Estimation Model Of Winter Wheat Chlorophyll Contents Based On Hyperspectral Data

Green vegetation plays an indispensable role in ecosystem,chlorophyll is an important factor characterizing vegetation growth status and development,becomes an important input parameter of hydrology,climate,soil,and ecology circulation process.It is widely used in climate change and ecological cycle changes,etc.The monitoring of vegetation chlorophyll content can provide decision-making information for agricultural management,agricultural variable water and fertilizer management and can improve the utilization rate of water.It is of great importance for sustainable development of modern agriculture.Based on the field experiments and laboratory measurement from 28 April to 2 May,2014 and April 25 to 26,2015,LAI,MTA,plant height,chlorophyll content,water contents and spectral reflectance data of winter wheat were collected.Canopy reflectance curves of wheat were simulated with PROSAIL model.The influence of content of chlorophyll,LAI,MTA,water content and observation zenith angle on canopy reflectance was simulated,and compared with the measured wheat reflectance curve.According to simulated reflectance curves,red paremeters,such as red edge area,NDVI,MCARI and CIred edge were calculated,and estimation models of chlorophyll contents were established using linear model,BPNN and SVM model,and the accuracy of estimation was verified.Estimation models of chlorophyll contents were built using NDVI,MCARI,CIred edge,BPNN and SVM based on different zenith angles.Conclusions were as follows.(1)Canopy reflectance curve simulated using PROSAIL model was consistent with field measured reflectance curve using SVC.The reflectance values were the same at visible band,and the reflectance value of model simulated was higher than that of field measured.(2)Sensitivity analysis of vegetation canopy reflectance showed that the influence of chlorophyll content on canopy reflectance was mainly at visible band,and canopy reflectance values decreased with the increase of chlorophyll contents.The impact of LAI on reflectance curve was mainly at near infrared band,and canopy reflectance values increased with the increase of LAI values.The impact of MTA on canopy reflectance was opposite to that of LAI,and canopy reflectance values at near infrared band decreased with the increase of average leaf inclination.The influence of vegetation water content on canopy reflectance was at near infrared band,and reflectance decreased with the increase of water content.(3)Canopy reflectance was simulated at different observation zenith angles using PROSAIL model,three observation zenith angles of 0 °,36 ° and 55 ° were selected.When vegetation biochemical component contents were constant,canopy reflectance increased with the increase of observation zenith angles at the same band.(4)The effects of chlorophyll contents on red edge amplitude,red edge area,NDVI,MCARI and CIred edge were analyzed.The results showed that red edge amplitude,red edge area,NDVI and CIred edge values also linearly increased with the increase of chlorophyll contents,while the value of MCARI gradually reduced with the increase of chlorophyll content.(5)The chlorophyll estimation models were established using red edge amplitude,red edge area,NDVI,MCARI and CIred edge.The results showed that the accuracy of MCARI and CIred edge were the highest in linear estimation models.The correlation coefficients R2 were 0.95 and 0.939,the root mean square errors were 2.789 and 2.806 respectively,and the relative errors were 0.45 and 0.048 respectively.The linear estimation model of chlorophyll contents using NDVI,MCARI and CIred edge were established based on different observation zenith angles.In linear estimation models,the estimation accuracy of model using CIred edge was the highest when observation zenith angle is 55 °,R2 was 0.953,and the root mean square error and the relative error were 7.088 and 0.094 respectively.(6)Chlorophyll estimation models using BP neural network and support vector machine were established.The inversion effects of BPNN models based on MCARI and CIred edge were the best,the mean square error of MCARI model was 2.809,the relative error was only 0.046.The square root error of CIred edge model was 2.600,the relative error was 0.927.In SVM inversion models,the inversion effect based on MCARI was the best,the root mean square error was 2.863,and the relative error was 0.045.(7)The RMS errors and relative errors using CIred edge and BPNN were 7.265 and 0.107 respectively at observation zenith angle of 0 °,while the RMS errors and relative errors using CIred edge and SVM model were 7.185 and 0.095 respectively at observation zenith angle of 55 °.