Spatio-Temporal Estimation of Rice Height Using Time Series Sentinel-1 Images

Rice height, as the fundamental biophysical attribute, is a controlling factor in crop phenology estimation and yield estimation. The aim of this study was to use time series Sentinel-1A images to estimate the spatio-temporal distribution of rice height. In this study, a particle filter (PF) was applied for the real-time estimation of rice height compared with a simplified water cloud model (SWCM) on the basis of rice mapping and transplanting date. It was found that the VH backscatter (sigma(vh)o) can potentially be applied to accurately estimate rice height compared with VV backscatter (sigma(o)(vv)), the sigma(vh)o/sigma(0)(vv) ratio, and the Radar Vegetation Index (RVI, 4* sigma(o)(vh)/(sigma(o)(vh)+sigma(o)(vv))). The results show that the rice height estimation by PF generated a better result with a root-mean-square error (RMSE) equal to 7.36 cm and a determination factor (R-2) of 0.95 compared with SWCM (RMSE = 12.59 cm and R-2 = 0.86). Moreover, rice height in the south and east of the study area was higher than in the north and west. The reason for this is that the south and east are near to the South China Sea, and there are higher temperatures and earlier transplanting. Altogether, our results demonstrate the potential of PF and sigma(o)(vh) to study the spatio-temporal distribution of crop height estimation. As a result, the PF method can contribute greatly to improvements in crop monitoring.

Automated summary

This study aimed to estimate the spatio-temporal distribution of rice height using time series Sentinel-1A images. VH backscatter was found to be potentially accurate for estimating rice height compared to VV backscatter, the VH backscatter to VV backscatter ratio, and the Radar Vegetation Index. The particle filter method generated better results compared to the simplified water cloud model, with higher rice height in the south and east compared to the north and west.