Study On Methods And Models For Vegetation Biochemical Information Retrieval By Remote Sensing

Vegetation is one of the most important components of the ecosystem. So it is important to obtain the content and spatial distribution of vegetation biochemical information over local to regional and eventually global scales. Remote sensing provides an easy and versatile tool to accurately estimate biochemical content information at different scales. Centered on biochemical information retrieval by remote sensing, this paper did a detailed study from retrieval by empirical and semi-empirical regression methods to retrieval by physical model inversion, and from method analysis to model development. The main contents and conclusions of this paper are summarized below.First, after different canopy and leaf physical models were reviewed, two traditional leaf models, PROSPECT and LIBERTY, were discussed in detail with respect to the aim of biochemical inversion. Leaf spectra were modeled theoretically and every parameter's sensitivity of these two models was analyzed.Second, empirical and semi-empirical vegetation biochemical information retrieval methods were discussed. l)Cellulose, lignin, total carbon and total nitrogen concentration of dry leaf were estimated by multiple regression. It was found out that good regression results can be obtained both by reflectance and first derivative of spectra, and biochemical concentration can be well estimated from validation data with the relationship derived by correction data. And the first derivative of reflectance performed better than reflectance. 2)Chlorophyll content retrieval methods at both leaf and canopy level were analyzed from theoretical as well as practical point of view. At leaf level, the applicability of different kinds of spectral indexes and their sensitivity to leaf types were analyzed, which explained why researchers can construct a good regression relationship between these indexes and chlorophyll content, but because these indexes are sensitive to leaf types, the estimation regression relationship made by some samples can not be applied to other samples. Considering as less sensitive as possible to leaf types and easy for application, this papergave a chlorophyll content estimation relationship made with index GM; and testing with experiment data, the estimation and real values were consistent, which showed that it is a possibly applicable model. At canopy level, with spectral index TCARI and soil adjusted index OSAVI, the background and LAI's effects can be effectively removed. By analyzing former model's merits and drawbacks, a new chlorophyll content retrieval model was put forward at canopy level, and through validation with experimental com data, it is thought to be applicable to crop canopy chlorophyll content estimation.Third, a detailed discussion of physical model inversion to estimate biochemical content and existing problems was made. l)This method is limited by models' invertibility: only those models that are invertible can be applied to estimate biochemical content. Leaf model PROSPECT is totally invertible while accurate inversion of biochemical content information by LIBERTY model relies on accurate a priori knowledge of the other model parameter. 2)Through analyzing the properties of noises, how to choose the cost function used in inversion process was discussed. It is pointed out that for specific application, different cost function should be chosen according to data's property in inversion: if there exists systematic noise, correlation coefficient can be used in cost function to reduce inversion results' sensitivity to systematic noise. 3)Aiming at application, the performance of different algorithms(traditional optimization algorithm SIMPLEX, NEWTON, Levenberg-Marquardt(LM) and genetic algorithm that is widely used in recent years) were compared in terms of inversion accuracy and CPU time cost. The study pointed out that LM algorithm is the best one. 4)Actual inversion problem is often very complex, multi-stage inversion can simplify complicated inversion problems. In this paper, the feasibility of multi-stage inversion was...