A Novel Principal Component Analysis Method For The Reconstruction Of Leaf Reflectance Spectra And Retrieval Of Leaf Biochemical Contents

Leaf is an important part of vegetation canopy,and the photosynthesis of leaves plays a vital role in regulating global climate change.Remote sensing retrieval of leaf biochemical contents has many important applications in precision agriculture and carbon nitrogen cycle.The optical properties of leaf are the significant basis for retrieving the physiological parameters of leaf.The traditional method of leaf biochemical contents retrieval includes three methods:empirical-statistical model,physical model method and hybrid retrieval method.The generality and universality of the empirical-statistical model method are low;The retrieval accuracy of physical model method is limited.The hybrid retrieval method will pass the error of model to the retrieval result.In addition to the data reduction in spectrum dimension,Principal component analysis get rid of the noise,target detection,etc,can also be used in atmospheric optical research in sample dimension for simulating atmospheric transmittance and retrieving the content of SO2.In this paper,we established a novel principal component analysis method for the reconstruction of leaf reflectance spectra and retrieval of leaf biochemical contents.The content of this paper contains as followed:Firstly,based on prior knowledge and the measured data of the leaf parameters,using the PROSPECT-5 model to simulate the vegetation leaf spectra reflectance data,two-thirds of spectra reflectance were randomly selected as the experimental training data set,the remaining one-third of the spectral data and ANGERS data set,LOPEX93 data set as a model validation data set.Secondly,principal component transformation was carried out in sample dimension and the ten leading principal components contained 99.998%of total information of the training dataset.The ten leading PCs were selected for the reconstruction of the leaf reflectance.Thirdly,to investigate whether the extracted PCs are able to account for the leaf biochemical properties.The coefficients of determination for the PC A data-driven model were 0.94,0.99,0.94 and 0.89 for SLW,EWT,Cab and Car,respectively.The PCA data-driven models were validated and compared to the traditional ?-based and physical approaches to the retrieval of leaf properties.Based on the PCA method,the spectral recovery of low noise-signal ratio was carried out using the information of effective bands.The vegetation indices calculated by the recovered band reflectance was employed to retrieve the measured leaf parameters.The results show that:1)The spectral RMSE between measured reflectance and reconstructed reflectance using the PCA method was found to be about 3-10 times smaller than that using the PROSPECT simulation method for the measured datasets.2)the PCA method gives similar or even better estimation of most of the leaf biochemical contents,including the SLW,EWT,Cab and Car.3)PCA method can restore the leaf spectral characteristics in low noise-signal ratio band,and the vegetation indices calculated by recovered spectral reflectance can be used to retrieve the leaf biochemical contents.