| The Horqin Sandy Land is the largest sandy land in China,and its ecosystem is complex and fragile.In recent years,under the influence of many factors such as global climate change and human activities,it is facing the threat of ecosystem health problems such as increasing desertification and landscape fragmentation,which has seriously restricted the sustainable development of regional ecological environment.There is an urgent need to carry out relevant basic research to support the scientific construction and management of regional ecology.Vegetation coverage is an important indicator of regional ecological health.Therefore,this paper takes the inland closed watershed of Horqin Sandy Land as the study area.Take one of the typical sand dune-meadow alternating geomorphological units as the key study area.A high precision vegetation coverage inversion model is constructed by using UAV,Worldview-2 and Landsat TM/OLI remote sensing image data and machine learning method.The dynamic characteristics and rules of vegetation coverage in the study area in the past 30 years were quantitatively analyzed by using the relevant methods of landscape ecology.The main conclusions of this paper are as follows:(1)The U-shaped neural network depth learning model for vegetation coverage extraction consists of four coding layers,two convolution layers,four decoding layers and one Softmax layer.The activation function used in the model is Relu function.On the whole,the model has good denoising and high accuracy.the accuracy of training set is 0.82 and that of verification set is 0.86.The vegetation coverage of different landforms and complex habitats can be extracted efficiently and conveniently.The high vegetation cover areas of semi-fixed and fixed sand dunes are mainly concentrated in the inter-dune depression,showing a crescent distribution,and the vegetation fragmentation is large,while the meadow area vegetation coverage boundary is significant,the distribution is regular.(2)Compared with artificial neural network and support vector machine model,stochastic forest model is more suitable for the inversion of vegetation coverage in Horqin dune-meadow area.The vegetation coverage inversion method of low-resolution remote sensing images is gradually realized by using multi-source remote sensing data,which can effectively improve the inversion accuracy of vegetation coverage.The accuracy of the small-scale(the key study area)vegetation coverage inversion model is very high,the training set R~2 is 0.94,MSE is 0.01,d is 0.98,and the test set R~2 is 0.84,MSE is 0.01,d is0.96.The large-scale(the study area)vegetation coverage inversion model has high accuracy,training set R~2 is 0.87,MSE is 0.01,d is 0.96,test set R~2 is 0.80,MSE is 0.04,d is 0.90.(3)In the past 30 years,the average annual vegetation coverage in the study area fluctuated in an interval,with strong interannual fluctuation and poor anti-interference ability,but showed a good trend as a whole.And the average vegetation coverage fluctuated between62.08%and 70.19%.Among them,the area of vegetation reduction is mainly distributed in the southeast and west of the study area,while the growth area is alternately distributed in the middle and boundary of the study area.The spatial heterogeneity of vegetation coverage in the study area is strong,the main types of vegetation cover are high,medium and high cover,followed by low and medium cover,and the proportion of low cover is the least.From1989 to 2019,the transition area of high cover class and medium-high cover class is the largest and frequent,and the center of gravity migration of each cover class has repeated characteristics,mainly in the southwest and northwest direction,and the change of standard deviation ellipse of each cover class is relatively stable.(4)The annual cumulative precipitation in the study area fluctuates greatly from 1989to 2019,showing wavy interval fluctuations,which are mostly distributed between 300 and500mm,and the precipitation is mainly concentrated from June to September.Precipitation is the main driving force of vegetation growth in arid and semi-arid areas.there is a significant correlation between annual vegetation coverage and cumulative precipitation from June to August,and a very significant correlation between monthly vegetation coverage and precipitation.On the whole,during the vegetation growing period,the effect of monthly precipitation on vegetation coverage in the same period was in the following order:semi-mobile sand dunes,semi-fixed sand dunes,fixed sand dunes,combination of farmland and meadow. |
PDF Full Text Request