Estimation Of Summer Maize Yield By Multispectral Remote Sensing With UAV And Crop Model Assimilation

With the progress of science and technology,China has gradually transformed from traditional agriculture to intelligent agriculture.Timely and accurate acquisition of crop growth,yield and other information is the basis for the development of smart agriculture and agricultural big data.Accurate prediction of maize yield is very important for agricultural production management,agricultural policy making and food security in China.In this study,an assimilation model for maize yield estimation was established based on UAV remote sensing data and crop model.The assimilation model can combine the advantages of remote sensing observation and model simulation.On the one hand,the remote sensing observation data can help the crop model to correct the simulation error and provide the way to obtain the model input variables in the region,on the other hand,crop model can make up for the shortcoming of remote sensing observation in mechanization and can better consider the internal mechanism of crop growth and development.In this study,by establishing a summer maize yield estimation model in arid area based on the coupling of UAV multi-spectral remote sensing and crop model,the accuracy and reliability of the assimilation system were verified and evaluated by using the measured summer maize data.The main research contents and conclusions are as follows:（1）In this study by UAV multispectral data 6 vegetation index are calculated,using linear regression,logistic regression,exponential regression,power function return to four kinds of mathematical model,in view of the vegetation index and leaf area index inversion model,analyzed different vegetation index and the inversion precision of mathematical model,finally selected the inversion precision of the highest leaf area index of the inversion model.The analysis results showed that there was a significant correlation between the six vegetation indices and LAI,and the linear regression model and vegetation index EVI had the highest accuracy（R~2=863）for LAI inversion.Therefore,it is selected as the inversion method of LAI.（2）The SAFY crop model was selected to simulate the growth process of summer maize.Combined with historical meteorological data and yield data,the influence ratio of the change of free parameters in the crop model on the results was compared,and the sensitivity analysis was carried out to determine the"localization"parameters of the model.The results showed that the sensitivity of sensitive parameters from high to low was Pla、Plb、R_s、STT、LUE.The values of parameters were as follows:leaf distribution coefficient Pla and Plb were 0.33 and 0.0027,respectively;light energy utilization LUE was 2.21 g/m J;the optimal accumulated temperature STT of leaves was 1024°C;and the optimal aging rate RS was 6785°C?day^-1.（3）Taking canopy LAI as the assimilation variable,LAI was retrieved from remote sensing images and simulated by crop models.Then,the Ensemble Kalman Filter algorithm was used to assimilate the two data,and the results of assimilation were returned to crop models to build an assimilation system.Based on the established assimilation system,the growth process of summer maize under the condition of insufficient irrigation was simulated,and the yield simulation results were outputted.The accuracy of the assimilation system under different irrigation conditions was verified and evaluated by the measured grain yield of summer maize.The results showed that the accuracy of the assimilation system for summer maize estimation was higher under TRT₂ and TRT₅ conditions.The highest accuracy was found in TRT₂ water treatment area（R~2=0.877,RMSE=551.8 kg/ha）.The second was TRT5 water treatment zone（R~2=0.803,RMES=742.2 kg/ha）.The estimation accuracy of summer maize under full irrigation condition was low（R~2=0.626,RMSE=897.1 kg/ha）.Under the condition of heavy water deficiency irrigation（TRT₃ and TRT₄）,the estimation accuracy of crop yield per unit area was the lowest,and the estimation accuracy of crop yield per unit area in TRT₃ was the lowest（R~2=0.355）.Under global conditions,the accuracy of the model per unit area yield was good（R~2=0.855,RMSE=692.8 kg/ha）.（4）Based on the constructed assimilation system,the spatial distribution map of maize yield was outputted.In order to reduce the complexity of calculation,combined with the actual production,the UAV multi-spectral remote sensing image was meshed to get the distribution map of summer corn leaf area index.The gridded LAI was used as the assimilation parameter to carry out point-by-point assimilation of crop yield,and finally the spatial distribution map of summer maize yield was output,which played a guiding role in irrigation.