Quantitative Inversion Of Crop Parameters Coupled With Multi-class Radiative Transfer Models

Maize is an important grain crop in our country.It is of great significance to monitor maize growth parameters in real time.In order to monitor the maize condition in southern China in real time,it is widely used to invert the maize parameters by remote sensing technology so as to know the maize height and yield.At present,there are few researches on the availability of optical images after cloud removal,and few researches on crop parameters inversion based on optical image cloud removal,image parameter extraction,and radiation transfer model.In this paper,Sentinel-1 and Sentinel-2 images and measured data were used as data sources,and Qingju Town was taken as the research area to study cloud removal from optical images,extraction of vegetation parameters from optical images with clouds,and inversion of maize height and biomass based on cloud removal images.These studies are conducive to reducing the adverse effects of cloud on optical images,timely obtaining the maize growth stage data under cloudy weather conditions,improving the availability of optical images with clouds,enhancing the collaboration between multi-source data,and exploring the possibility of collaborative biomass inversion with multi-modal data.These studies can provide a reference for crop parameter inversion in the cloudy and foggy areas in southern China.Based on previous cloud removal studies and crop parameter inversion studies,optical image cloud removal was realized in this paper,and combined with radiation transfer model,the height and biomass of maize in Qingju Town were retrieved.The research and results of this paper are as follows:(1)In the south of China,the summer is cloudy and foggy,and the optical image is easy to be polluted by cloud.As a result,the optical image that can be used to study the vegetation growth condition in the crop growth stage is less.In order to better real-time monitoring the growth condition of maize and improve the availability of Sentinel-2image polluted by cloud and fog.In this paper,Qingju Town is taken as the research area,and Sentinel-1 is used to construct six matching models.Then Sentinel-2 images are combined to remove clouds and fog from the images,and then the three-planting index is calculated.The three-planting index calculated by cloud-free images is used as validation data to evaluate the accuracy of vegetation index calculated by the six matching models and cloud-removing images respectively.The results show that the accuracy of the six matching models is good,among which the accuracy of 03＿VVd B＿VHd B＿RVI model is better than the other five models.The accuracy of the 03＿VVdb＿vhdb＿r VI model is the highest in SAVI,whose RMSE and MAE are 0.0736 and 0.0551,respectively.Among the three vegetation indices,the overall accuracy of SAVI is better than that of NDVI and EVI.It can be seen that the matching model built based on Sentinel-1 can be used for cloud removal of Sentinel-2 images,and the calculated vegetation index can also be used for follow-up studies,improving the availability of Sentinel-2 cloud images.(2)Based on the artificial neural network provided by SNAP,this paper calculated the LAI of the optical image with clouds,and then used the NDVI-LAI model and the LAI radiative transfer model respectively to restore the LAI of the cloud coverage area,and took the LAI extracted from 15 points data as verification.The results showed that the accuracy of NDVI-LAI model was better than that of LAI radiative transfer model,R~2 was 0.88,MAE and RMSE were 0.07 and 0.10,respectively.(3)In this paper,the planting range of maize was extracted based on the maize classification data in the study area,and the biomass and height of maize in the study area were respectively retrieved by combining the non-parametric model,the simplified MIMICS model based on the dynamic three-dimensional lookup table and the double-layer model.15 measured data were used as validation data to verify the accuracy of the three models.The accuracy of the two-layer model is better than that of the simplified MIMICS model and the non-parametric model.In the two-layer model,the biomass inversion result of VH polarization is better than that of VV polarization.The R~2 of VH polarization is 0.5530,RMSE and MAE are 0.3927 and 0.3081,respectively.In the inversion results of maize height,the accuracy of the two-layer model is better than that of the simplified MIMICS model,and the accuracy of VH polarization inversion height in the two-layer model is better than that of VV polarization.The RMSE and MAE of VH polarization are 0.4810 and 0.3648.In summary,Sentinel-2 and Sentinel-1 images were used in this paper to recover spectral curves of optical images polluted by clouds and fog,and to calculate vegetation index.Based on artificial neural network and two models,the leaf area index of cloud image is restored.Inversion of maize height and biomass based on three dynamic 3D lookup tables and three radiative transfer models.The quantitative inversion of crop parameters was realized,which further improved the acquisition method of crop parameters,increased the availability of cloud optical image,enhanced the collaboration between multi-modal data,and carried out beneficial exploration in the acquisition of crop parameters by remote sensing technology.