Phenology-based cropland retirement remote sensing model: a case study in Yan'an, Loess Plateau, China

Cropland retirement is a widespread phenomenon across the world. The conversion of inefficient cropland to forest or grassland is a policy aimed at restoring ecology, improving the environment, and promoting economic development. However, in most developing countries, the results of cropland retirement and land restoration are characterized by spatial fragmentation, and there are significant temporal differences as a result of poor agricultural intensification, human interference, and regional environmental differences. This substantially increases the difficulty of information extraction and reduces the extraction accuracy of remote sensing methods. In this paper, we developed a new phenology-based cropland retirement remote sensing (PCRRS) model to detect the extent and timing of cropland retirement. Considering the characteristic growth of crops, the normalized difference vegetation index (NDVI), at the start, middle, and end of the growth cycle, is the phenological metric to distinguish cropland from other vegetation types. In addition, the interannual variation of phenological metrics are significant after cropland retirement, which is the key to effectively identify retired cropland. High-resolution Google Earth images were used to verify the accuracy of the algorithm. The results suggested that the overall accuracy of our algorithm exceeded 85%, and was more suitable for sloping cropland. In comparison with other cropland retirement extraction methods, the PCRRS model had high sensitivity and stability. We found it was common for sloping cropland to be retired earlier, and we also identified the existing inter-planting phenomena between crops and shrubs in areas with gentle slopes. Overall, this study provided a basis for understanding the drivers of cropland retirement and evaluating their environmental effects.