Monitoring Winter Wheat Growth Conditions In The Northwest Region Of China By Using Hyperspectral Remote Sensing

Winter wheat is the most important grain crop in the northwest region of China.Agricultural production must be precious and efficient because of the limited natural resource and fragile ecological environment in this area in order to get the maximum economic performance with the minimum resource and protect the environment.Precisely Collecting the growth condition of crops and field ecological environment by quantitative remote sensing technology is a vital step to achieve this goal.This research took winter wheat in this area as study object and the experiments were carried out during 2014~2016.Hyperspectral images and spectral reflectance data were acquired by imaging spectrometer and non-imaging spectrometer on leaf scale,canopy scale and field scale.Agronomy parameters such as chlorophyll content,anthocyanin content,leaf area index and N,P,K content were measured and their spectral features were analyzed.The characteristic spectrum and sensitive spectral indices of each agronomy parameter were extracted which were then used to build the estimation models by means of least square regression,partial least squares regression and support vector regression.The visible and quantitative distribution map of each agronomy parameter on both leaf scale and canopy scale were made from the hyperspectral images by using the optimal estimation model.The result could provide basis for reasonable production decision.The main conclusions and achievements are as follows.(1)The agronomy parameters and their spectral features changed in different growth stages.From reviving stage to maturity stage,chlorophyll content and LAI first increased and then went down;Anthocyanin content decreased first and then went up;The N,P,K content kept decreasing.In visible region,the spectral reflectance of leave and canopy were higher in reviving stage and filling stage and lower in jointing stage,heading stage and flowering stage.In the near infrared region,the spectral reflectance of leave did not show obvious difference,while the spectral reflectance of canopy kept rising from reviving stage to the filling stage.In maturity stage,the spectra of leave and canopy no longer showed the characteristics of vegetation.(2)The spectral response to chlorophyll content of leaf and canopy was mainly in visible and near infrared region and showed similar regulation.In 350~680nm,the spectral reflectance decreased with the increase of chlorophyll content and the red edge position moved to the long wavelength and the red edge amplitude went up.The SPAD values of leaf and canopy had high negative correlation with the spectral reflectance in 750~1000nm,had positive correlation with the first derivative spectrum in 710~760nm and negative correlation in 670~690nm,and had negative correlation with continuum removed spectra in 350~750nm.Of all the leaf and canopy SPAD value estimation models,the SVR model using sensitive spectral indices as independent variables had the highest accuracy.E_GNDVI,MRENDVI and(SDr-SDb)/(SDr+SDb)were highly sensitive to SPAD values on both leaf scale and canopy scale.(3)The spectral response to anthocyanin content of leaf and canopy was mainly in visible and near infrared region and showed similar regulation.In 350~680nm,the spectral reflectance decreased with the increase of anthocyanin content and the red edge position moved to the short wavelength.The Anth values of leaf and canopy had high positive correlation with the spectral reflectance in 525~700nm,had positive correlation with the first derivative spectra in 480~550nm and 670~690nm,and negative correlation in 710~760nm,and had positive correlation with continuum removed spectra in 550~650nm and 680~750nm.Of all the leaf and canopy Anth value estimation models,the SVR model using sensitive spectral indices as independent variables had the highest accuracy.E_GNDVI,SDr/SDb,MRENDVI and(SDrSDb)/(SDr+SDb)were highly sensitive to Anth values on both leaf scale and canopy scale.(4)With the increase of LAI,the canopy spectral reference decreased in 350~680nm but decreased in 750~1150nm,and the red edge amplitude raised and the red edge position shifted to red.The correlation between LAI and spectral reflectance was significantly negative in 350~750nm and 1400~2500nm and positive in 760~1300nm.The correlation between LAI and the first derivative spectra was significant in multiple narrow bands.The correlation between LAI and continuum removed spectra was significantly negative in 250~750nm and 950~2450nm.The SVR estimation model of LAI using sensitive spectral indices as variables had the highest accuracy.DSI(776,801),RSI(776,801)and NDSI(776,801)were the most sensitive spectral indices of winter wheat LAI.(5)The correlation between N,P,K content and different types of spectra showed similar regulation.N,P,K content and the spectral reflectance were negatively correlated in 350~720nm and 1350~2500nm and positively correlated in 750~1150nm on 0.001 significant level.The correlation between N,P,K content and the first derivative spectra varied in different bands whit high coefficient.N,P,K content and continuum removed spectra were negatively correlated in 350~750nm、1300~1850nm and 1950~2250nm on 0.001 significant level.N,P,K content and multiple spectral indices were negatively correlated on 0.001 significant level.DSI(819,776)、DSI(918,790)and DSI(900,796)were sensitive to N,P,K content respectively with the correlation coefficient higher than 0.75.The SVR estimation models of N,P,K content using sensitive spectral indices as variables had the highest accuracy.(6)The spectra of the same target object acquired by SOC and SVC were highly coincident.SOC spectra had more noise than SVC in 750~1000nm.The SOC spectral reflectance fluctuated and decreased significantly in 900~1000nm.The spectral information on the hyperspectral image acquired by SOC was accurate in the range of 400~900nm.Different parts on wheat leaf had different spectral features.The reflectance of middle part was lower than the ends in 400~700nm but higher in 750~900nm.Different parts and levels of wheat plant had different spectral features.The reflectance of panicles and stalks were higher than leaves in 400~700nm but lower in 750~900nm.The reflectance of leaves in the bottom was higher than these in the middle and upper level in 400~700nm and had no significant difference in 750~900nm.The distribution maps of SAPD values and Anth values were made by solving the hyperspectral images of leaves and plants using optimal models.The range and distribution of predicted value of SPAD and Anth on the maps were consistent with measured values.(7)The spectra of the same target object acquired by UHD and SVC were highly coincident in the range of 450~850nm.Spectral signatures such as green peak,red valley and red edge extracted from UHD hyperspectral images were similar with these from SVC data.The UHD spectra in 850~1000nm were noisy and low in signal to noise ratio.The spectral information on the hyperspectral image acquired by UHD was accurate in the range of 450~850nm.Sensitive spectral indices of canopy SPAD value,Anth value,LAI and N,P,K content were extracted from UHD hyperspectral images and used to build estimation models.The distribution maps of SAPD values,Anth values,LAI,and N,P,K content were made by solving the UHD hyperspectral images using optimal models.The range and distribution of predicted value of each agronomy parameter on the maps were consistent with measured values,which could provide valuable information for production.