Growth Monitoring Of Winter Wheat In Guanzhong Area Based On Multi-Source Remote Sensing Data

Remote sensing technology is more and more widely used in wheat and its growth monitoring due to its macro,timely and dynamic characteristics.Wheat growth monitoring mainly includes wheat growth parameter inversion and growth condition assessment.With the development and maturity of remote sensing technology,the number,the positioning accuracy of calibration,the spatial-temporal resolution,and the setting of bands number of optical sensors,have made great progress,providing a large number of data sources for remote sensing monitoring of wheat growth.Different sensors have their own advantages in data quality,spatial and temporal resolution,time coverage and angle observation.Therefore,selecting suitable remote sensing data is the premise of accurate monitoring.In the remote sensing monitoring of winter wheat growth,the use of remote sensing data is mainly from a single source,mostly independent monitoring of a single indicator,lack of comprehensive monitoring index system and lack of multi-regional verification.In this paper,multi-source remote sensing data such as leaf hyperspectral,canopy hyperspectral,UAV hyperspectral image and satellite image were used to monitor the growth of winter wheat in multiple experimental areas and natural planting stages.The main contents and results are as follows:(1)In this paper,the field measured values of winter wheat from 150 plots in the natural planting area(DT area)of Guanzhong Plain were used to study the growth index of wheat in each growth period and its corresponding non-imaging and imaging hyperspectral characteristics from two scales of leaf and canopy.The results showed that from the jointing stage to the grain filling stage,the chlorophyll of winter wheat leaves gradually increased with the growth;the nitrogen balance index first increased and then decreased;the anthocyanin content first decreased and then increased.With the advancement of the growth period,the chlorophyll in the canopy of winter wheat fluctuated,with the lowest value at the heading stage and the highest value at the flowering stage;The distribution range of nitrogen balance index was between 13.3 and 42.8,the average value first decreased and then increased;the leaf area index increased first and then decreased;the water content of plants decreased gradually.The red edge position of the canopy reflectance spectrum shifted to short-wave(blue-shift)with the growth;the chlorophyll,anthocyanin and nitrogen balance index obtained at the leaf scale were higher than those obtained at the canopy scale.(2)For the hyperspectral diagnosis of wheat nutrient elements at leaf scale,this paper used the hyperspectral data of leaves in DT area and the measured values of chlorophyll(Chl),anthocyanin(Anth)and nitrogen balance index(NBI)in leaves.On the basis of screening the typical vegetation indexes and model methods in the existing literature,the optimal estimation models of leaf growth parameters of winter wheat were established based on the characteristic wavelength and vegetation index,respectively.The results showed that the R~2of leaf Chl inversion models based on characteristic wavelength at jointing stage,heading stage,flowering stage and filling stage were all greater than 0.773,and the maximum RMSE was 2.875,the simulation effect of the model is good.Anth estimation model based on the first derivative spectrum using SVR algorithm has the best effect,which can be used for Anth accurate inversion of winter wheat.The modeling R~2of wheat NBI estimation model constructed by vegetation indices GNDVI,CARI,MTCI and MRESR reached 0.7 in four growth stages,and the model accuracy and prediction ability were good.Overall,the model based on SVR algorithm is slightly better than that based on MLR method;GNDVI,MTCI,CARI and MRESR have good correlation with Chl,Anth and NBI in wheat leaves,which can be used as general sensitive vegetation indexes for three indicators.(3)In view of the hyperspectral diagnosis of wheat nutrient elements at canopy scale,for the hyperspectral diagnosis of wheat nutrient elements at canopy scale,the optimal estimation models of winter wheat canopy growth parameters were constructed based on the hyperspectral data of wheat canopy in DT area and the measured values of chlorophyll(Chl),anthocyanin(Anth)and nitrogen balance index(NBI)in this paper.The results show that the correlations of canopy Chl,Anth,NBI values with spectral reflectance and first derivative spectra in most periods were low in most periods due to the soil background information and more atmospheric interference.The multivariate estimation model constructed by MLR algorithm is difficult to obtain satisfactory accuracy.After the combination of spectral indices NDSI(525,557),RSI(525,557),DSI(793,815),SASI(793,815)and vegetation indices CARI,MTCI,NPCI,the four-stage Chl estimation model constructed by SVR algorithm was better(R~2>0.8).Similarly,wheat canopy Anth estimation based on NDSI(435,627),RSI(586,599),DSI(683,720),SASI(475,815)and vegetation indices CARI,MTCI,and MRESR has good model accuracy and stability.The wheat canopy NBI estimation model based on NDSI(481,597),RSI(472,604),DSI(799,831)and SASI(760,815)had good prediction ability.The estimation results of the canopy leaf NBI inversion models at flowering and jointing stages based on characteristic wavelengths are good,with modeling R~2exceeding 0.8,and verification R~2exceeding 0.7.(4)Since it is difficult for a single indicator to fully reflect the overall growth of wheat,this paper uses the UAV images obtained in BX area and the measured values of agronomic indicators,and uses the two weight calculation methods of equal weight and coefficient of variation weight to synthesize the six indicators representing the growth of wheat,namely,chlorophyll(Chl),nitrogen balance index(NBI),leaf area index(LAI),plant water content(PMC),plant height(PH)and biomass(AGB),and constructs the comprehensive growth index(CGMI)and its inversion model.The results showed that based on the UAV hyperspectral image,the R~2of the four growth period models constructed by the coefficient of variation weighting method were 0.76,0.79,0.76 and 0.81,respectively.On this basis,the R~2of CGMI inversion for wheat jointing stage,heading stage,flowering stage and filling stage were 0.77,0.72,0.66 and 0.77,respectively,indicating that the CGMI model had good accuracy and prediction ability.Using the same algorithm,the CGMI model based on coefficient of variation weighting is generally superior to the CGMI model based on equal weighting,mainly because the weight of coefficient of variation weighting is determined by the coefficient of variation,and the contribution of a single indicator to the difference of CGMI is considered,which is in line with the contribution difference of different indicators to the overall growth of wheat at different growth stages,and is more suitable as a calculation method to characterize the overall growth of wheat.(5)Based on Sentinel-2 satellite data,taking Chl,Anth,NBI and LAI of wheat canopy obtained in DT area as the research objects,the estimation models of wheat agronomic parameters in Sentinel-2 image at heading and flowering stages was constructed through satellite image wide-band reflectance and its vegetation indices.The sensitive vegetation indexes of Chl,Anth and NBI of winter wheat were CIre2,CIre1 and MCARI respectively;The correlation coefficient between LAI of wheat and Band B7 and B8a in red edge area are more than 0.7,the sensitive vegetation index is IDVI(r=0.66),and the correlation coefficient of Band B5～B8a are more than 0.5.Using Sentinel-2 satellite wide band and vegetation index,the multivariate estimation models of each index are constructed respectively.The best estimation models of Chl,Anth,NBI and LAI of wheat are Chl-ANN(validation R~2=0.45),Anth-ANN(validation R~2=0.46),nbi-ann(validation R~2=0.46),LAI-SVR(validation R~2=0.69).Taking winter wheat LAI in the study area as an example for regional mapping,the average LAI inversion value of the sampling points is overestimated by 0.53 and 0.37units respectively on heading and flowering stages compared with the measured value.The LAI estimation model based on Sentinel-2 satellite image and SVR algorithm can meet the needs of regional monitoring.