Crop Classification by Multitemporal C- and L-Band Single- and Dual-Polarization and Fully Polarimetric SAR

Abstract

Classification of crops and other land cover types is an important application of both optical/infrared and synthetic aperture radar (SAR) satellite data. It is already an import application of present satellite systems, as it will be for planned missions, such as the Sentinels. A multitemporal data set from the Danish airborne polarimetric EMISAR has been used to assess the performance of different polarization modes for crop classification. Both C- and L-band SAR data were acquired simultaneously over the Foulum agricultural test site in Denmark on a monthly basis during the growing season. Single- and dual-polarization and fully polarimetric data have been used in the analysis. The best result for a single-frequency system was a 20%-22% classification error, and the results for C-band and for L-band were very similar. The best result was obtained at C-band using the VVXP polarization combination (the dual-polarization mode, where the VV-channel and the cross-polarized channel have been combined) and at L-band using the fully polarimetric mode with the Hoekman and Vissers classifier. The best result for the combination of the C- and L-bands was 16%. In this case also, the VVXP polarization combination performed best. There is a tradeoff between the polarimetric information and the multitemporal information, where the best overall results are obtained using the multitemporal information.