Winter Wheat Grain Protein Content Prediction In Regional Through Remote Sensing

With the improvement of people’s living standard and food consumption,the consumption of grain processed products increases rapidly.Predicting the Grain protein content(GPC)which is an important factor to evaluate the quality of wheat quickly and accurately is beneficial to grade the quality of wheat.The results show that the quality of wheat grain is closely related to its seed character and environment.And the content of protein is especially affected by the growth environment of wheat.The protein quality of wheat is monitored by field sampling.This process is limited by time span,manpower and material resources,so it is difficult to collect data in large area.The difference in climatic conditions in different years can lead to a difference in quality.Therefore,the quality variation analysis based on field-based test samples will inevitably be limited by time and space.The development of remote sensing technology makes it possible to monitor the growth and quality of crops in a wide range.The sensors on the remote sensing platform can obtain the spectral information of a wide range of"surface"objects periodically,instantaneous and nondestructively,thus realizing the transformation from the traditional ground observation to the remote sensing regional observation.It provides a new method for quantitative monitoring the growth and quality of crops in the region.Based on the remote sensing monitoring experiment of winter wheat quality in Beijing area for many years,nitrogen agronomic parameters at flowering stage of wheat at different test sites in Beijing area were obtained,synchronized satellite remote sensing images,meteorological data from 8 meteorological stations,grain yield and grain protein content of wheat were obtained.Based on the data of many years,the key influencing factors of wheat GPC were screened.Secondly,the monitoring model of wheat grain protein content was constructed based on nitrogen parameters,spectral parameters and meteorological factors at flowering stage,and the optimal parameter combination of component model was determined by comparative analysis.On the basis of this,the prediction model of wheat GPC is constructed by means of multiple linear regression,extreme learning machine and geographical weighted regression,and the precision of the model is evaluated.Finally,the estimation and prediction of regional wheat GPC were carried out based on geographical weighted regression and spatial interpolation.The main findings are as follows:(1)Selection of influencing factors of GPC in Wheat.By analyzing the correlation between the key influencing factors and GPC,we can see that GPC is significantly correlated with most satellite spectral parameters,agronomic parameters and meteorological factors.According to the results of the analysis,the most relevant factors,such as the nitrogen parameters at wheat flowering stage,the wheat canopy spectral parameters,rainfall from 26May to 30 May,sunshine time from mid-May to early June and accumulated temperature from early March to early June,were selected as the independent variable to construct a prediction model for grain protein content of winter wheat.(2)Construction of GPC Prediction Model for Wheat.The linear regression method is used to compare the GPC prediction model of wheat with different influence factors as independent variables.The results show that each model has high modeling accuracy and verification accuracy.The modeling set R~2 of multiple linear regression models based on spectral parameters,agronomic parameters and meteorological factors is 0.598,and the test accuracies of its validation set were nRMSE=10.36%,RMSE=1.412 and MAE=1.091respectively.Compared with the GPC prediction model based on satellite spectral parameters and the GPC prediction model based on satellite spectral parameters and agronomic nitrogen parameters,the former is a better inverse prediction model.Using satellite spectral parameters,agronomic nitrogen parameters and meteorological factors as independent variables,a multivariate linear regression,extreme learning machine and geographically weighted regression were used to construct the prediction model of wheat GPC respectively.The coefficient of determination(R~2)of MLR model is 0.598,while the validation precision of Normalized root mean squared error(nRMSE)and Mean absolute error(MAE)are 10.36%and 1.091,respectively.The R~2 of the GPC model based on ELM model is 0.483,and the standard nRMSE and MAE of the Validation Set are 10.895%and 1.111,respectively.Based on the GWR model,the R~2 of the GPC GWR model is 0.616,and the standard nRMSE and MAE of the validation set are 8.58%and 0.956,respectively.Through comparison,it can be concluded that according to the precision evaluation index of the comprehensive analysis model,when using the same method to construct a model,the precision of the multi-variable model is better than that of the single-variable model,and when constructing a model with the same number of independent variables,the prediction accuracy of the geographically weighted regression model is higher than that of the global regression model.Therefore,adding geolocation information when constructing GPC model can more accurately predict the grain protein content of winter wheat,and realize the prediction of GPC regionally and annually of winter wheat.(3)Prediction of regional wheat GPC.On the basis of the above research,introducing spatial interpolation method and combining geographical weighted regression model to predict GPC of wheat regionally.In the GPC inversion process of regional wheat,combined with geographically weighted regression and Kriging interpolation method,using the regression coefficient that can follow the change of geographical location to establish the dynamic regression relationship between wheat GPC and six auxiliary variables.The spatial distribution of winter wheat GPC predicted by this method is basically consistent with the field investigation.Based on this,the GPC spatial distribution map of winter wheat in Beijing in 2010 was drawn.The results show that ecological factors such as meteorology have a great influence on the formation of grain protein of winter wheat.The method of combining geographical weighting and Kriging interpolation can better explain the influence of each influence factor on the spatial difference of GPC of winter wheat,and improve the prediction precision of the model.The GPC prediction model of winter wheat based on geographical weighted regression can provide higher prediction accuracy than other methods.It not only eliminates the influence of non-stationary spatial data on modeling accuracy,but also improves the efficiency of using remote sensing data to predict GPC of winter wheat.It can accurately reflect the variation of grain protein content with spatial geographical position of winter wheat.The Kriging interpolation method only considers the influence of geographical coordinates on the GPC of winter wheat,and it is difficult to take into account the influence of factors such as meteorological factors.The method proposed in this paper can effectively analyze the effects of different spatial locations and different influencing factors on the grain protein content of winter wheat,which has certain reference value for realizing the rapid prediction of regional winter wheat GPC by remote sensing.