Estimation Of Canopy Water Content Based On Sentinel-1 And Sentinel-2 In Wheat

Water is the main component of plants;it has a direct effect on the physiological and biochemical processes and morphological structure of plants,on their yield and quality.Real-time and accurate monitoring of crop water status can offer effective guide for crop fertilizer and water management,which is of great significance for agricultural production.At present,the methods for estimating canopy water content(CWC)based on optical images are mainly the empirical model and the physical model.The physical model method can realize the fast inversion of physical model parameters by optimization,Look-up Table(LUT)and machine learning.However,the extrapolation and stability of the empirical model method are poor,and the efficiency of Look-up Table is low;which is disadvantageous in the estimation of vegetation moisture on a regional scale.Compared with Look-up Table,the machine-learning method is more efficient and fast.On the other hand,its Synthetic Aperture Radar(SAR)can acquire multi-polarization,multi-incidence-angle,multi-frequency,multi-time and other multi-dimensional data,which provides abundant information for monitoring crop growth and becomes a powerful supplement to optical remote sensing.Therefore,in this study,field investigation experiments were conducted in the Xinghua experimental area in 2019,using the Sentinel-2multispectral satellite image with red-edge bands and the Sentinel-1 SAR satellite image,at the same time;and the data of water content in wheat canopy were obtained.In order to improve the precision of wheat canopy water content estimation,a hybrid method based on optical image was proposed in here,and the estimation of canopy water content of wheat on a regional scale was studied.In order to solve the problem of lack of sufficient sample data,the PROSAIL-5B physical model was used here to simulate the data set.The models for estimating the canopy water content of wheat were constructed based on Neural network(NN),Gauss Process Regression(GPR)and Kernel Ridge Regression(KRR).In order to reduce the difference between the satellite spectral reflectance and the simulated spectral reflectance generated by the physical model,we added Gauss noise to the simulated spectral reflectance data set.The results showed that the three wheat CWC estimation models all produced better effect when Gauss noiseη(0,0.06~2)with a standard deviation of 0.06 and a mean of 0 was added to the simulated spectral reflectance data set.Of the three models,the model based on PROSAIL and GPR presented the highest precision(NN:Rv~2=0.52,RMSE=0.0105 g/cm~2,RRMSE=26.9%;KRR:Rv~2=0.47,RMSE=0.099 g/cm~2,RRMSE=24.3%;and GPR:Rv~2=0.50,RMSE=0.096 g/cm~2,RRMSE=22.2%).In addition,this method was compared with the canopy water content(CWC)produced by Snap Toolbox(R~2,RMSE and RRMSE were 0.5,0.027 g/cm~2 and 44.1%,respectively).And the result was that that the hybird model method proposed in this study for a single wheat crop exhibited a higher estimation accuracy for the water content in the wheat canopy.In order to solve the problem of spectral saturation and the limitation of obtaining vegetation information,the Sentinel-1 satellite image and the Sentienl-2 satellite image were used in this study to monitor the growth of crops for the purpose of enhancing the precision of estimating canopy water content of wheat;and a combination index(SAR-Optical index)was constructed by combining the radar index and the optical index.The results showed that four composite indexes(SAR-Optical)performed well using the Simple linear regression methodand that the wheat CWC estimation model constructed on the basis of VH*NDWI,VH*GVMI,VV*NDWI,VV*GVMI produced Rc~2,Rv~2,RMSE and RRMSE of 0.58,0.55,0.0097 g/cm~2,22.81%;0.57,0.55,0.0097 g/cm~2,22.93%;0.57,0.55,0.0097 g/cm~2,22.99%;and 0.56,0.54,0.0098 g/cm~2,23.18%,respectively.Two composite indexes(SAR-Optical)performed well using the non-linear regression method,and the Wheat CWC estimation model constructed based on VH*NDWI and VH*SRWI presented Rc~2,Rv~2,RMSE,RRMSE of 0.58,0.57,0.0087 g/cm~2 and 20.82%;and 0.56,0.57,0.0090 g/cm~2,21.24%,respectively.The above results indicated that the collaboration of the Sentinel-1 and Sentinel-2 data could integrate the advantages of the optical satellite and the radar satellite to some extent,and provide more abundant and more accurate information for crop growth monitoring.