How accurately can minimum temperatures at the cold limits of tree species be extrapolated from weather station data?

Abstract

Most plant physiological processes act on micro-geographic scales within meters or less and on temporal scales of minutes or less. Yet, most studies relating species distribution to climate used typical resolutions of kilometers and months at best. Commonly available climate records from weather stations or freely available coarse-resolution geographic climatic layers thus, do not reflect local climatic conditions. In this study we selected sites where eight temperate deciduous tree species are growing at their cold upper elevational and latitudinal limits in the Swiss Alps (from 1165 m a.s.l. to 1804 m a.s.l.) and in Sweden (from 58°18′N to 59°27′ N). At each site, temperature was recorded for 1–2 years in different conditions: at understorey height (50 cm), 2-m above ground, in the top of tree canopies and at 10 cm depth in the soil. We compared these biologically meaningful temperatures with the closest weather station data after correction for elevation. The data evidence that in mountain terrain, scaling from weather station data to on-site forest conditions requires month-specific lapse rates of temperatures, separated for means and extremes (e.g. minima). Besides best elevation-correction procedures, monthly absolute minimum temperatures predicted from near weather stations remained 1.4 ± 0.2 K (mean ± se, 12 sites) cooler than in situ conditions during growing season (2.0 ± 0.2 K cooler during the non-growing season). At the time when 2-m air temperature reached its absolute monthly minimum, the top of the tree canopy was found 0.4 ± 0.1 K cooler (mean ± se, 12 sites) during growing season and 0.9 ± 0.1 K during the non-growing season. These systematic deviations of low temperature extremes from those predicted from weather stations close the gap between geographical range limits of species, their physiological limits (e.g. freezing resistance) and meteorological information. The “thermal niche” concept of species range limits needs to account for such deviations of life conditions from meteorological data, should the niche boundaries have a functional meaning rooted in plant biology.

Full text

Agricultural
and
Forest
Meteorology
184 (2014) 257–
266
Contents
lists
available
at
ScienceDirect
Agricultural
and
Forest
Meteorology
jou
rn
al
hom
ep
age:
www.elsevier.com/locate/agrformet
How
accurately
can
minimum
temperatures
at
the
cold
limits
of
tree
species
be
extrapolated
from
weather
station
data?
Chris
Kollas∗,
Christophe
F.
Randin,
Yann
Vitasse,
Christian
Körner
Plant
Ecology
Unit,
Botany,
Department
of
Environmental
Sciences,
University
of
Basel,
CH-4056
Basel,
Switzerland
a
r
t
i
c
l
e
i
n
f
o
Article
history:
Received
29
April
2013
Received
in
revised
form
12
September
2013
Accepted
7
October
2013
Keywords:
Deciduous
trees
Scandinavia
Swiss
Alps
Elevation
gradient
Temperature
profile
Microclimate
a
b
s
t
r
a
c
t
Most
plant
physiological
processes
act
on
micro-geographic
scales
within
meters
or
less
and
on
temporal
scales
of
minutes
or
less.
Yet,
most
studies
relating
species
distribution
to
climate
used
typical
resolutions
of
kilometers
and
months
at
best.
Commonly
available
climate
records
from
weather
stations
or
freely
available
coarse-resolution
geographic
climatic
layers
thus,
do
not
reflect
local
climatic
conditions.
In
this
study
we
selected
sites
where
eight
temperate
deciduous
tree
species
are
growing
at
their
cold
upper
elevational
and
latitudinal
limits
in
the
Swiss
Alps
(from
1165
m
a.s.l.
to
1804
m
a.s.l.)
and
in
Sweden
(from
58◦18#N
to
59◦27#N).
At
each
site,
temperature
was
recorded
for
1–2
years
in
different
conditions:
at
understorey
height
(50
cm),
2-m
above
ground,
in
the
top
of
tree
canopies
and
at
10
cm
depth
in
the
soil.
We
compared
these
biologically
meaningful
temperatures
with
the
closest
weather
station
data
after
correction
for
elevation.
The
data
evidence
that
in
mountain
terrain,
scaling
from
weather
station
data
to
on-site
forest
conditions
requires
month-specific
lapse
rates
of
temperatures,
separated
for
means
and
extremes
(e.g.
minima).
Besides
best
elevation-correction
procedures,
monthly
absolute
minimum
temperatures
predicted
from
near
weather
stations
remained
1.4
±
0.2
K
(mean
±
se,
12
sites)
cooler
than
in
situ
conditions
during
growing
season
(2.0
±
0.2
K
cooler
during
the
non-growing
season).
At
the
time
when
2-m
air
temperature
reached
its
absolute
monthly
minimum,
the
top
of
the
tree
canopy
was
found
0.4
±
0.1
K
cooler
(mean
±
se,
12
sites)
during
growing
season
and
0.9
±
0.1
K
during
the
non-growing
season.
These
systematic
deviations
of
low
temperature
extremes
from
those
predicted
from
weather
stations
close
the
gap
between
geographical
range
limits
of
species,
their
physiological
limits
(e.g.
freezing
resistance)
and
meteorological
information.
The
“thermal
niche”
concept
of
species
range
limits
needs
to
account
for
such
deviations
of
life
conditions
from
meteorological
data,
should
the
niche
boundaries
have
a
functional
meaning
rooted
in
plant
biology.
© 2013 Elsevier B.V. All rights reserved.
1.
Introduction
The
distributional
limits
of
tree
species
at
high
latitude
or
high
elevation
are
likely
associated
with
particular
manifestations
of
low
temperature
(Sakai
and
Larcher,
1987;
Von
Humboldt
and
Bonpland,
1807;
Woodward,
1987).
While
mean
growing
season
temperatures
are
critical
for
the
position
of
the
climatic
tree-
line
(Körner,
2012;
Körner
and
Paulsen,
2004),
the
range
limits
of
non-treeline
species
are
more
likely
associated
with
their
freez-
ing
tolerance
(Sakai
and
Larcher,
1987)
and
thus,
the
occurrence
of
freezing
events.
However,
temperature
data
are
rarely
recorded
Abbreviations:
meteoTAir,
2-m
air
temperature
of
the
weather
station;
meteoTAir∗,
elevation
corrected
2-m
air
temperature
of
the
weather
station;
on-
siteTAir,
2-m
air
temperature
on
the
tree;
on-siteTUstorey,
temperature
in
the
understorey;
on-siteTSoil,
temperature
in
the
soil;
on-siteTCrown,
temperature
in
the
top
of
the
crown.
∗Corresponding
author.
Tel.:
+41
61
267
3506;
fax:
+41
61
267
3504.
E-mail
address:
chris.kollas@zalf.de
(C.
Kollas).
where
species
reach
their
distributional
limits.
Hence
any
infer-
ences
of
species-range-climate
associations
rely
on
climate
records
from
surrounding
weather
stations
or
spatially
interpolated
cli-
matic
layers
derived
from
station
networks.
Absolute
minima
of
temperature
or
other
expressions
of
extremes
are
rarely
reported
(but
see
Zimmermann
et
al.,
2009)
and
not
offered
by
geographic
climatic
layers,
such
as
from
WorldClim
(Hijmans
et
al.,
2005),
Climate
Research
Unit
(New
et
al.,
2002)
or
Tyndall
Centre
for
Cli-
mate
Change
Research
(Mitchell
et
al.,
2004):
It
remains
unclear,
to
which
extent
such
measurements
from
weather
stations
or
data
obtained
from
climatic
layers
reflect
the
actual
life
conditions
that
trees
experience
at
their
range
limit.
Three
major
constraints
must
be
overcome
for
accurately
inferring
temperatures
acting
on
trees
at
their
limits:
(1)
the
geo-
graphical
position
of
weather
stations
which
is
often
distant
from
species
boundaries,
(2)
the
deviation
of
forest
microclimate
from
weather
station
conditions
and
(3)
the
spatial
and
temporal
resolu-
tion
of
the
temperature
parameter
provided
by
e.g.
the
geographic
climatic
layer.
0168-1923/$
–
see
front
matter ©
2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.agrformet.2013.10.001

258 C.
Kollas
et
al.
/
Agricultural
and
Forest
Meteorology
184 (2014) 257–
266
First,
weather
stations
are
sparsely
distributed
and
are
often
restricted
to
low
elevation.
This
is
critical
when
predicting
climate
variables
for
higher
elevation
sites
(e.g.
at
a
tree
species’
upper
ele-
vational
limit),
as
the
error
is
expected
to
increase
with
elevational
distance.
Second,
due
to
their
intrinsic
structure,
forests
generate
their
own
microclimate
(Geiger
et
al.,
2003).
The
structure
of
tree
canopies
affects
the
radiation
regime
and
the
wind
profile
(Van
Eimern
and
Riedinger,
1986).
The
resistance
of
the
associated
aero-
dynamic
boundary
layer
causes
air
temperatures
in
the
forest
to
deviate
from
those
measured
under
standardized
conditions
in
a
weather
station,
commonly
placed
on
an
open
field
(Larcher,
1975).
The
same
tree
individual
can
be
exposed
to
very
different
ther-
mal
conditions
from
roots
to
the
top
of
the
canopy.
Similarly,
tree
seedlings
and
saplings
in
the
understorey
may
experience
differ-
ent
temperatures
from
both
the
tree
canopy
and
standard
weather
station
data.
In
addition
to
heat
convection,
radiative
cooling
can
cause
temperatures
of
exposed
tissues
to
depart
from
air
tem-
peratures
measured
two
meters
above
ground
in
a
shelter
(Jones,
1992).
The
third
problem
arises
when
using
geographic
climatic
layers
due
to
broad
spatial
and
temporal
resolution
of
the
climate
data
obtained
from
them.
Here,
accuracy
decreases
because
tempera-
tures
were
interpolated
across
a
gridded
landscape
and/or
averaged
to
mean
conditions
of
a
large
area
(e.g.
a
pixel
that
represents
1
km2or
more):
this
has
been
shown
to
provide
inaccurate/false
species
range
shifts
in
mountain
regions
where
elevational
gra-
dients
are
important
(Randin
et
al.,
2009;
Scherrer
et
al.,
2011).
Further,
many
processes
related
to
tree
survival
operate
at
a
minute
to
an
hourly
time
scale
(e.g.
freezing
damage).
In
con-
trast,
climatic
layers
offer
long-term
series
of
climate
records
as
monthly
averages
only,
whereas
in
reality,
extreme
events
(sin-
gularities)
exert
significant
ecological
impact
(Easterling
et
al.,
2000;
Parmesan
et
al.,
2000;
Stenseth
et
al.,
2002).
Most
likely,
these
unknown
extremes
are
decisive
for
species
boundaries.
For
instance,
late
frost
events
in
spring
affect
new
leaves
and
active
meristems
of
trees.
A
single
frost
event
in
a
50-year
period
can
kill
all
trees
in
an
area
and
thus,
might
constrain
a
species
limit.
Similarly,
absolute
temperature
minima
during
winter
can
set
the
species
limit
(Sakai
and
Larcher,
1987;
Till,
1956).
Unsurprisingly,
Zimmermann
et
al.
(2009)
showed
that
species
distribution
models
improved
their
predictive
power
by
including
standard
devia-
tion
of
climate
variables
as
a
proxy
of
extreme
climatic
events.
There
is,
thus,
an
urgent
need
for
providing
a
proper
assess-
ment
of
deviations
between
weather
stations/climatic
layers
and
in
situ
temperatures
that
trees
experience
at
their
distributional
limits
(as
done
for
the
soil
surface
microclimate
by
Graae
et
al.,
2012).
We
focused
on
monthly
mean,
monthly
mean
minima
and
monthly
absolute
minimum
temperatures
because
mean
tempera-
tures
are
likely
to
control
growth
rate
whereas
absolute
minimum
temperatures
affect
survival.
The
broad
spectrum
of
taxa
(and
thus
locations)
included
here
should
buffer
our
findings
against
local
climatic
peculiarities.
Thus,
we
were
addressing
the
following
tasks:
(1)
A
quan-
tification
of
the
extent
of
deviations
between
in
situ
minimum
temperatures
recorded
directly
on
trees
growing
at
their
pre-
sumed
thermal
limit
and
temperatures
derived
from
weather
stations
which
were
scaled
to
the
same
elevation
using
regional
lapse
rates
(as
commonly
used
in
most
high-resolution
model-
ing
studies).
(2)
An
assessment
of
the
temperature
differences
across
a
vertical
profile
in
the
forest
at
the
same
species-
specific
elevational
and
latitudinal
limit.
By
means
of
(1)
and
(2)
we
provide
(3)
factors
for
scaling
meteorological
informa-
tion
to
temperatures
experienced
by
trees
from
roots
to
the
top.
2.
Methods
2.1.
Species
and
study
sites
We
selected
eight
widely
distributed
deciduous
tree
species:
Acer
pseudoplatanus
L.,
Fagus
sylvatica
L.,
Fraxinus
excelsior
L.,
Labur-
num
alpinum
(Mill.)
Bercht.
&
J.
Presl,
Prunus
avium,
Quercus
petraea
(Matt.)
Liebl.,
Sorbus
aria
L.
and
Tilia
platyphyllos
Scop.
We
used
data
from
the
Swiss
National
Forest
Inventory
(NFI)
from
two
inven-
tory
periods,
which
were
sampled
during
the
years
1983–1985
(NFI1)
and
1995–1997
(NFI2)
on
a
regular
1-km
grid
(1.4
km
grid
for
NFI2).
Additional
tree
occurrences
in
Switzerland
were
derived
from
the
forest
plots
database
(Wohlgemuth,
1992).
This
resulted
in
n
=
22,130
observations
for
all
Switzerland.
This
database
allowed
us
to
identify
the
highest
elevation
reached
for
each
study
species.
All
eight
taxa
reached
their
highest
elevation
in
two
regions,
one
in
the
Western
Alps
of
Switzerland
centered
on
Martigny
(46◦6#N,
7◦4#E)
and
the
other
one
centered
on
Chur
in
the
Eastern
Swiss
Alps
(46◦51#N,
9◦32#E,
Fig.
1).
Based
on
the
Swedish
national
for-
est
inventory
data
(Nilsson
and
Cory,
2011)
and
explorations
in
the
field,
we
identified
the
latitudinal
limits
of
a
subset
of
four
species
in
Sweden
(South
Scandinavia)
between
Göteborg
(57◦42#N,
11◦57#E)
and
Arvika
(59◦39#N,
12◦36#E,
Fig.
1).
For
each
species,
we
selected
one
adult
tree
among
others
within
the
regional
uppermost
(Alps)
or
northernmost
(South
Scan-
dinavia)
margin
of
distribution
and
within
closed
forest
stands.
So
we
did
not
select
the
single
highest
individual
(such
outposts
may
reflect
a
peculiar
microclimate),
but
rather
placed
our
sensors
where
several
reproductive
individuals
marked
the
limit.
When
two
or
more
species
shared
the
same
distributional
limits
(within
10
m
of
elevational
difference
or
1
km
in
latitude),
we
selected
only
one
common
site
for
our
measurements
related
to
these
species.
This
resulted
in
four
replicate
measurement
sites
(trees)
in
each
of
the
three
regions
(Table
1).
A
later
population
study
in
the
Alps
(Vitasse
et
al.,
2012)
revealed
that
some
isolated
individuals
may
be
found
in
certain
microhabitats
at
(several
tenth
of
meters
higher)
elevation,
but
their
reproductive
success
is
unknown.
2.2.
Temperature
records
and
experimental
design
At
each
monitored
site,
we
placed
four
temperature
loggers
(Tid-
biT
v2
Temp
UTBI-001,
Onset
Computer
Corporation)
at
different
positions
within
the
forest:
One
logger
was
placed
on
the
north-facing
side
of
the
stem
of
a
tall
tree
(d.b.h.
>0.3
m)
at
2
m
above
the
ground,
under
a
major
branch,
and
hence
completely
shaded
by
the
host
tree’s
canopy
and
stem
during
the
growing
season
and
by
the
stem
and
surround-
ing
branches
during
the
coldest
part
of
the
year
(called
hereafter
on-siteTAir).
We
were
using
the
loggers
to
obtain
minimum
tem-
peratures
of
the
air
inside
the
forest.
However,
by
choosing
such
location
we
avoided
direct
effect
of
sunshine
and
night-time
radia-
tive
cooling
during
the
non-growing
season.
Further,
low
solar
elevation
angle
and
short
day
length
make
radiation
errors
due
to
direct
sunshine
impossible.
This
is
particularly
true
in
Sweden
and
in
the
valleys
in
the
Alps
where
surrounding
mountains
screen
the
horizon.
These
air
temperature
records
were
used
to
compare
with
2-m
air
temperature
from
the
nearest
standard
weather
sta-
tions.
A
second
data
logger
was
buried
10
cm
under
the
soil
surface,
10
cm
north
of
the
stem
(called
hereafter
on-siteTSoil)
to
reflect
temperature
in
the
upper
root
zone.
Such
soil
temperatures
col-
lected
in
deep
shade
are
known
to
correlate
well
with
weekly
mean
air
temperature
records
(Körner
and
Paulsen,
2004).
A
third
log-
ger
was
positioned
at
c.
50
cm
above
ground
in
the
understorey
to
reflect
temperature
at
the
seedling/young
sapling
height
(called
hereafter
on-siteTUstorey).
This
logger
was
intentionally
not
shel-
tered
and
may
periodically
have
been
exposed
to
solar
radiation

C.
Kollas
et
al.
/
Agricultural
and
Forest
Meteorology
184 (2014) 257–
266 259
Fig.
1.
Locations
of
the
selected
study
sites
(white
circle
around
black
dots)
and
weather
stations
(black
triangle)
in
South
Sweden
(c),
in
the
Western
Alps
(d)
and
Eastern
Alps
(e)
within
Switzerland
(b)
and
Europe
(a).
or
nighttime
radiative
cooling
as
such
young
trees
are.
The
fourth
logger
was
placed
at
the
top
of
the
tree
to
reflect
canopy
temper-
atures
(called
hereafter
on-siteTCrown).
This
logger
was
attached
to
one
of
the
uppermost
lateral
branches
of
c.
2
cm
diameter
(above
leaves),
oriented
south,
so
that
this
logger
also
could
experience
nighttime
radiative
thermal
cooling
but
also
periodic
direct
solar
radiation.
These
loggers
were
mounted
between
6
and
14
m
above
ground
and
exposed
to
free
atmospheric
convection
(a
proxy
of
branch
night
time
temperature).
By
exposing
a
logger
this
way,
we
intentionally
allowed
for
violation
of
rules
of
radiation
protec-
tion.
In
the
ideal
case
we
would
have
inserted
tiny
sensors
into
buds
or
cambial
tissue
but
this
was
impractical
in
tree
tops
at
the
geographical
scale
we
were
operating.
Loggers
used
have
sim-
ilar
dimensions
(2.5
cm
diameter),
volume
density
(1.5
g
cm3)
and
color
(gray)
as
the
uppermost
branches.
Hence,
we
assume
that
by
the
nature
of
the
data
logger
and
the
way
we
mounted
it,
we
approximate
the
actual
life
conditions
at
tree
tops.
This
assump-
tion
was
tested
in
experiments
with
thin
wire
thermocouples
and
found
to
confirm
the
assumption
(see
Supplementary
Material
A).
Temperature
was
recorded
hourly
from
August
2009
to
October
2011
in
the
Alps
and
from
mid
September
2010
to
mid
September
2011
in
Scandinavia.
None
of
the
sites
were
prone
to
temperature
inversions.
Due
to
vandalism,
data
of
3
data
logger
in
the
Alps
were
available
for
1
year
only.
Before
and
after
the
logging
period,
devices
were
totally
immersed
in
an
ice-water
bath
for
0◦C
calibration
and
cross-checked
again
for
identical
readings
The
maximum
devia-
tion
was
0.17
K,
which
falls
within
the
manufacturers
specification.
Hence,
no
numerical
adjustments
were
made.
2.3.
Data
analysis
To
meet
our
objectives,
we
applied
a
three-step
analysis
(Fig.
2),
namely
(1)
verifying
regional
lapse-rates
in
order
to
scale
weather
station
records
to
the
elevations
of
our
sites
and
calculating
of
contrasts
between
scaled
weather
station
data
and
forest
2-m
air
Table
1
Location
and
elevation
of
the
study
sites
selected
inside
the
uppermost
margin
of
tree
species
distribution
as
well
as
the
coordinates,
the
elevational
(!Elev)
and
horizontal
(!Dist)
distances
to
the
nearest
weather
stations
used
for
predictions
in
the
three
regions.
Region
Elevation
(m)
Latitude
Longitude
Study
site
(affiliated
species)/Station
name
!Elev
(m)
!Dist
(km)
Sites
Western
Swiss
Alps
1374
46◦8#N
7◦4#E
T.
platyphyllos,
Q.
petraea
1592
6.9
1460
46◦11#N
6◦59#E
F.
sylvatica
1506
7.1
1523
46◦8#N
7◦4#E
F.
excelsior,
P.
avium
1443
7.1
1804
46◦8#N
7◦4#E
A.
pseudoplatanus,
S.
aria,
L.
alpinum
1162
5.9
Eastern
Swiss
Alps 1165
46◦50#N
9◦35#E
T.
platyphyllos
1525
14.7
1320
46◦50#N
9◦35#E
F.
excelsior,
Q.
petraea
1370
15.1
1540
46◦51#N
9◦23#E
F.
sylvatica
1150
31.1
1547
46◦55#N
9◦19#E
A.
pseudoplatanus
1143
36.6
South
Scandinavia
159
58◦21#N
12◦29#E
Q.
petraea
18
32.1
85
58◦30#N
11◦57#E
A.
pseudoplatanus
92
17.8
90
58◦46#N
11◦48#E
P.
avium
87
29.2
83
59◦1#N
12◦18#E
F.
sylvatica
94
25.5
Weather
stations
Western
Swiss
Alps
2966
46◦20#N
7◦12#E
Les
Diablerets
Eastern
Swiss
Alps
2690
46◦50#N
9◦48#E
Weissfluhjoch
South
Scandinavia 69
59◦40#N
12◦38#E
Arvika
177
59◦13#N
12◦4#E
Blomskog
177
58◦36#N
12◦11#E
Kroppefjäll-Granan

260 C.
Kollas
et
al.
/
Agricultural
and
Forest
Meteorology
184 (2014) 257–
266
Fig.
2.
Conceptual
design
of
the
study.
(1)
contrast
between
scaled
weather
station
temperature
and
air
temperature
in
the
forest,
(2)
contrasts
between
three
temper-
atures
within
the
canopy
profile
(crown,
understorey,
soil)
and
air
temperature
in
the
forest
and
(3)
scaling
factors
from
weather
station
temperature
to
temperatures
within
the
canopy
profile.
temperature,
(2)
the
calculation
of
contrasts
between
this
local
air
temperature
and
temperatures
acting
across
the
vertical
canopy
profile.
Finally,
(3)
we
developed
the
resultant
scaling
factors
to
predict
biologically
meaningful
inner
forest
temperatures
from
weather
station
data.
2.3.1.
Air
temperature
comparisons
between
weather
station
and
forest
climate
We
first
assessed
the
deviation
between
nearest
weather
station
temperature
(meteoTAir)
and
our
in
situ
measured
2-m
air
tem-
perature
(on-siteTAir)
at
all
our
sites.
In
order
to
demonstrate
the
effect
of
lapse
rate,
3
types
of
lapse
rates
were
applied
to
correct
weather
station
data
for
elevation:
(1)
the
commonly
used
mean
annual
−0.55
K
100
m−1(Körner,
2007),
(2)
we
calculated
regional
monthly
lapse
rates
from
weather
stations
(source:
MeteoSwiss)
at
contrasting
elevations
for
monthly
mean
temperature
for
both
the
Western
and
the
Eastern
Swiss
Alps
(linear
regression
model
with
n
=
8
and
n
=
7
weather
stations,
18
years
of
monthly
data)
and
(3)
we
additionally
calculated
lapse
rates
for
mean
minimum
tem-
peratures
by
using
the
same
data
source.
Notably,
we
found
almost
identical
monthly
lapse
rates
in
both
regions
in
the
Alps
(mean
monthly
sd
of
±
0.02
K).
We
therefore
used
the
monthly
mean
lapse
rates
for
both
regions
combined
(Western
and
Eastern;
Fig.
3)
for
the
next
step
of
the
analysis.
To
underline
the
validity
of
these
lapse
rates
for
the
time
period
used
for
our
analysis,
we
also
calcu-
lated
short-term
regional
lapse
rates
for
the
years
of
measurements
(2009–2011)
using
data
from
the
same
15
weather
stations.
Second,
we
identified
the
weather
station
that
was
the
clos-
est
to
each
of
our
test
sites
(Table
1
and
Fig.
1),
excluding
stations
located
at
the
bottom
of
the
Rhône
and
Rhein
valleys,
since
win-
ter
temperature
inversion
occurring
at
low
elevation
might
bias
the
coherence
between
weather
stations
and
our
sites
(Beniston
and
Rebetez,
1996;
Bolstad
et
al.,
1998).
Next
we
tested
the
rela-
tionships
between
both
the
temperature
parameters
from
our
0.35
0.40
0.45
0.50
0.55
0.60
0.65
0.70
Month
J
F
M
A
M
J
J
A
S
O
N
D
Temperature lapse rate (K 100 m1)
Mean temperature
Minimum temperature
1994
2011
2009
2011
Fig.
3.
Seasonal
variation
of
elevational
lapse
rates
for
mean
and
mean
minimum
temperatures,
showing
the
long-term
average
(black
symbols)
and
the
study
period
(gray)
in
the
Swiss
Alps.
Values
are
the
mean
lapse
rate
of
both
regions
(Western
and
Eastern
Swiss
Alps),
because
data
did
not
differ
between
regions
(n
=
15
weather
stations).
sites
(on-siteTAir)
and
the
corresponding
nearest
weather
station
(meteoTAir)
by
calculating
the
coefficient
of
determination
R2of
the
linear
regression.
This
relationship
was
not
used
in
further
analysis.
Instead,
for
every
climate
station
used
we
corrected
hourly
temperature
records
by
calculating
the
elevational
differ-
ence
between
meteoTAir and
on-siteTAir and
adjusting
meteoTAir
separately
by
(1)
−0.55
K
100
m−1,
(2)
the
above
mentioned
mean
regional
monthly
lapse
rate
and
(3)
the
separate
mean
regional
lapse
rates
for
mean
minimum
and
mean
temperature
for
the
time
of
measurements
(09/2009–10/2011,
Fig.
3).
In
South
Scan-
dinavia
we
used
the
regional
monthly
lapse
rate
provided
by
Christensen
et
al.
(1998),
although
elevational
differences
between
weather
stations
and
sites
were
<94
m.
We
did
not
correct
weather
station
data
for
horizontal
distance
by
latitudinal
lapse
rates
in
South
Scandinavia
since
weather
stations
were
located
very
close
to
our
sites
(maximal
horizontal
distance
40
km).
All
adjusted
weather
station
temperature
parameters
are
referred
hereafter
as
meteoTAir∗.
Third,
for
each
site
we
obtained
the
following
tem-
perature
parameters
for
both
meteoTAir∗and
on-siteTAir:
monthly
mean
temperature
(mean
of
daily
means
per
month),
monthly
mean
minimum
temperature
(mean
of
daily
minima
per
month,
from
hourly
records)
and
monthly
absolute
minimum
(coldest
hour
of
the
month).
Differences
between
altitude-corrected
weather
station
temperature
parameters
and
in
situ
measured
temperature
parameters
(!meteoTAir∗–on-siteTAir,
!meteoTAir∗–on-siteTCrown,
!meteoTAir∗–on-siteTUstorey)
were
calculated.
Last,
monthly
val-
ues
of
2
years
were
pooled
and
each
temperature
parameter
was
averaged
(n
=
4
sites)
per
region.
2.3.2.
Vertical
temperature
profile
within
the
forest
For
each
site
and
each
month
during
the
measurement
cam-
paign
(09/2009–10/2011),
we
selected
monthly
absolute
minimum
temperature
(coldest
hour
of
the
month)
and
calculated
monthly
mean
minimum
temperature
(mean
of
daily
minima
per
month)
from
on-siteTAir (2
m-air
temperature
within
the
forest)
and
concurrent
temperatures
from
on-siteTSoil,
on-siteTUstorey and
on-
siteTCrown.
For
each
month
and
each
site
we
then
calculated
the
differences
on-siteTSoil–TAir ,
on-siteTUstorey–TAir and
on-
siteTCrown–TAir to
relate
soil-,
understorey-
and
crown-temperature

C.
Kollas
et
al.
/
Agricultural
and
Forest
Meteorology
184 (2014) 257–
266 261
to
our
standard
2-m
air
temperature.
These
monthly
differences
(on-site!TSoil–Air,
on-site!TUstorey–Air and
on-site!TCrown–Air)
of
2
years
were
pooled
and
averaged
(n
=
4
sites)
for
each
of
the
three
regions
(n
=
4
sites
per
region).
To
test
if
these
differences
reflect
general
patterns
or
were
site
specific,
we
explored
effects
of
crown
height,
elevation,
air
temper-
ature
and
slope
exposure.
We
calculated
correlations
between
the
differences
(of
all
12
sites)
and
the
site-specific
parameters:
(i)
on-site!TCrown–Air and
crown
height
(ii)
on-site!TCrown–Air,
on-site!TUstorey–Air,
on-site!TSoil–Air and
elevation
(Alps)
or
latitude
(South
Scandinavia)
(iii)
on-site!TCrown–Air,
on-site!TUstorey–Air,
on-site!TSoil–Air and
on-siteTAir
(iv)
on-site!TCrown–Air,
on-site!TUstorey–Air,
on-site!TSoil–Air and
slope
exposure
2.3.3.
Spatial
and
temporal
resolution
of
temperature
records
To
underline
the
importance
of
spatially
and
temporally
pre-
cise
temperature
records,
we
calculated
the
following
temperature
differences:
First,
we
tested
the
agreement
between
temperature
data
of
the
WorldClim
dataset
(that
offers
a
spatial
resolution
of
30
arc
s,
cor-
responding
to
∼1
km
in
temperate
latitudes;
Hijmans
et
al.,
2005)
with
records
from
weather
stations.
Thus,
we
calculated
differ-
ences
between
monthly
mean
temperature
(1950–2000)
recorded
at
weather
stations
and
reported
by
WorldClim
(where
World-
Clim
grid
cells
overlay
the
weather
station
location
in
both
study
regions
in
the
Swiss
Alps,
n
=
8,
from
482
m
to
2966
m
a.s.l.).
The
matching
of
a
grid
cell
with
a
weather
station
is
not
necessarily
at
exactly
the
same
elevation.
Second,
we
explored
deviations
related
to
the
temporal
resolution.
For
each
month
(1981–2010)
and
each
weather
station
in
the
two
examined
regions
(n
=
15,
from
381
m
to
2966
m
a.s.l.),
we
calculated
the
following
temperature
differences
per
weather
station:
monthly
mean
temperature
versus
monthly
absolute
minimum
temperature
and
monthly
mean
minimum
tem-
perature
versus
monthly
absolute
minimum
temperature.
3.
Results
3.1.
Air
temperature
comparisons
between
weather
station
and
forest
climate
3.1.1.
Temperature
lapse
rates
The
temperature
lapse
rates
calculated
from
monthly
data
of
15
weather
stations
over
18
years
showed
a
pronounced
seasonality:
rates
were
lowest
in
winter
and
highest
in
spring.
Lapse
rates
dif-
fered
widely
among
monthly
mean
and
monthly
mean
minimum
temperatures
(Fig.
3),
with
mean
temperature
lapse
rate
having
the
larger
seasonal
amplitude.
They
are
close
to
the
expected
moist
adiabatic
lapse
rate
(−0.3
to
−0.4
K
100
m−1)
and
comparable
to
those
found
in
the
Italian
and
Austrian
Tyrol,
in
British
Columbia
and
in
the
Appalachian
Mountains
(Bolstad
et
al.,
1998;
Rolland,
2003;
Stahl
et
al.,
2006).
The
temperature
lapse
rates
during
the
study
years
2009–2011
were
similar
to
long
term
means
as
shown
in
Fig.
3.
In
December
only
the
short
term
lapse-rates
depart
from
the
long-term
lapse-rate
which
is
explained
by
exceptional
warm
winters
during
2009–2011.
Over
the
two-year
test
period,
we
found
strong
linear
relations
between
hourly
values
for
meteoTAir and
on-siteTAir for
all
sites,
with
a
mean
R2of
0.91
(range:
0.82–0.96).
Correlations
calculated
separately
for
each
month
showed
a
minimum
R2in
January
of
0.61
and
a
maximum
value
of
0.81
in
November,
with
most
values
around
0.72,
without
any
clear
seasonal
pattern
(data
not
shown).
Overall,
we
obtained
higher
agreement
with
in
situ
measured
air
temperatures
within
the
forest
when
scaling
meteoTAir by
two
sep-
arate
month-specific
lapse
rates
for
mean
and
mean
minimum
tem-
perature
(applying
root
mean
square
error,
Table
2).
Applying
these
two
temperature
lapse
rates
substantially
improved
the
predictions
of
most
temperatures,
e.g.
reducing
the
root
mean
square
error
from
2.0
to
1.4
K
in
mean
minima
compared
to
applying
an
annual
mean
temperature
lapse
rate
of
−0.55
K
100
m−1.
However,
for
mean
minimum
temperatures,
differences
between
meteoTAir∗and
on-
siteTAir were
slightly
smaller
when
the
annual
mean
lapse
rate
was
applied
(Table
2).
Nevertheless,
we
used
the
month-specific
lapse
rates
for
mean
and
mean
minimum
temperature
in
all
further
results.
3.1.2.
Minimum
temperatures
In
the
Alps,
the
monthly
absolute
minimum
temperatures
(coldest
hour
of
the
month)
derived
from
weather
stations
and
scaled
for
the
elevation
of
our
test
site
(2
m
above
ground)
were
always
colder
than
in
situ
temperatures
(−1.5
to
−2.5
K
during
the
growing
season,
−2.5
to
−4.2
K
during
the
non-growing
season;
Fig.
4).
Similarly,
in
South
Scandinavia,
monthly
absolute
min-
ima
of
meteoTAir∗were
colder
than
absolute
minima
of
on-siteTAir
(−0.5
to
−2
K,
except
for
February
and
December,
where
meteoTAir∗
absolute
minima
were
slightly
warmer
than
on-siteTAir;
Fig.
4).
In
all
regions,
standard
errors
were
smaller
during
the
growing
season
(on
average
0.5
K,
ranging
from
0.1
to
1.3
K)
compared
to
the
non-growing
season
(on
average
0.8
K,
ranging
from
0.4
to
1.8
K).
Averaged
across
the
3
regions
and
all
months,
monthly
abso-
lute
minima
of
meteoTAir∗were
−1.7,
−1.3
and
−1.5
K
colder
than
in
situ
2-m
air,
crown
and
understorey
temperature,
respectively,
with
more
discrepancy
in
winter
than
during
the
growing
season
(Table
4).
Similar
to
the
absolute
minima,
monthly
mean
minima
of
meteoTAir∗(mean
of
all
daily
minima
per
month)
in
the
Alps
were
colder
than
mean
minima
of
on-siteTAir for
all
except
one
month.
Although
monthly
mean
minima
show
smaller
deviations
between
meteoTAir∗and
on-siteTAir than
monthly
absolute
minima,
the
smaller
deviations
during
the
growing
season
were
found
for
mean
minima
as
well.
3.1.3.
Mean
temperatures
Differences
in
monthly
mean
temperature
between
meteoTAir∗
and
on-siteTAir were
the
smallest
among
all
types
of
temperature
measures
(Fig.
4),
and
ranged
between
<2.2
K
in
the
Alps
and
<1
K
in
South
Scandinavia.
As
described
for
minimum
temperatures,
smaller
!meteoTAir∗–on-siteTAir was
found
during
the
growing
season
(Fig.
4).
In
summary
means
and
minima
of
weather
station
data
scaled
to
on-site
elevation
(meteoTAir∗)
by
month-specific
lapse
rates
were
almost
always
colder
than
on-siteTAir at
2-m
height
with
larger
differences
during
the
non-growing
season,
when
the
deciduous
tree
canopy
bears
no
foliage.
3.2.
Vertical
temperature
profile
within
the
forest
3.2.1.
Temperature
profile
At
the
elevational
and
latitudinal
cold
limits
of
deciduous
tree
species
(Alps
and
Scandinavia),
both
monthly
absolute
minimum
and
monthly
mean
minimum
of
soil
temperatures
were
always
warmer
than
the
corresponding
2-m
air
temperature.
They
showed
a
seasonal
pattern
that
reflects
insulating
snow
cover
effects
in
winter.
Monthly
absolute
minima
of
soil
temperature
were
on
aver-
age
8.4
K
(ranging
from
4.2
to
15.1
K
in
the
Alps)
and
8.5
K
(0
to
20.9
K
in
Scandinavia;
Fig.
5)
warmer
than
the
corresponding
2-m
air
temperature.

262 C.
Kollas
et
al.
/
Agricultural
and
Forest
Meteorology
184 (2014) 257–
266
Table
2
Results
of
elevation-corrections
of
weather
station
data
for
the
Swiss
Alps
and
South
Scandinavia
using
3
types
of
lapse
rate
(mean
annual
lapse
rate,
month-specific
lapse
rates
for
mean
temperature
and
month-specific
lapse
rates
separately
for
mean
and
mean
minimum
temperature).
Numbers
show
the
RMSE,
root
mean
square
error
(in
K)
and
the
Bias
(mean
difference
from
actual,
with
direction,
in
K)
between
elevation-corrected
weather
station
data
and
in
situ
measured
forest
climate
(2-m
air
temperature)
averaged
over
all
sites
in
the
3
study
regions
(n
=
12
locations).
Growing
seasonaNon-growing
season
Full
year
RMSE
Bias
R2RMSE
Bias
R2RMSE
Bias
R2
Mean
Mean
annual
lapse
rate
1.2
−0.9
0.90
1.1
−0.4
0.93
1.1
−0.6
0.96
Month-specific
lapse
rate
for
mean
temperature
0.6
−0.5
0.90
1.2
−1.0
0.94
1.0
−0.8
0.98
Mean
minima
Mean
annual
lapse
rate
1.8
−0.5
0.88
2.1
−0.1
0.92
2.0
−0.3
0.97
Month-specific
lapse
rate
for
mean
temperature
1.0
−0.1
0.92
1.4
−0.7
0.92
1.2
−0.4
0.97
Month-specific
lapse
rate
for
mean
minimum
temperature
1.2
−1.1
0.93
1.5
−1.3
0.94
1.4
−1.2
0.97
Absolute
minima Mean
annual
lapse
rate 1.9 −0.9 0.83 3.0
−1.0
0.86
2.7
−1.0
0.93
Month-specific
lapse
rate
for
mean
temperature
1.8
−0.4
0.89
2.9
−1.6
0.88
2.5
−1.1
0.94
Month-specific
lapse
rate
for
mean
minimum
temperature
1.6
−1.4
0.89
2.7
−1.4
0.89
2.3
−1.4
0.96
aGrowing
season
refers
to
the
months
May
to
September.
Crown
temperatures
were
always
colder
than
2-m
air
tempera-
tures
when
these
reached
their
monthly
absolute
minimum
in
the
Alps,
except
in
July
in
the
Western
Alps
(Fig.
5),
and
this
differ-
ence
was
smaller
during
the
growing
season
(−0.1
to
−0.6
K)
than
during
the
non-growing
season
(−0.6
to
−1.8
K).
Similarly,
crown
temperatures
in
South
Scandinavia
were
always
colder
than
2-m
air
temperatures,
when
these
reached
their
monthly
absolute
min-
imum
(−0.1
to
−1.7
K),
with
the
strongest
deviations
occurring
in
spring
and
autumn.
For
all
understorey
air
temperature
measure-
ments
we
found
that
devices
had
been
installed
above
snow
cover.
Surprisingly,
understorey
minimum
temperatures
were
also
colder
than
2-m
air
temperature,
and
(with
few
exceptions)
differences
4
2
0
2
Western Swiss Alps
Minima
absolute minimum
mean minimum
Eastern Swiss Alps
South Scandinavia
4
2
0
2
JFMAMJJASOND
Mean
JFMAMJJASOND
Month
JFMAMJJASOND
Fig.
4.
Seasonal
variation
of
the
deviation
of
in
situ
forest
air
temperature
(TAir)
from
closest
weather
station
data
(TMet)
corrected
for
elevational
difference
by
month-specific
lapse
rates
(for
details
see
Section
2).
Negative
values
indicate
lower
temperatures
predicted
from
weather
station
data
than
those
measured
in
situ.
Error
bars
denote
standard
errors
of
4
sites
per
region.
Horizontal
gray
bars
represent
the
mean
duration
of
the
thermal
growing
season.

C.
Kollas
et
al.
/
Agricultural
and
Forest
Meteorology
184 (2014) 257–
266 263
Fig.
5.
In
situ
monthly
absolute
and
mean
minimum
temperature
deviations
of
crown
and
−10
cm
soil
temperature
from
air
temperature
at
2-m
above
ground.
Error
bars
denote
standard
errors
of
4
sites
per
region.
Horizontal
gray
bars
represent
the
mean
duration
of
the
thermal
growing
season.
In
the
majority
of
cases
crown
top
is
cooler
than
the
local
2-m
air
temperature
inside
the
deciduous
forest.
showed
the
same
seasonal
pattern,
but
with
lower
amplitude
than
in
the
top
canopy
(Table
3).
Absolute
minimum
temperatures
in
the
understorey
and
at
2-
m
showed
only
small
differences
during
the
growing
season
(+0.2
to
−0.3
K),
but
the
understorey
was
markedly
cooler
during
the
non-growing
season
(−0.4
to
−0.6
K
in
the
Alps).
In
South
Scandi-
navia,
understorey
temperatures
were
on
average
0.4
K
cooler
than
2-m
air
temperatures
when
these
reached
their
absolute
monthly
minimum.
Interestingly,
in
all
regions,
strongest
negative
tempera-
ture
deviations
in
the
understorey
occurred
at
the
beginning
and
at
the
end
of
the
growing
period,
where
absolute
minima
are
most
critical
for
plant
physiology
(on
average
−0.7
K).
3.2.2.
Effect
of
local
topography
on
the
vertical
temperature
profile
The
above
temperature
profiles
(temperature
differences)
were
neither
correlated
with
elevation
(Alps)
nor
with
latitude
(South
Scandinavia),
nor
had
slope
exposure
or
canopy
height
any
influ-
ence
(data
not
shown).
Hence
the
patterns
observed
are
robust
across
all
12
sites
against
local
conditions
as
long
as
there
was
a
near
to
closed
deciduous
forest
stand.
An
exception
are
soil
temper-
atures
that
became
decoupled
from
atmospheric
conditions
under
snow,
an
effect
that
increased
with
elevation
because
of
increasing
snow
duration.
3.3.
Scaling
factors
By
joining
the
two
above-mentioned
analysis
we
are
able
to
provide
a
widely
applicable
formula
for
predicting
biologically
meaningful
inner
forest
temperatures
from
weather
station
data
(crown
temperature
as
an
example):
on-siteTCrown =
meteoTAir (◦C)
+
!meteoElevation–on-siteElevation (m)
∗
regional
month-specific
lapse
rate
(K
m−1)
−
season-specific
!meteoTAir–on-siteTAir (K)
+
season-specific
!on-siteTAir–on-siteTCrown (K)
For
parameters
of
lapse
rate
and
temperature
differences
see
Tables
3
and
4
and
Fig.
1.
4.
Discussion
Characterizing
and
predicting
the
position
of
the
cold
limit
of
tree
species
requires
temperature
data
that
reflect
the
cur-
rent
proximal
life
conditions
of
trees.
Here
we
explored
how
to
adequately
infer
temperature
at
forest
sites
located
on
mountain
slopes
in
the
Alps
and
in
the
lowlands
of
South
Scandinavia
by
using
weather
stations.
We
showed
that
the
temperature
lapse
rates
differ
consistently
between
means
and
extremes
of
temperature
with
the
one
calculated
on
mean
temperatures
being
higher.
We

264 C.
Kollas
et
al.
/
Agricultural
and
Forest
Meteorology
184 (2014) 257–
266
Table
3
Understorey
and
crown
temperature
deviations
(on-siteTUstorey,
on-siteTCrown,
in
K)
from
2-m
air
temperature
(on-siteTAir)
in
the
forest
during
the
growing
season,
the
non-growing
season
and
the
whole
year.
Temperature
deviations
are
shown
for
monthly
absolute
minimum
and
monthly
mean
minimum
temperatures.
Numbers
show
the
means
(3
regions
pooled)
and
standard
errors
(standard
deviation
divided
by
the
root
of
sample
size)
of
data
for
4
sites
per
region.
Western
Swiss
Alps Eastern
Swiss
Alps South
Scandinavia
Growing
seasonaNon-growing
season Full
year Growing
season
Non-growing
season
Full
year
Growing
season
Non-growing
season
Full
year
!on-siteTCrown–on-siteTAir Absolute
minima −0.3 ±
0.1 −1.0 ±
0.1 −0.7 ±
0.1 −0.6 ±
0.1 −1.1 ±
0.1 −0.9 ±
0.1 −0.3 ±
0.1 −0.7 ±
0.2 −0.6 ±
0.1
Mean
minima
−0.5
±
0.1
−0.6
±
0.1
−0.6
±
0.0
−0.6
±
0.1
−0.8
±
0.1
−0.7
±
0.0
−0.1
±
0.1
−0.3
±
0.1
−0.2
±
0.1
!on-siteTUstorey–on-siteTAir Absolute
minima
0.2
±
0.2
−0.4
±
0.2
−0.2
±
0.1
−0.3
±
0.1
−0.6
±
0.2
−0.5
±
0.1
−0.2
±
0.1
−0.4
±
0.1
−0.4
±
0.1
Mean
minima −0.3 ±
0−0.5
±
0.0
−0.4
±
0.0
−0.5
±
0.1
−0.6
±
0.1
−0.6
±
0.1
−0.4
±
0.1
−0.5
±
0.1
−0.5
±
0.0
aGrowing
season
refers
to
the
months
May
to
September.
further
showed
that
monthly
absolute
minimum
air
temperatures
from
lapse
rate
scaled
meteorological
data
are
systematically
colder
than
temperatures
measured
2-m
above
ground
within
the
forest
(−1.4
±
0.2
K
during
growing
season,
−2.0
±
0.2
during
non-growing
season;
mean
±
se).
Furthermore,
both
top
canopy
and
understorey
monthly
absolute
minimum
temperatures
were
cooler
than
2-m
air
temperature
across
all
sites
and
seasons
(crown
0.4
±
0.1
K
cooler
during
growing
season,
0.9
±
0.1
K
cooler
hereafter).
We
largely
attributed
these
cooler
temperatures
to
radiative