Monitoring And Diagnosis Of Cotton Water And Nitrogen Status Under Plastic Film-Mulched Drip Irrigation In Southern Xinjiang Based On UAV Remote Sensing

Water and nitrogen are two important factors affecting cotton growth,yield and quality.Timely and accurate monitoring and diagnosis of cotton water and nitrogen information is the basis for precision water and nitrogen management.The experiment was carried out from April to September 2021 in the Second Company,Thirty-one,Second Division,Yuli County,Bayingoleng Mongol Autonomous Prefecture,Xinjiang Uygur Autonomous Region.Two factors,irrigated water amount and nitrogen fertilization amount,were set in the experiment.Three levels of water irrigation were set:0.6,0.8 and 1.0 times the water requirement of the cotton crop,named W0.6,W0.8 and W1.0 respectively,and four levels of N application:no urea,200kg/hm2,300kg/hm2 and 400kg/hm2,named N0,N200,N300 and N400 respectively,with a total of 12 treatments.The UAV flight system was composed of a DJI M200 V2 quadcopter with Micasense Altum multispectral sensors from the USA,which was used to acquire multispectral remote sensing images of the cotton canopy during the critical growth period of cotton,and to collect ground truth data such as canopy equivalent water thickness,leaf nitrogen weight,leaf area index,above-ground dry matter mass and canopy temperature simultaneously.The effects of different irrigation and nitrogen applications on cotton growth,the changes of water and nitrogen parameters with the process of growth period and the response of canopy spectrum were analyzed to establish the monitoring and diagnostic model of cotton water and nitrogen conditions and to determine the optimal model.The spatial maps based on crop water stress index(CWSI)and nitrogen nutrition index(NNI)were developed to provide a theory basis and technical support for water and nitrogen diagnosis and accuracy management of cotton.The main results of this study are as follows:(1)The pattern of variation of water nitrogen characterization parameters and the response of canopy spectrum of cotton under different water nitrogen treatments were revealed.The increase of water irrigation significantly improved the canopy equivalent water thickness and leaf N weight,and the interaction between water irrigation and N application had a significant effect on the canopy equivalent water thickness and leaf N weight of cotton.It reached maximum values at treatment W1.0N300 for CEWTupper,CEWTall,LNWupper,and LNWall with 0.142 cm,0.149 cm,10.73 g/m2,and 18.54 g/m2,respectively.,and the trends of canopy equivalent water thickness and leaf N weight of all treatments were basically the same.Under the same irrigation rate,the canopy equivalent water thickness and leaf N weight both showed a trend of increasing and then decreasing with the increase of N application;under the same N application rate,they both showed an increasing trend with the increase of irrigation rate.The changes of canopy equivalent water thickness and leaf N weight at the upper half leaves level and the all leaves level were basically the same.The canopy spectral reflectance of each band showed an overall trend of increasing and then decreasing with the advancement of fertility,and reached the maximum at flowering stage,which was basically consistent with the trends of canopy equivalent water thickness and leaf N weight of cotton.(2)The cotton water stress monitoring and diagnostic model combining texture information from UAV multispectral remote sensing images and vegetation indices was constructed.Most of the spectral variables were well correlated with the equivalent canopy water thickness at the upper half leaves level(CEWTupper)and the equivalent canopy water thickness at the all leaves level(CEWTall),and were better correlated with CEWTupper.The combination of vegetation indices and texture features could improve the canopy water prediction accuracy of the model,and the extreme gradient boosting(XGB)model was the best among the three machine learning methods.The prediction accuracy of both CEWTupper and CEWTall is the early growth period>the whole growth period>the late growth period.The prediction accuracy of CEWTupper was higher than that of CEWTall.The optimal monitoring models for both CEWTupper and CEWTall were XGB models with VIs+TIs as input variables in the early growth period,with validation set R2 of 0.74 and 0.69,RMSE of 0.02 cm and 0.03 cm,and RE of 19.86%and 28.56%,respectively.The CWSI decreased with increasing irrigation amount at the same N application rate,and the CWSI decreased with increasing N application amount at the same irrigation rate.Most of the spectral information had good correlation with CWSI.The combination of vegetation indices and texture features could improve the CWSI prediction accuracy of the model,and the XGB model performed the best among the three machine learning methods.The differences in CWSI prediction accuracy were small for the early growth period,the whole growth period,and the late growth period.The optimal diagnostic model for CWSI was an XGB model with VIs+TFs+TIs as input variables for the whole growth period,with a validation set R2 of 0.81,RMSE of 0.06,and RE of 8.87%.A spatial distribution map based on CWSI was drawn to provide a basis for cotton water stress diagnosis.(3)The cotton nitrogen nutrition monitoring and diagnosis model combining texture information from UAV multispectral remote sensing images and vegetation indices was constructed.Most spectral information showed well correlation with leaf nitrogen weight at the upper half leaves level(LNWupper)and leaf nitrogen weight at the all leaves level(LNWall),and better correlation with LNWupper.The combination of vegetation indices and texture features could improve the prediction accuracy of the model for cotton nitrogen nutrition,and the XGB model performed the best among the three machine learning methods.The prediction accuracy of both LNWupper and LNWall is early growth period>the whole growth period>the late growth period;the prediction accuracy of LNWupper was higher than that of LNWall.The optimal monitoring models for both LNWupper and LNWall were XGB models with VIs+TFs+TIs as input variables in the early growth period,with validation set R2 of 0.86 and 0.75,RMSE of 0.98 g/m2 and 2.20 g/m2,and RE of 19.65%and 22.45%,respectively.NNI increased with the increase of N application under the same irrigation amount;NNI increased with the increase of irrigation amount under the same N application.Most of the spectral information had a good correlation with NNI.The combination of vegetation indices and texture features could improve the NNI prediction accuracy of the model,and the XGB model performed the best among the three machine learning methods.The differences in NNI prediction accuracy between early growth period,the whole growth period,and the late growth period were small.The optimal diagnostic model for NNI was an XGB model with VIs+TFs+TIs as input variables for the whole growth period,with a validation set R2 of 0.65,RMSE of 0.09,and RE of 7.81%.A spatial distribution map based on NNI was drawn to provide a basis for cotton nitrogen deficiency diagnosis.