Abandoned land identification in karst mountain area based on time series SAR characteristics at geo-parcels scale

Mapping abandoned land is very important for accurate agricultural management. However, in karst mountainous areas, continuous high-resolution optical images are difficult to obtain in rainy weather, and the land is fragmented, which poses a great challenge for remote sensing monitoring of agriculture activities. In this study, a new method for identifying abandoned land is proposed: firstly, a few Google Earth images are used to transform arable land into accurate vectorized geo-parcels; secondly, a time-series data set was constructed using Sentinel-1A Alpha parameters for 2020 on each farmland geoparcel; thirdly, the semi-variation function (SVF) was used to analyze the spatial-temporal characteristics, then identify abandoned land. The results show: (1) On the basis of accurate spatial information and boundary of farmland land, the SAR time-series dataset reflects the structure and time-series response. The method eventually extracted abandoned land with an accuracy of 80.25%. The problem of remote sensing monitoring in rainy regions and complex surface areas is well-resolved. (2) The spatial heterogeneity of abandoned land is more obvious than that of cultivated land within geo-parcels. The step size for significant changes in the SVF of abandoned land is shorter than that of cultivated land. (3) The SVF time sequence curve presented a strong peak feature when farmland was abandoned. This reveals that the internal spatial structure of abandoned land is more disordered and complex. It showed that time-series variations of spatial structure within cultivated land have broader applications in remote sensing monitoring of agriculture in complex imaging environments.