Spruce Forest Ecosystem Health Assessment In Tianshan Based On Support Vector Machine

Mountain forests are not only one of the important components of an ecosystem,but also an important barrier for humans to face many environmental issues.According to data from the second survey of forest resources in 2014,the spruce of the Tianshan Snow Mountain accounts for approximately 59%of the total forest area of the mountainous region of Xinjiang.As a dominant species of mountain forests,its health status maintains the stability of the forest ecosystem in the entire Xinjiang and its arid and semi-arid regions.The sustainable development has an important role.Therefore,this paper takes 44 fixed sample plots in the study area of Banfanggou Forest Farm as the research object,and selects 33 indicators from four aspects of structure,vitality,anti-jamming and ecological service functions to conduct surveys,through factor analysis and data packages.The establishment and evaluation of forest health model based on network method and support vector machine(SVM),combined with the data of each class in the investigation of the forest resource in the second class of the 2014 survey of the forest at the forest of the Banfanggou,further evaluated the health of the forest ecosystem of the snow spruce at the Banfanggou Forest Farm..The specific results are as follows:(1)The result of factor analysis showed that the number of healthy and high-quality forest plots was 3,which accounted for 6.82%of the total sample plots;there were 14 healthy forests in the health class,accounting for 31.82%of the total sample plots;the forest health grade was sub-healthy sample plots.There were 15 of them,which accounted for 34.09%of the total number of plots.There were 12unhealthy plots with forest health grades and 27.27%of the total sample plots.(2)The weights of the 21 indicators obtained by using factor analysis method are:vitality indicators(net primary productivity,leaf area index,NDVI,and biomass of shrubs);structural indicators(average tree height,average forest age,mean diameter at breast height,Live stock volume,litter thickness,available phosphorus,organic matter,available nitrogen,soil thickness,soil bulk density,canopy density,herb coverage,available potassium);anti-interference indicators(cattle and sheep feeding,disease severity,fire)Dangerous);ecological function value.(3)The database and support vector machine theory(SVM)built on 44 fixed plots were trained by multiple cross validations to determine the final model.The prediction accuracy of the test set sample is 73.33%,the penalty parameter c of the radial basis kernel function is 90.51,and the width g of the kernel function is 9.19.(4)The health status of 1802 small classes(with a total area of 58,336.11 hm~2)was predicted using the trained model,resulting in a high-quality forest area of18,765.20 hm~2,accounting for 32.17%of the total area of the study area;a healthy forest area of 32,726.51 hm~2,accounting for the study 56.10%of the forest area in the district;the sub-health forest area is 6416.71hm~2,accounting for 11.0%of the total forest area in the study area.The unhealthy forest area is 427.69hm~2,accounting for0.73%of the total forest land area in the study area.(5)The analysis of the traditional data envelopment analysis(DEA)model and super efficiency data envelope analysis(SDEA)model shows that the input/output efficiency of 44 fixed sample plots is mainly its own scale efficiency,ie its own internal environment Quantitative allocation of indicators;and in the natural state,in the stands of healthy,high-quality grades,the input/output efficiency is not necessarily the highest.