Study On Spectral Scale Effect Of Remote Sensing Estimation Of Winter Wheat Aboveground Biomas

Crop biomass is one of the important biophysical parameters in the study of agricultural ecosystems,which is used to reflect crop growth and is closely related to crop yield.The remote sensing estimation of crop biomass is affected by various factors.The existing studies focus more on improving the accuracy of remote sensing estimation of biomass but ignore the effect of spectral scale on this process.In order to explore the spectral scale effect on the remote sensing estimation of above-ground dry biomass of winter wheat,this thesis is based on the above-ground dry biomass and corresponding canopy height spectral reflectance data of winter wheat at four key growth stages(jointing,heading,flowering and filling stage)and three nitrogen treatments levels(N1,N2 and N3).Meanwhile,six regression algorithms of machine learning and deep learning were used to estimate biomass,including random forest regression(RF),support vector regression(SVR),partial least squares regression(PLSR),deep neural network(DNN),long-and short-term memory recursive neural network(LSTM)and onedimensional convolutional neural network(1D-CNN).two parts of research were conducted.the first part analyzed the effect of spectral scales on the correlation between vegetation indices and winter wheat biomass,as well as the performance of the six algorithms in predicting biomass.The second part analyzed the correlation between arbitrarily two-band normalized vegetation index(NDVI-like)constructed from the spectral scale reflectance of satellites(Sentinel-2,Rapid Eye,and World View-2)with red edge bands and biomass,as well as the performance of the six algorithms in predicting biomass using satellite spectral scale reflectance data.The results showed that the correlation between the constructed vegetation index and biomass,as well as the accuracy of the six algorithms in estimating biomass,were affected by the spectral scale.The best NDVI constructed from satellite spectral scale reflectance was superior to the NDVI constructed from traditional bands.When the six algorithms were trained using satellite spectral reflectance,they could accurately predict winter wheat biomass.In both parts of the research,the prediction accuracy of the DNN algorithm was better than the other 5algorithms.Among the three satellites with red edge bands,Sentinel-2 had the best accuracy in the inversion of biomass.In conclusion,the results of this thesis showed that the effect of spectral scale should be considered when using hyperspectral reflectance data to estimate winter wheat biomass,and the accuracy of estimating biomass could be improved by using the spectral data at the best scale.The thesis also demonstrated the potential of the satellite with red edge band to invert biomass at different growth stages and nitrogen treatments levels.