Hyperspectral Remote Sensing Monitoring Of Growth,Physiology And Yield Of Winter Wheat Under Drought Stress

As one of the three major food crops in China,wheat production plays a vital role in national food security.The growth rate of crops under different conditions can be monitored with hyperspectral remote sensing technology.Among various stresses in the field faced by crops,drought is the most important limiting factor which impacts the grain production.It is of great significance to employ the hyperspectral remote sensing technology in wheat growth monitoring to predict the yield and grain quality in the arid plateau of the Loess Plateau.In the present study,four varieties of winter wheat(Jinnong 190,Chang 4738,Jintai 182,linmai 7006)at different drought stress levels and under different irrigation conditions were investigated in different years.With respect to different growth stages of winter wheat,several monitoring models were constructed based on the correlations between canopy spectrum,vegetation indices and winter wheat growth,physiology,yield,and quality using regression analysis of both single-and multi-vegetation indices.The main purpose was to provide scientific evidences and technical support for the precise field management of winter wheat.The research results were as follows:1.The growth indicators of winter wheat at different drought levels and under different irrigation conditions complied with the general growth and developmental rules of crops,and the differences between different treatments were significant.2.The spectral characteristics of the "green peak,red valley and near-infrared platform" and “red edge parameters” of the canopy spectrum of winter wheat were changed regularly at different growth stages and under various drought stresses.In the visible light range,a continuum of the highest to the lowest spectral reflectance were observed at maturing stage,filling stage,heading stage,flowering stage,jointing stage,and booting stage,in that order;while it was heading stage,flowering stage,booting stage,filling stage,jointing stage,and maturing stage in the near infrared range.The more severe the drought stress was,the less obvious the green peak was,and the shallower the red valley was with a lower near-infrared platform.On the contrary,the milder the drought stress was,the more green peaks were observed,and the deeper the red valley was with a higher near-infrared platform.The red edge amplitude was gradually decreased with the increase of drought stress levels.The red edge position was steadily moved toward the blue edge with the increase of drought stress level,and the red edge area was declined with the increase of drought stress.3.Vegetation index could be used to effectively monitor the growth of winter wheat and to predict the yield and grain quality under different drought stresses at different developmental stages.The optimal vegetation index varied at different growth stages.Compared to the models based on the single vegetation index(VI)of winter wheat,models based on the vegetation Indices(VIs)were of greatly improved accuracy and stability.The determination coefficient(R~2)of LAI(Leaf Area Index)models of winter wheat was ranged between 0.57 and 0.71 at different growth stages,and all of the root mean square errors(RMSEs)of LAI were less than 0.86.The monitoring model of LAI established at the flowering stage was the best with R~2 at 0.705 and RMSE at 0.7706.The R~2 of above ground dry biomass(AGDB)models was ranged between 0.52 and 0.83 at different growth stages,and RMSE of AGDB was between 7.2 and 1.7.The optimal monitoring model of AGDB was determined as the one established at the flowering stage with R~2 at 0.834 and RMSE at 2.5655.The R~2 and RMSE of chlorophyll content(SPAD)models at different growth stages were between 0.58 and 0.80 and between 1.0 and 2.0,respectively,with the best model of SPAD established at the flowering stage(R~2=0.797,RMSE=1.6613).The R~2 and RMSE of Plant Water Content(PWC)models at different growth stages were between 0.55 and 0.81 and between 1.1 and 2.1,respectively,with the optimal model of PWC established at the flowering stage(R~2=0.806,RMSE=1.5889).The R~2 and RMSE of the winter wheat yield predicting models at different growth stages were between 0.55 and 0.69,and between 889 and 991,respectively,with the optimal predicting model of yield established at the heading stage(R~2=0.693,RMSE=889.5460).The R~2 of GPC(Grain Protein Content)predicting models of the winter wheat at booting stage,heading stage,and flowering stage were 0.786,0.897,and 0.874,respectively,and the RMSE were 0.5401,0.3563,and 0.4087 in the same order,with the highest accuracy of the GPC model established at heading stage.The R~2 of WGC(Wet Gluten Content)predicting models at booting stage,heading stage,and flowering stage were 0.851,0.921 and 0.830,respectively,and the RMSE were 0.9219,0.7207 and 0.9376,respectively,with the highest accuracy of the WGC model established at heading stage.All developed optimal models were tested with 1:1 verification method using the same numbers of measured and predicted values based on irrigation data at different growth stages and good results were achieved in all mentioned models,which indicated that the winter wheat growth monitoring models,yield and quality prediction models based on multiple vegetation indices(VIs)were more precise and of stronger stability.4.Due to the significant difference of growth indices and the spectral reflectance of winter wheat canopy at different growth stages coupled with the impact of excessive data volume at different growth stages,the accuracy and stability of segmentation monitoring models were significantly greater than the full-breeding hybrid monitoring ones.