Spatial Variation Of Typical Plant Communities In Desert Hinterland Oasis Based On Low-altitude Remote Sensing

Oasis is a non-zonal landscape with a desert-based matrix and a high primary productivity.The oasis exists in the desert and is different from the desert.Under certain conditions,the two transform each other and form different community structures in the process of transformation.The Keriya River Owl is a “primitive” oasis of largescale human activities in the rare desert hinterland of the Tarim Basin,with important scientific research and oasis cultural historical value.The spatial differentiation characteristics of plant communities in this oasis are significant,it is of great significance to explore the spatial variation of plant community structure in this oasis and its environmental change trend and sustainable protection.However,the oasis is large and scattered,and there is no fixed road.It is difficult to accurately grasp the spatial distribution characteristics of typical plant community structure in a survey period,and it is difficult to carry out research on its distribution law.Remote sensing technology,especially low-altitude UAV remote sensing,provides vegetation data with a resolution of up to centimeter,providing a precise and feasible method for large-area vegetation surveys in desert areas.However,desert oasis has relatively poor health and vegetation.The pixel and non-vegetation pixels are more confusing.The spectral characteristics of different species(such as Populus euphratica and Tamarix chinensis)are not obvious.The traditional classification method based on pixel spectral information is not universal,and the extraction effect is not ideal.Machine learning can realize human learning behavior through computer simulation.The computer obtains information by itself and organizes the existing knowledge structure to continuously improve computing performance.This method has been introduced into the remote sensing classification,showing a wide application prospect,and also provides a means to solve the spatial characteristics of the typical plant community structure in the natural oasis of the desert hinterland.This study takes the natural oasis in the desert hinterland of Yu Tian County,Xinjiang China as the target area,based on multi-scale data such as MODIS products,Landsat images and low-altitude high spatial resolution UAV remote sensing images,combined with field vegetation survey,comprehensive utilization of CNN-CPMF(Convolutional Neural Network-Class Paired Median Filter)model,Vegetation index threshold method,K-means classification method and Triangulated Irregular Network(TIN)space model,the method of rapid extraction of vegetation coverage(FVC),plant community coverage and population distribution parameters is proposed.The longterm sequence variation of vegetation cover in natural oasis in desert hinterland was analyzed.The spatial variation of vegetation coverage-plant community-individual structure parameters was explored.The main conclusions are as follows:(1)In view of the complex features and the blurring of inter-class boundaries in the desert area,the CNN and CPMF are used based on high resolution remote sensing images and enables automatic classification of plant communities.The results show that the sample size and scale have a significant impact on the classification accuracy.After the generalization ability is improved,the modeling accuracy of Res Net50 network,VGG16 and VGG19 networks is improved from 76.67%,86.67% and 85.33%to 89.80%,90.80%,90.40% on the 5×5 pixel scale.On the 10×10 pixel scale,the modeling accuracy increased from 86.00%,80.00%,and 83.33% to 92.56%,88.22%,and 90.29%,respectively.Compared with the traditional supervised classification method,the classification accuracy of deep convolutional neural networks has been greatly improved.When the number of training samples is not less than 200,the Nest Net50 model based on CNN shows better classification results.This method is suitable for the extraction of spatial structure parameters of plant communities in arid regions.Using the convolutional neural network(CNN)Res Net50 structure and the irregular triangulation TIN space model,the spatial distribution parameters of the oasis Populus euphratica in the desert hinterland were extracted.Calling the Sklearn-cluster classifier for K-means clustering algorithm can accurately locate the centroid of Populus euphratica.The Voronoi polygon and the Delaunay triangulation model can quickly obtain the buffer and spacing of Populus euphratica.(2)The typical section is selected.The direction of the vertical channel is divided into the vicinity of the river(within 100 m),the river bank is 05 km,1 km,2 km,and 3km.Vegetation information was extracted by Ex GR index,and vegetation coverage was quickly extracted by threshold segmentation using OTSU method.The results showed that the average vegetation coverage in the near-channel area was 53%,and the vegetation coverage near the river bank(3 km)was 28%.The near-river group was composed of Phragmites communis and Tamarix chinensis,and the average coverage was about 22.68% and 72.63%.The 3 km area of the river bank was planted with Populus euphratica.The overall health status of Populus euphratica was poor,with an average coverage of about 25.75%.In the vertical river bank direction,the vegetation showed the characteristics of Phragmites communis-Tamarix chinensis-Populus euphratica replacement.The difference of vegetation water strategy under drought stress is the main reason for the difference of vegetation distribution.(3)On the long-term sequence,the changes of LAI and FVC in the Deryaboyi Oasis in the past 11 years were analyzed.During the year,LAI and FVC peaked from mid-July to early August,and then slowly decrease month by month,indicating that the vegetation in Deryaboyi Oasis was in the best condition between mid-July and early August.According to the internal variation,since 2009,the Oasis leaf area index LAI(spatial resolution of 1 km)and vegetation coverage FVC(spatial resolution of 500 m)have increased slightly.(4)From the sample survey,the average plant height of Populus euphratica was7.9 m and the highest was 17.4 m.Grade I Populus euphratica accounted for 17.8%,Grade II Populus euphratica accounted for 55%,Grade III Populus euphratica accounted for 25%,and Populus euphratica with height above 15 m accounted for 2.1%.Populus euphratica in the range of 5 m～15 m accounted for 80% of the total.Populus euphratica population was dominated by mature forests,and young forests accounted for less,reflecting the slow renewal rate of Populus euphratica.The overall health of Tamarix chinensis is poor,9.3% of Tamarix chinensis is in a state of sudden death,the number of degraded Tamarix is as high as half,accounting for 52.4% of the total,moderate health willow is about 29%,and the good state of Tamarix chinensis is only6.3%.Healthy yam is only 3%.The health of Tamarix chinensis and Populus euphratica is not good,and the population of Populus euphratica is declining and the population is unstable.(5)Using the high spatial resolution data of the drone,the vegetation coverage(FVC)was quickly extracted.The maximum value of the Oasis FVC was 47.53%,the minimum value was 3.57%,the average value was 24.17%,and the coefficient of variation reached 51.84%.The difference in vegetation coverage is large,and the degree of dispersion is high;the vegetation coverage in the southern and central oasis is good,and the vegetation coverage in the north and east is gradually decreasing.The spatial distribution of plant community coverage based on UAV images showed that the maximum coverage of Populus euphratica in the core region was 28.28%,and the maximum coverage of Tamarix chinensis was 69.47%.The core region of Oasis was mainly Tamarix chinensis.According to the overall coverage analysis,Tamarix chinensis is the dominant species in the oasis center;the maximum coverage of Populus euphratica in the southern region is 6.08%,the maximum coverage of Tamarix chinensis is 39.18%,and the maximum coverage of Phragmites communis is 44.71%.The coverage of Tamarix chinensis is larger;the maximum coverage of Populus euphratica in the northern region is 7.19%,the maximum coverage of Tamarix chinensis is 30.67%,and the maximum area of litter is 22.60%.The northern part of Oasis is mainly Populus euphratica and Tamarix,due to perennial surface.The decrease of water led to the gradual death of Populus euphratica and Tamarix chinensis population,and a certain area of dead population appeared in the northern part of the oasis.This paper explores the feasibility of deep convolutional neural networks in the automatic classification of high spatial resolution remote sensing image plant communities through machine learning techniques.The significance lies in the extraction of vegetation information.The network structure of VGN16,VGG19 and Res Net50 of the basic CNN abstracts and learns the features of the block image respectively,so as to realize the automatic extraction of plant community information;in terms of spatial distribution parameters of the population,it will be combined with CNN.The TIN model obtained the distribution parameters of Populus euphratica population,which further revealed the spatial distribution trend of Populus euphratica population and its response to ecological environment.It provides an auxiliary method for the rapid vegetation survey in the field,improving the efficiency of field work,and realizing the extraction of more plant community information at the regional scale.Overall oasis,the southern part of the oasis is mainly reed and tamarisk,and the core area is the dominant species.The northern part has more dead populations,mainly populus euphratica and Tamarix chinensis.The population of Tamarix chinensis and Populus euphratica are obviously degraded.The coverage of oasis plant community on surface water The response is extremely significant.The research results provide a useful reference for vegetation survey and data processing and analysis of desert drones,which is helpful to further explore the evolution mechanism of desert vegetation pattern and lay a foundation for the study of oasis vegetation protection.