Cotton Verticillium Wilt Surveillance And Yield Loss Estimation Based On Landsat8 Image

[Objective]Cotton Verticillium wilt began to appear in China in 1935,and then flooded in major cotton areas,especially in Xinjiang,which has become one of the most important diseases in cotton production.The use of satellite remote sensing technology to monitor diseases and insect pests makes up for the shortcomings of traditional monitoring methods,can efficiently and labor-saving to achieve real-time monitoring of large areas,and better guide agricultural production.In this study,cotton Verticillium wilt was used as the research object to monitor cotton Verticillium wilt and estimate its yield and yield loss by Landsat8 satellite remote sensing.[Methods]The best original single band,the best vegetation index and the best combination band related to Verticillium wilt were selected by correlation coefficient method and best index factor method.Through the analysis of the difference of ground point reflectivity and the incidence area,andeconomic performance,the best monitoring period of Verticillium wilt was selected.The best classification image and the best classification method based on Landsat8 satellite are selected through the overall classification accuracy and Kappa coefficient.The monitoring model of cotton Verticillium wilt was constructed and the best monitoring model was selected according to the evaluation index.The cotton yield loss was estimated according to the plot yield measurement data,the yield and yield prediction model of Verticillium wilt was constructed,and the best yield and yield loss prediction model was selected.[Results]1.The best single band,the best vegetation index,the best combination band the best monitoring period for monitoring cotton Verticillium wilt by Landsat8 satellite remote sensing.After correlation analysis with disease index,the single band with the top three correlation coefficients are mainly B5,B7,B4?|r| are 0.78,0.65,0.46 respectively.The top three vegetation indices with correlation coefficients are mainly DVI,SAVI and EVI.|r| are 0.837,0.834 and 0.833 respectively.The top three combination bands with OIF value are mainly B5-B6-B7,B4-B5-B7 and B4-B5-B6,which are 2170.44,2038.19 and 2009.82 respectively.The best period for monitoring Verticillium wilt by Landsat8 satellite is from August 9th to August 25th,and the best monitoring period is August 9th.2.The best classification image and method of cotton Verticillium wilt monitoring by Landsat8 satellite and screening of Verticillium wilt severity monitoring model.The best supervised classification method for cotton Verticillium wilt in the original band(B1-B7)of satellite images is the minimum distance classification(R2=92.70%,Kappa=0.8831).The best supervised classification method for cotton Verticillium wilt in the best combination band of satellite images(B5-B6-B7)is Mahalanobis distance classification(R2=96.28%,Kappa=0.9330).The best supervised classification method for the supervised classification of cotton Verticillium wilt by superimposing the best single band(B5)and the best vegetation index(DVI)from satellite images is support vector machine linear kernel function classification(R2=94.04%,Kappa=0.9080).The best supervised classification method for cotton Verticillium wilt supervised classification based on the best combination of satellite images(B5-B6-B7)and optimal vegetation index(DVI)is support vector machine(SVM)polynomial kernel function classification(R2=95.16%,Kappa=0.9237).The best supervised classification method for cotton Verticillium wilt supervised by satellite image original band(B1-B7)and optimal vegetation index(DVI)is the minimum distance classification(R2=94.04%,Kappa=0.9046).3.Prediction Model of optimal Cotton yield and yield loss based on Landsat8 Satellite.According to the yield measurement data of the plot and the area occupied by each disease grade,the seed cotton yield loss in the experimental area was 202.111t,the seed cotton yield loss rate was 21.77%,the lint yield loss was 102.075t,and the lint yield loss rate was 25.67%.The partial least squares regression model constructed by the superposition of original band(B1-B7)and optimal vegetation index(DVI)data is the best(training set:R2=0.8763,RMSE=0.1215;verification set:R2=0.8619,RMSE=0.1343);the principal component regression model constructed by raw band(B1-B7)data is the best(training set:R2=0.9116,RMSE=0.1027;verification sets:R2=0.849,RMSE=0.1498).The multivariate linear regression model constructed by the best combination band(B5-B6-B7)data is the best(training set:R2=0.9258,RMSE=0.0975;validation set:R2=0.8636,RMSE=0.1339);the support vector machine regression model constructed by the optimal combination band(B5-B6-B7)data is the best(training set:R2=0.9424,RMSE=0.0843;validation set:R2=0.8927,RMSE=0.1264).Taking the spectral reflectance of the best combined band(B5-B6-B7)as the independent variable,the yield and yield loss model were constructed,in which the support vector machine regression model was the best yield prediction model(training set:R2=0.9341,RMSE=0.0856;prediction set:R2=0.8279,RMSE=0.1495).The support vector machine regression model is the best prediction model of yield loss(training set:R2=0.9454,RMSE=0.0843;verification set:R2=0.8466,RMSE=0.1585).[Conclusion]The best single band,the best vegetation index,the best combinati band and the best monitoring period for cotton Verticillium wilt based on Landsat8 satellite remote sensing are B5,DVI,B5B6-B7 and August 9th,respectively.The best image based on Landsat8 satellite remote sensing image supervised classification is the best combined band(B5-B6-B7)image,and the best supervised classification method is Mahalanobis distance method.The best model for monitoring the severity of Verticillium wilt based on Landsat8 satellite is the support vector machine regression model constructed by the best combination band(B5-B6-B7)data.In the experimental area,the loss of seed cotton yield was 202.111t,the loss rate of seed cotton yield was 21.77%,the loss rate of lint yield was 102.075t,and the loss rate of lint yield was 25.67%.The support vector machine regression model based on the best combination band(B5B6-B7)image data is the best cotton yield prediction model.The support vector machine regression model based on the best combination band(B5-B6-B7)image data is the best cotton yield loss prediction model.