An orchard mapping index and mapping algorithm coupling orchard phenology and green-holding characteristics from time-series sentinel-2 images

Abstract

Agroforestry crops such as apples, peaches and pears are horticultural crops, which are an important part of modern agriculture and are of great economic and social importance. Accurate crop data at large scales (e.g., regional) are critical for effective agricultural management and resource regulation. However, existing orchard statistics, survey data, and expert knowledge are often lagging and of low confidence, lacking detailed data on the spatial distribution of orchards. The sparse distribution and indefinite characteristics of orchards compared to field crops, as well as the large intra-class variance of fruit tree spectra, make large-scale mapping of orchards a major limitation and huge challenge. To address these challenges, we developed an orchard mapping index (OMI) based on the phenology and green-holding characteristics of fruit trees, and automated orchard mapping algorithm using sentinel-2 time-series imagery and the Google Earth Engine platform (GEE). Fruit trees have unique phenological and greening characteristics: fruit tree canopies turn green earlier, turn yellow later, and have a long greenness saturation time in annual growth cycles. The proposed OMI index significantly captures the difference in green-holding between orchards and non-orchards [1.5*Interquartile Range (IQR): 0.72–39.5 for orchards, 0.10–3.36 for non-orchards]. The mapping algorithm successfully mapped 10 m-resolution orchard maps in the Loess Plateau region of China from 2020 to 2022, with an overall accuracy of 89.95–93.51 % and a kappa of 0.80–0.87. We have additionally identified that the implementation of a fine-grained agricultural plantation zoning mapping strategy exhibits the potential to enhance the performance of orchard mapping. Our study demonstrated the potential of a phenology-based approach, sentinel image data, and the GEE platform for orchard mapping, and for the first time developed a large-scale map of orchards in the Loess Plateau region of China. This study not only fills the gap of large-scale orchard mapping algorithm and products but also provides valuable spatial information for fruit tree flowering prediction, disease prevention and yield prediction.