Cross-frequency SAR indices for rice mapping using multi-frequency and full-polarization data

Timely and accurate mapping of rice planting distribution is essential for early yield estimation, food security assessment, agricultural policy-making, water resource management, climate change impact analysis, and achieving the United Nations Sustainable Development Goals 2030, particularly in cloudy and rainy regions. Optical remote sensing is significantly constrained by cloud cover, while single-frequency SAR faces difficulties in complex farmland mosaics. Although multi-frequency and full-polarization SAR data are becoming increasingly available, their potential for rice mapping has not been fully exploited. In this study, we developed a set of Cross-frequency SAR-based Rice Indices (CFSRIs) using high-resolution multi-frequency (Ka/X/C/S/L) and full-polarization SAR data acquired from the Chinese Aeronautic Remote Sensing System (CARSS). The indices were constructed through a systematic analysis of scattering characteristics of rice and easily confused vegetation across multiple frequencies and polarizations. Their discriminative ability was assessed through an integrated qualitative and quantitative evaluation, enabling the optimal selection of CFSRI. Rice fields were then extracted using threshold segmentation, and the results were compared with single-frequency SAR-based indices to validate the performance of CFSRI. The effects of different frequency/polarization combinations, as well as incidence angles, on rice separability were also discussed. Results show that rice exhibits stronger separability from easily confused vegetation in high-frequency bands (Ka/X) and from water in mid-frequency bands (C/S). Among the proposed indices, the cross-frequency Normalized Difference Index NDI(C_HH, Ka_HH) achieved the highest classification accuracy, with an overall accuracy of 91.07%, outperforming traditional single-frequency indices such as RSI and RVI. The combination of C-band with Ka/X-band yielded the best rice mapping results, with polarization combinations ranked HH-HH > HH-VH > HH-HV. Moreover, the proposed index remained stable across a wide incidence angle range (38°–53°).