SUPPLEMENTARY MATERIAL

Associated with: Linder M, Fitzgerald J, Zimmermann NE, Reyer C, Delzon S, van der Maaten E, Schelhaas M-J, Lasch P, Eggers, J, van der Maaten-Theunissen M, Suckow F, Psomas, A, Poulter B, Hanewinkel M Climate Change and European Forests: What do we know, what are the uncertainties, and what are the implications for forest management?

S1.Regional climate models and downscaling

The global climate is simulated using so-called general circulation models (GCM), which project the climate future on physics-based processes and first-principles. In order to get more realistic climate projections at a regional to local scale, two types of downscaling are often combined. First, so-called regional climate models (RCM) are calculated to certain larger regions of the world (e.g. all or parts of Europe). These models contain the same physical mechanisms as the GCMs, are fed by GCM output, and simulate the climate development within the study region by using GCMs data as boundary input to the study region. RCMs give a much better spatial representation of the climate in regions and the output is somewhat sensitive to the topography of mountains and their effects on the climate system, though often the output is still too coarse for management and decision-making. A further statistics-based downscaling procedure can be applied (e.g. Pielke & Wilby, 2012) in order to scale the output from RCMs to finer spatial resolution ranging from e.g. 100m to 2km, which can be considered well-suited for management applications. Below, we present climate trends from an ensemble of RCMs at a European scale.

S2. Current trends from regional models - Selection of regional climate models

An ensemble output of six climate models is presented, which have been further downscaled to 1km and finer for climate impact studies, and trends over Europe for these model combinations are reported. Five different RCMs are driven by four GCMs resulting in six GCM/RCM combinations which can then be used to study the impact of likely climate changes on forest species and ecosystems. Table S1 gives an overview of the models used, which originate mostly from the ENSEMBLES EU project, using GCM runs that were calculated for the 4th IPCC assessment report (IPCC 2007). Several scenarios are employed - each representing an assumption on how human actions might influence greenhouse gases, and thus the climate forcing of the future. These so-called “emission scenarios” are labelled A1, A2, B1 and B2) (Nakicenovic and Swart, 2000), and represent conditions that cause high to low warming of the atmosphere.

Each model simulation can be considered one possible representation of how the climate might evolve during the 21st century. The ensemble mean of global or regional trends are sometimes erroneously presented as the most

Model RCM/GCM      Scenario: A1B   A2   B1   B2 CLM/ECHAM5, run by MPI         x    x    x    – RACMO2/ECHAM5, run by KNMI     x    –    –    – HADRN3/HadCM3, run by HC       x    –    –    – HIRHAM5/Arpège, run by DMI     x    –    –    – RCA30/CCSM3, run by SMHI       x    x    –    x RCA30/ECHAM5, run by SMHI      x    x    x    –Table S1 - Climate models used to assess the impact of climate change on forest ecosystems and tree species ranges in the MOTIVE project. RCM models are labeled in bold face, while the GCMs used to feed the RCMs are in normal font.

12

3456

7

89

1011121314151617181920

21

22232425262728293031

3233343536

likely response of the forcing scenario compared to the baseline climate. However, because averaging over global scales is smoothing internal variability more than averaging over smaller domains it is important to realize that the smaller the spatial domain for which simulations or projections are being prepared, the less confidence we can have in the projections (Field et al., 2012). For example is has been shown that the model-simulated magnitude of changes in land precipitation trends is smaller than what has been observed from 1925-1999 in one region (Zhang et al., 2007). The smaller projected changes in model simulations may be due in part to averaging precipitation trends from different model simulations, as spatial patterns of trends simulated by different models are not exactly the same. This dampening effect of averaging different model outputs is crucial to consider in regional scale projections. It is therefore strongly recommended to use different model projections in regional climate change impact studies, and not just the ensemble mean trends.

S3. Climate trends for Europe

Figure S1 illustrates the projected climate change trend from the six used RCM simulations for the A1B scenario representing the summer and winter season means, and the uncertainty in projected summer climates. Summer is represented by the months April to September, while winter is represented by the remaining six months. Uncertainties are expressed as standard deviation among the six models per season.

123456789

1011

12

131415161718

Figure S1 – Climate anomalies for the A1B scenario by 2080 (deviations of the 2051-2080 period from the current, i.e. 1961-1990 climate) averaged over the six RCM models used to assess the impact of climate change on forest ecosystems and tree species ranges in the MOTIVE

project. First row: Anomalies for winter and summer temperature (in °C), and uncertainty (in °C) of summer temperature among all 6 RCMs used; Second row: Anomalies for winter and summer precipitation (in ratio compared to current), and uncertainty (in ratio) of summer

precipitation among all 6 RCMs used.

For temperature, we observe a general warming trend in the range of 1.1 to 4.1 °C, with least warming in the winter half in Atlantic regions (+1.5°C), and highest warming in the Boreal North (winter; +4°C) and Mediterranean South (summer; +3.5°C). The Alps generally face higher warming trends than the surrounding mainland, specifically in the winter months. The uncertainty among the six models is lowest in South-western Europe, increases towards North and East, and is highest in Eastern Europe. For precipitation, the trends show a similar and even clearer segregation between North and South. Winters are projected to become significantly wetter in Northern Europe (+30%), and to a lesser degree also in Central Europe (+15%), while Southern Europe is projected to become slightly drier (-15%). Summers are projected to become significantly drier in Southern Europe (-35%), and to a lesser degree also in Central Europe (-20%), while Northern Europe is projected to become slightly wetter (+20%). The uncertainty among the six models is highest in the (sub-) Mediterranean regions, in the Alps and in the far North of Europe. Regional summaries of projected trends are summarized in Table 1 of the main text, while country-wide summaries are given in Table S4).

S4. What changes have occurred in the 5th IPCC assessment report?

The periodic reports by the Intergovernmental Panel on Climate Change (IPCC) summarize the state of the art of how scientists see the development of the future climate and the associated impacts on ecosystems, economy and society. At the time of writing, the 5 th assessment report is approaching, and some comparisons to the last two reports are possible. The 4 th assessment report (IPCC, 2007) provided a more narrow range of the likely future climate stating that temperatures will likely be between 2.0 and

123456789

10111213

14

15

1617181920

4.5°C warmer than during the 1961–1990 period (with a likelihood of 66%). It also said that temperature increases by more than 4.5°C cannot be excluded (see Rogelj et al. 2012), but that the most likely temperature increase by 2100 is 3.0°C. First comparisons from global climate modelling studies for the 5th IPCC assessment report project an increase of 2.4–4.9°C as medians from three different scenarios of radiative forcing (following different emission scenarios that are similar to those used in earlier reports). A fourth scenario is added that assumes a more rigorous and rapid reduction of greenhouse gases than was ever used before, predicting a median temperature increase of only 1.1°C during the 21 st century. Overall, the model simulations for the 5 th IPCC assessment report project that the likelihood of having global temperature increase exceeding 4.9°C is 14%, thus also likely, but the average of the investigated warming scenarios at the global scale is still indicating a 3.0°C warming compared to 1961–1990. Thus, in general, the newest scenarios do project similar average warming trends as we have seen in the 4 th IPCC assessment report, although some scenarios point to somewhat higher warming trends than were calculated for the 4th report (see Rogelj et al. 2012). Figure S2 shows European climate data simulations for the 4th and 5th assessment report.

Figure S2 – Comparison of European average annual temperature change projected for 2071-2100 compared to 1980-1999 using the SRES scenarios A1B, A2, and B1 of the AR4 (IPCC 2007) compared to the new set of RCP climate scenarios. The SRES A1B scenario that was

most widely used for recent European climate change impact assessments lies in the middle of the range of temperature change projections. RCP means representative concentration pathway and the scenario names refer to the radiative forcing values in these scenarios, ranging from

2.6 to 8.5 W/m 2.

However, it is important to stress that the latest carbon dioxide emissions continue to track the high end of emission scenarios and this means that without fast and ambitious emission reduction efforts there is high likelihood that climate change will exceed the political target of maximum 2 degrees warming above the pre-industrial level. The current emission trajectory is in line with climate change projections

123456789

1011121314

15

1617181920

21

22232425

of 3.5–6.2°C globally (Peters et al., 2013), and consequently it is possible that the climate change projections presented in this paper could be considerably exceeded. Whether or not this will happen, depends strongly on the future development of emissions. The A1B scenario that was used in this study anticipates quite strong emission reductions after 2030.

It was decided to look at changes in extremes as well as the usual climate change variable such as precipitation and temperature. Extremes often have a greater influence on forests than average climate conditions. Variable such as continuous dry days, days with average temperature > 30 °C and various aridity and moisture indices were analysed and compiled for Europe.

The baseline climate data in the following pages was from the Climate Research Unit at University of East Anglia, whereas 2070–2099 projections were derived from four regional climate models from the ENSEMBLES EU project namely: CLM/ECHAM5, run by Max Planck Institute for Meteorology; RACMO2/ECHAM5, run by KNMI - The Royal Netherlands Meteorological Institute); HADRN3/HadCM3, run by the Hadley Centre at the UK Meteorological Office; and HIRHAM3/Arpège, run by DMI - The Danish Climate Centre at the Danish Meteorological Institute.

1234

5

6789

101112131415

16

17

18

Figure S3: Change in mean annual temperature in 2070–2099 period as compared with 1961–1990 for four climate models under A1B

emissions scenario, their mean and the standard deviation which shows greatest uncertainty in South-East Europe.

1

234

Figure S4 - The maps represent the relative change in annual precipitation according to four of the six models from Table 1 between 1961–1990 and 2070–2099. The maps show large uncertainties both in the Boreal and Mediterranean regions, where much higher increases or

decreases can be expected than is represented by the mean ensemble trend among models.

1

2

345

S5. Aridity IndicesAridity indices provide a simple means of expressing the ratio of precipitation to evaporation. Two aridity indices are described below:

1. Moisture index = Precipitation – Potential Evapotranspiration (mm/year)

2. The UNEP has adopted another index of aridity, defined as:

where PET is the potential evapotranspiration and P is the average annual precipitation. Here also, PET and P must be expressed in the same units, e.g., in millimetres. In this latter case, the boundaries that define various degrees of aridity and the approximate areas involved are as follows:

Classification Aridity indexHyperarid AI < 0.03Arid 0.03 < AI < 0.20Semi-arid 0.20 < AI < 0.50Dry subhumid 0.50 < AI < 0.65

Table S2 - IUNEP aridity index classifications.

Source: (UNESCO, 1979)

All models show an increase in moisture index (Figure 1 main text) and hence generally wetter climates in Northern Europe. Only one model shows a decrease in moisture index (METO) and hence a drier climate over northern central Europe, the others predict little to no change (grey areas).

All models except for DMI forecast a decrease in moisture index over the Mediterranean periphery. This is also where there is greatest disagreement amongst the models (higher standard deviation).

Overall, the 4 model mean indicates the area which will experience the greatest increase in aridity around the Mediterranean, and into France, Austria, Slovenia, Croatia, Greece, Bulgaria, Romania.

1

234

5

6

789

1011

12

13

14

151617

1819

2021

Figure S5 - IUNEP Index for 1961–1990 and 2070–2099. The IUNEP index is designed for use mainly in arid areas. Model results for averages for 1961–1990 and 2070–2099 are shown. The maps show an expansion of semi-arid and dry sub-humid areas into South Eastern

Europe and Spain by 2100.

1

2

345

Figure S6 - MAX CDD represents the average length of the longest dry spell which occurs in one year. It is a thirty year average. This graphic shows the average change in length between 1961–1990 and 2070–2099 from four models of the longest dry spell.

1

2

34

30 year Average 30 year MaximumLongitude

Latitude 1961-90 2070-99 1961-90 2070-99

Chamusca, Portugal -8.367 39.267 85 117 141 171North Karelia, Finland 29.617 62.650 14 14 24 24Kronoberg, Sweden 14.667 56.950 16 18 27 39North Wales -3.667 53.067 14 19 30 41Veluwe, Netherlands 5.833 52.067 16 25 29 55Black Forest, Germany 8.267 48.900 20 27 38 52Montafon Valley, Austria 9.933 47.033 13 15 28 35Prades, Spain 1.000 41.317 54 75 118 144Pangiurishte, Bulgaria 24.183 42.050 42 53 83 114Frasin, Romania 25.800 47.533 19 27 34 54

Table S2 - Changes in 30 year average and 30 year maximum of the longest annual number of continuous dry days at ten regional locations between the periods 1961–1990 and 2070–2099.

1

2

34

Figure S7 - Absolute change in number of frost days between 1961-1990 and 2070-2099.

1

2

3

4

5

COUNTRY Temperature (℃) Precipitation (relative change)

Max Annual CDD (Days)

No. of days Temp rises Above 30 Degrees

Moisture Index (yy) (mm/yr)

IUNEP (yy)

Austria +3.55 +0.00 +4.92 +17.53 -128.06 -0.82Belgium +2.94 -0.03 +8.06 +10.94 -111.09 -0.29Bosnia and Herz. +3.67 -0.07 +15.16 +19.32 -219.50 -0.34Bulgaria +3.59 -0.12 +19.34 +15.65 -203.56 -0.18Croatia +3.57 -0.04 +15.84 +22.86 -165.90 -0.22Cyprus +3.41 -0.30 +14.31 +9.35 -164.46 -0.08Czech Rep. +3.23 +0.02 +4.85 +21.66 -72.60 -0.18Denmark +2.88 +0.03 +3.54 +1.54 -41.26 -0.16Estonia +3.67 +0.09 +0.45 +2.63 -5.31 -0.11Finland +4.19 +0.17 -0.24 +0.48 +67.59 -0.06France +3.34 -0.08 +13.06 +20.37 -196.61 -0.39Germany +3.11 +0.00 +5.53 +17.06 -83.14 -0.24Greece +3.55 -0.22 +25.14 +21.06 -254.76 -0.20Hungary +3.43 -0.01 +11.28 +26.31 -114.35 -0.10Ireland +2.14 +0.02 +3.04 -0.06 -22.98 -0.20Italy +3.67 -0.12 +17.78 +18.94 -210.88 -0.38Lithuania +3.43 +0.06 +2.90 +7.61 -25.86 -0.11Luxembourg +3.14 -0.03 +6.83 +15.36 -119.90 -0.31Latvia +3.55 +0.08 +1.50 +4.60 -16.09 -0.11Netherlands +2.80 -0.01 +8.28 +8.25 -78.37 -0.19Norway +3.43 +0.13 -0.22 +0.17 +111.85 -1.76Poland +3.23 +0.02 +4.02 +16.83 -63.44 -0.15Portugal +3.41 -0.19 +26.03 +17.02 -313.87 -0.32Romania +3.57 -0.06 +12.48 +20.74 -165.46 -0.21Spain +3.67 -0.19 +21.90 +20.54 -288.01 -0.28Serbia +3.59 -0.05 +16.16 +22.88 -161.36 -0.15COUNTRY Temperature (℃) Precipitation Max Annual Days Above 30 Moisture IUNEP (yy)

(relative change) CDD (Days) Degrees (No. Of Days)

Index (yy) (mm/yr)

Slovakia +3.39 -0.02 +6.71 +19.82 -118.34 -0.23Slovenia +3.57 -0.04 +8.98 +22.13 -171.53 -0.34Sweden +3.61 +0.13 +0.34 +0.81 +51.09 -0.48Switzerland +3.77 -0.04 +6.42 +15.53 -223.48 -2.12United Kingdom +2.39 +0.02 +4.27 +0.43 -33.71 -0.25

Table S3 - Change in various climate parameters in 2070-2099 as compared with 1961-1990 for four models. All changes are absolute changes except for precipitation which is a relative change.

1

2

34

5

67

1

S6: Species Distribution Models

Description of the dependent variable.

This description follows Zimmermann et al. (2013) to describe the development of species distribution models (SDMs) for the MOTIVE project, which were used in this paper. We used presence/absence information from the ICP Forest Level I inventory (Lorenz, 1995, Fischer et al. 2010) to build a database of tree species presences and absences across Europe. We compiled data for 38 tree species at a total of >6’000 inventory plots. We compiled a series of climate maps under current and potential future climate from downscaled RCM models for future climates. Additionally, we compiled some topographic variables that additionally may influence the spatial patterns of trees. Prior to selecting the variables, we executed a variable importance analysis for each tree species separately. This was done in order to adjust the variable selection to those that have a strong effect on the range dynamics of the species. We refrained from using all possible climate variables on order to avoid too high correlations per species, and in order to keep control over the number of variables we maximally allowed entering the models.

Pre-selection of predictor variable.

We finally designed the variable pre-selection as follows. First, no more than five variables were allowed per species. Second, for species with less than 50 observations, a minimum of ten observations per variable added was required. Third, among similar and highly correlated variables, we selected variables that (a) showed high predictive performance, and (b) did correlate <0.7 to any other variable selected. We only modeled species with at least 25 presence observations, meaning that such a species would only be fitted against two predictor variables. Most species were fitted against five (24 spp) or four (12) variables. Four and three species were fitted against three and two variables, respectively, due to the low availability of presence points in the ICP Forest Level I data set.

Final selection of dependent variable

We selected predictor variables from the following groups of environmental predictors: (1) temperature – either degree days with a 5.56°C threshold or minimum temperature of the coldest month; (2) precipitation – we computed seasonal means and selected among those per species; (3) moisture index (difference between potential evapotranspiration and precipitation) – we computed this index for spring (March-May) and summer (June-August) and selected one of the two according to its predictive power; (4) potential global radiation – we computed winter or annual mean radiation values and selected according to predictive power; (5) slope angle in degree – we added this variable, if it was among the top 5 uncorrelated variables for a given species. By this we allowed to simulate habitat suitability based on predictors that are (a) relevant for a given species, (b) not highly correlated, and (c) we did not include too many variables into models, which might cause problems in a non-analogue climate due to changes in correlation among current to future climate variables. Winter and spring conditions were generally speaking more important for Mediterranean trees, while summer conditions were more powerful predictors for Central and Northern European trees.

12

3

456789

101112131415

16

171819202122232425

26

2728293031323334353637383940

Choice of statistical models for ensemble modelling

Several statistical models were used, since the choice of a specific model has been shown to significantly contribute to uncertainty in projections (Buisson et al. 2010). More specifically, we used the following statistical models: (1) Classification and regression trees (CART), (2) Flexible discriminant analysis (FDA), Generalized linear models (GLM), Generalized additive models (GAM), Artificial neural networks (ANN), and Generalized boosted regression trees (GBM). Therefore, given the use of six statistical models and six future climate model runs (Table S1), we model 36 different possible futures per species and time slice. This allows for assessing the projection uncertainty from both the variability in climate models and the variability originating from the choice of statistical methods.

We optimized each statistical model following procedures described in Thuiller et al. (2009) and we maximized kappa to select a threshold to split probabilistic projections of species presences into simulated presence and absence values. We therefore had one presence/absence map per climate model/statistical model combination available. We then built ensembles of these model projections and classified these as follows: (1) a species is unlikely to find a suitable habitat if less than 30% of the projections indicated presence of a species; (2) a species in moderately likely, associated with high uncertainty, if 30-60% of the projections suggested that the species is there; (3) a species is most likely present with rather low uncertainty under projected climates if in >60% of the 36 model projections presence of a species is simulated to occur. Such a simple classification avoids an over-interpretation of the results from the simple model approach. In order to generate biodiversity maps, we used the >60% threshold among the 36 model combinations to assign “simulated presence” of a species in this ensemble model framework.

References

Pielke R.A., Wilby R.L., 2012. Regional climate downscaling: what's the Point? EOS 93 (5), 52–53.

United Nations Educational, Scientific and Cultural Organization (UNESCO), Map of the World Distribution of Arid Regions: Map at scale 1:25,000,000 with Explanatory Note, MAB Technical Notes 71979, UNESCO; Paris.

1

23456789

10

111213141516171819202122

23

24

252627