Open AccessFinding essential scales of spatial variation in ecological data: a multivariate approach
Author(s) -
Jombart Thibaut,
Dray Stéphane,
Dufour AnneBéatrice
Publication year - 2009
Publication title -
ecography
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 2.973
H-Index - 128
eISSN - 1600-0587
pISSN - 0906-7590
DOI - 10.1111/j.1600-0587.2008.05567.x
Subject(s) - ordination , scale (ratio) , multivariate statistics , identification (biology) , principal component analysis , spatial ecology , spatial analysis , computer science , ecology , structuring , data mining , spatial variability , geography , statistics , cartography , mathematics , artificial intelligence , machine learning , biology , finance , economics
The identification of spatial structures is a key step in understanding the ecological processes structuring the distribution of organisms. Spatial patterns in species distributions result from a combination of several processes occuring at different scales: identifying these scales is thus a crucial issue. Recent studies have proposed a new family of spatial predictors (PCNM: principal coordinates of neighbours matrices; MEMs: Moran's eigenvectors maps) that allow for modelling of spatial variation on different scales. To assess the multi‐scale spatial patterns in multivariate data, these variables are often used as predictors in constrained ordination methods. However, the selection of the appropriate spatial predictors is still troublesome, and the identification of the main scales of spatial variation remains an open question. This paper presents a new statistical tool to tackle this issue: the multi‐scale pattern analysis (MSPA). This ordination method uses MEMs to decompose ecological variability into several spatial scales and then summarizes this decomposition using graphical representations. A canonical form of MSPA can also be used to assess the spatial scales of the species‐environment relationships. MSPA is compared to constrained ordination using simulated data, and illustrated using the famous oribatid mites dataset. The method is implemented in the free software R.