Determining Vegetation Parameters From UAV Imagery In Loess Region

Vegetation is an essential information for reflecting the changes in the ecology and environment.The traditional forest inventory survey mainly including the field sample survey,which is labor-cousming and time-cousming.Satellite remote sensing is restricted in practical applications due to low resolution,long re-entry period,and cloud occlusion.In recent years,unmanned aerial vehicle(UAV)technology has developed rapidly.UAV remote sensing images are not disturbed by weather factors during the acquisition process.The low cost of using it,short image acquisition cycle,and higher resolution provide new way for the rapid extraction of vegetation information.With the purpose of exploring quick and accurate methods for forest inventory survey,Caijiachuan Watershed in the Loess Region of Western Shanxi is selected as the study area to examine the applicability of UAV photogrammetry in extracting forest species and stand characteristics.The information of forest species in agricultural sub-watershed and plantation sub-watershed were identified and extracted from visible images of UAV using nearest neighbor classification of the object-oriented method,and the accuracy were evaluated by comparison with vegetation sampling plots investigation.The information of the tree height in study area were identified and extracted from the local maximum method and the crown height model(CHM).The information of crown width was identified and extracted from using nearest neighbor classification of the object-oriented method,and the accuracy was evaluated by comparison with vegetation sampling plots investigation.And the methods presented in this study was used to analyzetree species,tree crown characteristics and tree height values under the background of ecological restoration in the Caijiachuan watershed in the loess region of western Shanxi.The results show that:(1)The method for forest species and crown width extraction and its accuracy.The Kappa coefficient of the confusion matrix for forest species classification is 0.898 and 0.728 in the agricultural watershed and in the planted forests watershed,respectively,suggesting higher accuracy of extracting forest stands under the condition of low canopy closure than multiple vegetation types with high canopy closure.According to the determination coefficient between extracted crown width from images and field measured crown width,the accuracy of crown width extraction of planted forests and economic forest(R~2>0.7)were higher than that of natural secondary forests(R~2=0.422).(2)The method for tree height extraction and its accuracy.The tree height extraction results based on the local maximum method and the CHM of different types of plots are good,the value of R~2ranges from 0.683 to 0.939,and the value of the root mean square error(RMSE)ranges from 0.544m to2.595m,it has the value of application promotion.In addition,the tree height extraction accuracy of the sample plots with flat terrain and gentle slope were relatively higher.(3)Forest information in the study area based on the methods presented in this paper,Robinia pseudoacacia,Pinus tabulaeformis and Platycladus orientalis are the main tree species in afforestation sub-watershed,most of them are mixed forest,and the economic forest,Malus pumila and Pyrus spp.The forest density of Robinia pseudoacacia,Pinus tabulaeform and Platycladus orientalis were 997plant/ha,1744 plant/ha and 1862 plant/ha.The average crown width of them were 3.52m,2.24m and1.95m.The tree height of them were 1.41m,7.23m and 4.68m.A large part of orchards was including Malus pumila and Pyrus spp.The forest density of them were 382 plants/ha and 502 plants/ha,the average crown width of them were 4.26m and 4.01m.There are 912 Malus pumila with density of 439plant/ha and in the agricultural watershed and average canopy width of 3.84m.There are 2155 Robinia pseudoacacia with density of 1051 plant/ha in the agro-pastoral sub-watershed and average canopy width of 2.74m.This research shows that the UAV remote sensing would improve the efficiency of forest inventory surveys accurately with the capability of forest species classification,tree number counting and crown index estimation,providing references for forest resource survey and forest sub-space pattern allocation construction of large-scale afforestation forest ecological engineering.