Hyperspectral Remote Sensing Monitoring And Yield Estimation Of Winter Wheat Growth Based On Comprehensive Index

With the rapid development of remote sensing technology,agricultural remote sensing application is becoming more and more extensive,mainly reflected in the monitoring and evaluation of crop growth,yield,and disasters.Crop growth monitoring and advance prediction of yield play a crucial role in realizing field production management and ensuring food security.This paper takes winter wheat in Guanzhong area as the object,canopy spectral data collection and agronomic parameters determination were carried out in the key growth period of winter wheat.Through spectral transformation processing,spectral parameters such as“trilateral”parameters and vegetation index were selected for correlation analysis with agronomic parameters(aboveground biomass,plant nitrogen content and chlorophyll content),and estimation models were constructed respectively.In addition,in order to better reflect the real growth situation of winter wheat,this paper constructed a comprehensive growth index,and then conducts multi-angle and multi-faceted monitoring research of the winter wheat growth.Based on the comparison of winter wheat yield and effective wheat yield in different growth periods,this paper proposed a more accurate prediction method of winter wheat yield.The main research contents and conclusions are as follows:(1)Aiming at the inversion of single agronomic parameter,using the ground canopy hyperspectral data and the measured agronomic parameter data,on the basis of analyzing the correlation between each spectral parameter and each agronomic parameter,the effective variables were selected by multiple stepwise regression,and the optional estimation model of each single growth parameter was constructed.The results showed that the R2 range of the modeling set of the aboveground biomass estimation model was 0.45 to 0.89,and the average R2 was 0.70,reached a significant level,and the overall fitting effect was good.The RPD of the validation set was between 1.08-1.66,and the prediction ability of jointing stage was the best,in which the RPD was more than 1.4.There are models with RPD less than 1.4 in the rest of the growth period,so it was impossible to predict the samples.Among them,the random forest model using the first derivative brand by band combination of vegetation index in jointing stage was the best,which can be used for the accurate inversion of aboveground biomass of winter wheat.The R2 range of the modeling set of the plant nitrogen content estimation model was from 0.20 to 0.88.In each growth period,except for the partial least squares method with poor fitting accuracy,the rest of the modeling methods have reached a significant level;the RPD range of the validation set was 0.97-1.84.The RPD of 63%models was greater than or equal to 1.4,which could be predicted.Among them,the prediction result of the winter wheat plant nitrogen content estimation model constructed based on the XGBoost model of combined spectral parameters in the filling stage was the best.The R2 range of the modeling set of the chlorophyll content estimation model rangeed from 0.16 to 0.98,and the average R2 was 0.70.Except for the three models based on the red edge parameters at flowering stage,the other models have reached a significant level.The RPD range of the validation set was 0.73-3.86,the RPD 52%of the models was above 1.4.Among them,the estimation model of chlorophyll content based on the red edge vegetation index RF algorithm was the best and had excellent prediction ability.(2)To a certain extent,the inversion of a single growth parameter can only reflect the local information of the crop,which was difficult to reflect the real growth situation of the crop in many aspects.In this paper,the three single growth parameters of aboveground biomass(AGB),plant nitrogen content(PNC)and chlorophyll content(SPAD)were integrated by entropy weight and principal component analysis(PCA)to construct a comprehensive growth index(CGI)of winter wheat,and PLS,RF,and XGBoost algorithms were used to construct the remote sensing inversion model of comprehensive growth index.The results showed that the R2 range of the CGI estimation model based on the entropy method was 0.42-0.87,and the average R2 was 0.63,all of which had a good fitting effect;the RPD range of the validation set was 1.49-2.13,and the RPD of 17%of the models was more than 2.0.The CGI estimation based on PC A model modeling set R2 range was 0.5-0.86,the average R2 was 0.66,the fitting effect was better than the CGI constructed by the entropy weight method;the RPD range of the validation set was 1.42-2.17,and 33%of the models had an RPD more than 2.0,the prediction ability of CGI constructed by entropy weight method has been improved.In conclusion,the CGI estimation model based on PCA method can monitor the actual growth of winter wheat more comprehensively.The best model was the random forest model at the grain filling stage,with a robust RPD of 2.17 and excellent prediction ability.The coefficient of determination R2 was 0.81,and the root mean square error RMSE was 0.04.(3)Based on the ground hyperspectral data,this paper selected the spectral index empirical model yield estimation method and the continuous wavelet transform-based yield estimation method to construct yield estimation model of winter wheat in different growth stages.The results showed that for the spectral index yield estimation model,the R2 range of the modeling set was 0.67-0.88,the average R2 was 0.78,the RPD of the validation set was greater than 1.4,and the average RPD was 1.56.The estimation effects of final yield from high to low were filling stage>joining stage>flowering stage>heading stage.The verification RPD of the optimal yield estimation model was 1.69,R2 was 0.65,and RMSE was 362.5 kg/ha;R2 was 0.75,the RPD of the validation set was greater than 1.4,and the average RPD was 1.52.The estimation effect of the final yield from high to low was flowering stage>jointing stage>grain filling stage>heading stage.The verification RPD of the optimal yield estimation model was 1.53 and R2 was 0.65,with an RMSE of 290.4 kg/ha.Comparing the two methods,the yield estimation method based on the spectral index empirical model was generally better than that based on continuous wavelet change.