Rice Phenology Estimation And Parameter Retrieval Based On Polarimetric Synthetic Aperture Radar(SAR)

Rice is one of the most important grains in the world,of which the production is essential for the global food security,and the social prosperity.Thus it is urgent for the government and the farm managers to monitor the rice growth timely and accurately.Recently,remote sensing(including the optical and radar remote sensing)has been used popularly by the farm managers for monitoring rice growth,due to the advantages of large-cover and frequent revisits.Considering most paddy rice grows in warm,humid environments distributed in semi-/tropical area with heavy cloud cover and rainfall,it is difficult to acquire regular optical remote sensing data in rice growing regions.Thus Synthetic Aperture Radar(SAR),with the advantages of all-weather,day and night imaging,is the effective method for rice growth monitoring and yield prediction.With the development of SAR,the full-/compact-polarimetric SAR sensors emerge based on the single-and multi-polarimetric SAR sensors.The full-/compact-polarimetric SAR can attain the scattering characteristics of rice canopy in different polarization channels,including the amplitude and phase information,which is beneficial for monitoring the moisture,morphology structure and growth of rice canopy.Thus the exploration of the potential of polarimetric SAR in the rice growth monitoring is very important,including the improvement of methods,and the analysis about the advantages and limitations of polarimetric SAR in the rice growth monitoring.This paper aimed at rice phenology estimation and growth parameter retrieval,based on multi-temporal full-/compact-polarization SAR data and the multi-spectral optical remote sensing as the subsidiary data.Five main research aspects are included as follows.1)Based on multi-temporal compact polarimetric(CP)SAR data,the transplanted indica rice field(TRF)and the direct-sown japonica rice field(DRF)were classified with accuracies of 85% above.And seven phenological stages of two types of rice field were estimated based on the CP SAR data,respectively,without any auxiliary external information.2)A novel feature selection algorithm was proposed based on the Monte Carlo random experiments and correlation limitation(MCCL).Then the optimal feature subset was attained,based on which the phenology estimation was done automatically with the overall accuracy of 86.59%.3)Some key steps were discussed in the rice phenology estimation.First,we discussed the optimal scheme for rice phenology estimation,based on the multitemporal optical and PolSAR data.Then we found that it is significant to consider the differences between the TRF and DRF among the processes of the rice phenology estimation.The overall accuracy of rice phenology estimation was above 85% when the two types of rice field are considered repectively,which was 16% higher than that when the two types of rice field are considered as a whole.In addition,the estimated accuracy was lower than 80% when just the optical data or the polarimetric SAR data was used.The main reason is the optical Vegetation Indices(VIs)are not sensitive to the variation of rice canopy between the period of the tillering stage and the dough stage,and the polarimetric signatures are not sensitive to the variation of rice canopy from the milk stage to the mature stage.4)The improved polarimetric decomposition method was proposed,which can decrease the overestimation of the volume scattering and the negative pixels effectively.During different phenology,the scattering mechanism of rice canopy is different.The general volume scattering model(GVSM)in the improved polarimetric decomposition can denote the scattering mechanism of rice canopy during different phenological stages,through considering the variation of the ratio between the backscattering coefficients of HH and VV polarimetric channels.5)The Modified Water Cloud Model(MWCM)was proposed,which considered the horizontal heterogeneityof the rice canopy,and the double-bounce scattering.Then we built the scheme for interacting the MWCM and the polarimetric decomposition,and fulfilled the rice parameter retrieval during the whole rice growth season.Compared with the traditional Water Cloud Model(WCM),the proposed method in our study can be used to get much more accurate results of rice parameter.Future work focuses on two main aspects.First,it is noted that many SAR signatures are functions of the incidence angle.In future work,the performance of the phenological estimation for different incidence angles will be analyzed and compared,and then the most appropriate imaging mode(incidence angle)for identifying rice phenology using SAR data will be selected.Second,the Modified Water Cloud Model(MWCM)in this study is aimed to the transplanted rice field,and it is necessary to consider the improvements of the retrieval model for the direct-sown rice field.