Research On The Estimation Model Of Field Maize Soil Moisture Content Based On UAV Multispectral Remote Sensing

Soil moisture content?SMC?in field crop root zone is the basic condition for the growth of crops,and it is obviously significant for drought supervision and precision irrigation.With the advantages of low cost,high flexibility and high real-time,an unmanned aerial vehicle?UAV?multispectral remote sensing system could obtain images with high spatio-temporal resolution to monitor soil moisture content at a farm scale in a real time.In this study,in order to accurately and timely monitor soil moisture content at a farm scale,the images of field maize canopy with different levels of deficit irrigation treatments were collected in key growth period through the six-rotor UAV equipped with 5-band multispectral camera in 2018.Besides,physiological and biochemical parameters,canopy temperature?Tc?and the soil moisture content values at various soil depths were collected at the same time.The support vector machine?SVM?,an image classification method,was used to eliminate the multispectral images of soil background,as well as the maize canopy spectral reflectance was extracted and a number of vegetation indices?VIs?were calculated.Then,the spectral response characteristics of maize under different deficit irrigation treatments were analyzed,the correlation among vegetation spectrum,canopy temperature and soil moisture content at different depths were explored.Finally,the inversion models of soil moisture content at various soil depths in different growth period were established by using different characteristic spectral screening methods and mathematical modeling methods,and the models were evaluated comprehensively according to different accuracy evaluation indices.The main results are listed as follows:?1?The machine learning models of soil moisture content in field crop root zone in different growth period are established and the best retrieval model is selected.The selected typical VIs are used to determine the sensitivity of different VIs to soil moisture content based on gray correlation analysis are various,including NDVI,MSR,RVI,SIPI,RVI2,NDVI_g-band GI.It is found that the changes of vegetation indices based on different spectral bands are different with the growth period,and the sensitivity of different VIs to moisture stress condition are also not the same.In model inversion,the accuracy and robustness of the support vector machine regression?SVR?model is optimal among the three machine methods in different growth period.The effect of the back propagation neural network?BPNN?model followed,and the Cubist model was relatively the worst.From the model inversion effects in different growth period,the correlation between VIs and soil moisture content is better at V stage and M stage than that of R stage.However,the model inversion of the whole stage also has high modeling and validation accuracy,and correlation coefficient R²reaches more than 0.68,showing that the method of combining the gray correlation analysis with machine learning could estimate soil moisture content.It is found that the optimal model was the SVR model at M stage,the modeling R_c²and validation R_v²were 0.851 and 0.875,respectively,and the root mean square error?RMSE?both were 0.7%,and the normalized root mean square error?NRMSE?were 8.17%and 8.32%,respectively.?2?The inversion models of soil moisture content at various soil depths based on best subset selection-extreme learning machine?ELM?method are established.The spectral reflectance of maize canopy under different water treatments all showed a trend of“increasing first,then decreasing and then rising”,that is,the reflectance was lower in the visible spectrum and higher in the near infrared spectrum.The best subset selection method can effectively select the optimal spectral subset,and it is found that the optimal spectral subsets under different condition are of great difference.However,the selected variables have generally passed the significance test and the number of independent variables is small?2-6?,and it shows that the best subset selection method is simple and efficient.In model inversion of soil moisture content,the effect of the ELM model outperformed the ridge regression?RR?model almost under all the same conditions.It is found that the optimal monitoring soil depth is not the same in different growth period,and the optimal monitoring soil depth of maize during Jointing stage,Tasseling-silking stage is 0-20 cm,and the optimal monitoring soil depth of Milk-mature stage is 20-45 cm.The ELM inversion model at 20-45 cm soil depth at Milk-mature stage performs the best,with the modeling R_c²=0.825and validation R_v²=0.750,respectively,and the modeling RMSE_c=1.00%and the validation RMSE_v=1.32%,respectively,as well as the modeling NRMSE_c=10.85%and the validation NRMSE_v=13.55%,respectively.?3?An inversion model for the joint diagnosis of soil moisture content by vegetation indices and canopy temperature were constructed.A significant correlation existed between vegetation indices and physiological and biochemical parameters of maize,and the correlation coefficients are all above 0.51,and the correlation coefficient between each vegetation index and plant height even reaches about 0.8.Vegetation indices and canopy temperature have significant positive correlation and negative correlation with soil moisture content at different depths,respectively.The correlation coefficients of the two with 0-20cm depth soil moisture content are both above 0.69,and the correlation coefficients of the two with 20-45 cm depth soil moisture content are both above 0.53.The estimation models of soil moisture content based on both vegetation indices and canopy temperature is better than using vegetation indices or canopy temperature alone in estimation accuracy.It shows that the combined method of vegetation indices and canopy temperature have great potential to estimate soil moisture content in maize field.In addition,there are good correlation coefficients among vegetation indices,canopy temperature,physiological and biochemical parameters,as well as soil moisture content.