Thesis

Adopting a multidisciplinary approach based on 1) the content of spatial and multivariate analyses, 2) the use of a geographic information system, 3) experimental objectives defined by ecologists, this thesis gives new elements for the analysis of spatially referenced and multivariate ecological data. The first part presents the elements of the dialogue. It consists of 1) a synthesis, based on the duality diagram, of the main methods of multivariate analysis, 2) a description of written interfaces between three widely used software (R, ADE-4 and ArcView), 3) a presentation of the collaborations with ecologists, particularly N. Pettorelli, which, more than a simple statistical consulting, offers a number of methodological problems.

In the second part, the results obtained in these collaborations are presented with English publications (6 accepted and 4 submitted). The use of canonical correlation analysis of occurrences lists allows studying the spatial structure of forests at regional or local scale. Modeling the prevalence of a vector-transmitted disease, with statistical methods and spatial analyses, is proposed. The study of roe-deer population dynamic in relation with habitat leads to develop two new methods of data analysis. The first one (spatial RLQ analysis) links two datasets recorded from two different samplings of the same area, the second one (spatial NIPALS analysis) consists in PCA of data from k different samplings of the same area. Lastly, a compromise between co-inertia analysis and procrustean rotation is introduced for studying the co-structure of two datasets.