Upload
others
View
12
Download
0
Embed Size (px)
Citation preview
Nature Climate Change 4, 806–810 (2014)
Increasing forest disturbances in Europe and their impact on carbon storageRupert Seidl, Mart-Jan Schelhaas, Werner Rammer and Pieter Johannes Verkerk
In the version of this supplementary file previously published, the values given in Table 6 for net ecosystem productivity and net primary productivity were incorrect; this has no impact on the reported results. These errors have been corrected in this file 4 September 2014.
CORRECTION NOTICE
© 2014 Macmillan Publishers Limited. All rights reserved.
SUPPLEMENTARY INFORMATIONDOI: 10.1038/NCLIMATE2318
NATURE CLIMATE CHANGE | www.nature.com/natureclimatechange 11
Supplementary Information 1
2
Increasing forest disturbances in Europe and their impact on carbon storage 3
4
Rupert Seidl, Mart-Jan Schelhaas, Werner Rammer, Pieter Johannes Verkerk 5
6
We here combined empirical models of disturbance damage with scenario simulations of 7
forest development under a range of future climate scenarios and management strategies in 8
order to quantify trajectories of forest disturbance damage in Europe. Subsequently, an 9
analytical model of disturbance effects on forest C storage capacity was used to quantify the 10
impact of future disturbance regimes on forest C stocks in Europe. The main components of 11
our assessment are summarized in Supplementary Figure 1. 12
This Supplementary Information first gives additional details on the tools and methods 13
applied, and provides further information on the scenarios studied (Supplementary Methods: 14
Tools and scenarios). Subsequently, an in-depth evaluation and uncertainty analysis of crucial 15
assessment steps is presented (Supplementary Methods: Evaluation and uncertainty analyses). 16
Finally, additional details on key results are given in order to aid the interpretation of the 17
findings reported in the main text (Supplementary Results). 18
19
© 2014 Macmillan Publishers Limited. All rights reserved.
2
20
Supplementary Figure 1. The materials and methods applied to estimate future disturbance 21
trajectories and their impact on ecosystem C in Europe's forests. GCM: global circulation 22
model, RCM: regional climate model, EFISCEN: the European Forest Information Scenario 23
Model 1,2
, SEM: structural equation models of disturbance damage 3, REGIME: an analytical 24
model of disturbance effects on forest C storage 4. 25
26
27
Supplementary Methods – Tools and scenarios 28
The EFISCEN model 29
The European Forest Information SCENario model EFISCEN is a large-scale forest scenario 30
model that projects forest resource development on regional to European scale 2,5–7
. In 31
socio-economic
scenarios
climate
scenarios
scenarios of forest
development
future
disturbance regimes
carbon impact
of disturbance regimes
GCMs / RCMs
EFISCEN
SEM
REGIME
© 2014 Macmillan Publishers Limited. All rights reserved.
3
EFISCEN, the forest is described as an area distribution over age- and volume-classes in 32
matrices, based on national forest inventory data. Transitions of area between matrix cells 33
during the simulation represent different natural processes such as growth and mortality, and 34
are influenced by external drivers such as management strategies and changes in forest area. 35
The aging of forests is simulated by moving area proportions to a higher age class, while 36
growth is simulated by an area transition to a higher volume class. The latter transition 37
probabilities are derived from forest inventory data or yield table information. The effects of 38
climate change are implemented in EFISCEN based on assessments with detailed process-39
based models 8. For this study, we in particular used the results of Reyer et al.
9, who 40
estimated productivity changes for several European tree species with the process-based 41
model 4C (see Schelhaas et al. 10
for details), as well as the approach and data of Veroustraete 42
et al. 11
. 43
Management strategies are specified at two levels in EFISCEN. First, a basic management 44
regime defines the period during which thinnings can take place and a minimum age for final 45
fellings. These management regime parameters can be regarded as constraints on the total 46
harvest level. Second, the demand for wood is specified separately for thinnings and final 47
fellings, and EFISCEN may harvest the demanded wood volume if available under the 48
previously defined constraints. Harvest residues from thinning and final felling are either 49
extracted or left on site in management operations. 50
EFISCEN calculates stem wood volume, increment, age-class distribution, removals, forest 51
area, natural mortality and deadwood at five year time-steps. With the help of biomass 52
expansion factors, stem wood volume is converted into whole-tree biomass and subsequently 53
whole tree carbon stocks. Information on litterfall rates, felling residues and natural mortality 54
is used as input into the soil module YASSO 12
, which is dynamically linked to EFISCEN and 55
delivers information on forest soil carbon stocks. EFISCEN has been validated previously, 56
© 2014 Macmillan Publishers Limited. All rights reserved.
4
using long-term forest inventory data for Finland and Switzerland 13,14
. These validation 57
studies showed that EFISCEN is capable to reliably project the development of forest 58
resources for periods up to 50–60 years. Differences between projected and observed forest 59
structure (e.g., growing stock, age-class distribution) are largest at the regional or species 60
level as a result of uncertainties in the distribution of harvest removals over regions and 61
species. Here, following the European Forest Sector Outlook Study II 15
, we thus relied on 62
robust national-level projections. Data of EFISCEN projections are available from the 63
UNECE website (http://www.unece.org/efsos2.html). For a detailed technical documentation 64
of EFISCEN we refer to Schelhaas et al. 1. 65
66
The empirical models of disturbance damage 67
The empirical models of wind, bark beetle, and forest fire damage were developed using 68
country-scale disturbance data for the period 1958-2001 16
. The observational basis for these 69
relationships is a compilation of >29,000 disturbance records across Europe, summarized in 70
the European Forest Disturbance Database DFDE 17
. The data were logarithmically 71
transformed (after adding a value of +1) to remedy the nonnormality inherent in disturbance 72
data. All explanatory and response variables were standardized by subtracting their time-73
series mean and dividing by the standard deviation. The list of potential explanatory variables 74
considered in disturbance modeling is given in Supplementary Table 1. 75
76
77
© 2014 Macmillan Publishers Limited. All rights reserved.
5
Supplementary Table 1. Potential explanatory variables considered in empirical disturbance 78
modeling (see Seidl et al 3). 79
factor
group description
disturbance agent
wind bark
beetles
forest
fire
forest forest area × × ×
growing stock × × ×
proportion of conifers on growing stock × × ×
median age × × ×
proportion of forest area >100 years × × ×
skewness of age class distribution × × ×
climate mean annual temperature × ×
seasonal temperature (MAM, JJA, SON, DJF) × × ×
lagged annual temperature × ×
lagged seasonal temperature (MAM, JJA, SON, DJF) × ×
annual precipitation × ×
seasonal precipitation (MAM, JJA, SON, DJF) × × ×
lagged annual precipitation × ×
lagged seasonal precipitation (MAM, JJA, SON, DJF) × ×
percentage of DJF precipitation as rain ×
daily peak wind, annual aggregation ×
daily peak wind, seasonal (MAM, JJA, SON, DJF) ×
monthly peak wind, annual aggregation ×
monthly peak wind, seasonal (MAM, JJA, SON, DJF) × ×
interaction wind damage ×
lagged wind damage ×
MAM= spring, JJA= summer, SON= fall, DJF= winter 80
81
From these potential explanatory variables those with a significant influence on the respective 82
disturbance agent were selected by means of unsupervised machine learning using the 83
Random Forest algorithm 18
. Variables were eliminated iteratively, starting from the full set of 84
potential predictors (see Supplementary Table 1), and only variables reducing the mean 85
square error over random permutations of the same variable were retained. Subsequently, the 86
selected country- and agent-specific explanatory variables were related to disturbance damage 87
by means of structural equation modeling (SEM, 19
). The a priori path model used modeling 88
© 2014 Macmillan Publishers Limited. All rights reserved.
6
is given in Supplementary Figure 2, and shows that we applied a structure where climate 89
change and changes in forest structure, composition, and extent were combined in two main 90
latent variables. For every country and disturbance agent the measurement model for these 91
latent variables consisted of the manifest variables (i.e., drivers) selected by means of the 92
above described Random Forest analysis. The disturbance agents wind, bark beetles, and 93
forest fire were represented by independent models at the country-scale. However, 94
interactions between wind and bark beetle damage were considered where such an interaction 95
term significantly improved the mean square error in the Random Forest analysis. More 96
details on the derivation of these SEMs as well as on their goodness of fit are given by Seidl 97
et al. 3. An in-depth evaluation in the context of the current study is presented in the section 98
Supplementary Methods: Evaluation and uncertainty analyses below. 99
Based on the fitted SEM parameters, the skewness of age-class distribution was found to be 100
the most influential forest change predictors of wind disturbance across all countries 101
considered. For bark beetle damage, this age-related variable was found to be equally 102
influential as growing stock and species composition (i.e., proportion of conifers), with the 103
latter being significant in 50% of all the individual SEMs. For wildfire, the most frequently 104
retained variable over all SEMs was growing stock, which had a significant influence in two 105
thirds of the SEMs. The most influential variable on forest fires was median age. With regard 106
to climate variables, wind damage was most strongly driven by peak wind speeds, with a 107
particular influence of the wind climate in the winter season. For bark beetle damage and 108
forest fires, the interplay between low precipitation and high temperatures was found to be the 109
main climatic driver. Spring and summer precipitation was found to be particularly influential 110
on bark beetle damage, while fire was in addition also related to the temperature and 111
precipitation regimes at annual timescales 3. Supplementary Table 2 lists the top five climate- 112
and forest-related drivers of continental-scale disturbance damage in Europe over all agents 113
and SEMs. 114
© 2014 Macmillan Publishers Limited. All rights reserved.
7
115
Supplementary Figure 2. The generic design of the structural equation models used for 116
disturbance modeling. Ellipses indicate latent variables, boxes represent manifest variables 117
(note that the three indicators shown for the latent variables climate and management 118
graphically represent the full set {1,2,…,n} of drivers selected from the potential list by 119
means of Random Forest, cf. Supplementary Table 1), single-headed arrows denote a direct 120
influence in the model, double-headed arrows represent interactions between variables, ε 121
denote error terms. C1-Cn= climate indicators, M1-Mn= management indicators, D= 122
disturbance indicator, A= auxiliary interaction indicators. Dashed lines: not included for all 123
disturbance agents. Source: Seidl et al. 3. 124
125
We here used the thus parameterized disturbance models to estimate future forest disturbance 126
levels in Europe, based on input from climate scenarios and their effect on forest resources as 127
estimated with EFISCEN (Supplementary Figure 1). It has to be noted, however, that these 128
disturbance SEMs are purely data-driven, empirical models. A downside in applying 129
empirical models in this context is their inherent limitation with regard to the prediction of 130
novel future conditions 20
. In order to acknowledge this limitation we applied a number of 131
restrictions to our disturbance modeling. We restricted our disturbance SEMs to a prediction 132
climate
management
disturbance
C1
C2
Cn
M1
M2
Mn
D
1
1
1
10
βC
βM
εC1
εC2
εCn
εM1
εM2
εMn
αC1
αC2
αCn
αM1
αM2
αMn
γCM
A2 εA2
λA1
λA2
A1 εA1
interaction
© 2014 Macmillan Publishers Limited. All rights reserved.
8
period of 20 years (2011-2030), i.e., a considerably shorter time period than the one used for 133
model parameterization (1958-2001). We furthermore chose to use the SEMs in a 134
distribution-based mode of prediction, i.e., future changes in the disturbance regime were 135
estimated in terms of changes in the quantiles of the empirical distributions underlying the 136
models. This means that extrapolation beyond the parameterization data was omitted, and 137
future predictions are constrained by the most severe observation of 1958–2001. The future 138
disturbance levels calculated here are thus conservative estimates, as climatic changes might 139
lead to novel conditions (e.g., novel host agent combinations) that have the potential to 140
transgress the observed system boundaries of the past 21
. Furthermore, we restricted our 141
modeling in space, focusing on the core area of the respective disturbance agent. If an agent 142
was not reported with sufficient frequency and quality in the calibration period in a certain 143
ecoregion, we did not include it in our assessment of that particular region (see also 144
Supplementary Table 5 below). The fact that climate change might also alter the distribution 145
of disturbance agents, both with regard to their trailing and leading edges 22,23
, is thus not 146
considered in our estimates. Furthermore, while we here used annual and seasonal aggregates 147
of climate as drivers, future developments towards the use of process-based disturbance 148
models could improve e.g., the representation of climatic extremes in modeling 20,24
. 149
150
© 2014 Macmillan Publishers Limited. All rights reserved.
9
Supplementary Table 2. The most influential forest- and climate-related drivers of 151
disturbance damage in Europe. Values denote multiplicative combinations of the standardized 152
path coefficients α and β (see Supplementary Figure 2), averaged over all SEMs in which the 153
variables were significant (percentage of models in parenthesis). Influence ranks were 154
determined by weighting path coefficients with number of SEMs, and averaging over all three 155
disturbance agents. 156
factor
group rank variable
disturbance agent
wind bark beetles forest fires
forest 1 skewness of age-class distribution 0.250
(53.8%)
0.373
(37.5%)
0.286
(47.6%)
2 proportion of conifers 0.206
(53.8%)
0.376
(50.0%)
0.179
(47.6%)
3 growing stock 0.081
(38.5%)
0.371
(37.5%)
0.231
(66.7%)
4 median age 0.208
(30.8%)
0.270
(37.5%)
0.357
(38.1%)
5 proportion of forest area >100 years 0.215
(23.1%)
0.161
(37.5%)
0.271
(47.6%)
climate 1 mean annual precipitation - 0.07
(50.0%)
0.282
(47.6%)
2 precipitation in spring (MAM) - 0.346
(25.0%)
0.166
(38.1%)
3 precipitation in summer (JJA) - 0.265
(25.%)
0.206
(28.6%)
4 daily peak wind speed in winter (DJF) 0.228
(30.8%) - -
5 mean annual temperature - - 0.165
(38.1%)
157
158
The REGIME model 159
Modeling approach 160
The REGIME model 4 provides a general theoretical framework for quantitatively assessing 161
the effects of disturbances on ecosystem carbon storage at large spatial scales. The main 162
constituents of the C cycle in REGIME are net primary production (NPP), the size of biomass 163
and soil carbon pools, and their residence times. The disturbance regime is characterized via 164
© 2014 Macmillan Publishers Limited. All rights reserved.
10
disturbance interval and disturbance severity, assuming Poisson-distributed disturbance 165
events. The disturbance effect on ecosystem C is expressed as Equation 1, 166
(1)
with X the expected total ecosystem carbon, U the net primary productivity, the carbon 167
residence time in the ecosystem, the mean disturbance interval (years), s the fraction of 168
biomass removed by a disturbance event (severity), and the residence time of carbon in the 169
biomass pool. Equation 1 illustrates that high C uptake U (i.e., a swift recovery from 170
disturbance) and long C residence times positively affect ecosystem C storage, while high 171
disturbance levels ( and s) reduce the C stored in the system. The disturbance effect is 172
furthermore modulated by the residence time of the C pool directly affected by disturbance 173
( ). The longer-lived the C that is lost through disturbance, the bigger is the impact on the 174
total ecosystem C budget (cf. the “slow in, rapid out” nature of forest C 25
). Weng et al. 4 175
showed that disturbance frequency and severity can be combined into a single disturbance 176
index (σ) for the purpose of large scale modeling, with σ defined as the fraction of live 177
biomass C removed by disturbance per unit of time (Equation 2), 178
(2)
Through transformation and substitution we obtain a four-parameter model of C stocks as 179
affected by disturbance (Equation 3): 180
(3)
Here we made two modifications to the original REGIME model formulation: First, while 181
forest management is not explicitly considered in the original REGIME model, we expanded 182
the model to include management, as our focus here is primarily on managed forest 183
ecosystems. Conceptually, management reduces the flux from living biomass to the litter and 184
© 2014 Macmillan Publishers Limited. All rights reserved.
11
detritus pools via the removal of biomass C, with denoting the biomass fraction entering 185
the litter and soil pools (and, inversely, 1- the harvest fraction). Second, since we did not 186
have separate information on European-scale soil and litter pools available for this study, we 187
lumped these two compartments into a single pool of dead organic matter, with its C 188
residence time. The modified version of Eq. 3 was thus rendered as Equation (4): 189
(4)
190
Model parameters and drivers 191
Residence time parameters for REGIME were estimated using country-level EFISCEN 192
predictions for biomass and soil carbon stocks in combination with NPP estimates. Residence 193
times for the live biomass pool and the combined detritus and soil organic matter pool 194
were derived via Equation 5: 195
(5)
with 196
Also the harvest fraction ( was estimated from EFISCEN simulations of removals 197
from harvest (xH) and litterfall (xL) (equation 6). 198
(6)
© 2014 Macmillan Publishers Limited. All rights reserved.
12
These REGIME parameters were determined separately at the country-level for each of the 199
studied scenarios. Exemplarily, parameterization data and parameter estimates for the 200
reference management strategy are given in Supplementary Table 3. It has to be noted that 201
dynamic feedbacks of disturbances on forest structure and composition were not explicitly 202
considered in this study, as we used scenarios of (undisturbed) forest development as external 203
drivers in our modeling (see also Supplementary Figure 1). 204
The REGIME disturbance index σ (i.e., the fraction of live biomass removed by disturbance 205
per year) corresponds well to the disturbance percentages p (i.e., damaged timber volume as a 206
fraction of growing stock) estimated using SEMs (see Figure 2). However, in order to convert 207
p into σ additional assumptions on salvage harvesting and the fate of foliage, branch, and root 208
biomass after disturbance had to be made. As salvage harvesting is the default operation after 209
wind and bark beetle disturbance in managed forest ecosystems in Europe 26,27
we assumed 210
stemwood C to be removed from the ecosystem after disturbance by these two agents. C in 211
foliage, branch and root compartments, on the other hand, were assumed to remain on site in 212
our calculations. For wildfire, area-based disturbance estimates were converted to volume-213
based values using the ecoregion-specific conversion factors estimated by Schelhaas et al. 16
. 214
Higher consumption rates of foliage and branches can be expected for fire compared to wind 215
and bark beetles 28,29
, while the salvage proportion is likely to be lower for this agent. As 216
parsimonious baseline we assumed full compensation between these two effects and 217
implemented the same ratio of p/σ for fire as for wind and bark beetles. In order to test the 218
effect of these assumptions regarding consumption and salvage after wildfire on our overall 219
results we conducted a local sensitivity analysis for these parameters (see below). 220
221
© 2014 Macmillan Publishers Limited. All rights reserved.
13
Supplementary Table 3. REGIME model parameters used to estimate the effects of 222
disturbance on ecosystem C storage. All values are for the period 2021-2030 and pertain to 223
the reference management strategy under stable climate. 224
Xa X1a Ua
b b
b
(Mg·C·ha-1) (Mg·C·ha-1) (Mg·C·ha-1) (years) (years) (dim.c)
Alpine 274 137 9.6 28 14 0.21
Pannonic 206 101 7.2 29 14 0.22
Sub-Atlantic 202 104 7.4 27 14 0.23
Northern 149 58 5.5 27 11 0.21
Atlantic 144 70 5.9 24 12 0.20
Central Mediterranean 135 76 4.2 33 18 0.19
Mediterranean East 116 42 3.7 38 13 0.16
Mediterranean West 89 36 3.3 28 12 0.17
Europe 162 75 5.8 29 13 0.21
a EFISCEN simulation results 225
b REGIME parameter estimates (Equations 5-6) 226
c dimensionless 227
228
229
Climatic forcing and tree growth response 230
As the main approach to address future uncertainty in projecting trajectories of disturbance 231
damage scenario analysis was used. We studied 14 scenarios of future climate change and tree 232
growth (Supplementary Table 4). These scenarios comprise three different IPCC storylines of 233
future development 30
. Storyline A1B and B1 both describe a world of low population growth, 234
but differ with regard to economic growth (higher in A1B) and technological development 235
(geared towards a service and information technology in B1). While A1B and B1 assume 236
global convergence, storyline B2 emphasizes local solutions and more diverse technological 237
change 30
. The corresponding changes in the climate system were derived from runs with 238
three different sets of global circulation models (GCM) and regional climate models (RCM). 239
Furthermore, in order to address uncertainties with regard to tree growth changes, two 240
© 2014 Macmillan Publishers Limited. All rights reserved.
14
alternative responses to climatic changes were included. These variants differ with regard to 241
the underlying assumptions of a CO2 fertilization effect on forest productivity, with one 242
assuming acclimation of photosynthesis to rising CO2 (at a level that is functionally 243
equivalent to a concentration of 350 ppm, i.e., no CO2 fertilization effect on tree growth), 244
while the other simulates a persistent stimulation of photosynthesis through increasing CO2 245
concentration 9. Furthermore, for one climate model combination (ECHAM5-CCLM) and two 246
storylines (A1B and B1) a second realization of the climate model runs was also available for 247
analysis. These are runs that assume an identical forcing as in the first realization, but differ 248
with regard to the assumed starting conditions in the climate simulations (e.g., concerning the 249
state of oscillating climate phenomena such as ocean circulation). In order to achieve a 250
balanced ensemble of scenarios, and to not overemphasize any particular storyline or climate 251
model, only the first realization of all climate model runs was included in the ensemble 252
analysis reported in the main text. We used the median and interquartile range (IQR) to 253
describe the ensemble’s central tendency and spread, respectively. A no climate change 254
scenario was also simulated, serving as a baseline for the assessment of climate change 255
effects. 256
For each climate scenario, relative signals were derived by standardizing the prediction period 257
to a past baseline period. We used annual and seasonal (March-May, June-August, 258
September-November, December-February) aggregates of temperature and precipitation 259
anomalies in the prediction of disturbance damage (see Supplementary Table 1). In addition 260
we also calculated maximum daily and monthly wind speed anomalies from the climate 261
model data, and derived an index of winter wetness (percentage of winter precipitation falling 262
as rain). These variables were subsequently used as input in the structural equation models 263
used to estimate the climate change impact on disturbance damage in Europe. Climatic effects 264
on forest growth were estimated by means of simulation analyses with process-based tree 265
© 2014 Macmillan Publishers Limited. All rights reserved.
15
growth models 9,11
, and were implemented into EFISCEN via relative growth change values 266
per decade 6,31
. 267
268
Supplementary Table 4. The 14 scenarios of future climate change and tree growth, and 269
their changes in 2021-2030 relative to the baseline period 1971-2001. Please note that only 270
the first realization scenarios where used in the ensemble analysis reported in the main text, 271
while results of the second realization were used to quantify within-scenario uncertainties. 272
story-
line 30
climate modeling growth response modeling changes relative to 1971-2001
source models
(GCM-RCM)
reali-
sation model
CO2
fertilization
effect
MATa
(°C)
MAPb
(%)
MWSc
(%)
A1B ECHAM5-CCLM 1 4C acclimated +0.84 +2.79 +2.78 9,32
A1B ECHAM5-CCLM 1 4C persistent +0.84 +2.79 +2.78 9,32
A1B HadCM3-HadRM3 1 4C acclimated +1.67 +3.60 +1.02 9,33
A1B HadCM3-HadRM3 1 4C persistent +1.67 +3.60 +1.02 9,33
A1B Arpège-HIRHAM3 1 4C acclimated +1.06 -2.19 +0.73 9,33
A1B Arpège-HIRHAM3 1 4C persistent +1.06 -2.19 +0.73 9,33
B1 ECHAM5-CCLM 1 4C acclimated +0.55 +4.66 +0.26 9,34
B1 ECHAM5-CCLM 1 4C persistent +0.55 +4.66 +0.26 9,34
B2 HadCM3-HadRM3 1 C-Fix acclimated +1.33 +0.71 ±0.00 11,15
no climate change ±0.00 ±0.00 ±0.00 -
A1B ECHAM5-CCLM 2 4C acclimated +0.66 +4.81 +0.93 9,35
A1B ECHAM5-CCLM 2 4C persistent +0.66 +4.81 +0.93 9,35
B1 ECHAM5-CCLM 2 4C acclimated +1.11 +1.69 -0.94 9,36
B1 ECHAM5-CCLM 2 4C persistent +1.11 +1.69 -0.94 9,36
a mean annual temperature 273
b mean annual precipitation sum 274
c mean annual maximum daily wind speed. 275
276
The spatial grain of our analysis was determined by the resolution of the Europe-wide 277
disturbance data compiled in the DFDE 17
, which is the country scale. This is also the spatial 278
resolution for which the structural equation models (SEMs) of forest disturbance damage have 279
been parameterized previously 3. All gridded and point-based climate information was 280
© 2014 Macmillan Publishers Limited. All rights reserved.
16
interpolated to and/ or aggregated to the country level. Disturbance SEMs were applied 281
annually in order to capture the interannual variability in climate. Since EFISCEN does not 282
operate on annual time step, however, the state of the forest was kept constant for disturbance 283
projections within a given period, and results were aggregated to decadal scale averages for 284
analysis. Following previous works 3,16
, the 29 countries investigated were grouped into eight 285
ecoregions for continental-scale analysis (see Supplementary Table 5). Data gaps for a limited 286
number of countries and time steps were filled via the disturbance percentage of the 287
respective ecoregion. For all ecoregions we focused on the forest area that is available for 288
wood supply as defined by UNECE and FAO 15
, i.e., forest area where any legal, economic, 289
or specific environmental restrictions do not have a significant impact on the supply of wood. 290
Our study thus focuses on forests that are (potentially) managed, and exclude reserves and 291
areas that are exempt from management due to difficult terrain or strict forest protection. 292
Disturbances and C storage in designated wilderness areas or in the macchia shrublands of the 293
Mediterranean ecoregions, for example, are thus not included in the figures reported by this 294
study. Overall, the forest area available for wood supply according to this definition was 295
estimated to be 82.5% of the total forest area in our study region in 2010. 296
297
© 2014 Macmillan Publishers Limited. All rights reserved.
17
Supplementary Table 5. The 29 countries of the study region were grouped into eight 298
ecoregions. Of the three disturbance agents wind, bark beetles, and forest fires only those 299
agents for which sufficient data were available in an ecoregion (depending on the relevance of 300
an agent in an ecoregion as well as on the historical data quality) were considered for analysis. 301
ecoregion countries disturbance agents
Atlantic United Kingdom, Republic of Ireland wind, forest fire
Northern Norway, Sweden, Finland wind, forest fire
Sub-Atlantic Denmark, Germany, France, the Netherlands,
Belgium, Luxemburg
wind, bark beetles, forest
fire
Alpine Austria, Switzerland wind, bark beetles, forest
fire
Pannonic Czech Republic, Slovak Republic, Poland,
Hungary, Romania
wind, bark beetles, forest
fire
Mediterranean West Spain, Portugal forest fire
Central
Mediterranean
Italy, Slovenia, Croatia, Serbia, Bosnia,
Macedonia, Albania forest fire
Mediterranean East Greece, Bulgaria forest fire
302
303
Management strategies 304
For each of the 14 scenarios of future climate and tree growth (Supplementary Table 4) four 305
different management strategies were simulated. The reference strategy describes a 306
continuation of business-as-usual forest management 15
. It assumes that total wood demand 307
increases by 1.51% per year on average in Europe from 2010 to 2030, with stemwood 308
removals increasing accordingly in order to meet the growing wood demand. Also forest area 309
increases moderately by 0.11% per year in the reference scenario. Forest area and growing 310
stock reach 138.8·106 ha and 25.3·10
9 m³ by the end of the study period under the reference 311
management strategy (+5.5% and +18.8% relative to 2005, ensemble median). The proportion 312
© 2014 Macmillan Publishers Limited. All rights reserved.
18
of conifers on growing stock increases slightly to 65.4%, while median age decreases by 5 313
years (Supplementary Table 6). Overall, Europe’s forest resources are slowly but steadily 314
expanding under the reference management strategy. A detailed description of the 315
development of the reference strategy is given in UNECE and FAO15
; Supplementary Table 4 316
shows the state of Europe’s forest ecosystems as projected for the period 2021-2030 under the 317
reference strategy. 318
319
Supplementary Table 6. The state of Europe's forest ecosystems in 2021-2030 under the 320
reference management strategy. Measures of central tendency and spread over the studied 321
ensemble of climate change scenarios are given. Note that the effect of natural disturbances is 322
not considered here. 323
criteria indicator median 25th – 75th percentile
structure and
composition
forest area (ha) 138.8·106 138.8·106 – 138.8·106
growing stock (m³) 25.3·109 25.1·109 – 25.5·109
median age (years) 45.2 45.2 – 45.2
conifer share (% of growing stock) 65.4 65.3 – 65.4
C stocks total ecosystem carbon (Mg) 22.4·109 22.3·109 – 22.5·109
live biomass C (Mg) 10.4·109 10.3·109 – 10.5·109
detritus and soil C (Mg) 12.0·109 12.0·109 – 12.0·109
C fluxes net primary productivity (Mg yr-1) 799·106 783·106 – 805·106
removal by harvests (Mg yr-1) 153·106 153·106 – 153·106
net ecosystem productivity (Mg yr-1) 92.2·106 79.1·106 – 95.4·106
324
The first alternative management strategy studied puts a focus on maximizing the C stored in 325
forest biomass (in short henceforward referred to as carbon strategy). This strategy assumes 326
that there is an incentive for the forest owner to maximize carbon in the forest, for example 327
through a subsidy or carbon market at a sufficient level to cover the extra costs of the 328
modified management regime. Rotation lengths were increased in 5-year steps to a maximum 329
© 2014 Macmillan Publishers Limited. All rights reserved.
19
increase of 25 years in the carbon strategy. The maximum age of thinning was increased 330
accordingly. The final rotation length and removal from thinnings were optimized to yield the 331
highest in situ C storage at the country level, balancing increased growth of remaining trees 332
with C loss from removals. On average over all countries, the rotation age was increased by 333
11.6 years in the carbon strategy until 2021-2030. Thinning removals were on average 53.1% 334
of the total harvested volume 15
. As a result, both the growing stock and median age increased 335
considerably under the carbon strategy, compared to reference management (Supplementary 336
Table 7). 337
As second alternative management strategy a scenario prioritizing the conservation of 338
biodiversity was implemented (henceforward in short referred to as biodiversity strategy). 339
This management strategy assumes that political decision makers give priority to the 340
protection of biological diversity, and shape the political framework for the forest sector 341
according to the goal of conserving and enhancing biodiversity. In particular, it is assumed 342
that an additional 5% of the forest area currently under management will be set aside by 2030. 343
Since we here study only managed forests (forests available for wood supply sensu UNECE 344
and FAO 15
) this means that the forest area studied decreases under the biodiversity strategy 345
(see also Supplementary Table 7). In the remaining managed forests, rotation ages are 346
increased by 10 and 20 years for broadleaved species and conifers, respectively. Overall, 347
these measures result in 12% lower timber removals in 2021-2030 under the biodiversity 348
management strategy compared to reference management. Furthermore, harvesting residues 349
remain on site after all harvesting operations. In order to promote a convergence of the tree 350
species composition with the potential natural vegetation, 50% of conifer-dominated forests 351
are converted to broadleaved forest types after clearfelling. Overall, the biodiversity strategy 352
results in older forests and – despite a decreasing forest area in focus – an increase in total 353
growing stock and C storage due to lower biomass removals 15
. 354
© 2014 Macmillan Publishers Limited. All rights reserved.
20
The third alternative management strategy investigated aims at promoting the use of forest 355
biomass for energy production 37
(henceforward referred to in short as wood energy strategy). 356
The assumptions underlying this strategy relate to the ambitious goals for green energy use in 357
the European Union until 2020, and extrapolate this trend until the end of the study period in 358
2030. While this strategy assumes a considerable shift in the utilization of the extracted forest 359
biomass towards energy production, changes in forest structure and composition are only 360
moderate compared to the carbon and biodiversity strategies. Since removals are capped with 361
the maximum sustainable harvest in the EFISCEN simulations for all scenarios, the rising 362
demand on biomass for bioenergy only moderately reduces growing stock in this scenario 363
(Supplementary Table 7). The strong implications of this scenario on trade and material use 364
are discussed elsewhere 15
, and are of no direct relevance here. 365
366
© 2014 Macmillan Publishers Limited. All rights reserved.
21
367
Supplementary Table 7. Percent changes in forest extent, structure, and composition in the alternative management strategies (period 2021-2030) 368
relative to reference management (cf. Supplementary Table 6). All values relate to median changes over the ensemble of studied climate change 369
scenarios. 370
carbon strategy biodiversity strategy wood energy strategy
forest
areaa
growing
stock
median
age
conifer
shareb
forest
areaa
growing
stock
median
age
conifer
shareb
forest
areaa
growing
stock
median
age
conifer
shareb
Alpine ±0.0 +10.4 +15.2 +1.6 -4.9 -2.7 +1.7 -1.5 ±0.0 -0.6 -1.3 -0.1
Pannonic +0.1 +3.7 +6.3 +0.4 -3.4 +15.1 +26.9 +1.6 ±0.0 -0.3 -0.4 ±0.0
Sub-Atlantic ±0.0 +11.2 +17.8 -0.2 -5.0 +6.2 +14.1 1.1 ±0.0 -0.7 -1.1 ±0.0
Northern ±0.0 +3.6 +6.4 -0.4 -3.0 +5.7 +10.8 -1.1 ±0.0 -1.4 -1.1 ±0.0
Atlantic ±0.0 +8.9 +5.5 +0.6 -4.2 +34.8 +24.4 +2.3 ±0.0 -1.2 -2.1 ±0.0
Central Mediterranean +0.2 +1.6 +5.7 -1.4 -3.7 -1.3 +9.2 -3.0 ±0.0 -0.4 -0.1 -0.2
Mediterranean East ±0.0 +3.3 +26.3 -3.5 -3.5 +2.9 +31.6 -3.6 ±0.0 ±0.0 ±0.0 ±0.0
Mediterranean West ±0.0 +2.9 +17.3 +2.2 -5.3 +12.2 +31.4 -5.0 ±0.0 -1.5 -4.5 +0.1
Europe ±0.0 +5.2 +10.2 -0.1 -3.9 +6.2 +15.3 -0.4 ±0.0 -0.7 -1.0 ±0.0
a relates to forest area available for wood supply as defined by UNECE and FAO
15. 371
b relative to growing stock 372
373
374
© 2014 Macmillan Publishers Limited. All rights reserved.
22
Supplementary Methods: Evaluation and uncertainty analyses 375
Evaluation of forest disturbance projections 376
We conducted an in-depth evaluation of our continental-scale disturbance modeling 377
framework to test its predictive ability against data from the calibration period 1958-2001 as 378
well as against independent data (i.e., data that have not been used in fitting the models). We 379
first compared observed and predicted disturbance damage at decadal scale for the period 380
1958-2001. Subsequently, we evaluated the predicted disturbance change between the periods 381
1958-2001 (calibration period) and 2002-2010 (independent evaluation period). For the latter, 382
the European Forest Disturbance Database DFDE 17
was updated with observations for the 383
period 2002-2010. This second exercise aims particularly at testing the ability of the models 384
to estimate disturbance levels under novel conditions. This performance is a crucial factor for 385
the predictive application of these models in the current study. Thirdly, we tested if the 386
empirical disturbance models are able to reproduce the observed decadal-scale variation in 387
damage from the three disturbance agents wind, bark beetles, and forest fires. To that end we 388
related the damage levels for each period to the mean damage level 1958-2010, and compared 389
the predicted relative differences to the observed data. 390
We used reanalysis climate for the years 1958-2010 38
for our model evaluation. In addition to 391
the forest data used in model fitting 3 information for the period 2002-2010 was compiled 392
from recent reports on the state of Europe’s forests 39
. Predictions were done for all years in 393
the calibration and evaluation periods, and for all countries for which disturbance data for 394
model calibration and evaluation were available from the updated DFDE 17
. These cover 395
81.8% (wind), 74.0% (bark beetle), and 87.0% (fire) of the total area considered for the 396
respective agent in the year 2005. All analyses were done at the level of ecoregions, however, 397
conversely to the predictions presented in the main manuscript, no gap-filling was done in the 398
evaluations, in order not to impair the assessment of model performance. 399
© 2014 Macmillan Publishers Limited. All rights reserved.
23
Our evaluation exercise revealed that the SEMs were generally well able to reproduce the 400
observed patterns in disturbance damage in the calibration period (Supplementary Figure 3). 401
The models explained between 40.3% and 69.5% of the variation in decadal-scale disturbance 402
damage at the level of ecoregions. In the independent evaluation period 2002-2010, predictive 403
success did not deteriorate significantly compared to the calibration period (mean R² over all 404
agents of 0.625). This underlines the utility of our empirical disturbance models also for 405
short- to mid-term predictions under novel conditions. Furthermore, we found the models to 406
be generally well able to reproduce the observed changes in disturbance levels between the 407
periods 1958-2001 and 2002-2010 (Supplementary Figure 4). 408
409
410
Supplementary Figure 3. Evaluation of decadal-scale mean annual disturbance damage by 411
(A) wind, (B) bark beetles, and (C) forest fire at the level of ecoregions. White symbols 412
indicate the calibration period 1958-2001 whereas grey symbols show results for the 413
independent evaluation period 2002-2010. Note that the axes are scaled logarithmically. 414
© 2014 Macmillan Publishers Limited. All rights reserved.
24
Supplementary Figure 4. Observed (white) and predicted (grey) disturbance damage by (A) wind, (B) bark beetles, and (C) forest fires in the 415
periods 1958-2001 (calibration period) and 2002-2010 (independent evaluation period). Predictions are made using SEMs 3, and error indicators 416
give the 5% to 95% confidence interval estimated via a Monte Carlo simulations over the parameter space of the empirical models (n=5000). Note 417
that horizontal axes are scaled logarithmically. 418
© 2014 Macmillan Publishers Limited. All rights reserved.
25
Observed annual bark beetle damage, for instance, increased in all ecoregions between these 419
two periods, a trend that was also captured by the disturbance models. Furthermore, for five 420
out of the six ecoregions for which increases in area burnt were reported for the most recent 421
observation period the model predictions agreed with this observed trend. Despite acceptable 422
overall performance the evaluation exercises also revealed that some of the SEMs were biased 423
in their predictions (e.g. Alpine ecoregion for wildfire, Supplementary Figure 4). The largest 424
divergence between observed and predicted disturbance damage, as well as the largest 425
uncertainty ranges, were found for wind disturbance. 426
Also with regard to decadal-scale variation the wind models performed poorest among the 427
three considered disturbance agents. The observed levels of wind damage were particularly 428
underestimated in four selected periods and ecoregions (Supplementary Figure 5). This 429
suggests that not all aspects contributing to highly damaging wind events might be captured in 430
the respective SEMs, an issue that seems to particularly concern the Northern and Alpine 431
ecoregions (see also Supplementary Figure 4). Overall, however, also the variation in damage 432
levels between periods was well captured by the disturbance SEMs (Supplementary Figure 5). 433
For bark beetles and forest fires (as well as for wind when the above mentioned outliers were 434
omitted), our analysis revealed unbiased predictions of decadal-scale damage levels (p-values 435
of 0.376 and 0.323 for bark beetles and forest fires, respectively). Over all agents and periods, 436
the correlation between predicted and observed variation was 0.263 (p=0.044), and improved 437
to 0.495 (p<0.001) when the above described outliers for wind were omitted. 438
© 2014 Macmillan Publishers Limited. All rights reserved.
26
439
Supplementary Figure 5. Comparison of predicted decadal-scale variation in disturbance 440
damage (relative to their 1958-2010 mean value) to observed values at the level of ecoregions. 441
A value of 1 (dotted gridlines) indicates that the observed and/ or predicted damage in the 442
period equals the long-term mean damage from that agent. Note that the axes are scaled 443
logarithmically. 444
445
446
Effects of uncertain starting points in climate scenarios 447
The ensemble analysis conducted to assess possible future disturbance trajectories and their 448
implications for forest C storage considered a range of different scenario storylines, climate 449
models, and growth responses to changing environmental conditions. To investigate the 450
uncertainties within a set of climate model runs we also analyzed a second realization of 451
ECHAM5-CCLM simulations for the storylines A1B and B1 (see Supplementary Table 4). 452
© 2014 Macmillan Publishers Limited. All rights reserved.
27
This analysis examined the uncertainties introduced by unknown initial conditions in climate 453
modeling (e.g., with regard to the state of oscillating ocean circulation patterns), and how they 454
translate into climate impact modeling. As shown in Supplementary Figure 6, the disturbance 455
levels estimated for the second realization are generally within the ensemble of values 456
obtained with the first realization of the same climate model. A pairwise test between 457
realizations indicated no significant differences for wind and forest fire (p=0.507 and 458
p=0.416, paired Wilcoxon signed rank test) for the period 2021-2030. With regard to bark 459
beetle damage, however, realization 2 differed significantly from realization 1 (p<0.001). This 460
indicates that not only different scenario storylines and climate models but also different 461
starting points in climate modeling can have a significant influence on climate change impact 462
assessments. In the particular case of our analysis these findings highlight that our ensemble 463
analysis could be underestimating ensemble spread due to within-scenario uncertainty (e.g., 464
with regard to different starting points in climate modeling). However, since different 465
realizations were not available for the other scenarios studied here, a broader inclusion of this 466
aspect in the ensemble analysis was not possible. 467
© 2014 Macmillan Publishers Limited. All rights reserved.
28
468
Supplementary Figure 6. Disturbance damage in scenario storylines A1B and B1 as 469
projected with ECHAM5-CCLM in four management strategies and two scenarios of growth 470
response to CO2 fertilization (see Supplementary Table 4). Solid lines are for the first 471
realization of climate model runs (used also in the ensemble analysis reported in the main 472
text); dashed lines indicate results for the same scenario assumptions and models but 473
considering alternative starting points in climate modeling (second realization). Note that the 474
y-axis is logarithmically scaled. 475
476
Sensitivity analyses of the REGIME model 477
To corroborate the plausibility of modeling the disturbance impacts on C cycling and gain 478
insights into the effects of parameter uncertainty we conducted a set of sensitivity analyses of 479
the REGIME model. While the general approach was analyzed in depth previously 4, we here 480
focused on the performance of the model in the specific context of this study. Supplementary 481
© 2014 Macmillan Publishers Limited. All rights reserved.
29
Figure 7 shows the sensitivity of total ecosystem C stocks to different disturbance levels 482
relative to undisturbed conditions. As expected, C stocks decrease with increasing level of 483
disturbance σ. However, the magnitude of the disturbance impact on C storage also depends 484
on NPP (as an indicator of recovery speed) and the level of living biomass stock (i.e., the 485
level of C affected by disturbance), decreasing with the former and increasing with the latter. 486
This model behavior is in line with theoretical considerations and previous observations of 487
disturbance impacts on forest ecosystem C 40
. 488
489
Supplementary Figure 7. Sensitivity of the disturbance impact on total ecosystem carbon 490
stocks (relative to undisturbed systems) derived with the REGIME model under different 491
levels of disturbance (i.e., σ of 0.1%, 0.5%, and 1.0%) over (A) NPP, and (B) live biomass 492
stocking levels. All other parameter values were kept constant at the level of their continental-493
scale averages for this sensitivity analysis (see Supplementary Table 3). 494
495
REGIME calculates the C effect of disturbances under the assumption of equilibrium 496
conditions 4. As managed forests rarely are in equilibrium we examined the effect of this 497
© 2014 Macmillan Publishers Limited. All rights reserved.
30
assumption in conjunction with the above described parameterization routine, conducting a 498
sensitivity analysis of the C lost from disturbance to different (fluctuating) levels of biomass 499
(X1) and soil (Xs) carbon. To test the sensitivity of the disturbance effect on forest ecosystem 500
C storage (estimated to 503.4 Tg C in 2021-2030 with the default model parameterization 501
under climate change, see Table 1) to the equilibrium assumption underlying REGIME we 502
varied the living biomass and soil C stocks by ±20% and recalculated the continental-scale 503
effect of disturbances. This local sensitivity analysis (i.e., with all other parameters remaining 504
at their default values) aims to gauge how sensitive our results are to disequilibria and 505
transient changes in C pools in Europe's forests within our ten-year assessment periods. The 506
continental-scale C effect of disturbance in 2021-2030 varied by ±10.1% for a change in XS of 507
±20%. Effects of similar changes in live biomass stocks were more distinct, with the C 508
reduction through disturbance ranging from 372 to 669 Tg C for an X1 of ±20%. The relative 509
differences between alternative management strategies (cf. Table 2), however, remained 510
unchanged in these sensitivity tests. 511
We furthermore tested the effect of our assumptions with regard to salvage and consumption 512
from wildfires. Since uncertainties remain with regard to the C lost in fires 28 and local 513
differences in fire consumption 29 had to be lumped into a single, country-specific factor in 514
our large-scale analysis, it was of interest to assess the sensitivity of our overall results to this 515
particular parameter. We thus set up a local sensitivity analysis varying the share of biomass 516
C lost in fires by ±20%. Using the total continental-scale C effect of disturbance in 2021-2030 517
as evaluation criterion (cf. Table 1) we found that a ±20% variation in the ratio p/σ for forest 518
fires affected the overall results only moderately (±2.8%). This finding adds confidence that 519
our large-scale estimates of disturbance effects are robust despite the remaining uncertainties 520
on salvage and consumption as well as the high local variability in fire regimes. 521
© 2014 Macmillan Publishers Limited. All rights reserved.
31
Supplementary Results 522
Projected future disturbance regimes in relation to past observations 523
The continental-scale wind damage of 44.5·106 m³ yr
-1 predicted for 2021-2030 (ensemble 524
median under reference management) corresponds to the 93nd
percentile of the observed 525
continental-scale wind damage in 1971-2001 (Supplementary Figure 8). According to this 526
analysis, the wind disturbance levels that were on average exceeded only every 15 years in the 527
last decades of the 20th
century can be expected to occur every other year by 2030. The 528
ensemble median of predicted bark beetle damage for 2021-2030 (17.9·106 m³ yr
-1) even 529
corresponded to the 97th
percentile of the historically observed damage levels, conforming to 530
a historic once-in-32-years event. The area of forest predicted to burn on average every other 531
year in 2021-2030 (406,700 ha yr-1
) was historically only exceeded once in 18 years. Values 532
for the individual scenarios in the ensemble varied, but were significantly greater than the 533
historical values (p<0.01, Wilcoxon signed rank test). 534
535
Regional trajectories of future disturbance damage 536
Forest fires were projected to increases most strongly in the western Mediterranean ecoregion 537
(Figure 1). The main driver behind this predicted increase was climate change, in particular a 538
combination of increasing temperatures and decreasing precipitation. This interaction between 539
changes in temperature and precipitation was particularly severe in the scenarios for the 540
period 2011-2020, but persisted throughout the projection period for the western 541
Mediterranean ecoregion. In contrast, such an amplifying interaction of changes in 542
temperature and precipitation was not projected for the central and eastern Mediterranean 543
region by the studied climate scenarios. In these parts of the Mediterranean the climate was 544
predicted to warm more slowly in most scenarios, and precipitation sums even increased 545
© 2014 Macmillan Publishers Limited. All rights reserved.
32
slightly compared to the reference period, resulting in a less pronounced increase in the 546
projected forest fire trajectories compared to the western Mediterranean ecoregion. 547
Bark beetle damage was projected to increase most strongly in the Alpine ecoregion (Figure 548
1). The most prominent driver behind this projected trend was a strong increase in growing 549
stock in the area, inter alia mediated by climate change-related improvements of growing 550
conditions in the mountain forests of this ecoregion. Concurrently, climate was projected to 551
warm disproportionally in the Alpine ecoregion, and summer precipitation declined slightly, 552
additionally contributing an increasing level of damage from bark beetles. In absolute terms, 553
the highest levels of bark beetle damage were predicted for the Sub-Atlantic ecoregion. Also 554
there, rising temperatures and higher growing stocks contributed to increasing bark beetle 555
damages in the projections. 556
Wind disturbance showed the highest temporal variation of all three disturbance agents, 557
underlining the fact that individual extreme events are strongly driving the overall damage 558
from wind in Europe. This is, for instance, evident for the observation period in the Atlantic 559
and Northern ecoregions, which show strong peaks in relation to “the Great Storm of 1987” in 560
the UK and the storm “Gudrun” (January 2005) in Scandinavia. While making predictions 561
about the future occurrence of such extreme events remains difficult, some studies indicate 562
the possibility of increasing cyclone activity in the northern Atlantic in the future 41
. Based on 563
daily wind speed data from the employed climate models we here found that wind disturbance 564
increased particularly in mid-latitude ecoregions (Figure 1). Growing stock increased 565
significantly in these regions, and was a main driver of the projected increase in wind damage. 566
In addition, more frequent wet winter conditions (i.e., and indicator for wet and unfrozen 567
soils) in combination with moderately increased maximum daily winter windspeeds 568
contributed to rising wind disturbance levels (please note that all trajectories and changes 569
discussed in this section refer to the reference management strategy). 570
© 2014 Macmillan Publishers Limited. All rights reserved.
33
571
Supplementary Figure 8. Distribution of the observed annual disturbance damage in Europe 572
in the period 1971-2001 for (a) wind, (b) bark beetles, and (c) forest fires, approximated by a 573
lognormal distribution. The mean damage 1971-2001 is indicated by a dashed vertical line. 574
Shaded areas indicate the ensemble spread of predictions for the period 2021-2030 (min-max 575
range: light grey; interquartile range: dark grey). The predicted ensemble median is indicated 576
as a white vertical line. Predictions relate to simulations under reference management. 577
© 2014 Macmillan Publishers Limited. All rights reserved.
34
Climate sensitivity of future disturbance regimes 578
In order to gain insight into the climate sensitivity of the forest disturbance regime, and to test 579
whether the predicted disturbance changes are attributable to changes in the climate system 580
(and not, e.g., to age-related changes or changes in forest structure 42
), we also made 581
projections for a scenario assuming stable climatic conditions (see Supplementary Table 4). 582
Simulations under this scenario resulted in significantly lower damage levels for all three 583
disturbance agents, compared to climate change runs (p<0.01, Wilcoxon signed rank test). 584
The no climate change simulation lay outside the ensemble of studied climate change 585
scenarios for all agents (Supplementary Figure 9). The predicted disturbance damage in 2021-586
2030 was between 13.5% and 49.3% lower when the effects of climate change were 587
disregarded, compared to the ensemble median under climate change. However, projected 588
damage levels for all agents did not return to the significantly lower values of the 1970s under 589
the no climate change scenarios, but remained at levels that were observed for the 1990s and 590
2000s. 591
592
The effect of alternative management strategies on disturbance regimes 593
The sensitivity of disturbance regimes to the management-induced changes in forest structure 594
and composition in the three alternative management strategies differed by agent 595
(Supplementary Table 8). Overall, wind damage increased most strongly under the 596
biodiversity strategy (+37.6%) relative to reference management. In contrast, under this 597
strategy the area burned by wildfires decreased by -14.2%. While the wood energy scenario 598
had the lowest wind disturbance level in 2021-2030, bark beetle damage was projected to be 599
lowest in the carbon strategy. This indicates that none of the four management strategies is the 600
single best strategy with regard to reducing disturbance damage. Consequently, the locally 601
© 2014 Macmillan Publishers Limited. All rights reserved.
35
and regionally changing importance of disturbance agents and their drivers 43
as well as the 602
heterogeneity in starting conditions needs to be accounted for when developing large scale 603
risk management strategies for Europe's forests 44
. Nonetheless, the considerable sensitivity of 604
disturbance regimes to management changes documented here underlines the potential of 605
management to mitigating intensifying disturbance regimes 45,46
. 606
607
608
Supplementary Figure 9. The effect of climate change on trajectories of future disturbance 609
damage in Europe. Solid lines and envelopes indicate the results of the climate change 610
ensemble analysis (Supplementary Table 4), while dashed lines are projections for a scenario 611
assuming stable climate. For all runs a continuation of business-as-usual forest management 612
was assumed (reference strategy). The ensemble envelopes indicate the ensemble median, 613
interquartile range (dark grey), and minimum – maximum range (light grey). Please note that 614
the y-axis is logarithmically scaled. 615
© 2014 Macmillan Publishers Limited. All rights reserved.
36
Forest carbon implications 616
The highest future C pools in Europe were simulated for the Northern ecoregion 617
(Supplementary Table 9). EFISCEN simulations (without disturbance) showed that alternative 618
forest policies could either strongly increase the C stored in Europe's forests (carbon strategy) 619
or reduce it as a result of an intensified utilization of forest resources (wood energy strategy 620
15). With regard to increasing the in situ C storage (carbon strategy), for instance, the Alpine 621
and Sub-Atlantic ecoregions were the most responsive regions 47
, with C storage increases 622
exceeding 5% compared to reference management by 2021-2030. Disturbance damage 623
considerably reduced forest C storage in all management strategies, with a disturbance effect 624
of between 494 and 584 Tg C (compared to the respective undisturbed runs, see also Table 2). 625
The most negatively affected areas in terms of carbon were the Sub-Atlantic and Eastern 626
Mediterranean ecoregions, in which disturbances reduced ecosystem C stocks by 5.4% and 627
3.8%, respectively, compared to undisturbed simulations (reference management). 628
629
© 2014 Macmillan Publishers Limited. All rights reserved.
37
Supplementary Table 8. Relative changes (%) in disturbance percentage (i.e., the mean annual percent damage relative to growing stock) in 630
alternative management strategies relative to reference management (period 2021-2030). 631
wind bark beetles wildfire
carbon
strategy
biodiversity
strategy
wood
energy
strategy
carbon
strategy
biodiversity
strategy
wood
energy
strategy
carbon
strategy
biodiversity
strategy
wood
energy
strategy
Alpine +49.9 -62.2 +0.5 -7.6 -5.0 +0.4 +6.5 -26.5 +0.4
Pannonic +56.7 +54.7 -3.7 +10.6 +23.8 -0.2 -57.4 -91.2 +2.1
Sub-Atlantic +15.2 +35.2 -0.9 -9.9 -4.2 +0.7 -9.3 -41.5 +0.7
Northern -2.3 +63.9 +7.6 ±0.0a ±0.0a ±0.0a -3.4 -37.0 +1.4
Atlantic +12.0 +132.5 -0.2 ±0.0a ±0.0a ±0.0a -3.4 +2.6 -0.9
Central Mediterranean ±0.0a ±0.0a ±0.0a ±0.0a ±0.0a ±0.0a -2.6 +3.2 +0.4
Mediterranean East ±0.0a ±0.0a ±0.0a ±0.0a ±0.0a ±0.0a -2.0 -2.0 ±0.0
Mediterranean West ±0.0a ±0.0a ±0.0a ±0.0a ±0.0a ±0.0a -19.9 -19.9 +1.6
Europe +24.7 +37.6 -0.3 -3.9 +3.7 +0.5 -6.5 -14.2 +0.8
a disturbance agent not relevant/ not modeled in this ecoregion. See also Supplementary Table 3. 632
633
634
635
© 2014 Macmillan Publishers Limited. All rights reserved.
38
Supplementary Table 9. Total forest ecosystem C stocks (biomass and soil, in Tg C) under alternative management strategies in Europe. Shown 636
are results for simulations without disturbance (EFISCEN) and for runs including the effect of disturbance from wind, bark beetle, and wildfire for 637
the period 2021-2030. For both renderings, the median over the scenario ensemble is presented. 638
without disturbance with disturbance
reference
strategy
carbon
strategy
biodiversity
strategy
wood energy
strategy
reference
strategy
carbon
strategy
biodiversity
strategy
wood energy
strategy
Alpine 1,247 1,313 1,206 1,234 1,209 1,266 1,189 1,196
Pannonic 4,097 4,147 4,250 4,063 4,000 4,016 4,102 3,967
Sub-Atlantic 5,352 5,625 5,349 5,294 5,063 5,323 5,005 5,009
Northern 6,662 6,768 6,661 6,551 6,648 6,755 6,643 6,538
Atlantic 499 516 552 495 498 515 550 494
Central Mediterranean 2,404 2,433 2,348 2,374 2,393 2,423 2,338 2,364
Mediterranean East 879 890 868 872 846 856 834 839
Mediterranean West 1,270 1,279 1,324 1,256 1,251 1,262 1,308 1,237
Europe 22,421 22,983 22,574 22,152 21,917 22,428 21,990 21,658
639
640
641
© 2014 Macmillan Publishers Limited. All rights reserved.
Supplementary references 642
1. Schelhaas, M. J. et al. Model documentation for the European Forest Information 643
Scenario model (EFISCEN 3.1.3). 118 (Alterra report 1559, Alterra, and EFI technical 644
report 26, European Forest Institute, 2007). 645
2. Sallnäs, O. A matrix model of the Swedish forest. Stud. For. Suec. 183, 23 (Studia 646
Forestalia Suecica, Swedish University of Agricultural Science, 1990). 647
3. Seidl, R., Schelhaas, M.-J. & Lexer, M. J. Unraveling the drivers of intensifying forest 648
disturbance regimes in Europe. Glob. Chang. Biol. 17, 2842–2852 (2011). 649
4. Weng, E. et al. Ecosystem carbon storage capacity as affected by disturbance regimes: A 650
general theoretical model. J. Geophys. Res. 117, G03014 (2012). 651
5. Nabuurs, G. J., Pussinen, A., Brusselen, J. & Schelhaas, M. J. Future harvesting pressure 652
on European forests. Eur. J. For. Res. 126, 391–400 (2007). 653
6. Eggers, J., Lindner, M., Zudin, S., Zaehle, S. & Liski, J. Impact of changing wood 654
demand, climate and land use on European forest resources and carbon stocks during the 655
21st century. Glob. Chang. Biol. 14, 2288–2303 (2008). 656
7. Verkerk, P. J., Anttila, P., Eggers, J., Lindner, M. & Asikainen, A. The realisable potential 657
supply of woody biomass from forests in the European Union. For. Ecol. Manage. 261, 658
2007–2015 (2011). 659
8. Nabuurs, G.-J., Pussinen, A., Karjalainen, T., Erhard, M. & Kramer, K. Stemwood volume 660
increment changes in European forests due to climate change-a simulation study with the 661
EFISCEN model. Glob. Chang. Biol. 8, 304–316 (2002). 662
9. Reyer, C. et al. Projections of regional changes in forest net primary productivity for 663
different tree species in Europe driven by climate change and carbon dioxide. Ann. For. 664
Sci. 71, 211–225 (2014). 665
10. Schelhaas, M. J. et al. Adaptive forest management to account for climate change-induced 666
productivity and species suitability changes in Europe. Reg. Environ. Chang. submitted 667
(2014). 668
11. Veroustraete, F., Sabbe, H. & Eerens, H. Estimation of carbon mass fluxes over Europe 669
using the C-Fix model and Euroflux data. Remote Sens. Environ. 83, 376–399 (2002). 670
12. Liski, J., Palosuo, T., Peltoniemi, M. & Sievanen, R. Carbon and decomposition model 671
Yasso for forest soils. Ecol. Modell. 189, 168–182 (2005). 672
© 2014 Macmillan Publishers Limited. All rights reserved.
40
13. Nabuurs, G.-J., Schelhaas, M.-J. & Pussinen, A. Validation of the European forest in- 673
formation scenario model ( EFISCEN ) and a projection of finnish forests. Silva Fenn. 34, 674
167–179 (2000). 675
14. Thürig, E. & Schelhaas, M.-J. Evaluation of a large-scale forest scenario model in 676
heterogeneous forests : a case study for Switzerland. Can. J. For. Res. 36, 671–683 677
(2006). 678
15. UNECE and FAO. The European Forest Sector Outlook Study II. 2010-2030. 107 (United 679
Nations Economic Commission for Europe, 2011). 680
16. Schelhaas, M.-J., Nabuurs, G. & Schuck, A. Natural disturbances in the European forests 681
in the 19th and 20th centuries. Glob. Chang. Biol. 9, 1620–1633 (2003). 682
17. DFDE. Database on Forest Disturbances in Europe. Eur. For. Inst. (2013). at 683
<http://www.efi.int/databases/dfde/> 684
18. Breiman, L. Random Forests. Mach. Learn. 45, 5–32 (2001). 685
19. Grace, J. B. Structural Equation Modeling and Natural System. 378 (Cambridge 686
University Press, 2006). 687
20. Seidl, R. et al. Modelling natural disturbances in forest ecosystems: a review. Ecol. 688
Modell. 222, 903–924 (2011). 689
21. Raffa, K. et al. Cross-scale drivers of natural disturbances prone to anthropogenic 690
amplification: the dynamics of bark beetle eruptions. Bioscience 58, 501–518 (2008). 691
22. Jönsson, A. M., Appelberg, G., Harding, S. & Bärring, L. Spatio-temporal impact of 692
climate change on the activity and voltinism of the spruce bark beetle, Ips typographus. 693
Glob. Chang. Biol. 15, 486–499 (2009). 694
23. Marini, L., Ayres, M. P., Battisti, A. & Faccoli, M. Climate affects severity and altitudinal 695
distribution of outbreaks in an eruptive bark beetle. Clim. Change 115, 327–341 (2012). 696
24. Jentsch, A. & Beierkuhnlein, C. Research frontiers in climate change: Effects of extreme 697
meteorological events on ecosystems. Comptes Rendus Geosci. 340, 621–628 (2008). 698
25. Körner, C. Slow in, rapid out--carbon flux studies and Kyoto targets. Science 300, 1242–699
1243 (2003). 700
26. Lindenmayer, D. et al. Salvage harvesting policies after natural disturbance. Science 303, 701
1303 (2004). 702
27. Lindenmayer, D. B., Burton, P. J. & Franklin, J. F. Salvage logging and its ecological 703
consequences. 246 (Island Press, 2008). 704
© 2014 Macmillan Publishers Limited. All rights reserved.
41
28. Vilén, T. & Fernandes, P. M. Forest fires in Mediterranean countries: CO2 emissions and 705
mitigation possibilities through prescribed burning. Environ. Manage. 48, 558–567 706
(2011). 707
29. Campbell, J., Donato, D., Azuma, D. & Law, B. Pyrogenic carbon emission from a large 708
wildfire in Oregon, United States. J. Geophys. Res. 112, G04014 (2007). 709
30. IPCC. Emissions Scenarios. 570 (Intergovernmental Panel on Climate Change, Cambridge 710
University Press, 2000). 711
31. Nabuurs, G., Pussinen, A. R. I., Karjalainen, T. & Erhard, M. Stemwood volume 712
increment changes in European forests due to climate change - a simulation study with the 713
EFISCEN model. 304–316 (2002). 714
32. Lautenschläger, M. et al. Climate Simulation with CLM, Scenario A1B run no.1, Data 715
Stream 3: European region MPI-M/MaD. (World Data Center for Climate, 2009). 716
doi:10.1594/ WDCC/CLM_A1B_1_D3 717
33. Van der Linden, P. & Mitchell, J. ENSEMBLES: climate change and its impacts: 718
summary of research and results from the ENSEMBLES project. (Met Office Hadley 719
Centre, 2009). 720
34. Lautenschläger, M. et al. Climate simulation with CLM, Scenario B1 run no.1, Data 721
Stream 3: European region MPI-M/MaD. (World Data Center for Climate, 2009). 722
doi:10.1594/ WDCC/CLM_B1_1_D3 723
35. Lautenschläger, M. et al. Climate Simulation with CLM, Scenario A1B run no.2, Data 724
Stream 3: European region MPI-M/MaD. (World Data Center for Climate, 2009). 725
doi:10.1594/ WDCC/CLM_A1B_2_D3 726
36. Lautenschläger, M. et al. Climate Simulation with CLM, Scenario B1 run no.2, Data 727
Stream 3: European region MPI-M/MaD. (World Data Center for Climate, 2009). 728
doi:10.1594/ WDCC/CLM_B1_2_D3 729
37. Böttcher, H., Verkerk, P. J., Gusti, M., HavlÍk, P. & Grassi, G. Projection of the future EU 730
forest CO2 sink as affected by recent bioenergy policies using two advanced forest 731
management models. GCB Bioenergy 4, 773–783 (2012). 732
38. Kalnay, E. et al. The NCEP/NCAR 40-year reanalysis project. Bull. Am. Meteorol. Soc. 733
77, 437–471 (1996). 734
39. MCPFE. State of Europe’s forests 2007. The MCPFE report on sustainable forest 735
management in Europe. Europe 247 (Ministerial Conference on the Protection of Forests 736
in Europe, 2007). 737
© 2014 Macmillan Publishers Limited. All rights reserved.
42
40. Turner, M. G. et al. A revised concept of landscape equilibrium : Disturbance and stability 738
on scaled landscapes. Landsc. Ecol. 8, 213–227 (1993). 739
41. Donat, M. G., Leckebusch, G. C., Pinto, J. G. & Ulbrich, U. European storminess and 740
associated circulation weather types: future changes deduced from a multi-model 741
ensemble of GCM simulations. Clim. Res. 42, 27–43 (2010). 742
42. Vilén, T. et al. Reconstructed forest age structure in Europe 1950–2010. For. Ecol. 743
Manage. 286, 203–218 (2012). 744
43. Thom, D., Seidl, R., Steyrer, G., Krehan, H. & Formayer, H. Slow and fast drivers of the 745
natural disturbance regime in Central European forest ecosystems. For. Ecol. Manage. 746
307, 293–302 (2013). 747
44. Jactel, H. et al. A multicriteria risk analysis to evaluate impacts of forest management 748
alternatives on forest health in Europe. Ecol. Soc. 17, 52 (2012). 749
45. Seidl, R., Rammer, W., Jäger, D. & Lexer, M. J. Impact of bark beetle (Ips typographus 750
L.) disturbance on timber production and carbon sequestration in different management 751
strategies under climate change. For. Ecol. Manage. 256, 209–220 (2008). 752
46. Jactel, H. et al. The influences of forest stand management on biotic and abiotic risks of 753
damage. Ann. For. Sci. 66, 701–701 (2009). 754
47. Nabuurs, G. J. et al. Hotspots of the European forests carbon cycle. For. Ecol. Manage. 755
256, 194–200 (2008). 756
757
© 2014 Macmillan Publishers Limited. All rights reserved.