Study On Growth Physiology And Yield Monitoring Model Of Winter Wheat Based On Hyperspectral Analysis

Winter wheat is one of the most important food crops in China.Natural precipitation cannot meet the water demand of winter wheat.It is necessary to ensure the water demand of winter wheat growth through timely irrigation.The timely acquisition of various growth physiological parameters during the growth of winter wheat can help farmers adjust the irrigation time and irrigation volume to achieve a high yield.Therefore,using hyperspectral remote sensing technology to monitor the growth physiological parameters and yield of winter wheat in real-time is of great significance to timely adjust the irrigation measures and stabilize the food market security.In this study,the winter wheat variety "Jintai 182" was used as the material.Five treatments were set up,including no irrigation,irrigation once at jointing stage,irrigation once at jointing stage and flowering stage,irrigation once at jointing stage and filling stage,and irrigation once at jointing stage,flowering stage,and filling stage.The changes of aboveground dry biomass(AGDB),leaf area index(LAI),chlorophyll density(CHD),plant water content(PWC),and yield of winter wheat were analyzed.The original spectrum(R)was preprocessed using five preprocessing algorithms,including reciprocal logarithm(Lg),multiple scattering correction(MSC),standardized normal variate(SNV),first derivative(FD),and second derivative(SD).The concentration gradient method(CG),Kennard-Stone method(KS),and sample subset partition based on joint X-Y distances method(SPXY)were used to divide all samples into calibration sets and validation sets according to 1:1(Ratio1),3:2(Ratio2),2:1(Ratio3),5:2(Ratio4),and 3:1(Ratio5).On this basis,based on the full-spectrum band and the characteristic band screened by the uninformative variable elimination(UVE),hyperspectral monitoring models for growth physiological parameters of winter wheat AGDB,LAI,CHD,and PWC were constructed using partial least squares regression(PLSR),stepwise multiple linear regression(SMLR),artificial neural network(ANN),and support vector machine(SVM).The effects of five modeling factors on hyperspectral monitoring models were analyzed.Finally,hyperspectral monitoring models of yield were constructed based on the two modes of "hyperspectral-growth physiological parameters-yield" and "hyperspectral-yield",hoping to realize the hyperspectral monitoring of winter wheat yield.The main results were as follows:(1)Due to the differences in sowing date,precipitation,and temperature during the two-year experiment,there were certain differences in the level and the change trend of AGDB,LAI,CHD,and PWC of winter wheat in the two-year experiment.Irrigation contributed to the improvement of AGDB,LAI,CHD,PWC,and yield of winter wheat,and the growth physiological parameters and yield of winter wheat after two or three irrigation treatments were generally higher than those of only one irrigation.In addition,the results of variance analysis also showed that the irrigation had different effects on the growth physiological parameters and yield of winter wheat.Taking yield as the final standard,the best irrigation treatment was to irrigate once at jointing stage and flowering stage respectively.(2)The five preprocessing algorithms,Lg,MSC,SNV,FD,and SD,can affect the change trend and value of the original spectral reflectance curve to varying degrees,and improve the correlation between spectral reflectance and winter wheat AGDB,LAI,CHD,and PWC.The correlation coefficients between the root mean square error of the validation set(RMSEv)of the winter wheat AGDB,LAI,CHD,and PWC models and the factors of the preprocessing algorithm were 0.3923,0.1482,0.2070,and 0.5094,respectively,which indicated that the accuracy of the growth physiological parameter model based on R spectrum may be higher than that of Lg,MSC,SNV,FD,and SD spectra.When dividing the total data sets of different growth physiological parameters into calibration sets and validation sets according to different sample division methods and sample division ratios.CG method and SPXY method all divided the maximum and minimum values of each growth physiological parameter into calibration sets.The KS method only divided the minimum values of AGDB,LAI,and CHD into the calibration set,when dividing the data set according to Ratio1,the maximum and minimum values of PWC were divided into the validation set.The average value,standard deviation,kurtosis,and skewness of the divided data set also had some differences,but most of them were close to the average value,standard deviation,kurtosis,and skewness of the total data set.The correlation coefficients between RMSEv of winter wheat AGDB,LAI,CHD,and PWC models and the factors of sample division method were-0.4629,-0.5137,-0.4967,and-0.1745,respectively,and the correlation coefficients between RMSEv and the factors of sample division ratio were-0.1804,-0.0155,0.0188,and-0.1651,respectively,which indicated that the model constructed based on the calibration set and validation set divided by SPXY-Ratio5 may have higher accuracy.The UVE algorithm had a good ability to remove bands.For winter wheat AGDB,LAI,CHD,and PWC,it can remove 45.9503% to 99.4600%,14.3629% to98.0022%,37.1490% to 96.9762%,and 45.1404% to 96.5983% of bands,respectively,significantly reducing the redundancy of spectral information.The correlation coefficients between RMSEv of winter wheat AGDB,LAI,CHD,and PWC models and the factors of dimension reduction were-0.1345,-0.1038,-0.1219,and-0.2749,respectively,which indicated that the UVE algorithm can reduce the number of bands,and at the same time obtain higher model accuracy compared with full spectrum band.The correlation coefficients between the RMSEv of winter wheat AGDB,LAI,CHD,and PWC models and the factors of modeling methods were-0.1316,-0.0317,-0.0427,and-0.0619,respectively,which indicated that the SVM algorithm may have better modeling ability and effect than PLSR,SMLR,and ANN when constructing the hyperspectral monitoring model of winter wheat growth physiological parameters.According to the universality of the data,when constructing hyperspectral monitoring models of different growth physiological parameters of winter wheat,four models of R-SPXY-Ratio4-UVE-SVM,R-SPXY-Ratio5-UVE-SVM,R-SPXY-Ratio5-UVE-SVM,and MSC-SPXY-Ratio3-UVE-SVM were used respectively to obtain accurate and stable hyperspectral monitoring models of AGDB,LAI,CHD,and PWC.According to the particularity of the data,in this study,the four models of FD-CG-Ratio4-Full-SVM,SD-SPXY-Ratio4-UVE-SVM,SD-SPXY-Ratio4-UVE-SVM,and SNV-CG-Ratio5-UVE-SMLR had the highest monitoring accuracy for winter wheat AGDB,LAI,CHD,and PWC,respectively.The determination coefficient of calibration set(R2c),the root mean square error of calibration set(RMSEc),the determination coefficient of validation set(R2v),RMSEv,and relative analysis error(RPD)were 0.9487,0.1663,0.7335,0.3600 and1.9226,0.8462,1.5742,0.8120,1.3151 and 2.2891,0.8462,1.5742,0.8120,1.3151 and 2.2891,0.9533,2.4567,0.9633,2.0875 and 5.1775,respectively.(3)There was a good correlation between the yield of winter wheat and the growth physiological parameters at the regreening stage,jointing stage,flowering stage,late grain filling stage,and maturation stage.Among the spectra at different growth stages,the correlation between the filling stage spectrum and the yield was the highest.The preprocessing algorithm can improve the correlation between the original spectral reflectance and the yield to varying degrees.The two yield hyperspectral monitoring modes of "hyperspectral-growth physiological parameter-yield" and "hyperspectral-yield" can realize the monitoring of winter wheat yield.In the "hyperspectral-growth physiological parameter-yield" mode,only the model constructed based on the LAI of the whole growth stage had higher accuracy,and its R2 c,RMSEc,R2 v,RMSEv,and RPD were 0.6773,49.3717,0.7613,33.9421,and 1.9297,respectively.In the “hyperspectral-yield” mode,the difference spectral index(DSI)constructed based on R and FD spectra at the late grain filling stage can predict the yield at the late growth stage of winter wheat,and its R2 c,RMSEc,R2 v,RMSEv,and RPD were0.7016,37.1143,0.7721,33.1638,1.9570,and 0.8358,27.5305,0.7381,35.5520,and 1.8423,respectively;The DSI constructed based on the MSC spectrum at the regreening stage can predict the yield at the early growth stage of winter wheat.Its R2 c,RMSEc,R2 v,RMSEv,and RPD were 0.5978,43.0893,0.7103,37.3898,and 1.7518,respectively.In contrast,the "hyperspectral-yield" mode has greater flexibility than the "hyperspectral-growth physiological parameters-yield" mode.