Introduction
Climate change alters wood production by modifying the rates of
photosynthesis and respiration rates of trees . Changes in forest productivity have been observed
in past decades all over the world . The carbon stock of forests and their role as carbon sinks are
therefore changing. These findings have stimulated discussions about whether
forest management strategies can be adapted to reduce forest vulnerability to
climate change, support recovery after extreme events and foster the
carbon sink function of forests .
Wood production is influenced by several factors, such as CO2
fertilization, nitrogen deposition, precipitation and temperature
. For instance, rising CO2 increases water use
efficiency of forests , which could compensate negative
effects of climate change on European forest growth .
Another important process is fertilization . Due to depositions of nitrogen in the second half of the last
century, wood production had increased in European forests
. However, temperature modifies photosynthesis,
respiration and growth rates of trees . In the temperate biome, positive effects
on wood production e.g. as well as negative ones have been found
e.g.. However, it remains
unclear why forests react differently to temperature change.
In addition to the influence of climate variables, wood production is also
affected by internal forest properties. These properties can be grouped into
two types: properties which describe forest structure and those which
describe species composition (Fig. ). For instance, changes in
productivity can result from changes in basal area , in leaf
area index or in the heterogeneity of tree heights within
a forest . Furthermore, wood production often increases with
the increasing number of species .
Overview of drivers influencing wood production. External variables in
this study are temperature, radiation and precipitation. Forest properties
are divided into two groups: species composition properties (e.g. Rao's Q
as a measure of functional diversity and species distribution index
ΩAWP) and forest structure properties (e.g. forest height,
leaf area index and tree height heterogeneity).
Forest stands, which differ in their forest properties, might respond
differently to the same climate change . For instance, the
positive effect of increasing temperature on wood production fades with
forest age in temperate deciduous forest e.g., and showed that higher diversity buffers
the effect of inter-annual variability on wood production. However, these
studies include only a few forest properties and rarely include properties
related to both species composition and forest structure. Hence, it is
unclear how these forest properties influence wood production change due to
temperature rise and which forests will benefit from rising temperatures.
As far as we know, there is no data set available that covers forests,
differing in structure and diversity, under almost identical climatic
conditions. Even if a larger number of forest stands were available, it would
be difficult to manipulate, for instance, temperature while keeping all other
climate variables constant. Forest simulation models offer an alternative to
the analysis of field experiments. Such models are able to estimate wood
production under different climate conditions e.g.. For instance, investigated the effect of
climatic change on forests by simulating 30-year time slices of a range of
different future climates for 135 inventoried forest stands. There are also
model-based studies, which systematically analysed the effect of species
diversity on productivity and stability over long periods . However, disturbed or managed forest stands and the influence of
climate change have not been included in these analyses.
In this study, we therefore propose a new simulation-based approach. First, we
generate a large number of forest stands covering various forest structures
and species compositions (for up to eight temperate tree species). Annual
aboveground wood production (AWP) is then calculated for all forest stands
based on climate time series. These time series differ in the mean annual
temperature and the intra-annual temperature amplitude. We aim to analyse
how productivity of forest stands (AWP) is influenced by (i) increasing mean
annual temperature and (ii) increasing intra-annual temperature amplitude.
Furthermore, we address the question of which forest stands will
benefit most from rising temperatures.
Method
To analyse the effect of temperature on the productivity of forest stands, we
applied the “forest factory” model approach . The forest
factory generated 370 170 different forest stands (see Sect. ) and
allowed the estimation of AWP under various
climate time series (see Sect. ). The 320 scenarios differed in mean
annual temperature and annual temperature amplitude. Finally, we calculated
the forest-stand-specific sensitivity of productivity to temperature change
as the relative change of wood production per temperature change of
1 ∘C (see Sect. ). To relate these sensitivities to forest
structure and species composition, we characterized every forest stand with
five properties (see Sect. ). We analysed the influence of the five
forest properties on temperature sensitivity using boosted regression trees
(see Sect. ). Finally, we analysed which combination of forest
properties resulted in the highest sensitivity values for different
successional stages (see Sect. ). All analysed data are available in the Supplement to this manuscript.
The forest factory approach
The forest factory creates forest patches based on different stem size
distributions and species mixtures. We used 15 stem size distributions
covering a gradient from young to old and disturbed to undisturbed forests.
Species mixtures included all 256 possible combinations of Pinus sylvestris, Picea abies, Fagus sylvatica, Quercus robur, Fraxinus excelsior, Populus x canadensis,
Betula pendula and Robinia pseudotsuga. We used the species
parameter set and algorithms of the FORMIND model version for temperate
forests within the forest factory . A total of 100 forest
patches of each combination were built.
To generate forest patches, the forest factory randomly chose trees from the
stem size distribution, assigned a species identity and planted them within a
patch of 400 m2 size. To place a tree within a patch, the following
rules must be met: (i) there must be enough space available for crowns of
every tree, and (ii) every tree in the forest must have a positive
productivity under its environmental conditions (light, temperature, water).
We used climate time series from the year 2007, measured at Hainich National
Park, central Germany. We assumed this time series to be a typical example
for a temperate year (in principle, it is possible to use climate data from any
other location). In contrast to an artificially generated climate, this
climate is perfectly physically consistent (with regard to light, air temperature
and precipitation).
In a few cases, not all species of the mixture could be placed within a patch
by the algorithm, so we rejected such forests. We ended up with 370 100
forest stands. For more details regarding the forest factory, see
.
Wood production
The calculation of AWP of trees was based on
algorithms of the model FORMIND . In this
model, the wood production of a single tree is calculated as the difference
between climate variables driven respiration rates and photosynthesis. The
photosynthesis rate (Ptree) results from the crown size,
self-shading within the crown and available light at the top of the tree. The
available light depends on the radiation above the canopy, reduced by the
shading of larger trees within the forest stand. Furthermore, productivity
can be limited due to air temperature and available soil water, which is
expressed by the photosynthesis-limiting factor ϕ for each tree
. Available soil water within
the stand results from precipitation, interception and evapotranspiration of
trees and runoff.
One part of the photosynthesis production of a tree (Ptree) is
allocated to its maintenance respiration (and to non-wood tissues;
Rm). Maintenance respiration depends on tree biomass and
temperature ψ . The remaining organic carbon is
transformed into newly grown aboveground wood (AWPtree) and a
proportional growth respiration (rg).
AWPtree=(ϕPtree-ψRm)(1-rg)
AWPtree was summed over all trees to obtain the productivity of
the modelled forest stand – AWP (for a more detailed description of growth
processes, see ).
Climate sensitivity
To generate a set of 320 annual climate time series, we selected daily
climate measurements of the Hainich station in central Germany between the
years 2000 and 2004. This time series includes mean daily radiation,
precipitation and air temperature (see Appendix , Fig. ).
We separated these time series into five distinct time series of 1-year
length. First, we increased or decreased the mean annual temperature of each
year by adding or subtracting 0.5 ∘C steps between -1.5 and
+2 ∘C. Second, we changed the amplitude of the annual temperature
cycle for these time series variation of each year. To do so, we modified the
standard deviation of each year by 4 % steps between -12 and
+16 %. We ended up with five sets of climate time series (of 1-year
length) that differ in temperature, precipitation and radiation. Each of
these five sets includes 64 time series, which differ only in temperature
(see Appendix , Fig. ). Temperature change was quantified
using two indices: (i) mean annual temperature and (ii) annual temperature
amplitude, which described the 95 % interquartile range of all daily
temperature values of a given year. We did not model the effects of nitrogen
and CO2 fertilization (as both do not vary strongly within 1 year) or
extreme anomalies (e.g. pathogen attacks) on wood production.
Figure a–c show the AWP for
different annual temperatures for three different forest stands.
Overview of forest properties and resulting temperature sensitivity
of AWP of three exemplary forests:
(a) old even-aged spruce forest; (b) mature deciduous
forest; (c) a quite young mixed species forest. The middle
(panels d, e and f) shows the corresponding
stem size distributions and provides information on the highest tree in the
forest (Hforest) and species distribution index
ΩAWP (which quantifies the suitability of a species
distributed within the forest structure with regard to AWP). Each forest is
treated with 320 climate time series; the last column (panels g, h and i) shows the AWP as a
function of mean annual temperature (MAT). The colours indicate different
inter-annual temperature amplitudes (Q95) of the used time series. (The
coloured bands show the standard deviation due to the variability of the five
different time series that exist for each combination of mean annual
temperature and intra-annual temperature amplitude.)
We analysed the sensitivity of every forest stand to temperature change
following the approach of . For every forest stand, a
general linear model was fitted relating wood production mean annual
temperature (MAT) and intra-annual temperature amplitude (Q95), as well as
the nuisance parameter year.
AWP=αxMAT+βxQ95+γxyear+ϵ
For every forest, we calculated the relative change of productivity resulting
from an increase of 1 ∘C:
SIMAT=αAWP‾SIQ95=βAWP‾.
In our analysis, we excluded all forests stands for which AWP turns negative
if the temperature rises by 1 ∘C (this occurs in 2 % of all
stands).
We also determined the sensitivity of forests to temperature change using the
German forest inventory to validate our results. However, the inventory does
not include leaf area index (LAI) measurements. We therefore assumed the
basal area as a proxy for LAI, and we selected subsamples of forests stands
with similar structure (basal area, tree height heterogeneity, forest height
and same species mixtures). In addition, we used elevation as a proxy for
mean annual temperature, assuming temperature changes of 0.65 ∘C per
100 m on average . Only in the case of spruce and beech
monocultures did we find enough data to calculate SIMAT values
for several forest structures (for more details, see Appendix ,
Fig. ).
The comparison between the SIMAT estimation based on the German
forest inventory with SIMAT values of corresponding forests from
the forest factory showed quite good agreement (R2=0.65). However, the
simulated SIMAT values of the forest factory slightly
overestimated the sensitivity compared to the inventory-based values
(Fig. ). This might be explained by the difference in the methods
used because, in the case of the inventory, we used basal area instead of LAI and
altitude instead of temperature. Another explanation could be that in our
approach the climate time series showed relatively high and regular
precipitation. In the German forest inventory, warmer sites might be more
frequently exposed to water stress, which then reduced the SIMAT values.
SIMAT values of seven different forest types derived
from the analysis of the German forest inventory vs. SIMAT values
derived from corresponding forest types of the forest factory. Only those
SIMAT values of the field data are analysed which showed
p values smaller than 0.05.
Five forest properties to describe forest stands
We used three indices to describe the forest structure: LAI,
maximum forest height (Hforest), which corresponds to the
height of the largest tree in a forest stand, and tree height heterogeneity
(θ), which was quantified by the standard deviation of the tree
heights. To describe species composition, we used Rao's Q and species
distribution index (ΩAWP). Rao's Q quantified functional
diversity based on species abundances and differences in species traits
(Botta, 2005, for details, see Appendix ). ΩAWP
analysed the optimal location of species within the forest structure.
ΩAWP is defined as the ratio of the forest's productivity to
the maximum possible productivity of the forest without changing tree sizes
or number . Hence, the maximum productivity can be obtained
by varying only the species identities of trees in the forest stand. We
changed the assigned species of each tree until we found the optimal species
for each individual tree and its specific environmental condition. All five
indices were nearly uncorrelated for the investigated forest stands
(Appendix , Table ).
Boosted regression trees
We applied boosted regression trees to quantify the influence of the five
forest properties on SIMAT and SIQ95. Boosted regression
trees are a machine learning algorithm using multiple decision (or
regression) trees. It is able to address unidentified distributions
. Each model was fitted in a forward stage-wise
procedure to predict the response of the dependent variable on
(SIMAT or SIQ95) to multiple predictors (tree height
heterogeneity, forest height, LAI, Rao's Q and ΩAWP). To
omit an overfitting with regard to maximal forest height, we classified forest
stands into 18 classes. Each class had a width of 2 m, starting with 4 to
6 m and finishing with 36 to 38 m. The boosted regression trees tried an
iterative process to minimize the squared error between predicted SI values
and those of the data set. Hereby, part of the data were used for a fitting
procedure and the rest was used for computing out-of-sample estimates
of the loss function . This boosted regression tree
analysis was performed in the R package gbm 2.1.1 .
We used a quarter of the data (randomly sampled) for the machine learning
procedure. To get the best model, we varied the following four parameters of
the boosted regression tree algorithm: learning rates (0.1, 0.05 and 0.01),
the bag fractions (0.33, 0.5 and 0.66), the interaction depths (1, 3 and 5)
and the cross validation (3-, 6- and 9-fold) assuming a Gaussian error
structure (the default setting). The best-fitted boosted regression tree for
both SIMAT and SIQ95 showed a learning rate of 0.1, a
bag fraction of 0.66, an interaction depth of 5 and a 3-fold cross
validation. These two models were used for all further analyses. The
remaining 75 % of the data were used to validate the fitted boosted
regression tree algorithm.
Finding the forest stands for different successional stages that benefit the most increasing temperatures
Here, we assumed forest height as a proxy for the successional stage of a
forest. In every height class, we selected those 5 % of forests that
showed the highest sensitivity values (SIMAT and SIQ95). We
removed the forest height classes between 10 and 14 m, as they only
contained only 15 forests. For all
other classes, we analysed the relationship between height class and the
forest properties (ΩAWP, Rao's Q, LAI and tree height
heterogeneity).
Results
We analysed the sensitivity of productivity (AWP) to temperature for forest
stands that differ in forest properties (species distribution index
(ΩAWP), functional diversity (Rao's Q), tree height
heterogeneity (θ), forest height class and LAI). The annual
AWP was estimated for each forest stand using
320 different climate time series. We then quantified the changes in
productivity resulting from changes in mean annual temperature
(SIMAT) and intra-annual amplitude (SIQ95). For the analysed
forest stands, the average SIMAT is
1.5 % ∘C-1 and the average SIQ95 is
-5.4 % ∘C-1 (see also the frequency distribution in
Appendix , Fig. ).
With a boosted regression tree algorithm, we analysed how the five forest
properties influence the temperature sensitivity of forests. To validate the
fitted boosted regression tree algorithm, we compared SI values, which are
not used for the fitting, with the SI value predicted by the boosted
regression tree algorithm (Fig. ). The sensitivities to mean annual
temperature change (SIMAT) correlated very well (R2 of 0.84)
and showed a low root mean squared error (RMSE) of ±2.9 % ∘C-1 (see
Appendix , Fig. ). The RMSE even decreased to
±1.5 % ∘C-1 if a subset of the forest stands was
analysed that showed SIMAT values larger than
-5 % ∘C-1 (90 % of the data). The accuracy of the
sensitivities to temperature amplitude change (SIQ95) was even slightly
better. In addition, a subset that included SIQ95 values larger than
-15 % ∘C-1 (93 % of the data) showed a RMSE of only
±1.1 % ∘C-1 (see Appendix , Fig. ).
Partial dependency plots of the five forest properties –
ΩAWP (species distribution index), forest height class,
Rao's Q (functional diversity), tree height heterogeneity and LAI (leaf
area index) – for SIMAT (sensitivity to changes in the mean annual
temperature) and SIQ95 (sensitivity to changes in annual temperature
amplitude). Relative importance (RI) compares the influence of different
input variables on the variability of a target variable. Histograms show the
frequency of forest property values in the analysed data set. Note that
ΩAWP is the ratio of the current AWP of a forest and the
highest possible AWP obtained by shuffling only species identities without
changing the forest structure.
According to boosted regression tree analysis, ΩAWP was the
most relevant forest property to explain temperature sensitivities (relative
influence of 87 % for SIMAT and 89 % for SIQ95; see
also Appendix , Fig. ). However, the influence of
ΩAWP on temperature sensitivity flattened out for high
ΩAWP levels (Fig. ). The second relevant forest
property was forest height (Hforest). Forests with heights
between 25 and 30 m benefited the most from increasing mean annual
temperatures. The other three properties (LAI, Rao's Q and tree height
heterogeneity) had a low influence on SIMAT.
Analysis of those forests that show the highest 5 % of the
SIMAT values depending on forest height. Lines indicate mean values of the forest
subsamples which include the best 5 % with regard to SIMAT of
each hight class. The grey band indicates the interquartile range.
Panel (a) shows temperature sensitivity of aboveground wood production
over forest height, analysing only the best the forest subsample.
Panels (b) to (e) show the change of the remaining forest
properties within the forest subsamples (ΩAWP is the optimal
species distribution; LAI is the leaf area index; Rao's Q quantifies
functional diversity).
Both sensitivity indices showed similar relationships to the five forest
properties. However, an increase in annual temperature amplitude always
reduced productivity, whereas increasing mean annual temperature could result
in a positive effect on wood production. To detect those stands that benefit
the most from increasing temperature, we selected the 5 % of forest
stands that showed the highest SIMAT values in each forest height
class (Fig. ). In all forests classes, we found forest stands that
would benefit from increasing temperatures. Analyses of their forest
properties revealed that the ΩAWP levels were always high.
Young forests (low forest height), which had a positive temperature
sensitivity, showed low functional diversity and low tree height heterogeneity
(θ). For older forests (of intermediate and high forest height) with
positive temperature sensitivity, we found an intermediate level of
functional diversity. Interestingly, for three variables (Rao's Q, tree
height heterogeneity and LAI), the relationships changed their character
between young and intermediate forest heights. We obtained similar simulation
patterns for SIQ95 (Appendix , Fig. ).
Understanding the patterns
The influence of forest structure on temperature sensitivity
Forest structure affects the wood production of single trees in two ways.
First, it determines the amount of light available to each individual tree,
and second, the size of trees influences their photosynthesis and respiration
rates (Fig. ). Hence, based on the height of a tree and the amount
of light available to it, it was possible to calculate its SI values (for a
detailed discussion of these calculations, see Appendix ).
In even-aged forests, all trees have the same height and receive full light
(e.g. Fig. ; forest C). In our study, such forests showed a
bell-shaped relationship between forest height and temperature sensitivity
(Fig. ; SI values for 100 % available light depending on tree
height).
Analysis of the sensitivity index of AWP against mean annual
temperature (SIMAT) values of single trees within three different
forests. The diagram shows the calculated SIMAT value of
individual trees for every combination of tree height and available light
(for Pinus sylvestris between SIMAT levels of 6.5 and
-6.5; other species show similar patterns). The dots indicate the different
trees of the three forest examples. The white dots belong to trees with the
corresponding number of forest A, grey dots belong to the trees of forest B,
and dark grey dots belong to forest C. Note that, in the case of forest C, all
trees have the same height and the same light, so that all three dots are at
the same place in the diagram.
In the case of a forest consisting of trees of different heights, smaller trees
receive less light due to shading. Note that, even if trees received less
light, the bell-shaped relationship between tree height and productivity
persisted (Fig. ). Two cases will be discussed (assuming identical
LAI as forest C; Fig. ). In the first case, all trees have not yet
reached their maximal SI values (Fig. ; forest A); in the
second case, all trees have already passed their maximal SI values
(Fig. ; forest B). In the case of forest A, trees in the shade of
larger trees always had lower SI values if they belonged to the same species
(see Appendix ). Hence, the temperature sensitivity level of this
forest was lower than the sensitivity of an even-aged forest, whose trees
have the same size as the largest tree in forest A (Fig. ; tree 1).
Hence, if maximal SI values were not reached, increasing height heterogeneity
decreased SI values of a forest.
In forest B (Fig. ), SI values of the shaded trees can be similar
(or even higher) than the SI value of the largest trees in the forest (SI
values of tree 1 show similar levels to trees 2, 3 and 4 in forest B;
Fig. ). Hence, if maximal SI values were passed, increasing tree
height heterogeneity resulted in similar (or even more positive) temperature
sensitivity levels compared to even-aged forest trees (an even-aged forest
consisting only of trees similar to tree 1 of forest B in Fig. ).
These general considerations explain the change from low levels of height
heterogeneity in young forests to a more heterogeneous structure in the
analysis of those forests, which will benefit from increasing temperature
(see Fig. d).
The effect of species composition on temperature sensitivity
In this study, we use the new ΩAWP index called the species
distribution index . ΩAWP is the ratio
between current AWP and the highest possible AWP of the forest which can be
reached due to shuffling of species identities. Its huge importance for forest
temperature sensitivity might be illustrated by the following considerations.
If species are unfavourably distributed within the forest (low
ΩAWP), the AWP of the forest is low.
Panel (a) shows which species have the highest
productivity (ΩAWP value of 1) under the current climate for
different heights and different light conditions. Panel (b) shows
which species show the highest increase in productivity due to rising
temperatures for different heights and different light conditions. Red
colours indicate coniferous trees, whereas green colours indicate deciduous
trees. Darker colours indicate late successional species, whereas lighter
colours indicate pioneers. The dots indicate the different trees of the two
forest examples (A and B). The white dots belong to trees with the
corresponding number of forest A. Note that all trees have the same height
and the same light, so all five dots are at the same place in the diagram.
Grey dots belong to the corresponding trees with the same number of forest B.
Increasing functional diversity (Rao's Q) stabilized the forests'
sensitivity to temperature. This corresponds to results of
and the theoretical consideration of . The analysis of the
single species can give additional insight into the mechanisms behind those
species that benefited the most from temperature increase, which were
deciduous trees under most conditions. This is reasonable as warmer regions
host more deciduous species than needleleaf species. The highest functional
diversity (Rao's Q), on the other hand, occurred in mixtures of deciduous
and needleleaf trees (Appendix , Fig. ). As only two
needleleaf species were considered here in the species pool, low Rao's Q
values were dominated by mixtures of deciduous trees. Such deciduous tree
mixtures mostly benefited from temperature increases. In contrast,
mixtures with high Rao's Q values, which mostly included both functional
types, reacted more poorly (Fig. ; Appendix ,
Fig. ).
We developed two diagrams that show the species with the highest temperature
sensitivity and with the highest productivity for different conditions
(available light and height of a tree) (Fig. ). Interestingly, the
species with the highest productivity differed from the species that benefit
most from rising temperatures in many cases. This has important implications.
The highest benefit due to increasing temperatures was obtained by forests
with high but not maximal ΩAWP (Fig. ).
Additionally, deciduous trees benefited more than coniferous trees from
rising temperatures (Fig. , Appendix , Fig. ).
Hence, young forests should consist of deciduous trees (compare
Figs. and ; forest A), although the highest
productivity values are found for coniferous trees (Fig. ;
forest A). Forests including large trees obtained the highest sensitivity
values if intermediate-sized trees differed in their species identity from
the largest trees (Fig. ).
Discussion
The study design
In this theoretical study, we present a new climate sensitivity analysis
(with regard to temperature) of AWP. This approach extends field observations and
long-term model simulations, as it allows the analysis of existing forests
but also of those that might exist in the future due to management changes
and/or disturbances. Our approach includes only forest stands in which every
tree has positive productivity and enough space for its crown. Hence, it is
impossible, for instance, that light-demanding species grow below a closed
canopy or forests are overcrowded. However, the data set also includes a few
very unusual stand structures or species combinations, which cannot emerge
in a natural system, but may result from disturbances or management. In the
case of field observations, it is difficult to explore the influence of a
single climate variable (e.g. temperature) on one target variable (e.g.
AWP), as in most cases, several variables are altered at the same time (see
also Appendix ). Process-based models are one option to analyse such
relationships and separate these effects. The simulation of AWP with the
FORMIND model in temperate forests has been successfully compared to eddy
flux sites , the national German forest inventory
and European yield tables .
An advantage of the forest factory approach is the huge set of various
forests stands that can be analysed. The data set includes forest stands that
often occur in temperate forests (even-aged spruce, pine and beech stands).
However, it also includes hypothetical ones that could occur through
alternative forest management or disturbances (fire, bark beetles, etc.).
Hence, our data set of forest stands covers a much larger variety of forest
property combinations compared to long-term forest simulations with the focus
on natural forests in their equilibrium state e.g. or
on monocultures e.g.. Long-term simulations with
ecosystem models, which process modelled climate projections, face a
trade-off between cascade uncertainty and path dependency . The accumulations of model uncertainties over such a process
chain result in increasing uncertainty. Our study design tries to minimize
this uncertainty and omit path dependencies by including only those processes
that might be relevant for the research question. In this study, for
instance, we omit the effect of climate change on regeneration and mortality.
Furthermore, using several climate variables as model inputs but only
analysing the effect of one variable might lead to incorrect interpretations
of its effect. For example, temperature and radiation often correlate, and
both might increase productivity. Therefore, in this study, we only vary one
variable in all five sets of time series. This guarantees that there are no
relationships between the target climate variable and the remaining climate
variables.
As an increase in global mean temperature of 1.5 to 2 ∘C can
hardly be avoided, even under the Representative Concentration Pathway (RCP) 2.6 climate scenario ,
this study focuses on temperature change. This RCP scenario predicts only
small changes in annual precipitation levels for temperate regions. Hence,
our approach focuses only on the effect of temperature change on wood
production. However, this might be critical for the analysis of strong
temperature changes (e.g. RCP8.5) which will result in an increased
incidence of drought and changes in the annual temperature cycles and a
strong change in CO2. Such more complex scenarios should be analysed in
future studies. Further, we neglect the effect of time lags (e.g. bud
building in the previous year). However, it is possible to extend the used
time series to analyse the behaviour of the forest over longer time periods
and study not only productivity but also effects on regeneration or
mortality.
To characterize the annual temperature cycles, we used two variables: mean
annual temperature and intra-annual temperature amplitude. Both variables can
be varied independently. In the case of higher mean annual temperature, we observe
an elongation of the vegetation period. This leads to higher forest
productivity (if other resources are not limiting and
explains why SIMAT is often positive. However, warmer summer
temperatures can also lead to a decline in wood production due to an increase
in respiration. In the case of increasing intra-annual temperature amplitude,
more days with extreme temperatures will occur in a year. Thus, an increase
of 1 ∘C-1 of intra-annual temperature amplitude will increase
respiration more strongly compared to an increase of 1 ∘C-1 of
mean annual temperature. Hence, the increase of intra-annual temperature
amplitude normally has negative effects on the productivity (negative SI
values).
The temperature sensitivity values obtained here are in the same range as
those found for temperate ecosystems in heating experiments
4.4 ± 2.2 % ∘C-1. Within the
16 analysed studies reviewed by , the experimental plots show
almost identical environmental conditions (soil, radiation and
precipitation) and species composition. To heat the plots, greenhouses or
infrared heaters were used. Another study, based on natural forest stands in
New Zealand, found an AWP increase of between 5 and
20 ∘C-1 for forests, assuming no change in forest
structure and species composition . The analysed plots
were spread throughout New Zealand, and warmer temperatures coincided with
higher radiation . Hence, the analysed temperature effect also
includes the influence of radiation. In our setting, however, the influence
of temperature is independent of radiation as in. We also
found a good correlation between SI values derived from growth measurements
of the German forest inventory and simulated SI values based on the forest
factory (Fig. , Appendix , Fig. ).
Implications for forest management
Our findings might be relevant for future management strategies for temperate
forests. Specifically, our new understanding of which species benefit most
from rising temperatures (Fig. 6) suggests possible strategies, e.g.
replacing spruce monocultures with mixtures of deciduous trees. Further,
based on the analysis of which forest structure benefits most from rising
temperatures (Figs. 4, 5, 6), early-stage even-aged forests should include
mainly pioneer species. In the mature stage, we predict a positive effect of
temperatures on wood production for a mixture of climax species including
different tree sizes. These climax species could be planted below the canopy
of the pioneer species in young forests. In our approach, we do not simulate
the establishment of very young trees. However, during the conversion between
these two forest types, one big challenge might be the removal of the pioneer
trees without damaging the young trees that will build the mature forest.
Implications for global vegetation modelling
Most global vegetation models represent vegetation as fractional cover of
different plant functional types within a grid cell e.g.
Lund–Potsdam–Jena (LPJ);. Only a few global vegetation models include a more
detailed representation of vegetation structure and functional diversity
. It would be interesting
to perform the analysis presented here with global vegetation models which
include structure to better understand the mechanisms driving forest systems'
sensitivity to climate change.
Besides the global vegetation models, forest gap models, which have been
restricted to local stands, are now able to simulate forest dynamics in
regions or even entire continents . Studies
using global vegetation models or large-scale forest gap models simulate
natural succession. Our analysis indicates that natural and managed (or
disturbed) forest systems, which differ in forest structure, might react
differently to climate change. Hence, we suggest considering forest structure
in future analyses of global vegetation. Such information on forest structure
might be derived from remote sensing.