The Methods For Estimating Leaf Area Index(LAI)of Winter Wheat Using Hyperspectral Remote Sensing

Leaf area index (LAI) reflects multiple ecological functions of vegetation. Therefore, it serves as an important input variable in many land surface processes models. In recent years, hyperspectral remotely sensed data is increasingly widely applied for retrieving LAI. However, the following problems remain in LAI retrieval based on canopy reflectance models, using hyperspectral data. First, the advantages of hyperspectral data compared with multispectral data for LAI estimation remain controversial. The optimum scheme of band selection and parameterization of physical models for LAI estimation based on conventional look-up tableinversion techniques are still unclear. Secondly, the conventional methods of LAI estimation using physical models cannot effectively constrain"the curse of dimensionality" and the ill-posed problem. At last, the current pixel-based inversion techniques cannot effectively utilize the spatial information in hyperspectral data to constrain the ill-posed problem. Moreover, it is inaccurate to apply the same scheme of parameterization to different plots. Hence, this doctoral dissertation aims at the scientific problem that how to apply hyperspectral data effectively to improve LAI inversion. The conventional inversion techniqueswere improved toeffectively utilize hyperspectral data forinproving the accuracy of modle parameterization and constraining the ill-posed problem. The major contents of this research are concluded as follows.(1) A comprehensive literature review is performed. By doing this, the limitations in current studies can be concluded, together with their solutions.(2) A field campaign for measuring the canopy spectra and the corresponding parameters of winter wheat is performed. Studies are carried out to find the optimum band selection and schemes of parameterization for LAI retrieval. The experiments proved that when band selection has been properly performed, and the uncertainties in input parameters of CR model are relatively low, the hyperspectral data performs better than multispectral data for retrieving LAI.(3) The Multiple-step Inversion method was developed for constrain the ill-posed problem by mining and applying spectral details in hyperspectral data. The degree of fitting between simulated and measured spectra was separately evaluated in different wavelength. During the progress, the uncertainties of different parameters are gradually reduced and the input variables are gradually optimized. Our experiments show that theinnovated inversion approach constrains the ill-posed problem, and, to some extent, free the inversion procedure from prior-knowledge.(4) Based on the spatial distribution pattern of crop variables in farmland ecosystem, the method of Object-based Inversion was developed. Pre-inversion using conventional method were firstly conducted. With its results, well fitted parameterization schemes are separately made for each plot. The ranges of free variables are significantly narrowed in each plot. By doing this, the accuracy of parameterization can be incrieased; the ill-posed problem, and the sample points with significant errors can be constrained.(5) The values of N (Leaf mesophyll structure) in Plot 5 are different from other plots, what cause it fail to retrieve LAIs in Plot 5. In this study, the idea of Multiple-step Inversionwas adopted to retrieve N of each plot firstly. The retrieved N values are then substituted to object-based inversion. By the integrated application of Multiple-step and Object-based inversion approaches, the LAI values in Plot 5 are successfully estimated with good accuracy. The stability of LAI inversion is thus dramatically improved.In conclusion, in order to constrain the ill-posed problem in LAI inversion, an to improve the accuracy and stability of LAI retrieval using hyperspectral data, innovated inversion methods for mining and effectively utilizing both spectral and spatial information in hyperspectral remotely sensed data were developed in this research. The ideas and findings in this study provide useful references for future studies, which is aimed estimate vegetation parameters with hyperspectral data.