Monitoring On Insect-borers Infestation In Artemisia Ordosica Shrubs By Remote Sensing

Remote sensing monitoring technology has been widely applied in all kinds of disaster monitoring, such as the dynamic monitoring on forest biological disasters can be realized through remote sensing technology in large area, it is benefit for timely early monitoring of the occurrence of biological disasters, to grasp the biological disaster rules, warn disaster process, take control and prevention measures, and reduce loss disasters. Due to the wide distribution, large area of desert shrubs, strong similarities with the surface background and difficult to monitoring after plants being attacked from insect borers, fast and accurate monitoring of forest pests is of great significance for forecasting and evaluating the forest disasters by remote sensing technology.In this paper, this study was based on the theories of forest protection and quantitative remote sensing. Multispectral and hyperspectral image data were applied in detection, This paper, evaluates the identification and information extraction of A. ordosica community by Decision Tree Classifier based on the spectral characteristics and spatial distribution characteristics, and the annual change of A. ordosica community were analysized through overlying the eleven Landsat images based on density slice of Normalized Difference Vegetation Index (NDVI) during ten years, and by combining the field investigation and meteorological data. Due to no remarkable change after the target plant being attacked from insect borers, Firstly, the assessments of health degree on each plant were applied at hyperspectral data in field, Judgment matrix, four criteria (death ratio of crown area, pest density per plant, ground diameter, plant height) of evaluation factors for injure level of A. ordosica and membership function were established by fuzzy evaluation; Secondly, According to the comprehensive evaluation score, four health levels of health, slight, moderate and severe were determined; Finally, four damaged degree of spectral curves and twenty-two spectrum parameters were extracted by correlation analysis, also the model of estimated healthy level of A. ordosica were established. The main conclusion and innovations was included four points (as follows):(1) Compared the spectral of A. ordosica at different growth vigour, the obvious phenomenon was blue shift o f the red edge spectra after the health A. ordosica were damaged by insect borers. The insect mainly caused damage to roots of Artemisia spp., it was possible to use hyperspectral data to monitoring the occurrence of insect borer infestation, and it is safe to the plant. From fitting and prediction accuracy, the best one was semi-fixed sandy land, the shifting sandy land and the fixed sandy land were relative low among them. In the semi-fixed sandy land, it model formula was Y=1.292×Dr+0.008×Xr+13.277×Rg+2.506×(SDr-SDb)/(SDr+SDb)-0.395×SDg-7.510, its fitting coefficient of determination was 0.406, the predicted coefficient of determination was 0.562 (its fitting multiple correlation coefficient and predicted multiple correlation coefficient was 0.637 and 0.759 respectively), root mean square error was 0.192. Although their fitting and prediction accuracy were not very high, it conducted an availability study about remote sensing technology can be applied at detecting and monitoring insect borers.(2) In comparison of no auxiliary artificial interpretation and the involvement of auxiliary artificial interpretation after decision tree classification based on the spectral and spatial distribution characteristics, the overall user accuracy were 69.85% and 71.19% respectively, the improved accuracy was not particularly evident. But for A. ordosica community extraction, the user accuracy were 60.21% and 75.14% respectively, the accuracy was increased to nearly 15%, the mainly reason is that desert vegetation has very close-by the spectral characteristics, the phenomenon of one species growth in different vegetation types and mixed growth seriously among different species, it is hard to avoid the decline of extraction accuracy. Although the accuracy was not very high, the results show that using decision tree classification based on the spectral and spatial distribution characteristics to extract the vegetation information is feasible method in desert region. At the same time, the involvement of auxiliary artificial interpretation after classification is so important to the improvement of accuracy, especially the desertification areas of strong spatial heterogeneity and mixed growth seriously among different species.(3) The infestation plant has still at green attack stage, and no significant red attack stage, and no obviously change of leaf color and leaf litter, the leaves direct transition to gray attack stage and lead to plant dry death. Although the NDVI, vegetation coverage and damage degree are suitable for the loss assessment in traditional monitoring on pest remote sensing, these researches are not so good to detect and monitor insect borers of A. ordosica. Gradually it was observed that the damage degree increased; there was no distinct decrease in the vegetation coverage and NDVI.(4) The annual change of A. ordosica community were analysized through overlying the eleven Landsat images based on field investigation, meteorological data and density slice of Normalized Difference Vegetation Index(NDVI) during ten years. It was found that semi-fixed sandy land is a transitional stage from shifting sandy land to fixed sandy land, and semi-fixed sandy land was the worst severely damage in three sites. Therefore, it was obvious that semi-fixed sandy land is the most fragile habitat conditions and unstable transitional stage. So the transitional type should be attached the more importance to the succession processes, especially the monitoring and research on biological disasters in desert shrubbery.