| The abandonment of arable land has a great impact on social and economic development,ecological environmental protection and food security.It is of great significance to extract abandoned farmland and grasp the distribution of abandoned farmland timely for the country to formulate farmland protection policies.Initially,the abandoned farmland was extracted mainly through field investigation and manual analysis,which was inefficient and difficult to complete the survey of large areas.With the development of remote sensing technology,the spatio-temporal resolution of remote sensing image is getting higher and higher,which provides the data basis for long-term and large-scale research.In recent years,due to the development of computer technologies such as machine learning,image processing,parallel computing and cloud services,rapid interpretation,classification and comparison of a large number of remote sensing images have been realized,making it possible to rapidly extract abandoned farmland.In order to improve the extraction efficiency of abandoned farmland and solve the problem of long time scale and large area extraction,this paper relies on Google Earth Engine cloud computing platform,Pycharm,Matlab,Arcgis and other tools,using Python,Javascript and Matlab language,using Landsat remote sensing image data from1999 to 2021 as basic data,this paper studies the abandonment of farmland in Yunnan Province in recent 20 years.The main research contents and conclusions are as follows:(1)Land use type classification based on Random forest in Yunnan Province.According to the need of extraction of abandoned farmland and the actual situation of Yunnan Province,7 types of land use were set up,including building land,cultivated land,forest land,grassland,glacier and permanent snow,bare land and water body.A total of 14735 unchanged sample points were collected respectively.Based on the data of remote sensing image and digital elevation model(DEM),39 features in five categories including spectral features,textural features,topographic features,exponential features and tasseled hat transform features were calculated.In order to improve the classification accuracy and reduce the amount of calculation,the Relief F algorithm combined with single feature classification accuracy verification was used to optimize the above features,and the optimal special collection was determined to participate in the classification.Finally,the land use classification data of Yunnan Province from 1999 to 2021 were obtained by random forest method,and the land use type map was drawn.The results showed that the overall classification accuracy over the years was between 90.08%-94.17%,and the Kappa coefficient was between 0.8621-0.9202.The overall classification effect was good,meeting the requirements of subsequent extraction of abandoned farmland.(2)Extraction of abandoned farmland based on land use change reclassification in Yunnan Province.The land use types before and after years were compared,and the mapping table established according to the definition of abandoned land was reclassified into five types: no change in land use type,no change involved in cultivated land,stop farming,occupied land,reclamation or open tillage,that is,land use type change.After that,the changes of land use types before and after were compared and classified into four types: abandoned land,fallow land,reclamation land and no relation to abandoned land.After morphological processing,the distribution map of abandoned land in Yunnan Province from 2000 to 2020 was obtained.It was found that the abandoned farmland rate in Yunnan Province fluctuated in the range of 6%-14% from2000 to 2020,and the abandoned farmland was mainly distributed in the mountain area of Zhaotong,the karst landform area of Honghe and Wenshan,and the tropical mountainous area of southern and southwestern Yunnan.(3)Prediction of abandoned area in Yunnan Province based on Markov algorithm.By improving the Markov prediction model commonly used in LUCC research,combining with the extraction method of abandoned arable land,and using the sliding average method to improve it,the area of abandoned land was calculated by the predicted area of various land use types and the calculated probability matrix of land use type transfer,and then corrected by the correction formula.The predicted area of abandoned farmland in the future was obtained.The results showed that all types of land area remained stable on the whole before 2025 in Yunnan Province,and the area of abandoned farmland increased slightly compared with that in 2018-2020. |
