Extraction of multiple cropping information at the Sub-pixel scale based on phenology and MODIS NDVI time-series: a case study in Henan Province, China

Understanding and mapping multiple cropping patterns(MCP) is important to meet global food demand. Due to the random crops in smallholder farming, it is challenging to obtain phenology and location of MCP. Phenology exhibits stable rhythmic variation, and vegetation index(VI) forms time-series profiles reflecting rhythmic variations. Since most crops have different growth cycles, time-series are used as a fingerprint to distinguish multiple crops in a specific area. Moderate Resolution Imaging Spectroradiometer(MODIS) normalized-VI(NDVI) data were used to construct time-series, but different land cover types were often mixed in a 250 m MODIS pixel. Time-series were explored as input of spectral unmixing algorithms that could capture the spatial distribution of MCP and subpixels at large areas. Time-series profiles of MCP were extracted based on N-finder algorithm(N-FINDR). Considering the mixed-pixel, fully constrained least-squares(FCLS) was used to extract the spatial distribution of MCP. The accuracy of the algorithm was verified using high spatial resolution images with an overall accuracy of 85.65%. The results show that the algorithm simultaneously extracts MCP and spatial distributions at large scales.