ORI GIN AL PA PER
Accessibility predicts structural variation of Andean
Polylepis forests
Johanna M. Toivonen • Michael Kessler • Kalle Ruokolainen •
Dietrich Hertel
Received: 23 July 2010 / Accepted: 16 April 2011 / Published online: 26 April 2011
 Springer Science+Business Media B.V. 2011
Abstract High Andean mountain forests, formed almost purely by trees of the genus
Polylepis, occur nowadays as scattered remnant patches of a more continuous past dis-
tribution. Apparently, the destruction of Polylepis forests has mainly been caused by
millennia of human disturbance, although forest distribution may also have fluctuated
according to prevailing climatic conditions. Nowadays, the remaining Polylepis forest
stands are still threatened by anthropogenic disturbance, which gradually degrades the
forests. The aim of our study was to test if the structural variation of Polylepis forest
patches, as an indication of forest degradation, can be predicted by accessibility to humans.
The study was carried out in the Cordilleras Vilcanota and Vilcabamba, Cuzco, Peru. We
used indices of forest biomass and proportion of vegetative regeneration as forest structural
variables. First we examined the dependence of these variables on elevation with linear
regressions. We did this separately for different Polylepis species and combining the
species within humid and dry areas. Thereafter, we used the residual forest structural
variation to assess possible relationships with accessibility, quantified as geographical
distance to the nearest village, road or market centre. We found several significant rela-
tionships between the structural variables and accessibility, which may reflect different
landscape related preferences in forest use. The results suggest accessibility can be used for
rapid spatial prediction of Polylepis forest degradation, which facilitates identifying
Polylepis forests that are potentially the most degraded and therefore in the most urgent
need of restoration or conservation activities.
J. M. Toivonen (&)  K. Ruokolainen
Department of Biology, Section of Biodiversity and Environmental Science, University of Turku,
Turku 20014, Finland
e-mail: jomito@utu.fi
M. Kessler
Institute for Systematic Botany, University of Zurich, Zurich, Switzerland
D. Hertel
Plant Ecology, Albrecht-von-Haller Institute for Plant Sciences, University of Go¨ttingen,
Gottingen, Germany
123
Biodivers Conserv (2011) 20:1789–1802
DOI 10.1007/s10531-011-0061-9
Keywords Biomass  Forest degradation  Forest regeneration  High Andes 
Human disturbance  Peru
Introduction
Trees of the genus Polylepis (Rosaceae) typically form monospecific stands in the Andean
highlands above 3300 m. In the central Andes, Polylepis can reach up to 4800–5200 m,
thereby forming the highest treelines in the world. Polylepis forests can be regarded as a
key ecosystem in the high Andes and of outstanding importance for biodiversity, because
they provide irreplaceable habitat for several endemic plant and animal species (Fjeldsa˚
1993; Servat et al. 2002; Lloyd and Marsden 2008). Furthermore, they have an important
role in water circulation and prevention of soil erosion (Fjeldsa˚ and Kessler 1996; Fjeldsa˚
2002) and represent a potential carbon sink in the otherwise treeless landscape (Hoch and
Ko¨rner 2005). Polylepis forests also provide an important source of timber and firewood
for local people (Lazcano and Espinoza 2001; Aucca and Ramsay 2005; Hagaman 2006;
Jameson et al. 2007).
The high Andes have been subject to intensive human land use for millennia (Chepstow-
Lusty et al. 1996; Chepstow-Lusty and Winfield 2000), and as a consequence, Polylepis
forests have been strongly decimated and fragmented through timber extraction, anthro-
pogenic fires, and grazing by domestic animals (Ellenberg 1958; 1979; Laegaard 1992;
Kessler 1995, 2000, 2002; Purcell and Brelsford 2004; Coblentz and Keating 2008). This
human influence has been especially strong in the inner parts of the Andean mountain
ranges that have been densely and permanently populated for thousands of years and are
nowadays even more densely populated, in comparison to the sparsely populated Amazo-
nian side of the Andes that is still partly inaccessible (Dobyns 1966; Kessler and Driesch
1993; Etter and Villa 2000). Human impact may not be the only cause for the current
fragmented distribution of Polylepis forests. Palaeoecological studies have also shown that
the extent of Polylepis forests has strongly fluctuated already before the arrival of humans in
South America ([12,000 years ago), suggesting that Polylepis forests may partly be nat-
urally fragmented due to climatic factors (Gosling et al. 2009). In any case, Polylepis forests
are counted among the most endangered tropical and subtropical mountain forest ecosys-
tems in the world (UNEP-WCMC 2004) and the majority of the ca. 30 species of the genus
are classified as vulnerable (IUCN 2010).
Currently, one of the major threats for remaining Polylepis forest stands seems to be
gradual habitat degradation (Renison et al. 2006, 2010, 2011; Jameson et al. 2007;
Cingolani et al. 2008). Complete disappearance of the remaining forest patches also occurs,
but it seems to be a smaller problem (Jameson et al. 2007). The degradation is caused
especially by livestock grazing as well as wood harvesting for timber and fire wood
(Renison and Cingolani 1998; Renison et al. 2004). Grazing animals affect Polylepis forest
regeneration so that regeneration in general becomes more difficult and root suckers, being
more resistant to physical damage than seedlings, become the dominant form of regen-
eration. Excessive grazing and wood harvesting inevitably lead to a decrease in forest
biomass. This degradation has also been documented to manifest itself as a lack or dim-
inution of certain more heavily used age and/or size cohorts (Renison et al. 2011) and
overall decrease in canopy density (Jameson et al. 2007).
Our study is based on an assumption that forest degradation manifests itself as a
decrease in above-ground biomass and changes in forest regeneration patterns. However,
1790 Biodivers Conserv (2011) 20:1789–1802
123
these same changes can also be a consequence of natural factors, such as forest distur-
bance, diseases, herbivory or low temperature. Of these, temperature is a special case as it
is by necessity closely linked with elevation above sea level. It is a general rule for
mountain forests that above-ground biomass, and thereby canopy height and tree diameter
decrease with elevation (e.g. Stadtmu¨ller 1987; Young 1993; Rada et al. 1996; Garcia-
Nun˜ez et al. 2004; Hoch and Ko¨rner 2005; Kessler et al. 2007; Hertel and Wesche 2008).
There is also evidence of proportional increase in vegetative regeneration relative to sexual
reproduction with elevation (Cuevas 2000; Germino and Smith 2002; Cierjacks et al.
2007a, b, c; Hertel and Wesche 2008).
We assume that a statistical effect of elevation on above-ground biomass and relative
abundance of vegetative regeneration can be taken as a biological necessity. Therefore,
interesting variation in these features of Polylepis forests should be sought only in residuals
after the effect of elevation has been taken into account. We are specifically interested in
seeing if the residual variation can be explained by human disturbance. The amount of
human disturbance in a forest patch is difficult to measure directly and therefore we use
accessibility as its indicator. We hypothesize that accessibility is negatively related to
above-ground biomass and positively related to the proportion of vegetative regeneration.
We test this hypothesis by correlating indicators of Polylepis forest biomass and regen-
eration type with measures of geographical distance to road or human settlement. As
human influence has historically been stronger in the central, drier parts of the Andes,
where the centers of pre-Hispanic cultures were located and which are also nowadays the
most densely populated areas, we hypothesize that structural variation of Polylepis forests
is currently more strongly associated with human influence in dry areas than in more
distant and inaccessible humid areas on the Amazonian side of the Andes.
The degradation of Polylepis forests can be counteracted by improving legislation and
increasing public awareness. However, rapidly increasing forest degradation and frag-
mentation needs explicit spatial identification of potentially the most endangered and
therefore critical areas for conservation. As resources for field inventories to identify the
most critical Polylepis forest patches for conservation are necessarily limited, simple
geographical measures would be useful for making spatial predictions of the degree of
Polylepis forest degradation. Our study tests if indices of accessibility can provide such a
measure.
Methods
Study area and species
We carried out our field work in the Cordilleras Vilcanota and Vilcabamba in Urubamba,
Cuzco, south-eastern Peru (13070–13170 S and 72020–72290 W) in 2006 (Fig. 1). The
climate of the region varies from semi-humid to semi-dry with a clear wet season
(October–March) and very strong diurnal temperature fluctuations especially in the dry
season. The complex orography of the region directs clouds to the area from the north-
west. When the clouds hit high Andean ridges, they discharge much of their water as rain.
This process creates strong geographical gradients in precipitation so that areas behind the
high ridges remain in the rain shadow. This situation is exemplified in our study area by the
contrasting precipitation records at the climate stations of Urubamba (2863 m) with
454 mm mean annual precipitation and Win˜aywayna, protected area of Machupicchu
(2800 m), with 1606 mm (records of former INRENA, National Institute of Natural
Biodivers Conserv (2011) 20:1789–1802 1791
123
Resources of Peru and SENAMHI, National Service of Meteorology and Hydrology of
Peru). Based on this gradient and the geographical variation in Polylepis species compo-
sition (Kessler 1995; Fjeldsa˚ and Kessler 1996), we separated the studied valleys as dry
and humid ones as shown in Fig. 1. Drier valleys correspond to relatively densely inhabited
inner parts of the Andes and humid valleys to less populous Amazonian side of the Andes.
Steep topography and remarkable elevational differences of the region have created
diverse climatic conditions, which have enabled the evolution, diversification and spe-
cialization of many plant and animal groups, including Polylepis (Fjeldsa˚ 1992; Fjeldsa˚
and Kessler 1996). In our study area, at least five Polylepis species occur in close proximity
to each other, representing the highest concentration of species of the genus found any-
where (Fjeldsa˚ and Kessler 1996). Species are segregated by elevation and humidity, and
almost always form monospecific stands: P. pauta, P. sericea and P. pepei in humid areas,
and P. racemosa and P. subsericans in drier areas (Kessler 1995; Fjeldsa˚ and Kessler
1996).
Data collection
Our study area comprised nine different valleys at elevations between 3070 and 4550 m
(Fig. 1). Five of the valleys were located in relatively humid and four in dry areas. In
each valley we selected one to four forest patches for further study. The number of
selected forest patches in each valley depended on the total area of the forest cover and
the number of species in the valley. The forest patches were selected from different
elevations to cover as well as possible the whole altitudinal range of each species in the
study area. The altitudinal difference between two forest patches of the same species in
the same valley was at least 100 m. Within each forest patch, we established one study
plot of 10 m 9 10 m. The plots were located in an accessible point in the middle of the
Fig. 1 Map of the study area with nine studied valleys. Valleys 1–5 are in humid and 6–9 in dry climate
1792 Biodivers Conserv (2011) 20:1789–1802
123
forest patch, at least 25 m from the edge so as to give—according to a subjective
assessment—as representative sample as possible of the structure of the forest patch in
general. In three valleys, we established two plots within the same forest patch with only
small elevational difference between them, because of the problems to find accessible
forest patches far enough from each other. In these cases, we used average values of
measured forest characters of the two plots to avoid pseudo-replication. After the aver-
aging, we had in total 13 forest patches in humid areas and 11 in dry areas of which four
were formed by P. pauta, four by P. sericea, five by P. pepei, six by P. racemosa and
five by P. subsericans.
In each plot, we measured the circumference and visually estimated the height of all
Polylepis trees (C10 cm of circumference at breast height). We also counted the number of
all smaller individuals. Among them, we separated vegetatively generated root suckers
from sexually produced individuals by inspecting if the individual sprouted from the roots
or branches of a bigger tree. Root suckers were counted as separate individuals when
clearly rooted fine roots were observed between the sprouts.
One forest patch always included just one Polylepis species and in 16 out of the 24
studied forest patches the given Polylepis species was the only observed tree species. In the
mixed patches, Polylepis was always the most abundant tree species.
Forest structural variables
One of our aims in the field observations was to obtain measures that would give us
information about the amount of Polylepis forest above-ground biomass, because a general
effect of excessive forest use is reduction in biomass. For this end, we measured mean and
maximum Polylepis tree height and circumference per plot. Maximum tree height and
circumference were defined as average values of the three highest or thickest trees in a plot.
Basal area per plot was not applicable as the given Polylepis species was not the only tree
species in one third of the plots. Another forest feature that is often observed to be affected
by human activities is the proportion of vegetative regeneration. As measures of this
feature we used the percentage of root suckers of the combined number of seedlings,
saplings and root suckers.
For very basic biological reasons, we take it for granted that forest above-ground
biomass decreases and proportion of vegetative regeneration increases with elevation. In
other words, we have a situation in which the variance of our variables of interest is
associated with elevation. Therefore, before addressing our research question about forest
degradation, we first removed the statistical dependence of these forest structural variables
with elevation. We did this by calculating linear regressions in which elevation was the
explanatory variable and each of the five forest structural variables separately the
dependent variables. The residuals of each of the dependent variables were then taken as
the new dependent variables to be explained by our indices of accessibility. We calculated
the regressions both across all species taken together within dry and within humid area, and
separately for each species.
Taking residuals from an overall regression in which all species are included can be
justified if the decline in tree size in our study area is uniform over all species. However, if
each species can have somewhat different mathematical function describing the decline
and/or if different tree species have different sizes when growing at the same spot, then the
residuals should be taken separately for each species. The available knowledge of natural
size and its variation of the Polylepis species included in our study is not conclusive
enough to define which way to deal with the residuals should be used.
Biodivers Conserv (2011) 20:1789–1802 1793
123
1794 Biodivers Conserv (2011) 20:1789–1802
123
Because of the strong biological justification, we took residuals only of regressions
having the assumed sign of regression coefficient (i.e., positive or negative slope of the
regression line, Fig. 2). In cases where the sign of the regression coefficient was
actually observed to be opposite to the one expected, we took the original values
(effectively the same as forcing the regression coefficient to be zero). In the case of
calculating regressions for individual species, the residuals in each species–specific
regression, or original values, were then separately standardised so as to keep each
species’ contribution to the final dependent variable directly proportional to the number
of plots in which the species had been observed. Finally, the species–specific stand-
ardised residuals of each forest structural variable were put together for the dry and
humid valleys separately.
Indices of accessibility
As indices of accessibility of forest patch, we used measures of geographical distance from
the study plot (1) to the nearest village, (2) to the nearest paved road or railway station, and
(3) to the nearest market centre (Table 1). Distances were measured along the easiest
walking or driving routes deduced from topographic information extracted from the
satellite images of Google Earth.
Data analysis
We tested the hypothesis that accessibility explains the degree of Polylepis forest degra-
dation by calculating how well the indices of accessibility explain, in a linear regression,
the residual variance of each of the measures of forest structure and regeneration.
Forest structure and regeneration can be affected by several naturally occurring
biological factors, like herbivores, pathogens and symbiotic organisms. These factors
are difficult to quantify in practice, but typical for them is a spatial structure in which
sites that are nearby are more similar to each other in terms of the factor than sites that
lie far apart (Dale et al. 2002). Such spatial structure can also complicate the inter-
pretation of other statistical analyses. We tested this biologically justified spatial pre-
diction separately in dry and humid valleys for each residual variable of forest structure
and regeneration. We first constructed matrices of Euclidean distance separately on the
basis of each residual variable. These distance matrices were then correlated in a
standardised Mantel test (Smouse et al. 1986) with the corresponding horizontal geo-
graphical distance matrix of plots in either dry or humid valleys. The statistical sig-
nificances of the Mantel tests were estimated through 999 permutations. We run
regression and correlation analyses with the statistical software SPSS 19.0, and calcu-
lated the matrices and performed Mantel tests with the statistical software R-package
(Legendre and Vaudor 1991).
Fig. 2 Forest structural characters in relation to elevation in Polylepis forest patches. Different symbols
present different species. Humid areas: filled circle = P. pauta; open circle = P. sericea; filled inverted
triangle = P. pepei. Dry areas: filled circle = P. racemosa; open circle = P. subsericans. Short solid lines
are species–specific linear regressions. Short dashed lines indicate the cases when species–specific
regression slope was opposite to the one expected and therefore the slope was set to zero and standardized
values were used for further analysis. Long solid lines are overall regression across all species. When these
lines were opposite to the expected elevational patterns, standardized values were also used for further
analysis (four out of five cases in dry areas). Thicker lines indicate statistically significant regressions
(P \ 0.05)
b
Biodivers Conserv (2011) 20:1789–1802 1795
123
Results
Elevational patterns of forest structure
The three Polylepis species (P. pauta, P. sericea and P. pepei) of humid areas were
observed in only slightly overlapping altitudinal ranges, and the two species (P. racemosa
and P. subsericans) of dry areas had no overlap (Table 1; Fig. 2). All forests were rather
low (tree height being roughly 10 m and below) and tree circumferences only rarely
exceeded 100 cm.
In humid areas, when taking all species into one analysis, we observed the expected sign
of regression coefficient in all forest structural features (Fig. 2). The regressions were
statistically significant for features of mean and maximum tree height and mean tree
circumference. In the similar regression analysis in dry areas, the expected sign was
observed only for vegetative regeneration—in all other cases the sign was contrary to the
biologically justified one, yet only in the case of maximum tree circumference the sign was
statistically significantly different from zero.
Looking each species separately, the species in humid areas showed in most cases the
expected sign of regression coefficient (Fig. 2). In dry areas, the expected relationships
with elevation appeared in three out of five cases for P. racemosa, but P. subsericans
consistently showed an opposite trend in relation with elevation. Due to the low number of
Table 1 Average values and standard errors of forest patch characteristics for each Polylepis species
Humid Dry
Species: P. pauta P. sericea P. pepei P. racemosa P. subsericans
Number of plots: 4 4 5 6 5
Forest patch character
Elevation (m) 3520 ± 156 3916 ± 111 4240 ± 47 4100 ± 48 4379 ± 31
Max tree height (m) 11.4 ± 2.1 6.8 ± 0.9 4.9 ± 0.4 8.5 ± 0.8 9.4 ± 0.9
Mean tree height (m) 8.6 ± 1.5 4.5 ± 0.8 3.3 ± 0.3 5.1 ± 0.8 6.9 ± 0.8
Max tree circ (cm) 128.3 ± 40.1 69.6 ± 8.4 61.8 ± 8.1 85.3 ± 9.2 119.3 ± 17.4
Mean tree circ (cm) 65.3 ± 6.6 30.7 ± 6.1 31.7 ± 3.6 45.8 ± 7.4 72.7 ± 14.5
Tree density
(ind./100 m2)
20.5 ± 10.4 46.3 ± 17.7 24.1 ± 3.9 25 ± 4.1 16.8 ± 4.8
Vegetative
regeneration (%)
57.7 ± 19.3 28.0 ± 14.6 66.2 ± 13.7 31.2 ± 14.6 62.5 ± 16.9
Total regeneration 11.0 ± 4.7 81.5 ± 29.1 332.3 ± 192.0 382.7 ± 93.8 312.5 ± 56.6
Distance to village
(km)
5.042 ± 0.82 3.914 ± 0.83 3.041 ± 1.00 4.374 ± 1.38 4.796 ± 1.06
Distance to road (km) 6.637 ± 2.16 3.939 ± 0.81 6.094 ± 2.39 8.503 ± 0.89 10.361 ± 0.97
Distance to market
(km)
36.694 ± 4.49 25.878 ± 6.09 24.01 ± 3.52 15.694 ± 1.69 17.213 ± 1.96
All characteristics except tree density are calculated only on the basis of individuals of the given Polylepis
species even if there were individuals of other species present in the study plots. Three plots of P. pauta
were mixed with other tree species and in these plots on average 57% of the trees represented P. pauta. In
the case of P. sericea, there were two mixed plots with on average 86% of the trees belonging to P. sericea.
In one plot of P. pepei, P. racemosa and P. subsericans there were also other tree species and the dominance
of Polylepis was correspondingly 96, 82 and 94% of the individuals
1796 Biodivers Conserv (2011) 20:1789–1802
123
plots per species, it was not meaningful to estimate statistical significance of the species–
specific regressions.
Accessibility in relation to forest structure
Each of the three explanatory variables had a statistically significant relation to at least one
of the dependent variables that describe residual forest structure or regeneration after
removing the statistical effect of elevation (Table 2). The relationships between dependent
and independent variables were somewhat different between dry and humid areas. Distance
to the nearest village was related rather similarly with the dependent variables in both dry
and humid areas. Distance to the nearest market centre explained remarkably well the
residual maximum tree height and circumference in dry but not in humid areas. Distance to
the nearest paved road or railway station was the explanatory variable that had most
notably a different relationship with dependent variables between the dry and humid areas.
In the dry areas, it had no statistically significant explanatory power over any of the
dependent variables, whereas in humid areas it was statistically significantly related with
most of the dependent variables. Because distance to the nearest village and nearest paved
road or railway station were strongly intercorrelated in humid areas, they should be
regarded as effectively a single independent explanatory variable. Consequently, it can be
summarised that there were four independent significant or nearly significant relationships
in both areas when the residuals of forest structural variation were extracted from species–
specific regressions. When the residuals were extracted from the regression line across all
the species within dry or humid areas there were four significant or nearly significant
relationships in the humid and three in the dry areas. The explanatory variables were
generally only weakly correlated with each other, but in humid areas distance to the nearest
village and distance to the nearest paved road or railway station were strongly related
(Pearson correlation coefficient r = 0.796, P \ 0.01).
None of the forest structural variables produced a statistically significant correlation
with geographical distances in the Mantel test (highest correlation was found for mean tree
circumference in humid areas when the residuals were extracted from the overall regres-
sion line across all the species, Mantel r = 0.241, P = 0.068).
Discussion
In both the climatically dry and humid areas, residual variation of the features on forest
structure was explainable by at least one of the indices of accessibility. The fact that this
statistical dependence was found in two independent data sets suggests that human
activities are causing variation in the studied forest features. At least it is difficult to invoke
other alternative explanatory variables that would covary with the indices of accessibility.
Our hypothesis was that the variation in Polylepis forest structure in the drier and more
densely inhabited valleys would be more strongly controlled by human activities than in
the relatively humid and more sparsely inhabited valleys. However, our results do not
support this hypothesis. The variance in the dependent variables was explained quite as
effectively by the indices of accessibility in both humid and dry valleys. This result is
somewhat surprising given that dry valleys have been historically much more affected by
humans (Binford et al. 1997; Etter and Villa 2000). The inner parts of the Andes have been
densely populated for thousands of years (Dobyns 1966; Kessler and Driesch 1993). Also,
it appears that the majority of centres of pre-Hispanic cultures were located in more
Biodivers Conserv (2011) 20:1789–1802 1797
123
T
ab
le
2
R
es
u
lt
s
o
f
li
n
ea
r
re
g
re
ss
io
n
s
b
et
w
ee
n
d
ep
en
d
en
t
v
ar
ia
b
le
s
o
n
fo
re
st
st
ru
ct
u
ra
l
ch
ar
ac
te
rs
,
an
d
in
d
ep
en
d
en
t
v
ar
ia
b
le
s
d
es
cr
ib
in
g
ac
ce
ss
ib
il
it
y
(d
is
ta
n
ce
to
th
e
n
ea
re
st
v
il
la
g
e,
p
av
ed
ro
ad
o
r
ra
il
w
ay
an
d
m
ar
k
et
ce
n
tr
e)
se
p
ar
at
el
y
fo
r
d
ry
an
d
h
u
m
id
ar
ea
s
D
is
ta
n
ce
m
ea
su
re
R
es
id
u
al
s
o
f
sp
ec
ie
s–
sp
ec
ifi
c
li
n
ea
r
re
g
re
ss
io
n
w
it
h
el
ev
at
io
n
fo
r
fo
re
st
st
ru
ct
u
ra
l
ch
ar
ac
te
rs
M
ax
tr
ee
h
ei
g
h
t
M
ea
n
tr
ee
h
ei
g
h
t
M
ax
tr
ee
ci
rc
u
m
f.
M
ea
n
tr
ee
ci
rc
u
m
f.
V
eg
.
re
g
en
er
.
H
u
m
id
V
il
la
g
e
(?
)0
.1
5
(?
)0
.2
7

(?
)0
.0
4
(?
)0
.3
5
*
(-
)0
.2
6

R
o
ad
(?
)0
.3
7
*
(?
)0
.5
0
*
*
(?
)0
.1
9
(?
)0
.5
2
*
*
(-
)0
.4
3
*
M
ar
k
et
(?
)0
.0
4
(?
)0
.0
0
(?
)0
.0
4
(?
)0
.0
0
(-
)0
.0
0
D
ry V
il
la
g
e
(?
)0
.1
3
(?
)0
.3
0

(?
)0
.2
0
(?
)0
.2
2
(-
)0
.4
3
*
R
o
ad
(-
)0
.0
0
(-
)0
.2
0
(?
)0
.0
2
(-
)0
.1
3
(?
)0
.0
3
M
ar
k
et
(?
)0
.5
4
*
(?
)0
.1
6
(?
)0
.6
9
*
*
(?
)0
.1
9
(-
)0
.0
6
D
is
ta
n
ce
m
ea
su
re
R
es
id
u
al
s
o
f
o
v
er
al
l
li
n
ea
r
re
g
re
ss
io
n
w
it
h
el
ev
at
io
n
ac
ro
ss
al
l
sp
ec
ie
s
fo
r
fo
re
st
st
ru
ct
u
ra
l
ch
ar
ac
te
rs
M
ax
tr
ee
h
ei
g
h
t
M
ea
n
tr
ee
h
ei
g
h
t
M
ax
tr
ee
ci
rc
u
m
f.
M
ea
n
tr
ee
ci
rc
u
m
f.
V
eg
.
re
g
en
er
.
H
u
m
id
V
il
la
g
e
(?
)0
.1
6
(?
)0
.1
5
(?
)0
.1
1
(?
)0
.2
0
(-
)0
.1
6
R
o
ad
(?
)0
.3
9
*
(?
)0
.3
4
*
(?
)0
.3
0

(?
)0
.5
0
*
*
(-
)0
.0
8
M
ar
k
et
(?
)0
.0
3
(-
)0
.0
4
(?
)0
.0
0
(?
)0
.0
3
(?
)0
.0
2
D
ry V
il
la
g
e
(?
)0
.0
9
(?
)0
.2
4
(?
)0
.1
2
(?
)0
.1
8
(-
)0
.3
5

R
o
ad
(?
)0
.0
0
(-
)0
.0
7
(?
)0
.1
2
(-
)0
.0
0
(?
)0
.0
6
M
ar
k
et
(?
)0
.3
3

(?
)0
.1
0
(?
)0
.6
6
*
*
(?
)0
.2
6
(-
)0
.0
5
N
u
m
b
er
s
ar
e
p
ro
p
o
rt
io
n
o
f
v
ar
ia
n
ce
ex
p
la
in
ed
in
re
g
re
ss
io
n
an
al
y
si
s,
(?
)
o
r
(-
)
sy
m
b
o
ls
in
d
ic
at
e
p
o
si
ti
v
e
o
r
n
eg
at
iv
e
re
la
ti
o
n
sh
ip
o
f
re
g
re
ss
io
n
,
*
*
in
d
ic
at
e
p
ro
b
ab
il
it
y
o
f
er
ro
r
(P
)
\
0
.0
1
,
*
P
\
0
.0
5
an
d

n
ea
rl
y
si
g
n
ifi
ca
n
t
P
-v
al
u
e
(P
=
0
.0
5
1
–
0
.0
8
2
)
1798 Biodivers Conserv (2011) 20:1789–1802
123
accessible and climatically favourable inner part of the Andes rather than in inaccessible
humid eastern side of the Andes, which is still nowadays sparsely inhabited (Etter and Villa
2000). However, one has to bear in mind that the statistical ability of accessibility to
explain the residual variation in forest structure is not necessarily directly related to the
degree of forest degradation. It may be that in one area forests are in general more
degraded than in another, and yet, in both areas the variation one observes in forest features
related to degradation can be equally strongly related to the level of human influence. In
other words, forest degradation may have advanced more in one area than in the other, but
the variance in it can be equally strongly associated to the degree of human influence in
both areas. This may be exactly the situation between the studied dry and humid valleys.
For this interpretation there is some evidence. The fact that only in dry areas there was a
statistically significant tendency of trees getting thicker towards higher elevation—a pat-
tern that is against ecological theory—and the statistically significant relation between this
tendency and an index of human accessibility, suggests that forests in dry areas are more
strongly degraded in absolute terms.
Even if the forests were equally degraded in climatically distinct areas, this is not to say
that there would not have been any difference between the two data sets from distinct
climatic conditions. In dry areas, the distance to nearest market centre was the only index
of accessibility that explained significantly variance in the dependent structural variables.
On the other hand, in the humid areas only distance to village or road were significant
explanatory variables. This may indicate that in the dry areas, Polylepis forests are used
more for providing products that can be sold, whereas in the humid areas the use of these
forests is directed more at local subsistence. However, in this interpretation one has to
realize that the sampled forest patches were not situated at equal distances from villages
and other human installations in the two data sets. In the data from humid valleys, average
distance to market centre was almost twice of that in the dry valleys. Therefore, it is
possible that the difference in the explanatory power of the distance to market centre is just
an indication of people having little interest to transport any products from Polylepis
forests to a market centre beyond a distance of ca. 20 km.
In the case of distance to village and in distance to roads, there is relatively little
difference between humid and dry areas in average distances as well as in their standard
deviations. Therefore, the observed difference in the explanatory power of these variables
appears to be indicative of a true regional difference in human socioeconomic behaviour.
However, the role of distance to the nearest road or railway station as a significant pre-
dictor is not clearly evident as it is strongly correlated with distance to the nearest village.
It is therefore quite risky to assume either one of these predictor variables more important
than the other only on the basis of the data analyzed here. In theory, distance to the nearest
village seems to be a more likely truly important variable explaining Polylepis forest
degradation, as Polylepis stands are most often managed and utilized exclusively by the
people who live in the village and own the forest (Hagaman 2006; Maxwell 2004).
However, road building has been shown to increase deforestation, and forest fragmentation
and exploitation in Amazonia and in highlands of Venezuela and Mexico (Fearnside 1980;
Dirzo and Garcı´a 1992; Young 1994; Allan et al. 2002), indicating a potential importance
of this variable explaining forest degradation.
Our findings of easily measurable geographical indices of accessibility predicting
Polylepis forest degradation are supported by earlier studies on P. pepei forests in northern
Bolivia, where forest stands near the villages were shown to be more impacted by livestock
grazing, in particular, than distant forest stands (Hagaman 2006). Indirect measures of
human influence based on geographical distances have also been used by Cingolani et al.
Biodivers Conserv (2011) 20:1789–1802 1799
123
(2008) and Renison et al. (2010) to explain variations in forest canopy cover and soil loss
in forests of P. australis in central Argentina. We used a very simple linear model and our
sample size was rather modest. Therefore, one would expect that it would be possible to
build more powerful predictive models by increasing model flexibility if more data points
were available. However, we do not think that building such better models would be
particularly cost-effective if the purpose is to guide conservation and restoration efforts
into the most critical areas. Improving the success of modelling would require considerable
effort in increasing the number of field inventories. This effort would probably be better
invested in looking for direct measures of disturbance such as fire scars, dung of domestic
animals and cut trees or stumps, together with interviews of local people on their use of
Polylepis forests. This kind of detailed field studies would most probably be necessary
anyway before starting with conservation and management activities. However, one of the
first steps in such activities would necessarily be the identification of the most vulnerable
forest patches. Our results add to the increasing evidence (Hagaman 2006; Cingolani et al.
2008; Renison et al. 2010) that such first priority spots for active management and con-
servation efforts are to be found in the relative proximity of villages and other human
infrastructure. Alternatively, if the aim of conservation action is to identify forest patches
with least human impact and hence lowest potential conflicts with conservation action,
then remote forest patches may be the best choice.
In conclusion, we found that purely geographical measures of accessibility of a
Polylepis forest patch are related to indicators of forest biomass and proportion of vege-
tative regeneration in dry and humid areas in the cordilleras Vilcanota and Vilcabamba, in
Cuzco, Peru. Use of these relatively easily obtainable variables can help to locate areas
where Polylepis forests are most degraded and therefore in most urgent need of protection
and sustainable management.
Acknowledgments We thank the NGO ECOAN for the collaboration in the field work. A special thanks
goes to the persistent field assistant Louella Puelles Linares. We also thank Alfredo Tupayachi Herrera for
his guidance and help in the field and Carlos Gonzales Inca for the study area map. We are also grateful to
INRENA—Machupicchu for a working permit in the protected area of Machupicchu. The study was funded
by Jenny and Antti Wihuri Foundation and Turku University Foundation.
References
Allan JD, Brenner AJ, Erazo J, Fernandez L, Flecker AS, Karwan DL, Segnini S, Taphorn DC (2002) Land
use in watersheds of Venezuelan Andes: a comparative analysis. Conserv Biol 16(2):238–527
Aucca C, Ramsay PM (2005) Management of biodiversity and land use in southern Peru. ECOAN’s
activities to help conserve Polylepis woodlands. Mt Res Dev 25:287–289
Binford MW, Kolata AL, Brenner M, Janusek JW, Seddon MT, Abbott M, Curtis JH (1997) Quatern Res
47:235–248
Chepstow-Lusty AJ, Winfield M (2000) Inca agroforestry: lessons from the past. Ambio 29(6):322–328
Chepstow-Lusty AJ, Bennett KD, Switsur VR, Kendall A (1996) 4000 years of human impact and vege-
tation change in the central Peruvian Andes—with events paralleling the Maya record? Antiquity
70:824–833
Cierjacks A, Wesche K, Hensen I (2007a) Potential lateral expansion of Polylepis forest fragments in central
Ecuador. For Ecol Manag 242:477–486
Cierjacks A, Ru¨hr NK, Wesche K, Hensen I (2007b) Effects of altitude and livestock on the regeneration of
two tree line forming Polylepis species in Ecuador. Plant Ecol 194:207–221
Cierjacks A, Iglesias JE, Wesche K, Hensen I (2007c) Impact of sowing, canopy cover and litter on seedling
dynamics of two Polylepis species at upper tree lines in central Ecuador. J Trop Ecol 23:309–318
Cingolani AM, Renison D, Tecco PA, Gurvich DE, Cabido M (2008) Predicting cover types in a mountain
range with long evolutionary grazing history: a GIS approach. J Biogeogr 35:538–551
1800 Biodivers Conserv (2011) 20:1789–1802
123
Coblentz D, Keating PL (2008) Topographic controls on the distribution of tree islands in the high Andes of
south-western Ecuador. J Biogeogr 35:2026–2038
Cuevas JG (2000) Tree recruitment at the Nothofagus pumilio alpine tree line in Tierra del Fuego, Chile.
J Ecol 88:840–855
Dale MRT, Dixon P, Fortin M-J, Legendre P, Myers DE, Rosenberg MS (2002) Conceptual and mathe-
matical relationships among methods for spatial analysis. Ecography 25(5):558–577
Dirzo R, Garcı´a CM (1992) Rates of deforestation in Los Tuxtlas, a neotropical area in southeast Mexico.
Conserv Biol 6:84–90
Dobyns HF (1966) An appraisal of techniques with a new hemispheric estimate (estimating aboriginal
American population). Curr Anthropol 7(4):395–416
Ellenberg H (1958) Wald oder Steppe? Die natu¨rliche Pflanzendecke der Anden Perus. Umschau
1958:645–681
Ellenberg H (1979) Man’s influence on tropical mountain ecosystems in South America. J Ecol 67:401–416
Etter A, Villa LA (2000) Andean forests and farming systems in part of the Eastern Cordillera (Colombia).
Mt Res Dev 20(3):236–245
Fearnside PM (1980) Land use allocation of the transamazon highway colonist of Brazil and its relation to
human carrying capacity. In: Barbira-Scazzochio F (ed) Land, people, and planning in contemporary
Amazonia. Centre of Latin American Studies, University of Cambridge, Cambridge, UK, pp 114–138
Fjeldsa˚ J (1992) Biogeography of the birds of the Polylepis woodlands of the Andes. In: Balslev H, Luteyn
JL (eds) Pa´ramo. An Andean ecosystem under human influence. Academic Press, Padstow, Great
Britain
Fjeldsa˚ J (1993) Avifauna of the Polylepis woodlands of the Andean highlands: the efficiency of basing
conservation priorities on patterns of endemism. Bird Conserv Int 3:37–55
Fjeldsa˚ J (2002) Polylepis forests—vestiges of a vanishing ecosystem in the Andes. Ecotropica 8:111–123
Fjeldsa˚ J, Kessler M (1996) Conserving the biological diversity of Polylepis woodlands of the highlands of
Peru and Bolivia. A contribution to sustainable natural resource management in the Andes. NOR-
DECO, Copenhagen
Garcia-Nun˜ez C et al (2004) Leaf gas exchange and water relations in Polylepis tarapacana at extreme
altitudes in the Bolivian Andes. Photosynthetica 42(1):133–138
Germino MJ, Smith WK (2002) Conifer seedling distribution and survival in an alpine treeline ecotone.
Plant Ecol 162:157–168
Gosling WD, Hanselman JA, Knox C, Valencia BG, Bush MB (2009) Long-term drivers of change in
Polylepis woodland distribution in the central Andes. J Veg Sci 20:1041–1052
Hagaman DJ (2006) Conservation and use of Polylepis pepei woodlands in the Apolobamba mountains of
Bolivia. MSc thesis, University of Wisconsin-Madison
Hertel D, Wesche K (2008) Tropical moist Polylepis stands at the treeline in East Bolivia: the effect of
elevation on stand microclimate, above- and below-ground structure, and regeneration. Trees
22:303–315
Hoch G, Ko¨rner C (2005) Growth, demography and carbon relations of Polylepis trees at the world’s highest
treeline. Funct Ecol 19:941–951
IUCN (2010) IUCN red list of threatened species. Version 2010.1. www.iucnredlist.org
Jameson JS, Weaver RE, Ramsay PM (2007) Changes in high-altitude Polylepis forest cover and quality in
the Cordillera de Vilcanota, Peru´, 1956–2005. Biol Conserv 138:38–46
Kessler M (1995) Present and potential distribution of Polylepis (Rosaceae) forests in Bolivia. In: Churchill,
SP, Balslev H, Forero E, Luteyn JL(eds) Biodiversity and conservation of neotropical montane forests.
New York Botanical Garden, Bronx, pp 281–294
Kessler M (2000) Observations on a human-induced fire event at a humid timberline in the Bolivian Andes.
Ecotropica 6:89–93
Kessler M (2002) The ‘‘Polylepis problem’’: where do we stand? Ecotropica 8:97–110
Kessler M, Driesch P (1993) Causas e historia de la destruccion de bosques altoandinos en Bolivia. Ecologı´a
en Bolivia 21:1–18
Kessler M, Bo¨hner J, Kluge J (2007) Modelling tree height to assess climatic conditions at tree lines in the
Bolivian Andes. Ecol Model 207:223–233
Laegaard S (1992) Influence of fire in the grass pa´ramo vegetation of Ecuador. In: Balslev H, Luteyn JL
(eds) Paramo. An Andean ecosystem under human influence. Academic Press, Padstow, Great Britain,
pp 151–170
Lazcano JM, Espinoza D (2001) Tendencia en el uso de la len˜a en dos communidades con bosques de
Polylepis con e´nfasis en variables econo´micas. Rev Bol Ecol Conserv Amb 9:61–77
Legendre P, Vaudor A (1991) The R-package: multidimensional analysis, spatial analysis. De´partement de
sciences biologiques. Universite´ de Montre´al, Montreal
Biodivers Conserv (2011) 20:1789–1802 1801
123
Lloyd H, Marsden SJ (2008) Bird community variation across Polylepis woodland fragments and matrix
habitats: implications for biodiversity conservation within a high Andean landscape. Biodivers Con-
serv 17:2645–2660
Maxwell K (2004) Lost cities and exotic cows: constructing the space of nature and culture in the Machu
Picchu Historic Sanctuary, Peru. Doctoral dissertation, Yale University School of Forestry and
Environmental Studies
Purcell J, Brelsford A (2004) Reassessing the causes of decline of Polylepis a tropical subalpine forest.
Ecotropica 10:155–158
Rada F, Azo´car A, Bricen˜o B, Gonza´les J, Garcia-Nun˜ez C (1996) Carbon and water balance in Polylepis
sericea, a tropical treeline species. Trees 10:218–222
Renison D, Cingolani AM (1998) Experiencias en germinacio´n y reproduccio´n vegetativa applicados a la
reforestacio´n con Polylepis australis (Rosaceae) en las Sierras Grandes de Co´rdoba Agrentina.
AgriScientica 15:47–53
Renison D, Hensen I, Cingolani AM (2004) Anthropogenic soil degradation affects seed viability in
Polylepis australis mountain forests of central Argentina. For Ecol Manag 196:327–333
Renison D, Hensen I, Suarez R, Cingolani AM (2006) Cover and growth habit of Polylepis woodlands and
shrublands in the mountains of central Argentina: human or environmental influence? J Biogeogr
33:876–887
Renison D, Hensen I, Suarez R, Cingolani AM, Marcora P, Giorgis MA (2010) Soil conservation in
Polylepis mountain forests of Central Argentina: is livestock reducing our natural capital? Aust Ecol.
35:435
Renison D, Hensen I, Suarez R (2011) Landscape structural complexity of high-mountain Polylepis aus-
tralis forests: a new aspect of restoration goals. Restor Ecol. doi:10.1111/j.1526-100X.2009.00555.x
Servat GP, Mendoza W, Ochoa JA (2002) Flora y fauna de cuatro bosques de Polylepis (Rosaceae) en la
cordillera del Vilcanota (Cusco, Peru). Ecolgı´a Aplicada 1(1):25–35
Smouse PE, Long JC, Sokal RR (1986) Multiple regression and correlation extensions of the Mantel test of
Matrix correspondence. Syst Zool 35:627–632
Stadtmu¨ller T (1987) Cloud forests in the humid tropics. The United Nations University, Tokio, p 81
UNEP-WCMC (2004) United Nations Environment Programme. World Conservation Monitoring Center.
http://www.unep-wcmc.org
Young KR (1993) Tropical timberlines: changes in forest structure and regeneration between two Peruvian
timberline margins. Arc Alp Res 25:167–174
Young KR (1994) Roads and the environmental degradation of tropical montane forests. Conserv Biol
8:972–976
1802 Biodivers Conserv (2011) 20:1789–1802
123