Improved Mapping of Perennial Crop Types Based on Patterns of Intra-Annual Variation in Land Surface Components

Perennial crops hold significant importance in global agricultural markets, local agricultural economies, poverty reduction, and biodiversity conservation. Accurate spatial distribution data of different perennial crops are crucial for local agricultural management, crop yield prediction, and sustainable agricultural development. However, precisely obtaining the spatial distribution of different perennial crops using remote sensing data, especially in regions with complex cropping patterns, remains a challenge. This study, focusing on Yantai City, China, and three counties in California, USA, aims to develop a method suitable for both smallholder and intensive production systems. On the Google Earth Engine (GEE) platform, we applied linear spectral mixture analysis (LSMA) to transform Sentinel-2 time-series data (2020-2022) from the original spectral space into a unified endmember space, including photosynthetic vegetation, non-photosynthetic vegetation, soil, and shadow, thereby characterizing the time-series land surface component information. Subsequently, based on the time-series endmembers data, we quantified seven sets of harmonic features and five phenological features. Relative phases were employed as phase indicators for the harmonic features. These seven harmonic features represent the intra-annual patterns of land surface component changes and inter-component interactions, while the phenological features quantify the timing of phenological events. Our findings reveal significant variations in the intra-annual patterns of land surface component changes among different perennial crops, attributable to differences in phenology and phenology-associated human management. Building on this, the study achieved precise mapping of different perennial crops, even in areas with complex cropping patterns. The overall classification accuracy for perennial crops in Yantai City and the three counties in California was 90.3% and 94.8%, respectively, with Kappa coefficients of 89.2% and 93.9%. Utilizing intra-annual time-series land surface component information for extracting the spatial distribution of perennial crops demonstrated advantages over traditional optical indices. This work provides a method that is applicable to both smallholder and intensive production systems, enabling precise mapping of perennial crop types. It represents an important step towards achieving large-scale mapping of perennial crop types.