MPRI: a multi-phenology index for rapeseed mapping using time-series Sentinel-1 SAR data

Accurate and timely rapeseed mapping is critical for assuring national food security and supporting agricultural regulation. Time series synthetic aperture radar (SAR) data is widely used to monitor the growth of rapeseeds and identify rapeseed distribution in cloudy and rainy regions. However, the spatial and temporal variations in plantation conditions keep presenting a challenge for SAR-based rapeseed mapping over large areas. Additionally, the temporal dynamic patterns and mechanism of SAR scattering could be investigated to enhance the mapping of rapeseed. Therefore, using time-series Sentinel-1 SAR data, this study proposed a general multi-phenology rapeseed index (MPRI) for automatic rapeseed mapping. Central to this method is establishing the linkage between time-series SAR backscattering (including intensity and polarimetric) patterns with rapeseed phenology. First, the rapid growth, flowering, and maturity stages of rapeseed were identified using a pseudo scattering-type parameter that was derived from polarimetric decomposition on Sentinel-1 SAR data. Second, the multi-phenology rapeseed index (MPRI) was developed by incorporating three stage-specific indices. Third, an optimal threshold was applied on the MPRI probability maps to generate binary masks of rapeseed distribution. The proposed MPRI approach is further evaluated by mapping rapeseed across three sites with varying planting conditions in China's Yangtze River Basin. Comparative experiments were conducted against with three available standard methods, including the RSG-OC, CARM, and SARM methods. Results show that the MPRI approach achieved the best performance for rapeseed mapping, with an overall accuracy and an F1 score above 0.90 at all three sites, which exceed these comparative methods by 0.02 ~ 0.10 in overall accuracy and 0.03 ~ 0.12 in the F1 score. Further discussion also reveals that by explicitly incorporating the temporal dynamics of microwave scattering throughout crop development, the proposed index offers a general and adaptable solution for large-scale automatic rapeseed mapping using time-series SAR data.