Spatial Variation in Tree Species Composition 23distance or habitat variables alone were 15 and
43%, respectively, strikingly similar to the results
of Phillipset al.
In contrast, two additional studies (Duqueet al.
2002, Pottset al. 2002) reported a small and
non-significant correlation between floristic sim-
ilarity (as measured by the Steinhaus index)
and linear geographical distance, and a very
high and significant correlation with environ-
ment (Mantel tests). For example, Duqueet al.
(2002) reported another elegant field study where
all trees greater than 2.5 cm diameter at breast
height (dbh) were identified in thirty 0.1 ha plots
alon gthe río Caquetá, southern Colombia (see
also Pottset al. 2002). Unfortunately, neither
study tested the expectation that floristic simi-
larity should decrease with the log-transformed
geographical distance. For this reason, the con-
clusion in both papers that habitat specialization
plays a far more important role than distance
in structurin gtropical tree communities remains
unconvincing.
Similar conclusions were reached by two stud-
ies aimed at examinin glar ge-scale biodiversity
patterns in selected plant groups. Tuomisto,
Ruokolainen and colleagues have worked on the
distribution of ferns and fern allies (Pteridophyta)
and shrubby plants in the family Melastomataceae
(henceforth melastomes).They have assembled an
unrivaled dataset comprisin gca. 300,000 ferns
and 40,000 melastomes in numerous neotropical
forest sites (Tuomisto and Ruokolainen 1994,
Tuomistoet al. 1995, 2003a). Vormisto and col-
leagues have worked on neotropical palm species
(Vormistoet al. 2004). Because ferns, melas-
tomes, and palms have such distinct biological
features (habitat specialization, dispersal syn-
dromes, growth form), it is particularly relevant to
contrast these studies. Both Vormistoet al. (2004)
and Tuomistoet al. (2003b) used similar floristic
diversity indices (Sørensen and Jaccard, respec-
tively). They found that space alone explained
22, 31, and 22% of the floristic variation (ferns,
melastomes, and palms, respectively), environ-
ment alone 34, 33, and 8%, and the interaction
between the two 15, 16, and 38%.
All of the abovementioned studies are based
on very large field inventories, and they show
remarkably convergent patterns.The contribution
of space to floristic variation appears roughly
constant across several groups of understory and
canopy plants, at around 20–30%. The contri-
bution of the environment appears much more
variable, perhaps because of the important differ-
ences in habitat characteristics measured across
studies, but there is little doubt that the environ-
ment explains 10–40% of the floristic variation
(Phillipset al. 2003, Tuomistoet al. 2003b). Thus,
any forthcomin gmechanistic theory of biodiver-
sity aimin gat predictin gpatterns of biodiversity at
the landscape scale should take into account both
dispersal and environmental factors. The above
results hold even though some of the studies have
employed statistical methods that do not really
partition beta-diversity (Legendreet al. 2005).
This is an obvious limitation of the above com-
parison, but should also be taken as an evidence
that the method based on distance matrices does
grasp the main trends in biological datasets.Measuring the environment of plantsAt present, there is no comprehensive review
on the quantitative relationships between soil
features and beta-diversity, but Sollins (1998) pro-
vided an excellent critique of studies focusing
on the detailed influence of soil on plant species
occurrence in tropical forests, which may serve
as a background in this discussion. He suggested
that the followin gfour soil factors may influ-
ence species occurrence, in decreasin gorder of
importance: (1) available P content; (2) free Al
content; (3) soil physical properties; and (4) avail-
ability of base-metal cations. It is unclear whether
the influence of soil on floristic diversity would
be similar to that on species occurrence. To my
knowledge, only Phillipset al. (2003) mentioned a
significant Mantel correlation between the floris-
tic diversity matrix and log-transformed P content
(reported in their figure 7). However, this corre-
lation could be spurious because of the contrast
between rich Holocene soils and poor Pleistocene
soils, which also show strikingly different floris-
tic composition. Phosphorus is generally limiting
in tropical soils (Vitousek 1984), it may be an
important driver of beta-diversity (Gartlanet al.
1986). Soil Al content may also prevent the