Research On Wheat Yield Prediction Model Based On Remote Sensing And Meteorological Data In Jiangsu Province

Food is the foundation of human survival and development.Food security is crucial to social stability and economic development.To ensure food security,reliable and effective estimation of crop yields is required.Traditionally,wheat yield acquisition methods are influenced by human factors,have low efficiency and poor timeliness,and are difficult to obtain spatial distribution information,which cannot be applied in a large area.Crop remote sensing identification and yield prediction have problems such as large amount of data,time-consuming and laborious data processing,and low accuracy.Therefore,this thesis takes Jiangsu Province wheat as the research object,uses GEE(Google Earth Engine)cloud platform to call massive meteorological and remote sensing data and other geographic information data online,uses cloud programming to carry out data mining features and simplify yield estimation model,and selects growth,precipitation and temperature indexes closely related to yield for yield prediction.By establishing stepwise regression,LASSO regression,support vector regression and DNN deep neural network models,the wheat yield of Jiangsu Province was predicted,and the prediction effect of the models was compared and analyzed.The specific work was as follows:The GEE cloud platform was used to retrieve satellite remote sensing images online for data analysis and information extraction.which realized the identification of wheat crop planting area distribution in Jiangsu Province.At the same time,based on the wheat identification results,spatial analysis of MODIS crop growth and ERA5 climate data products was carried out to automatically extract 18 characteristic index data of crop growth,monthly cumulative precipitation and monthly average temperature during the wheat growth period from January to June of 2018 to 2020 in the county area,and correlation analysis was conducted between index factors and yield per unit area.Finally,by establishing stepwise regression,LASSO regression model,support vector regression model and deep neural network to analyze the correlation between factors and yield,different yield estimation models were used to predict wheat yield per unit area.Their advantages and disadvantages and applications are compared respectively from the overall prediction accuracy,mean relative error and root mean square error.The results show that support vector regression model has higher prediction accuracy and better prediction effect.This thesis provides a more simple,effective and feasible yield estimation method combining remote sensing and statistical model.By using GEE cloud platform service to retrieve satellite remote sensing images online,data analysis and information extraction are realized through online programming,which effectively overcomes the problems of high processing pressure,long time and high data cost,verifies the feasibility and effect of various models,and can predict the yield information of the current year in advance by obtaining the evaluation index factors of the current year.It provides method guidance for the relevant departments to analyze the relevant influencing factors of grain yield and forecast the yield,and has important reference significance and value for the agricultural application of crop disaster loss assessment and other aspects.