基于变化和时序信息的耕地撂荒遥感识别方法

Abandoned Cropland Identification Based on Change Detection and Temporal Information from Remote Sensing

摘要:

及时准确获取耕地撂荒分布情况对耕地保护和农业生产管理具有重要意义。针对撂荒识别难、精度有限的难题，本文提出一种融合变化检测和土地覆被分类方法的新框架，以实现山东省南四湖区域撂荒地识别。选取2019—2024年Sentinel-2卫星遥感影像为数据源，引入物候知识改进变化向量分析方法，识别变化区域;在此基础上，利用随机森林算法对变化区域开展土地覆被分类;进而采用时间滑动窗口算法识别撂荒发生的时间和位置。结果表明：本文方法撂荒地识别精度（F1分数）为86.05%～89.89%，优于传统的基于土地覆被分类后比较方法和基于变化向量分析的撂荒识别方法;研究区2019—2024年撂荒面积为604.70 km2，以2019年耕地制图面积为基准，撂荒率为3.87%;撂荒分布呈“东北高、西南低”格局。本文方法在撂荒地识别中表现出较高的准确性和适用性，为推动耕地撂荒相关研究和耕地保护工作提供数据和方法支撑。

Abstract:

Obtaining timely and accurate information on the distribution of abandoned cropland is of crucial significance for cropland protection and agricultural production management. To address the challenges of difficult identification and limited mapping accuracy of abandoned cropland, a framework integrating change detection and land cover classification methods was proposed to identify abandoned cropland in the Nansi Lake region of Shandong Province. Using Sentinel-2 satellite remote sensing imagery from 2019 to 2024 as the primary data source, a phenology-enhanced change vector analysis method was introduced to detect changed areas. On this basis, the random forest algorithm was utilized to perform land cover classification within these changed areas. Subsequently, a temporal sliding window algorithm was adopted to identify the specific timing and location of the abandoned cropland. The results indicated that the F1 score of abandoned cropland using the proposed method stabilized between 86.05% and 89.89%, outperforming that of traditional methods based on land cover post-classification comparison and change vector analysis. From 2019 to 2024, the cumulative area of abandoned cropland in the study region was 604.70 km2, yielding an abandonment rate of 3.87% based on the 2019 cropland baseline map. Spatially, the distribution of abandoned cropland exhibited a distinct pattern of “high in the northeast and low in the southwest.” The framework developed demonstrated high accuracy and applicability for monitoring abandoned cropland, providing crucial data and methodological support for advancing research on cropland abandonment and facilitating cropland protection workflows.