Remote Sensing Estimation Of Crop Canopy Chlorophyll Content At Satellite Spectral Scal

Canopy chlorophyll content（CCC）is closely related to crop growth.Accurate estimation of canopy chlorophyll content is very important for agricultural production and management.Canopy chlorophyll content is affected by Leaf Area Index（LAI）and leaf inclination distribution.Canopy chlorophyll content is correlated with leaf area index,so leaf inclination distribution is the only structural parameter affecting canopy chlorophyll content.Vegetation Index（VI）is commonly used in vegetation inversion of chlorophyll content.At present,the influence of leaf dip distribution of various crops is rarely considered in the process of constructing vegetation index to estimate canopy chlorophyll content,and the noise effect of actual measurement data is also rarely considered in the process of constructing canopy chlorophyll content inversion model with simulated data.Therefore,this paper carried out a study to estimate crop canopy chlorophyll content based on satellite reflectance of different spectral scales.On the one hand,11 general vegetation index models were used to construct a vegetation index sensitive to CCC.On the other hand,Six algorithms（partial least squares regression,support vector regression,random forest regression,long and short term memory network,convolutional neural network and deep neural network）were used to investigate the effects of adding 1%,3%,5%and 10%Gaussian noise to satellite data at different spectral scales on the construction of canopy chlorophyll content estimation model.The main achievements and conclusions of this paper are as follows:（1）The CCC sensitive vegetation index considering leaf inclination distribution is constructed on satellite data at different spectral scales.Through the measured data and model data,the CCC sensitive vegetation index considering the distribution of leaf inclination angle was constructed on different spectral scales.The best sensitive vegetation indexes were:Sentinel-2 soil regulation index SAI（B6,B7）,World View-2 Verrelts three-band index BSI-V（NIR1,Red,Red Edge）,Rapid Eye Tian three-band index BSI-T（Red Edge,Green,NIR）and Gaofen 6 difference index DI（B6,B4）.The performance of the sensitive vegetation index constructed in the model simulation data is consistent with that in the measured data.The optimal sensitive vegetation index has a strong correlation with CCC(R_CCC² on the measured data is between 0.76-0.80,R_CCC² on the model simulation data is between 0.84-0.95),and has no correlation with MTA(R_MTA² on the measured data is 0.00,R_MTA² on the model simulation data is between 0.00-0.04).（2）The CCC inversion model is constructed by using different satellite spectral scale reflectance data.The best inversion effect is Sentinel-2 data（the R²of the measured data is between0.77 and 0.84,and the R² of the model simulation data is between 0.97 and 1）.The best inversion algorithm is support vector regression（the R² of the measured data is between 0.83 and 0.84,and the R² of the model simulation data is near 1）.（3）A CCC inversion model based on satellite reflectance of different noise and spectral scales is constructed.The R² of PROASAIL model is close to 1 when the reflectance data of different scales are not added noise.With the increase of data noise,R² gradually decreases,and the difference of CCC estimation from reflectance data of different spectral scales is more obvious.Sentinel-2 data inversion CCC performs well.When 3%noise is added,the inversion differences of spectral data at different scales begin to appear.The inversion model is constructed by adding 3%-5%noise data,and the effect is close to the accuracy of the inversion CCC model constructed by the measured data.（4）Properly adding noise can improve the accuracy of model inversion.The data simulated by PROSAIL model with 3%noise added,combined with the measured data,the optimal model of CCC inversion is mixed,which is 0.02 higher than the R² of the optimal inversion model constructed by the data without noise combined with the measured data,and the RRMSE is reduced by 0.03.