Remote Sensing Monitoring Of Spad Value Of Winter Wheat Under Waterlogging Stress

Waterlogging is a kind of agro-meteorological disaster caused by the disturbance of crop growth due to the soil being in the supersaturated state of water,heat,air and nutrient imbalance for a long time.In this study,to explore the optimal monitoring method of stained winter wheat based on hyperspectral and digital image technology,the correlation between the characteristic index of commonly used hyperspectral and digital image and the soil and plant analyze development(SPAD)and stratified SPAD value of stained winter wheat was analyzed based on micro-area experiment controlled by drainage and irrigation.Multiple linear regression model,support vector machine model,BP neural network model,decision tree model and random forest model were constructed based on the optimal monitoring characteristic index,in order to explore from leaf-canopy-field to provide a basis for the application of remote sensing technology in winter wheat waterlogging monitoring.The conclusions of this study were as follows:1)Compared with normal wheat,short-term waterlogging(≤7 d)had no obvious effect on average SPAD and SPAD values of each layer.When waterlogging time was more than 12 d,SPAD value decreased sharply with the increase of waterlogging time,and was close to 0 at the later growth stage.3 d of waterlogging could promote the growth of winter wheat,and the number of effective panicles,grains per panicle,1000-grain weight and actual yield were higher,while other effective panicles,grains per panicle,1000-grain weight and actual yield were decreased gradually with waterlogging time.2)Compared with L1,L2 and L3 layers,the average SPAD estimation model based on the image feature index was better.The R~2 between measured value and predicted value was 0.633,the root mean square error(RMSE)was 6.003,and the relative error(RE)was 4.080.The estimation results of average SPAD value based on random forest algorithm are good,and its R~2 was 0.809,RMSE was 4.660,RE was3.167.The average SPAD based on hyperspectral feature index could predict the SPAD value of winter wheat.The accuracy of the multiple linear regression model was less than that of the support vector machine model,less than that of the BP neural network model,less than that of the decision tree model,less than that of the random forest model,and the random forest model can better predict the SPAD value of winter wheat in different layers.3)By analyzing the canopy scale,SPAD estimation results of winter wheat were obtained based on digital image feature indexes such as GMR,EXR,NRI and EXG,which were basically consistent with the estimation results based on corresponding hyperspectral bands.The maximum R~2 between the measured and predicted values of the estimation model was 0.935,the root mean square error(RMSE)was 2.845,and the relative error(RE)was 1.746.Compared with the digital image feature index,the R~2 between measured value and predicted value of the winter wheat SPAD estimation model based on four hyperspectral feature indexes,namely carotenoid reflection index(Ctr2),Yellow edge amplitude(Dy),Normalized vegetation index(NDVI)and structural insensitivity index(SIPI),reached 0.945,and RMSE was as low as 2.463.The relative error was 1.621.It could be seen that both hyperspectral and digital image techniques can be used to estimate SPAD of winter wheat,and the BP neural network model based on hyperspectral feature index has a better estimation result.4)From the field scale analysis,it was concluded that the accuracy of multiple linear regression model was less than that of support vector machine model,less than that of BP neural network model,less than that of decision tree model,less than that of random forest model,and the random forest model can better predict the SPAD value of each plot.