Modeling Of Cotton Photosynthetic Parameters And Water Content Retrieval By Multi-spectral Remote Sensing Of UAV

The water content of crop plants can directly reflect the degree of water stress,and soil moisture is the basis for stable and high yield of crops.Photosynthetic parameters can reflect the drought and water use status of crops to a certain extent.UAV remote sensing can effectively monitor the photosynthetic parameters,plant moisture and soil moisture of crops in real time,and become an important technical means to realize precision irrigation.Taking cotton as the research object,this paper used six-rotor UAV to observe the canopy spectral characteristics of cotton in boll stage with a multispectral camera,and collected the photosynthetic parameters of cotton canopy under different water treatment,the water content of plant and the soil moisture content in different soil depths.The spectral response characteristics of cotton under different water treatment and the correlation between spectral reflectance,vegetation indexor vegetation supply water index and photosynthetic parameters,plant water content or soil moisture content in different soil depths were analyzed,and unary and multivariate regression models of cotton photosynthetic parameters and plant water content based on characteristic bandsor sensible vegetation supply water index were established,as well asthe machine learning retrieval models of soil moisture content.The main results obtained are as follows:(1)The retrieval models of different photosynthetic parameters are constructed and the best retrieval model is selected.The reflectance of the 6 bands shows a tendency to decrease first and then increase in one day.The blue band(490nm)and the red band(680nm)show low reflectance,and the changesare not obvious.However,the trend of green band(550nm),red edge band(720nm)and near infrared bands(800,900nm)is obvious.The trend of photosynthetic parameters of cotton canopy under different water treatment are basically the same,in which the net photosynthetic rate(Pn),stomatal conductivity(Gs)and transpiration rate(Tr)show an approximate parabolic change with increasing first and then decreasing,while the intercellular carbon dioxide concentration(Ci)is the opposite,showing the reverse parabolic change which decreases first and then increases.The 4 photosynthetic parameters have sensitive bands,and some correlation coefficients reach more than 0.8.It is found that the optimal model of net photosynthetic rate(Pn)is the linear model based on the reflectivity of blue band at 13:00,the best model for stomatal conductivity(Gs)is the unary linear model based on the reflectance of red band at 15:00,the best model of intercellular carbon dioxide concentration(Ci)is the RR model at 15:00,and the best model of transpiration rate(Tr)is the unary linear model based on red band reflectance at 15:00.(2)The retrieval models of water content in different parts of cotton plant are constructed and the best retrieval model is selected.The bandssensitive to LWC(Leaf water content)are the blue band with a wavelength of 490nm and the red band with a wavelength of 680nm.The band sensitiveto SWC(Stalk water content)is the near-infrared band with a wavelength of 900nm.The bands sensitive to BWC(Bud&boll water content)are the blue band of 490nm,the red band with a wavelength of 680nm and the near-infrared band with a wavelength of 900nm,showing a very significant negative correlation.The correlation betweenVSWIs based on different vegetation indicesand plant water content is higher than that of single vegetation index and plant water content.Among them,the correlation coefficient between VSWI_GI and LWC reaches-0.853,the correlation coefficients between VSWI_DATT2,VSWI_MTCI2 and SWC reaches-0.895,the correlation coefficient between MTCI2 and BWC reached-0.872.The best retrieval model of LWC is the unary linear model based on VSWI_GI,and SWC is the unary linear model based on VSWI_MTCI2,and BWC is also the unary linear model based on VSWI_MTCI2.(3)The retrieval models of soil moisture in cotton field are constructed and the best retrieval model is selected.The correlation coefficients between VSWI_VARI,VSWI_GI,VSWI_MCARI,VSWI_MTCI1 and SMC at 45cm soil depth are more than 0.8.There is a negative correlation between VSWI_MTCI1 and SMC,while VSWI_VARI,VSWI_GI and VSWI_MCARI are positively correlated.Machine learning models such as SVM(Support vector machines),BPNN(Back propagation neural networks)and RF(Random forests),constructed from 4 VSWIs such as VSWI_VARI,VSWI_GI,VSWI_MCARI and VSWI_MTCI1,have achieved higher accuracy in the retrieval of SMC in cotton fields.Modeling and validation R~2 are above 0.8,and RMSE is less than 0.02,RE is less than 5%.The comprehensive analysis of sensitive VSWI models and machine learning models shows that the prediction accuracy of machine learning model is obviously better than the sensitive VSWI models.Among them,the modeling decision coefficient R~2,RMSE and RE of the retrieval model constructed by RF are 0.906,0.010 and 0.719%,respectively,all of which are the optimal values in all models.This shows that the RF model is the optimal retrieval model of SMC in cotton field.