Temporal backscatter characterisation of ratoon rice crops based on Sentinel-1 intensity data

Ratooning, the cropping practice of harvesting a second crop from the stubbles of the primary harvest, is gaining renewed popularity as a resource-efficient alternative to increase rice production. Although current remote sensing-based rice monitoring systems have considered rice ratooning systems, nothing is known about the temporal backscatter response of ratoon rice, which is necessary for accurate rice ratooning detection in cloud-pervasive regions. Using backscatter time series from Sentinel-1A/B data, for the first time, we characterized the temporal backscatter signatures of ratoon rice crops in four features (VV and VH polarizations, the ratio of VH/VV, and the radar vegetation index (RVI)) to determine the optimal period and SAR features for main and ratoon rice discrimination. We also investigated the influence of harvesting methods on the backscatter of stubbles and the difference in backscatter between ratoon crops in irrigated and rainfed rice. We obtained data covering three growing seasons (2018–19), rice field boundaries and farmer interview data on cropping practices in the Philippines. The backscatter differences were assessed using the Mann–Whitney <i>U</i> and the Kruskal – Wallis test, while the classification was performed using partial least squares discriminant analysis (PLS-DA). We found that the observation during the peak of the growing season could best distinguish main and ratoon rice, specifically in the reproductive (VH, <i>p</i> = .010) and ripening phase (VH/VV, <i>p</i> = .089 and RVI, <i>p</i> = .089). The PLS-DA model at the reproductive phase performed better, with an overall accuracy of 68% (AUC = 0.70) than the model from the ripening phase (OA = 60%, AUC = 0.64). The backscatter of stubbles from mechanically harvested fields is not significantly different from that of manually harvested fields. We also found no significant backscatter difference in ratoon crops across different water managements throughout all growth phases. This study demonstrates the potential of SAR Sentinel-1 time series data to determine periods and SAR features for optimal main and ratoon rice discrimination, which offers advantages for future remote sensing-based rice ratooning mapping and rice production estimation.