Assessment of interferometric coherence contribution to corn and soybean mapping with Sentinel-1 data time series

Abstract

This work aims to evaluate the added value of interferometric coherence to backscatter information of Synthetic Aperture Radar (SAR) systems for soybean and corn mapping. First, SAR response to crop growth, and then accuracies for the classification using a combination of SAR variables were evaluated for scenarios that employ in-season or the entire season time series. Results showed that: i) using a single feature, the backscatter at vertical-horizontal (VH) polarization would be the most suitable variable; ii) the complementarity of coherence to single backscatter at vertical-vertical (VV) polarization was demonstrated, adding a significant contribution to late sown corns differentiation and iii) the combination of VV and VH backscatter would be the preferable variables for the proposed classification. In this case, the adding of coherence did not show a significant accuracy improvement, while a high computational cost is required. Finally, high general accuracies (until 90%) for early-season maps were achieved.