| High yield is one of the major objectives of soybean breeding.Soybean yield is influenced by so many factors,such as hereditary,environment,cultivation management and so on.Thus,the selection of high-yield lines also needs to monitor soybean above-ground dry mass and leaf area index(LAI)except the measured yield after harvest.We can assess photosynthetic efficiency and net primary production which can provide reference for selection of high-yield variety through this method.However,this work takes a lot of manpower,material and financial resources and retard soybean breeding process severely and further influence breeding efficiency.This study mainly explore a real-time,rapid and non-destructive tactics that can monitor soybean growth status and assess seed yield using hyperspectral remote sensing technology.And it can provide strong technical support for large-scale soybean breeding.This study aims to explore the relationship between canopy reflectance and soybean yield and growth traits based on two consecutive years of field trials.When choosing experimental material,two aspects are taken into consideration.One aspect is significant difference between seed yield and growth character,while the other one is the similarity of growth period among materials especially flowering so that it is convenient for us to collect canopy spectrum data during the same growth period.We have chosen 52 materials and planted with randomized complete block design with 3 repetition.Three major growth stages(R2,R4 and R5)were used to measure canopy reflectance spectra,above-ground dry mass and leaf area index(LAI)destructively and measure production after harvest.Variance analysis of yield and growth traits shows there exists significant difference(P<0.01)among materials and non-significant difference between materials and years.Meanwhile,there exists no difference between materials and years for canopy spectrum single waveband reflectance which means it is fit to analyze using combined 2 years data.The study plans to explain variation of target traits as many as possible using the fewest wavebands based on the combined data.In this way,we mainly do research on the relationship between vegetation indices and yield,growth traits of single waveband,two wavebands and three wavebands using spectrum analysis technology and statistical analysis method with the purpose of the selection of sensitive band and its allowable bandwidth and sensitive spectrum parameter which can stand for soybean yield,above-ground dry mass and LAI.The final destination is to establish an optimal prediction model for soybean yield and growth traits accurately.The selection of sensitive band and its allowable bandwidth can provide technology support for the development of portable spectrometer so that yield estimation and growth status monitoring with no damage in large-scale breeding can come into reality.Thousands of materials are used in the breeding trials with no repetition.Using canopy vegetation index to estimate yield indirectly and soybean dynamic changes in the field can help to select,reduce the workload,improve working efficiency.Yield comparison test is strictly used to test breeding production during post-breeding experiment.In contrast,canopy vegetation index can help to make an auxiliary selection before harvest.The main analysis steps and results of this study are as follows,(1)Analyze the correlation between canopy original spectrum,first derivative spectra and target traits,select waveband range with significant differences(P<0.05)and very significant differences(P<0.01)in visible and near infrared bands using student t-test,extract the maximum correlation coefficients of positive and negative correlation and its corresponding band is considered as the sensitive band and used in subsequent comparison.Correlation coefficients between grain yield and canopy spectral reflectance,first derivative spectra progressively increased from R2 to R5 stage in soybean and the maximum positive and negative correlation coefficients existed at R5 stage,which indicated R5 is the best yield estimation stage.Based on the canopy spectral reflectance at R5 stage,The corresponding wavelengths with maximum and minimum correlation coefficients between original and derivative spectrum and grain yield in soybean are 694 nm,750 nm,689 nm and 935 nm.Correlation coefficients between canopy spectral reflectance,first derivative spectra and above-ground dry mass,leaf area index have waveband range with correlated significantly or very significant difference.For above-ground dry mass,its sensitive bands are 652 nm,742 nm,779 nm and 1331 nm at R2 stage,605 nm,646 nm,688 nm and 875 nm at R4 stage and 633 nm,662 nm,689 nm and 818 nm at R5 stage.For leaf area index,its sensitive bands are 653 nm、743 nm、780 nm and 1302 nm at R2 stage,606 nm、661 nm、689 nm and 866 nm at R4 stage and 695 nm、689 nm、750 nm and 817 nm at R5 stage.(2)102 reported vegetation indices are collected,and these indices include construction form of different vegetation indices,most of which are combined with two wavebands.Then we construct these vegetation indices based on canopy spectrum.After correlation analysis with target traits,we extract the most relevant 10 existed vegetation indices.Meanwhile we can ascertain the construction form of these vegetation indices.In the most relevant 10 existed vegetation indices associated with grain yield,GNDVI,ND705,NDCI,NDVI(810,560),PSNDa,PSNDb and RNDVI are constructed in the form of normalized difference vegetation index(NDVI),while RVI(810,560),SR705 and SR(900,680)are in the form of ratio vegetation index(RVI);In the most relevant 10 existed vegetation indices associated with above-ground dry mass,GNDCI、NDCI、NDVI(810,560)and PSNDb are constructed in the form of NDVI,while GM-1、GM-2、PSSRb、RI_half、RVI(810,560)and SR705 are in the form of RVI;In the most relevant 10 existed vegetation indices associated with leaf area index,GNDCI、NDCI、NDVI(810,560)and PSNDb are constructed in the form of NDVI,while GM-1、GM-2、PSSRa、PSSRb、RVI(810,560)and SR(900,680)are in the form of RVI.Therefore,sensitive vegetation index in the form of NDVI and RVI were determined to indicate grain yield and growth traits in soybean.(3)Find out all the vegetation indices(NDVI and RVI)mentioned above in spectral range 350-2500nm,and then build linear model with target traits.After comparison and analysis,the best linear correlation vegetation indices form are selected.Then establish linear and non-linear models based on constructed vegetation indices and target traits in the whole spectral range and then select linear and non-linear models with maximum determination coefficient and finally determine the best regression model based on sensitive vegetation index according to the comparison and analysis of model test,parametric test,determination coefficients and standard error.These results showed that based on canopy spectral reflectance at R5 stage,linear correlation between the yield and NDVI was superior to that of RVI.Depending on the principle of largest R2,exponential regression y=2.57e7.88x of NDVI(938,642)at R5 stage to soybean yield was the best one with R2 of 0.711.Similarly,based on canopy spectral reflectance at different stage,linear correlation relationship of above-ground dry mass and leaf area indes to RVI was superior to that of NDVI.The best regression model of above-ground dry mass were power model y=0.14x1.63 on RVI(855,609)at R2 stage,power model y=0.04x2.24 on RVI(820,588)at R4 stage and exponential model y=6.34e0.11x on RVI(1000,604)at R5 stage with their determination coefficients of 0.771、0.762 and 0.783,respectively,and the best one for LAI were y=0.03x1.83 on RVI(825,586)at R2 stage,y=0.38e0.14x on RVI(763,606)at R4 stage and y=0.06x1.79 on RVI(744,580)at R5 stage with their determination coefficients of 0.677,0.639 and 0.664,respectively.(4)Find out all three-band vegetation indices in near infrared range of 760 nm to 1000 nm,red range of 620 nm to 760nm and blue-green regiom(blue range of 430 nm to 470 nm,green range of 500 nm to 560nm)in construction form of the reported three-band vegetation index.Then build linear and non-linear mode with target traits.After that,determine the linear and non-linear model with maximum determination coefficient.The optimal estimation mode could be decided based on three-band vegetation indices according to the comparison and analysis of model test,parametric test,determination coefficient and standard error.The optimal regression model could be ascertained after comparison and analysis with the selected model in procedure(3).These results showed that exponential regression y=6755.4le5.65x of NDVI(771,760,560)at R5 stage to soybean yield was the best one with R2 of 0.622.Similarly,the optimal regression model of above-ground dry mass were power model y=0.07x-2.18 on NDVI(768,756,519)at R2 stage,power model y=141.39x-12.94 on NDVI(765,671,518)at R4 stage and power model y=1.53x-1.48 on NDVI(770,756,570)at R5 stage with their determination coefficients of 0.752、0.716 and 0.750,respectively,and The best one of lear area index(LAI)were power model y=0.02x-2.24 on NDVI(770,757,520)at R2 stage,power model y=0.01x-3.05 on NDVI(894,757,528)at R4 stage and exponential model y=0.11x-1.89 on NDVI(771,756,560)at R5 stage with their determination coefficients of 0.660、0.601 and 0.654,respectively.Further comparison and analysis,the coefficient of determination between three-band vegetation indices and soybean yield,above-ground dry mass,leaf area index is were lower than that of two-band vegetation indices.(5)Rebuild existed vegetation indices in canopy spectral range and establish linear and non-linear models with target traits.Choose the top 5 regression models with vegetation indices with maximum determination coefficient.Compare and analyze determination coefficients and standard error,compare goodness of fit and root mean square error comprehensively in model validation and finally decide the optimal sensitive vegetation index and the best regression model based on the vegetation index.The results showed that in procedure(3),the predicted power of constructed regression models for soybean yield,above-ground dry mass and leaf area index(LAI)are better than model based on the reported vegetation index,so the constructed models in procedure(3)were determined as the best regression model.Exponential model y=2.57e7.88x of grain yield on sensitive vegetation index NDVI(938,642)at R5 stage was the best model.The best regression model for above-ground dry mass were power model y=0.14x1.63 on RVI(855,609)at R2 stage,power model y=0.04x2.24 on RVI(820,588)at R4 stage and exponential model y=6.34e0.11x on RVI(1000,604)at R5 stage respectively.The best regression model for LAI were y=0.03x1.83 on RVI(825,586)at R2 stage,y=0.38e0.14x on RVI(763,606)at R4 stage and y=0.06x1.79 on RVI(744,580)at R5 stage,respectively.(6)Define the sensitive band of target traits again.Compare the correlation of target traits with the optimal sensitive vegetation index and single band reflectance and then make sure the sensitive band with biggest correlation coefficient.For example,if the correlation coefficient of vegetation index is the largest,then the waveband combination consisted in the optimal sensitive vegetation index is the sensitive band.The result showed that the correlation coefficient of sensitive vegetation index NDVI(938,642)at R5 stage to soybean yield is better than that of single-band reflectance significantly,so the sensitive band of soybean yield was identified as 938nm and 642nm.Similarly,the sensitive band of above-ground dry mass were 855 nm and 609 nm at R2 stage,820 nm and 588 nm at R4 stage,1000 nm and 604 nm at R5 stage,and that of LAI were 825 nm and 586 nm at R2 stage,763 nm and 606 nm at R4 stage,744 nm and 580 nm at R5 stage。(7)In order to decide the effect of the change of spectral resolution of sensitive wave band on prediction accuracy of target traits,some work are done.Take the optimal sensitive vegetation index for example,expand the scope of the sensitive band(in steps of 1mm),fit the regression model with vegetation index and target traits,compare the determination coefficients and standard error of constructed model and finally select the allowable bandwidth to its sensitive band.These results showed that there are differences in the impact of spectral resolution on vegetation index(NDVI and RVI).The R2 of regression model based on sensitive spectral vegetation index NDVI(93 8,642)of soybean yield showed a decreasing trend in two directions 642 nm and 938 nm with the increase in bandwidth and better obvious change in 642nm.Based on selection criterion maximum of R2 and minimum of standard error,the resolution of 642nm less than 9nm and that of 938nm less than 15nm is better respectively.So the search for allowable bandwidths of broad bands indicated that 17 nm of 938 nm and 43 nn of 642 nm were the most practical range for the estimation of grain yield at R5 stage in soybean.Similarly,the allowable bandwidth of broad bands for above-ground dry mass indicated that 9 nm of 855 nm and 13 nm of 609 nm at R2 stage,15 nm of 820 nm and 9 nm of 588 nm at R4 stage,and 21 nm of 1000 nm and 9 nm of 604 nm at R5 stage,and that of broad bands for LAI were 13 nm of 825 nm and 19 nm of 586 nm at R2 stage,11 nm of 763 nm and 7 nm of 606 nm at R4 stage,and 7 nm of 744 nm and 17 nm of 580 nm at R5 stage.(8)This study further analyzed the common sensitive band combination among different agronomic traits.Set and change the threshold of coefficient of determination and shrink the searching region of common core band combination until obtaining common sensitive band combination of different agronomic traits based on the regression models listed above.The constructed vegetation indices based on common sensitive band combination not only can estimate several traits but also can provide more comprehensive information for selection of high-yield soybean lines.These results showed that the common sensitive core waveband combinations of LAI at different growth stages and above-ground dry mass were consistent with the sensitive band of recommended LAI monitoring model.Besides,RVI based on common sensitive waveband combination had good performance for the prediction of above-ground dry mass.Models y=0.14x1.68,y=1.36e0.15x and y=5.47e0.17x based on RVI(825,586),RVI(763,606)and RVI(744,580)at growth stages of R2,R4 and R5,respectively,showed that all have good prediction results to above-ground dry mass of its relative growth stages with its determination coefficient 0.760,0.757 and 0.775 respectively.Similarly,based on soybean canopy spectral reflectance at R5 stage,the common sensitive core waveband combinations of above-ground dry mass and grain yield were further identified as 938 nm and 642 nm which was also consistent with commended production prediction model.Model y=6.96e0.08x based on RVI(938,642)at R5 stage have good performance for the prediction of above-ground dry mass with its determination coefficients 0.745.Vegetation index constructed based on common sensitive core waveband combination can assess several traits simultaneously and can provide more comprehensive information for selection of high-yield soybean lines.In summary,we hope to introduce the hyperspectral technology to large scale breeding of soybean from the perspective of crop breeding based on the principle of simple,flexible,fast and accuracy with the purpose of analyzing the relationship between yield,growth traits and vegetation index of single band,two-band and three-band and further explaining the largest variation of target traits with relative fewer wavebands,sensitive band and its allowable bandwidth,sensitive vegetation index and optimal regression model of soybean yield and growth traits are ascertained.During the large-scale breeding of soybean,canopy vegetation index was used to estimate soybean yield at the R5 growth stage and to estimate field growth traits in various growth stages.In this way,high-yield advanced lines can be selected so workload can be reduced and efficiency can be improved.At the same time,the allowable bandwidth of sensitive band of different traits can provide inference for development of portable spectrometer and makes it more effective for obtaining the information of canopy vegetation index.In addition,more than 2 traits may be monitored using the same waveband combination so this study further define common sensitive waveband combination among different traits.Vegetation index constructed based on the method not only can monitor above-ground dry mass and LAI at different growth stages but can assess soybean production and above-ground dry mass at R5 stage.Selection of high-yield advanced lines with no destruction in large-scale breeding of soybean can increase the versatility of the portable spectrometer and improve its monitoring efficiency which plays an importance role in large-scale soybean breeding. |
PDF Full Text Request