Research On Estimation Of Crop Canopy Structure Parameters Based On Sentinel-2 Ban

Canopy structure parameters including Leaf Area Index（LAI）,Mean Tilt Angle（MTA）and Fractional Vegetation Cover（Fcover）can directly affect the proportion of solar energy captured by crop canopy,thus affecting photosynthesis.Meanwhile,the parameters are also the necessary data source to realize precision agriculture.As remote sensing technology upgrades,it is available to obtain canopy structure parameters in large area.The red-edge band are specified for Sentinel-2,and also can be used to convey the information of canopy structure parameters in a wide range.Renewed ML makes new breakthroughs in remote sensing inversion.Due to the lack of large-scale Sentinel-2 field data,this paper based on the aerial data and PROSAIL simulated data,studies the parameters of crop canopy structure in Sentinel-2 resampling band.The following show the main contents and conclusions of the research:（1）To investigate the ability of sentinel-2 band combined with machine learning to estimate canopy structure parameters.By studying the Sentinel-2 band and vegetation index resampling from regional aerial hyperspectral data,the estimation model of canopy structure parameters was constructed based on four machine learning regression algorithms（random forest,support vector machine,partial least squares and multilayer perceptron）,and verified by using PROSAIL simulation data.The results showed that sentinel-2 band 6 had a strong correlation with MTA（R²=0.87 for field measurements andR²=0.79 for model simulation）,but had little correlation with LAI（R²=0.06 for field measurements andR²=0.07 for model simulation）.Four indexes based on red-edge bands(NDVI_RE,CI_RE,WDRVI_RE,MSR_RE)were moderately correlated with LAI（R²=0.55～0.57）,but weakly correlated with MTA（R²=0.05～0.07）.Multilayer perceptron algorithm was used to accurately estimate MTA（RMSE=2.24°）combined with four 10m band data.The estimation of MTA（RMSE=1.50°）can be improved by using the MLP algorithm and the optimized index.When PROSAIL simulation data is used for verification,the prediction accuracy increases with the increase of samples,and finally tends to saturation,indicating that the spectral band in Sentinel-2 contains vegetation structure information.With sufficient measured data,the machine learning algorithm can retrieve canopy structure parameter information from spectral data.（2）The ability of PROSAIL simulation inversion and machine learning algorithm to estimate canopy structure parameters is discussed.The sentinel-2 resampling band data and vegetation index were established by simulating 5000 sets of data from crop parameters measured in the field and PROSAIL model.The estimation model of canopy structure parameters was constructed by four regression algorithms,and the model was verified by the measured data.The results show that the model accuracy tends to saturation due to the large number of training samples.The model has a very high accuracy in estimating the measured MTA（R=0.92 for field bands,R=0.89 for field Vegetation indices）,which further proves the feasibility of the Sentinel-2 spectral band combined with machine learning algorithm to estimate the large-scale crop MTA in the field,and provides a new idea for constructing the inversion model of canopy structure parameters when the ground measured data is few.