The study was conducted in the region of Caucaia do Alto (23°40' S,
47°01' W) about 80 km south-west of the city of São Paulo, in a
transition zone between dense ombrophilous forest and semi-deciduous
forest classified as "Lower Montane Atlantic Rain Forest"  in the State of São Paulo, Brazil. Elevation ranges from 800 to 1100 m .
Monthly mean temperature range from a minimum of 11°C to a maximum of
27°C. The annual precipitation level is 1300–1400 mm and fluctuates
seasonally with the driest and coldest months between April and August.
The area includes a fragmented landscape, composed of secondary
forest fragments embedded in an agricultural landscape, and an adjacent
10.000 ha forest reserve (Morro Grande Forest Reserve). Secondary
forest covers 31% of the fragmented landscape, which is dominated by
anthropogenic altered habitat (33% agricultural fields, 15% areas with
rural buildings or urban areas, 10% vegetation in early stages of
regeneration, 7% pine and eucalyptus plantations, 4% others). A more
detailed description of the study area can be found in Pardini et al. .
This study is part of a larger project investigating the effects of
forest fragmentation on the small mammal community in Caucaia do Alto [16,21].
In the overall project several forest fragments differing in size and
connectivity to other fragments as well as different plots in the
continuous forest of the Morro Grande Forest Reserve were studied. To
investigate the microhabitat preferences of small mammals, five forest
fragments and one study site in the continuous forest were chosen, so
that all sites were of secondary growth forest between 50 and 80 years
but varied widely in terms of fragment size. Two sites were located in
fragments of about 14 ha, two in fragments approximately twice as large
(30 ha), one in a 175 ha fragment and one within secondary forest
inside the Morro Grande Forest Reserve. The study sites were at least 8
km apart and movement of individuals between them was unlikely (and not
detected). However, other forest fragments were in the vicinity of the
study sites and inter-fragment movement between these fragments was
Regular trapping grids of one hundred trap locations 20 m apart from
one another were established in all six study sites. One small and one
large trap (23 × 9 × 8 cm and 38 × 11 × 10 cm, respectively; Sherman
Traps Inc., Tallahassee, USA) were set up at each trap location. One
trap was put on the ground, the other one at an approximate height of
1.0 to 1.5 m, alternating the positions of small and large traps from
one trap location to the next.
Data collection was carried out during two trapping sessions in each
of the study sites from March to June 2004 (1st session:
4.03.-7.04.2004, 2nd session: 18.05.- 26.06.2004). Each session
consisted of six consecutive nights of capture, totalling 2400 trap
nights per study site. The traps were baited with banana, and a mixture
of peanut butter, oat and sardines. Traps were left open for the night,
checked every morning and rebaited if necessary. Captured animals were
anaesthetized (Forene®, Abbott GmbH, Wiesbaden, Germany) for
1–2 minutes and marked individually by numbered ear tags (Fish and
small animal tag size 1, National Band and Tag Co., Newport, Kentucky,
USA). In addition to sexing and weighing, the length of their tibia was
measured to the nearest 0.5 mm (measurements were taken for another
study). All individuals were released at their respective trapping
Each of the 600 trap locations were characterized with respect to
vegetation structure within a 5-meter radius around the trap locations
according to August .
Vegetation characteristics were canopy height (estimated with the help
of a marked 5-meter pole in meters and grouped into seven categories:
< 6 m, 6.1–8.0 m, 8.1–10.0 m, 10.1–12.0 m, 12.1–15.0 m, 15.1–20.0 m
and 20.1–25 m), canopy cover (estimations were based on a scale from 1
to 4: 1 equals less than 25% cover while 4 counts for 76–100% cover),
vegetation density at 0 to 0.5 m, 0.5 to 1.5 m and 1.5 to 3 m
(estimations were based on a scale from 1 to 5: 1 indicates 20% of the
specific strata is covered by plants, 5 indicates 81–100% of the
specific strata is covered by plants), the amount of bamboo
(estimations were based on a scale from 1 to 4: 1 represents a
percentage of cover of up to 25 of bamboo while 4 represents 76–100
percent bamboo) and the quantity of horizontal structures as an
indicator of the connectivity of the vegetation above ground (1–4
scale: 1 means that few or no surrounding trees are connected by
horizontal structures, 4 means a 76–100% of the surrounding trees are
connected to each other by horizontal structures at any height).
The vegetation variables were standardized and reduced to three
principal components (PC1, PC2 and PC3) to minimize correlation and to
identify major traits of variation in vegetation structure (SPSS
Principal Component Analysis, default settings, Varimax rotation).
We investigated if species distribution is influenced by micro-scale
variation in vegetation structure and if these influences depend on the
fragment where animals were captured. To test these hypotheses, we
compared the scores of the principal components (dependent variables:
PC1–PC3) using a factorial ANOVA, with FRAGMENT (1–6) and USE of trap
locations (used and unused) as factors. Thus we investigated
differences in principal component scores among fragments and between
used and unused traps, as well as the existence of interactions between
these two factors. Significant interactions indicate that the influence
of micro-scale variation in vegetation structure is dependent on the
Only A. montensis and M. incanus were captured in all fragments. Other species were captured only in five (D. sublineatus, G. microtarsus) or four (O. nigripes)
fragments. The component scores of all principal components deviated
little from a normal distribution (Kolmogorov-Smirnov-Test). In
analysis of variance, non-normality of data can be tolerated,
especially when sample size is large .
The assumption of homoscedasticity was tested by Levene's test for
homogeneity of variances. Variances of PC1 and PC2 differed among
fragments. ANOVA is robust to deviation from this assumption where
sample size is large .
The comparison among fragments included six treatments with n >= 100
each. With many samples or treatments the effect of any one sample
variance on the estimated variances within treatments can only be small
. Therefore we continued with the analysis although Levene's test indicated heteroscedasticity in PC1 and PC2 among fragments.
We aimed to keep data independent of individual behavioural
responses to capture. Therefore, only first-captured individuals of
both capture sessions were considered . Trap locations were weighted by the number of individuals captured at the respective location.
All tests were conducted on STATISTICA 6.0 (StatSoft Inc., Tulsa) using a significance level of 0.05.