Characteristics Analysis and Classification of Crop Harvest Patterns by Exploiting High-Frequency MultiPolarization SAR Data

At harvest season, crops are often harvested using various methods at different times. Mapping and monitoring of the patterns of croplands during the harvest period provide information for farmers to help guide the harvest practices that are time critical and to support early warning of threats to food security. This study discusses the feasibility of high-frequency (C/X) polarimetric synthetic aperture radar (PolSAR) for the discrimination of crop patterns during harvest. The polarimetric signals gathered from a farmland area during harvest in Inner Mongolia, China, have been evaluated to investigate the properties of different harvest patterns by using the fully polarimetric Radarsat-2 and dual-pol TerraSAR-X images. A set of polarimetric parameters were derived from the datasets to interpret the radar signatures. The statistics show the sensitivity of the polarimetric parameters to the properties of the harvest patterns. The crop type, biomass, water content held by plants, crop swath direction, and crop state make a large contribution to the fluctuation of the polarimetric scattering characteristics. By exploring the polarimetric characteristics across different harvest patterns, a new method of mapping the harvest state is proposed by utilizing the decision tree algorithm. In the proposed method, GIS data are exploited to avoid the confusion of similar harvest patterns for different species. The harvest pattern mapping results by using the multipolarimetric data acquired over the study area in different years, demonstrate the feasibility and potential of polarimetric data of short wavelength for harvest pattern monitoring during harvest.