Monitoring rice leaf area index using time-series SAR data

Abstract: Rice is one of the most important food crops in China, so timely obtaining accurate rice growth information in regional scale is highly significant for crop management and decision making. However, there are plenty of rain and dense cloud cover in rice growth season, that makes it difficult to monitor rice paddy information by optical remote sensing data. It is an accepted fact that Synthetic Aperture Radar (SAR) data is a suitable alternative for crop monitoring in cloud-prone and raining area. In this paper, we investigated the potential of Radarsat-2 SAR data for rice paddy growth monitoring at regional scale. Four C-band SAR backscattering coefficients (VH, VV, HH, HV) and HH/VV ratio backscattering coefficient were employed for setting up the relationship with leaf area index (LAI) in rice growing season. The test site located in Leizhou (20°52′N, 110°05′E), Guangdong province. Four Radarsat-2 products in fine quad polarization mode were acquired during the critical rice growth stage. In situ measurement in year 2013 was also made concurrently with the satellite pass. 25 plots were selected for measuring the rice growth parameters, such as LAI, plant height, sowing date, etc. The backscattering coefficients with multi-incidence angle were extracted from multi-temporal SAR images and then normalized to same angle. Firstly, the study analyzed the temporal behavior of microwave backscattering coefficients (VH, VV, HH, HV, HH/VV) and LAI in rice growth season. Secondly, correlation analyses between the backscattering coefficients in different polarization and LAI were also carried out in vegetative stage, reproductive stage and whole growth period respectively, then picking the polarization and growth stage corresponding with high correlation coefficient which was above 0.8 to build water cloud model and evaluate the performance of each model. Finally, based on the previous result, the LAI distribution map in time domain was generated using the best model of entire growth period. The results showed that, (1) In flat area, the correlation coefficient in different stage from high to low is: vegetative stage, whole growth period, reproductive stage. There is a positive correlation between HH, HV (VH) and LAI, a negative correlation between VV, HH/VV and LAI. The correlation coefficients of VV, HH/VV and LAI were both above 0.8; (2) The VV, HH/VV water cloud models (R2=0.77, R2=0.87 respectively) in vegetative stage performed better than in full rice growth stage (R2=0.73, R2=0.8 respectively); (3) The better prediction model was applied in multi-temporal SAR image for computing LAI value in rice area. The LAI distribution map of each single date can point out the rice growth condition in different area, while multi-temporal LAI distribution maps in rice growth season can point out the LAI changes in the same region. In conclusion, the study proved the potential of timely rice monitoring by multi-temporal FQ mode SAR data in regional scale, and also provided a reliable approach for rice LAI prediction. As the development of precision agriculture, C-band SAR data can be used in quantitative rice crop monitoring and it will become an increasingly important data source.