ISSN 10674136, Russian Journal of Ecology, 2010, Vol. 41, No. 1, pp. 84–94. © Pleiades Publishing, Ltd., 2010.
Models predicting the spatial distribution of spe
cies (Boyce and McDonald, 1999; Guisan and Zim
mermann, 2000; Manly et al., 2002; Pearce and
Boyce, 2006)—sometimes called resource selection
function or habitat suitability models—are currently
gaining interest. As they often help both in understand
ing species niche requirements and predicting species
potential distribution, their use has been especially pro
moted to tackle conservation issues, such as managing
species distribution, assessing ecological impacts of var
ious factors (e.g. pollution, climate change), risk of bio
logical invasions or endangered species management
(Scott et al., 2002; Guisan and Thuiller, 2005).
The geographic distribution of a species is then pre
dicted by mapping the area where these environmental
requirements are met (Elith et al., 2006). Depending
on data quality and the application at hand, these
models can assist in identifying previously unknown
populations, determining sites of high candidacy for
The article is published in the original.
reintroductions, guiding additional surveys, and
informing selection and management of protected
areas (Graham et al., 2004).
The quantification of such species–environment
relationships represents the core of predictive geo
graphical modeling in ecology. These models are gen
erally based on various hypotheses as to how environ
mental factors control the distribution of species and
communities (Austin, 2002).
Recently, several approaches to predictive model
ing of species geographic distributions have been
developed in a geographic information system (GIS).
Main statistical approaches (to modeling) grouped into
seven categories: Multiple regression and its generalized
forms, Classification techniques, Environmental enve
lopes, Ordination techniques, Bayesian approaches,
Neural networks and a seventh category including other
potential approaches or approaches involving several
methods (mixed approach) [Tan, 2007].
Bio et al. (1998) tested the species response shapes
of 156 species in relation to six environmental vari
ables using GAM (Generalized additive model).
Multivariate Statistical Methods as a Tool for ModelBased
Prediction of Vegetation Types
M. A. Zare Chahouki, H. Azarnivand, M. Jafari, and A. Tavili
Department of Rehabilitation of Arid and Mountainous Regions, University of Tehran, Iran
Received November 3, 2009
—The current research was carried out to find the most effective environmental factors in plant spe
cies occurrence and providing their predictive habitat models. For this purpose, study was conducted in Posh
tkouh rangelands of Yazd province in the central of Iran. For modeling, vegetation data in addition to site
condition information including topography, climate, geology and soil were prepared. CCA and Logistic
regression (LR) techniques were implemented for plant species predictive modeling. To plants predictive
mapping, it is necessary to prepare the maps of all affective factors of models. To mapping soil characteristics,
geoestatistical method including variogram analysis and Kriging interpolation were used. Based on obtained
predictive models for each species (through LR method) and for whole species (through CCA method)
related predictive maps were prepared in GIS. The accuracy of predictive maps were tested with actual vege
tation maps. Vegetation modeling results with CCA indicates that predictive map of vegetation corresponds
with actual map (with high accuracy). Predictive maps of
Cornulaca monachantha, Ephedra strobilaceaZygo
phyllum eurypterum, Seidlitzia rosmarinus
, which have narrow amplitude, has high
accordance with actual vegetation map prepared for the study area. Among species of study area, predictive
, due to its ability to grow in most parts of Poshtkouh rangelands with relatively dif
ferent habitat conditions, is not possible. Comparing CCA and LR methods showed that each technique has
its advantages and drawbacks. In general, LR will provide better specificmodel, but CCA will provide a
broader overview of multiple species.
: Canonical Correspondence Analysis, Environmental factors, Geostatistical methods, Logistic
Regression, Poshtkouh rangelands, Predictive vegetation modeling.