| Rice is an important agricultural crop in China,and its planting area ranks the first in the world.The safety of rice production is the top priority for ensuring the stability of grain in China.Water is an important part of rice,and it is also the main participant in photosynthesis and respiration of rice.The real-time monitoring of rice water content,timely,effective and reasonable irrigation of rice is the most effective method to ensure the healthy growth of rice and alleviate the shortage of water resources.With the development and application of hyper spectral technology and UAV technology,it is possible to get water content information of rice rapidly,nondestructively and extensively.In this paper,leaves and canopy hyperspectral data and UAV multispectral data of rice were used to establish estimation model of rice leaf water content and plant water content.The main research contents include:(1)The relationship between hyper spectrum and leaf water content of rice leaves was analyzed.It was found that the sensitive bands of rice leaf water content were mainly located in the near infrared band.1450 nm,1833 nm,1930 nm,2220 nm and 2500 nm were the sensitive bands of rice leaf water content.The relationship between the hyper spectrum of rice canopy and the water content of plant was analyzed.It was found that the sensitive bands of rice plant water content were the same as the water content of leaf,mainly located in the near infrared band.The spectral reflectance of 1200 nm and its adjacent bands were also sensitive to the change of plant water content.(2)Using common vegetation index,tectonic vegetation index,principal component estimation,random forest and gradient boosting decision tree,rice leaf water contents of the four periods(jointing and booting stage,heading and flowering stage,grouting and filling stage Ⅰ,grouting and filling stage Ⅱ)were estimated.The results showed that in the jointing and booting stage,the linear regression model established by the structural vegetation index ND_FD(1062,1784)was the optimal estimation model for leaf water content of this period,the model validation coefficient was 0.58,the root mean square error was 3.50,the RPD was 1.61;in heading and flowering stage,the multiple linear regression model of 10 leaf spectral principal components extracted by principal component analysis was the best to estimate leaf water content,the determination coefficient of the model is 0.61,the root mean square error is 1.66,and the RPD is 1.73;In grouting and filling stage Ⅰ,random forest has the best effect,the model verification coefficient is 0.86,the root mean square error is 1.16,and the RPD is 2.33;in grouting and filling stage Ⅱ,the linear regression model established by the structural vegetation index DV_NR(2196,2255)is the best model to estimate leaf water content,the coefficient of determination is 0.62,the root mean square error is 2.56,and the RPD is 1.70.The optimal structural vegetation index of four stages of rice was NDVI_FD(1062,1784),DV_RC(1883,2300),SR FD(2087,2250)and DV_NR(2196,2255),which were all composed of near infrared band.The reflectance of near infrared band is closely related to leaf water content,and they are sensitive bands of leaf water content.(3)Using common vegetation index,tectonic vegetation index,principal component estimation,random forest and gradient boosting decision tree,rice plant water contents of the four periods(jointing and booting stage,heading and flowering stage,grouting and filling stage Ⅰ,grouting and filling stage Ⅱ)were estimated.The results showed that in the period of jointing and booting,random forest was the best to estimate plant water content in this period,the model validation coefficient was 0.56,the root error of the model was 0.68,the RPD was 1.58;in heading and flowering stage,the linear regression model established by the structural vegetation index DV_FD(590,1720)was the best,the mean square root error is 0.55,the root mean square error is 1.54,the RPD is 1.57;In grouting and filling stage Ⅰ,the gradient boosting decision tree is the best,the model verification coefficient is 0.53,the root mean square error is 2.05,the RPD is 1.48;in grouting and filling stage Ⅱ,estimation effect of random forest is best,the decision coefficient of model validation is 0.50,the root mean square error is 2.32,and the RPD is 1.41.Because of the influence of field moisture,the accuracy of estimation models of plant water content decreased with the decrease of LAI in four periods.The optimal structural index of plant water content in four periods is DV_FD(790,1250),DV_FD(590,1720),SR_RC(745,1114)and DV_FD(775,2040),most of which are near infrared wavelengths,because of the interference of air moisture in the canopy spectral collection,1450 nm,1930 nm and 2500 nm and their near water absorption bands are eliminated,the visible light band belongs to the secondary influence band of moisture and has a certain correlation with the plant water content,therefore,a few visible light bands appear in the optimal structure index.(4)using UAV multispectral data to estimate plant water content of rice at jointing and booting stage,the results showed that the multivariate linear model established by band2 and Band6 was the best,the modeling decision coefficient was 0.51,the root mean square error was 1.04,and the RPD was 1.52.In the estimation models of rice water content,the modeling effect of plant water content is worse than that of leaf water contant because of the different rice varieties and the effect of water in the field. |
PDF Full Text Request