Monitoring Rice Phenology Based On Multi-Temporal RADARSAT-2 Datasets

Vegetation phenology is a cyclical rhythm developed in a long-term mutual adaptation between plant growth and environment.Rice is an extremely important food crop for China and the world.Accurate and timely phenology monitoring is an important step for rice disease control and yield estimation.Rice is more likely to grow in hot and humid areas,where the weather is often cloudy and foggy.Optical satellites can hardly acquire cloud free images in these areas.Microwave remote sensing data provided by synthetic aperture radar?SAR?becomes an important alternative since it can penetrate clouds and fogs,and is nearly immune to bad weather.Space-borne SARs launched by several countries have provided researchers with microwave remote sensing data of various wavelengths,various resolutions,and various polarization modes.The backscattering coefficients and target polarimetric decomposition parameters derived from SAR data are widely used in crop monitoring for the analysis of scattering mechanism and the inversion of biophysical parameters.The rice planting area in Meishan City,Sichuan Province was selected as the research area to carry out rice phenology observation and biophysical parameters measurement.The RADARSAT-2 data was used to extract the quad-polarized backscattering coefficients?HH,VV,HV,and VH?and two kinds of widely used target polarimetric decomposition?Cloude-Pottier decomposition and Freeman-Durden decomposition?parameters.The variation of these parameters and their mathematical combinations during the rice-growing season was analyzed.The sensitivity of the backscattering coefficients or their combinations to rice phenology evolution was explored.The inversion model for rice phenology based on backscattering coefficients was established.The sensitivity of the Cloude-Pottier decomposition and Freeman-Durden decomposition parameters or their combinations to rice phenology evolution was explored.The inversion model for rice phenology combining two kinds of decomposition parameters was established.The correlations between rice leaf area index?LAI?and backscattering coefficients or decomposition parameters were investigated at different phenological phases and the entire growth cycle.The empirical inversion model for rice LAI based on polarimetric decomposition parameters was established.The capability of using backscattering coefficients for rice LAI inversion was also analyzed.?1?The multi-temporal four-polarization backscattering coefficients were extracted for 30 sample sites.The combinations?e.g.,addition,subtraction,multiplication,and ratioing?were performed on the different polarized backscattering coefficients.Backscattering coefficients and their combinations were grouped according to their corresponding rice phenological phases.It was found that VV/VH,HH/VH and HH/VV can be used to segment the four phenological phases.The segmentation thresholds were determined and the inversion model was established.The overall accuracy is 82.8%and the Kappa coefficient is 0.75.?2?The Cloude-Pottier and Freeman-Durden target polarimetric decomposition algorithms were performed on RADARSAT-2 data.The Cloude-Pottier and Freeman-Durden decomposition parameters were extracted for 30 sample sites.Decomposition parameters and their combinations were grouped according to the corresponding phenological phases.It was found that Anisotropy,Vol/Surf and Entropy can be used to segment the four phenological phases.The segmentation thresholds were determined and the inversion model was established.The overall accuracy is 93.1%and the Kappa coefficient is 0.89,which is better than the backscattering coefficients ones.?3?The rice LAI at each phenological phase and the whole growth cycle was correlated with the backscattering coefficients and the target polarimetric decomposition parameters.The results showed that the radar vegetation index based on Freeman-Durden decomposition(RVI_FD)has the highest correlation with the rice LAI for the whole growth cycle,and the empirical inversion model achieved high precision?RMSE=0.95?.The VH/VV based rice LAI inversion model had a slightly lower inversion accuracy?RMSE=1.33?.Based on the multi-temporal quad-pol RADARSAT-2 dataset,the responses of backscattering coefficients and target polarimetric decomposition parameters to rice phenology development were studied.The biophysical structure and electromagnetic scattering characteristics of rice plants in different phenological phases were analyzed.The feasibility of using C-band SAR data to retrieve rice phenology and LAI was investigated.The imaging problem caused by cloudy and foggy weather was solved.The target polarimetric decomposition parameters are better than the backscatter coefficients in rice phenology and LAI inversion,because the target polarimetric decomposition parameters contain the phase information of the radar echo.The inversion models based on backscattering coefficients are less effective than that of decomposition parameters,but it is simple and does not rely on quad-pol data.At present,Sentinel-1A/B provides free C-band single/dual-polarized SAR data.As a result,the backscattering coefficients based models have great potential for rice monitoring application.