Study On Estimation Of Maize Agronomic Parameters In Guan Zhong Area Based On Hyperspectral Remote Sensing

Hyperspectral remote sensing technology can quickly and accurately obtain the information of crop growth status in real time,and provide important technical support for precision agriculture.In this study,the maize in Guanzhong area of Shaanxi Province was selected as the research object.Ground-based canopy spectral data,chlorophyll(Soil and plant analyzer development,SPAD)values and leaf nitrogen content(LNC%)were measured at important growth periods of maize.Based on analyzing the characteristic of spectral data,and the correlation between the spectral reflectance and SPAD,LNC,the characteristic wavebands and best band combination of optimized spectral index were obtained.Thereafter,the estimation models of maize physiological parameters were constructed based on the characteristic wavebands,vegetation index,best band combination of optimized spectral index and BP neural network(BPNN).Combining hyperspectral imagery by UAV,SPAD and LNC were inverted in filed scale.The results of this research intended to provide a theoretical basis and technical support for dynamic monitoring of corn growth and field fertilization management.The main results are as follows:(1)The change of canopy spectral reflectance of maize at different growth stages were analyzed.The result showed that the spectral reflectance increased gradually in the 550nm band with the development of growth period,the spectral reflectance in the near infrared region were lager in the better tasseling stage and spinning stage,but smaller in the jointing stage of the early period and in the maturing stage of the later period.SPAD values gradually increased from the jointing stage to tasseling stage,while decreased from tasseling stage to maturing stage.LNC of leaves showed a decreasing trend with the development of growth period.(2)The correlation between spectral reflectance and SPAD values of canopy leaves was analyzed.It showed that during the jointing stage,tasseling stage,spinning stage,maturing stage the correlation between the original spectral and SPAD values was highest respectively at the bands of 709nm,714nm,553nm,716nm,and the correlation between first derivative spectrum and SPAD values was greatest respectively at the bands of 760nm,522nm,760nm,694nm.NIR/NIR,RVI2,ZTM,VLOPT2 of the fifteen traditional vegetation indices were significantly correlated with SPAD values at 0.01 level during all growth stages of maize.The correlation coefficients of five optimized spectral indices(Normalized Difference Spectral Index(NDSI),Ratio Spectral Index(RSI),Difference Spectral Index(DSI),Modified index(MSI),Soil Adjust Spectral Index(SASI))and SPAD values based on original spectral and first derivative spectrum were higher than 0.7.The determinant coefficient of test model based on BP neural network in all growth period is higher than 0.8,and had great universality.The results showed that the regression model based on BP neural network is the best estimation model of SPAD values,and the estimation model was constructed by BP neural network based on first derivative spectrum was best at the spinning stage,the R2 of regression equation with training dataset and test dataset,root mean square error(RMSE)values,the relative error(RE)were 0.810,0.817,1.254,1.021,2.218%and 2.084%.(3)At all growth period,the correlation between the original spectral and LNC values was highest respectively at the bands of 707nm,709nm,735nm,710nm,and the correlation between first derivative spectrum and LNC values was greatest respectively at the bands of760nm,760nm,860nm,665nm.NIR/NIR,RVI2,ZTM,VLOPT2 of the fifteen traditional vegetation indices were significantly correlated with LNC values during all growth stages of maize.The correlation coefficients of five optimized spectral indices and LNC values based on original spectral and first derivative spectrum were higher than 0.6.The determinant coefficient of test model based on BP neural network in all growth period is higher than 0.4,and had great universality.The results showed that the regression model based on BP neural network is the best estimation model of LNC values,and the estimation model was constructed by BP neural network based on first derivative spectrum was best at the jointing stage,the R~2 of regression equation with training dataset and test dataset,root mean square error(RMSE)values,the relative error(RE)were 0.790,0.802,0.154,0.117,4.549%and4.124%.(4)Combined with the hyperspectral images acquired by UAV,the spatial inversion of SPAD value and LNC were achieved at regional scale.The estimation model of SPAD value and LNC was constructed by BP neural network based on first derivative spectrum had the most accurate inversion results,the R~2 of the regression equation with test dataset,RMSE,RE were 0.627,0.493,2.759,0.254,13.264%,10.902%.