The pre-processing will take the characteristics of the individual scanners into account. Methods include moment or histogram matching of subgroups of data, data dependent roll corrections, transformations of data into a noise-ordered feature space followed by (possibly Fourier based) noise removal and back transformation.
In order to exploit the collected multi-source data fully it is tempting to resample coarse scale data to the grid of the fine scale data. This can be done by means of geostatistical methods as kriging which is based on the autocovariance or the semivariogram of the data or by means of techniques based on Markov random field (MRF) theory. As opposed to analyses of the individual groups of data statistical analyses of the joint multi-source data can reveal otherwise uncovered cross dependencies.
Classification of the relevant signals can be done by means of various classical forms of discriminant analysis. Also, spatial or contextual extensions to the classical methods and non-parametric classifiers such as Owen-Hjort-Mohn discriminant analysis and classification and regression trees (CART), a binary decision tree based method, can be applied.
Multi-temporal analysis in this context consists primarily of change detection methods. The methods can deal with i) simple differences (or ratios) of bi-temporal data, ii) orthogonal transformations of simple differences (or ratios) of bi-temporal data, and, iii) orthogonal transformations applied directly to bi- or truly multi-temporal data. Also the classifiers mentioned under heading 3) can benefit from the results of a multi-temporal analysis or the classifier itself can incorporate an autoregressive temporal part.
The results of the work will include better understanding of the nature of the physical and biological phenomena under study, and of the mathematical methods applied. In this context it is considered very important that a Ph.D. student is affiliated with both an application oriented institution and a mathematical methodology oriented institution. Hence the sharing of a Ph.D. student. Furthermore, the results will consist of healthy algorithms implemented as computer programs and the final end products from computer runs on the project data. These results can be used directly by the partners in the project, and the methods with examples from the project can be published in the international scientific literature.