole process change characteristics was not realized.In the future,we need to obtain data to analyze the whole process change characteristics and factors affecting the change characteristics to protect the safe and healthy development of forest ecosystem.