西南农业学报2025，Vol.38Issue(10)：2041-2051,11.DOI:10.16213/j.cnki.scjas.2025.10.001
基于机器学习与时序遥感的撂荒耕地识别与分布特征
Identification and distribution characteristics of abandoned farmland based on machine learning and time-series remote sensing:A case study of Bishan district,Chongqing city
摘要
Abstract
[Objective]Monitor the abandonment of cultivated land in real time and with precision,providing scientific insights and decision-making support for tracking regional agricultural production dynamics and ensuring national food security.[Method]Selecting Bishan district in Chongqing as the research sample,based on the GEE(Google Earth Engine)cloud platform,we integrated high-resolution remote sensing data such as Sentinel-2 and Landsat 8 to construct a time-series dataset of Normalized Difference Vegetation Index(NDVI).Based on this,three machine learning algorithms,namely Random Forest(RF),Support Vector Machine(SVM)and Classification And Regression Tree(CART),were employed to classify and evaluate the accuracy of the imagery,aiming to precisely extract information on abandoned farmland in Bishan district.[Result](i)The Random Forest algorithm exhibited optimal classification performance,achieving an overall accuracy of 89.28％and Kappa coefficient of 0.84,fully verifying the high consistency between the extraction results and the actual situation.(ⅱ)From a time series perspective,the area of abandoned farmland in the study area revealed an overall decreasing trend,with the abandonment rate had dropped to 13.25％by 2022.In terms of spatial distribution,abandoned farmland was mostly scattered along the edges of rivers.(ⅲ)Landscape pattern analysis indicated increasing spatial complexity and boundary fragmentation of abandoned parcels,with no large-scale con-tiguous abandonment observed.(iv)Topographical factors had a significant impact on the distribution of abandoned farmland.Specifically,o-ver 90％of the abandoned farmland was located within an elevation range of 100-700 meters.Notably,within the elevation range of 400-700 meters and a slope of 0°-15°,both the area and rate of abandonment showed a continuous downward trend.In addition,the distribution index of abandoned farmland continued to decrease across different terrain level grades.Terrain grades 1 and 2 areas were the dominant distri-bution areas for abandoned farmland,which is consistent with the overall trend of abandonment.[Conclusion]The study confirms that integra-ting machine learning technology with time-series remote sensing data can effectively enhance the identification accuracy of abandoned farm-land.The distribution characteristics of abandoned farmland in Bishan district exhibit significant terrain dependence,providing a strong scien-tific basis and practical guidance for the monitoring and management of abandoned farmland in the mountainous areas of Southwest China.关键词
撂荒耕地/机器学习/时间序列NDVI/遥感/璧山区Key words
Abandoned farmland/Machine learning/Time-Series NDVI/Remote sensing/Bishan district分类
管理科学引用本文复制引用
曾誉彤,周启刚,张晓媛,李辉,李昕燃..基于机器学习与时序遥感的撂荒耕地识别与分布特征[J].西南农业学报,2025,38(10):2041-2051,11.基金项目
国家社会科学基金项目(22XJY006) （22XJY006）
重庆市教育委员会科技项目(KJQN202302105) （KJQN202302105）
重庆市教育委员会人文社科项目(23SKGH400) （23SKGH400）
