Rice Yield Estimation At Field Scale By Using High-Frequency SAR At Panicle Stage

China is a large agricultural country and our government attaches great importance to the issue of food security.As rice is a major food crop in China,timely and accurate rice monitoring and yield estimation are of great significance to the country,government and agricultural producers.Historical monitoring methods are time-consuming,laborious and time-inefficient,the accuracy of the data is easily influenced by human activities,making it difficult to monitor dynamically.With the development of satellite remote sensing technology,rice monitoring has become more convenient.However,because rice is planted in tropical and subtropical regions,the conditions are often cloudy or rainy for a large part of the growth cycle making it difficult to obtain clear optical remote sensing data.Therefore,the use of radar remote sensing with its all-day,all-weather capabilities is the best means of carrying out effective rice monitoring and making estimates of rice yield.Radar remote sensing can obtain the radar response characteristics of the rice canopy under different polarization modes,which can better provide information about the rice canopy water content,morphological structure and growth situation it also strong complementarity with optical remote sensing.Therefore,radar remote sensing provides means of establishing a reliable and stable remote sensing-based rice monitoring system.The traditional low-frequency C and L bands can penetrate the ear layer,so that the echo contains a large number of stem layers,and even underlying surface information,which increases the modeling difficulty.The research on rice growth monitoring and mapping is concentrated.The study found that the backscattering coefficient of high-frequency microwaves is highly correlated with the ear weight,thus providing a basis for the direct inversion of rice ear biomass based on high frequency SAR images.In this paper,high-frequency SAR data is used to study the rice yield estimation methods on the field scale.Taking the rice at the ear stage of the grain production functional area of Xinshe Village,Xin’an Town,Deqing County,Zhejiang Province as the experimental object,the TerraSAR-X VV polarimetric SAR data in 2018,the UAV Ku band HH polarimetric SAR data in 2019,the UAV optical RGB data and rice growth parameter data collected synchronously with the sky,extracting image backscattering information and vegetation indices information,optimizing the traditional water cloud model,establishing the modified water cloud model based on spike layer,and improving the accuracy of rice yield estimation.It provides a feasible practical method for monitoring of rice growth and yield estimation at the field scale.The main research contents include:(1)According to the high-resolution characteristics of TerraSAR-X spotlight mode data,extracting the backscattering information of SAR images based on the field scale,the statistical regression empirical model related to rice panicle weight and the empirical model of ear grains microstructure were established respectively.Analysis of X-band VV polarimetric SAR radar backscattering based on empirical model to invert rice yield.The X-band VV single-polarization radar backscattering information based on the empirical model was analyzed to invert rice yield.Comparing the inversion results with the rice yields are estimated using a field survey together with the government statistical survey data for Zhejiang Province,it is found that the inversion results of the ear grains microstructure empirical model are superior to the inversion results of the statistical regression empirical model.Among them,the root-mean-square error(RMSE)of the ear grains microstructure model established by the measured data on September 18,2018 is the smallest.Compared with the field survey estimated yield and the provincial government statistical survey data:45.13kg/mu,116.1 lkg/mu,the highest accuracy of estimated yield(P)is 93.85%,82.46%,respectively.The difference between the model inversion yield and the measured yield is smaller.The two empirical models are simple and easy to operate.The prediction of rice yield at the ear stage is completed.(2)In order to improve the accuracy of yield estimation,using the traditional water-cloud model as the starting point and taking the ear layer of rice as the cut-in point to optimize the traditional water-cloud model,we established the modified water-cloud model based on the ear layer and combined with TerraSAR-X data to predict the yield of rice in the same plot,the inversion results are significantly higher than traditional empirical model.Similarly,using the Ku-band SAR data of the UAV,the inversion accuracy of the panicle wet weight and ear grain number based on the modified water-cloud model of the ear layer reached 84%and 79%,respectively.The change trend of the inversion value and the measured value is basically consistent.Therefore,the modified water-cloud model based on the panicle layer can complete the effective prediction of rice yield at the panicle stage at the field scale.(3)In the rice simultaneous experiment,the optical RGB image of the UAV was obtained and six kinds of UAV optical RGB vegetation indices were extracted.The experimental results showed that the green-red ratio index(GRRI)had significant correlation with ear weight.To establish the regression model between vegetation indices and yield to estimated output.Comparing and analyzing the yield of rice with vegetation indices inversion and yield based on the modified water-cloud model of panicle layer,it is further verified that the modified water-cloud model of panicle layer is feasible and effective for estimating rice yield at panicle stage.This paper provides a very effective technical means and feasible practical methods for inversion of rice biomass at the field-scale with high-frequency SAR.It also provides important data references and technical solutions for rice remote sensing monitoring and yield prediction in cloudy and rainy areas.