Research About The Normalization Of Multi-source NDVI

Remote sensing is an effective method to provide data and information for the global dynamicobservation. NDVI (Normalized Difference Vegetaion Index) is an important parameter for amountsof application. The trade-off between the spatial and temporal resolution makes the intergration ofmultiple source data much more meaningful. Differnce caused by the imaging environment, senorcharictristic and so on limites the integration of multi-source data, and makes the normalization toachieve the consistency and comparaity among multiple data very important. Thus, we studied thenormalization of multi-source NDVI. Works about the normalization, including the normalizationmethod, the evaluation method, comparison among the normalizing methods, are shown in this papar.The main work and conclusion are generalizing as following:(1) The local cluster-specific linear model method for normalization is proposed. Consideringthe relation between the sensor difference and the land cover type, and the spatial heterogeneity ofimfactor like the atmospheric condition, the view geometry and so on, the "local process" and "clusterspecified process" are introduced to the normalizing method. The proposed method could efficientlyeliminated the difference among multiple NDVI data. And M-estimation is ultilized to exlude theinfluence of outliers. What's more, experiments about the parameters setting are taken and someadvices are given.(2) The kernel regreesion based normalization method is propsed, taking the advantege of thenon-parametic estiomation, which do not need to give concrete form of model, to deal with theunknown complex relation among the multiple source NDVI needing normalizing. Good normalizingperformance was obtained by the method What's more, experiments about the parameters setting aretaken to give advices for parameter setting.(3) The normalization method based on the suppprt vector regression machine (SVR) isproposed. Taking use of the kernel function, SVR could solve the complex non-linear problem in thehigh-demension feature space, which is helpful for the non-linear relation between multiple NDVIMoreover, the workflow and the parameter setting method are introduced in the paper. Influence ofthe parameters setting is analyzed and suggestions for parameters setting are given.(4) The evaluation methods for the reference-based method is summarized, and some new?methods are proposed to asses the performance of the normalization method and give a generalevaluation about the normalizing methods. At the same time, comparison of the methods, includingthe three proposed methods and two existing methods, are give from different aspects, such as therequirement of input data, the efficiency of algorithm, and the general proformacc in eliminating thedifference among multi-source data. This work is helpful to make choice of the normalization methodsand valuable for the real application.Generally, complete work about the nomalzation of multi-source NDVIdata, utilizing the coarser resolution reference data based normalization method are introduced in this article. Different aspects ofthe normalization was shown, including the theoretical framework, the specific normalizing algorithm, the evaluation method, the performance comparison of different methods, and some other aspects. Thework is valuable for the downstream researches and applications based on multi-source NDVIdata.