BRDF Characteristics And Physicochemical Parameters Retrieval Of Vegetation Based On Hyperspectral Data From A UAV Platform

The accurate acquisition of vegetation canopy reflectance and retrieval of physicoparameters are of great significance for vegetation growth assessment,yield estimation and pest monitoring.In this study,unmanned aerial vehicle(UAV)was used as the remote sensing platform,and hyperspectral camera was used as the sensor to obtain remote sensing images of three kinds of crops,including corn(grain maturity period,wax maturity period),sweet potato(branch and tuber setting period,stem and leaf flourishing period),and peanut(pod setting period).The canopy reflectance of each crop was extracted after hyperspectral remote sensing images preprocessing.After that,the BRDF characteristics of crop canopy reflectance was analyzed in various growth periods with the help of anisotropy factor(ANIF)and anisotropy index(ANIX).The solar azimuth angles(SAA)include 110°,180°,250° and view zenith angles(VZA)include-40°,-20°,0°,20°,40°,including a total of 13 observation angles were analysed in this study.Moreover,based on canopy reflectance,the PROSAIL model,spectral index method and BP neural network method were used to retrieve leaf area index(LAI)and canopy chlorophyll content(CCC)of corn in wax maturity period and sweet potato in stem and leaf florishing period.The retrieval accuracy of physicochemical parameters of the three retrieval methods under 13 observation angles were explored and the optimal retrieval method of physicaochemical parameters at each angle was determined.The main conclusions are as follows:(1)Different growth periods and observation angles have big impact on crop canopy reflectance:from the perspective of growth periods,the canopy reflectance of crops in the early growth stage varies greatly,while the canopy reflectance of crops in the late growing stage varies little.From the perspective of observation angles,the canopy reflectance value of each crop showed that VZA-40°>VZA-20°>VZA0°>VZA20°>VZA40° at SAA110° and SAA180°;VZA0°>VZA-200>VZA20° at SAA2500.(2)Time phase and angle information will affect the BRDF characteristics of crop canopy.Results show that the sparse canopy can lead to significant anisotropy effect variation.By comparing different SAAs,it can be seen that the angle from strong to weak of canopy reflectance anisotropy is.SAA250°、SAA110°、SAA180°respectively.The trend of ANIX curves of different crops under the same observation angle is basically the same.The ANIF curves of crops showed an obvious scoop-shaped structure.In short,the density of the canopy structure,the solar azimuth angle,the view zenith angle and the wavelength are the essential factors that affects the anisotropy of the canopy.(3)Retrieval of LAI and CCC of sweet potato and corn at various angles in September was carried out by PROSAIL model,spectral index method and BP neural network method.It was found that the retrieval results of physicochemical parameters of forward angle data by PROSAIL model showed poor performance(R2<0.52),but it was ideal in the specific VZAs at SAA110° and SAA180°(the best LAI retrieval accuracies of sweet potato and corn are R2=0.9255,RMSE=0.0832m2/m2,R2=0.8139,RMSE=0.1360m2/m2,respectively.;the best CCC retrieval accuracies of sweet potato and corn are R2=0.9483,RMSE=1.026 g/m2,R2=0.7415,RMSE=1.236 g/m2,respectively).Based on the spectral index method,the optimal spectral indices for LAI retrieval of sweet potato and corn are plant senescence reflectance index(PSRI)(R2=0.8739,RMSE=0.1218 m2/m2)and photochemical reflectance index(PRI)(R2=0.9828,RMSE=0.0043m2/m2).The best spectral indices for CCC retrieval of sweet potato and corn were modified chlorophyll absorption reflectivity index(MCARI)(R2=0.711,RMSE=1.872g/m2)and visible atmospherically resistant index(VARI)(R2 2=0.9772,RMSE=0.7377g/m2)respectively.However,the best spectral index obtained by this method is not suitable for retrieval of crop physicochemical parameters at all angles.BP neural network method has the highest overall inversion accuracy,and its retrieval effect at all angles is relatively stable compared with the other two methods,and its retrieval effect on CCC is higher than LAI.The best accuracies of LAI retrieval of sweet potato and corn were R2=0.9937,RMSE=0.0239 m2/m2,R2=0.9619,RMSE=0.0048m2/m2,respectively;the best accuracies of CCC retrieval of sweet potato and corn were R2=0.9897,RMSE=0.325g/m2,R2=0.9175,RMSE=1.689g/m2.Therefore,for LAI and CCC retrieval of sweet potato and corn,the PROSAIL model and the spectral index method have higher retrieval accuracy at specific angles,while BP neural network has higher overall retrieval accuracy and can better eliminate the anisotropic effects caused by angles.