Assessing effects of forecasted climate change on the diversity and distribution of European higher plants for 2050 M . B A K K E N E S , J . R . M . A L K E M A D E , F . I H L E , R . L E E M A N S a n d J . B . L A T O U R National Institute of Public Health and Environment, P.O. Box 1, 3720BA Bilthoven, the Netherlands Abstract The rapidly increasing atmospheric concentrations of greenhouse gases may lead to significant changes in regional and seasonal climate patterns. Such changes can strongly influence the diversity and distribution of species and, therefore, affect ecosystems and biodiversity. To assess these changes we developed a model, called EUROMOVE. The model uses climate data from 1990 to 2050 as compiled from the IMAGE 2 model, and determines climate envelopes for about 1400 plant species by multiple logistic regression analysis. The climate envelopes were applied to the projected climate to obtain predictions about plant diversity and distributions by 2050. For each European grid cell, EUROMOVE calcu- lates which species would still occur in forecasted future climate conditions and which not. The results show major changes in biodiversity by 2050. On average, 32% of the European plant species that were present in a cell in 1990 would disappear from that cell. The area, in which 32% or more of the 1990 species will disappear, takes up 44% of the modelled European area. Individual responses of the plant species to the forecasted climate change were diverse. In reviewing possible future trends, we found that plant species, in general, would find their current climate envelopes further northeast by 2050, shifting ranges that were comparable with those ranges in other studies. Keywords: biodiversity, climate change, Kappa statistic, multiple logistic regression Received 3 July 2001 revised version received and accepted 19 August 2001 Introduction The increasing atmospheric concentrations of greenhouse gases could lead to significant changes in regional and seasonal climatic patterns (Houghton et al. 1996). Because of the major influence of climate on the distribution of plant species and vegetation types from continental to regional scale (Holdridge 1947 Walter 1979 Woodward 1987 Huntley 1999), we expect that climate change will alter plant distribution considerably, and could strongly influence the diversity of ecosystems and species (IGBP 1988 Smith et al. 1993 Watson et al. 1996). Consequently, climate change may affect various properties of natural ecosystems, such as biodiversity (Leemans & Halpin 1992) and ecosystem stability (Mooney 1997). Palaeobotanical studies (e.g. Davis & Botkin 1985 Prentice 1986 Huntley 1990 Prentice et al. 1991), simulation studies (e.g. Cramer & Leemans 1991), and observations and experiments (e.g. Hattenschwiler et al. 1996 Parmesan 1996 Molau & Alatalo 1998 Parmesan et al. 1999) also indicate climate- induced changes in ecosystems. The cost and difficulty of detesting climate-induced changes in the coming decades (Watson et al. 1996) make use of alternative ways to assess impacts of current and future climate change a valuable guide to observational work. Various methods and models can assess effects of cli- mate change on biodiversity (e.g. Peters & Lovejoy 1992). However, in many studies, changes are not quantified and spatially oriented over larger (regional) areas, or are only based on climate correlations with species counts, so that a change in local species composition cannot be assessed (e.g. O'Brien et al. 1998). Our study aimed at a geographically explicit quantification of the possible effects of forecasted climate change on the diver- sity of the European flora. Biodiversity or biological di- versity refers to variety within the living world. The term biodiversity is commonly used to describe the number, Global Change Biology (2002) 8, 390��407 390 �� 2002 Blackwell Science Ltd Correspondence: Michel Bakkenes, fax ��3130 27444 19, e-mail Michel.Bakkenes@rivm.nl

variety, and variability of living organisms. Biodiversity is usually defined in terms of genes, species, and ecosys- tems, corresponding to three fundamental and hierarch- ically related levels of biological organisations. As biodiversity can be species-dependent, we developed a species-based probabilistic model, called `euromove'. In euromove, occurrences of plant species with a known European distribution (Jalas & Suominen 1972��94) are correlated with climate data from the image 2 model (Alcamo 1994) using multiple logistic regression. The euromove calculations resulted in climate envelopes for nearly 1400 plant species. Euromove integrates calcu- lated regression equations to analyse the effects of cli- mate change on the European flora. With euromove, response curves and maps can be drawn for plant species diversity and species distributions for various climate scenarios. Climate envelopes, describing ranges of climate vari- ables over which species can occur, are derived from the niche theory of Hutchinson (1957). Many important stud- ies using climate envelopes have been carried out (e.g. Ellenberg 1974 Box 1981). Although climate envelopes of species have been analysed experimentally and mechan- istically, such analyses are not feasible for species grow- ing in natural environments (Sykes & Prentice 1995). Instead, climate envelopes based on empirical correlation between plant species distribution and climate variables have no such practical limitations (e.g. Woodward 1987 Woodward & Williams 1987 Booth et al. 1988 Huntley et al. 1995 Carey 1996). We assume that climate deter- mines the large-scale patterns in physiognomy and po- tential species distribution. Other factors, such as soil characteristics, are important as well, but they influence plant distribution on smaller (i.e. more local) scales. The ecophysiologically relevant variables that we applied to our study were all available in the output of the image 2 model (Alcamo 1994). By using principal component analyses and correlations between the vari- ables, we grouped highly correlated variables, and selected from each group one explanatory variable. Al- together, the resulting variables are assumed to reflect the main controlling factors on plant distribution. As a result, intercorrelative effects between initial explanatory vari- ables were reduced, thereby preventing overfitting. In our study we used observed climate for 1990 and simulated climate for 2050. By applying the climate en- velopes to the forecasted climate in 2050, resulting from scenario calculations from image 2, we were able to ana- lyse climate-induced changes in the diversity of the European flora. Large-scale patterns in species diversity vary geographically (Walter 1979 Grabherr & Kojima 1993 O'Brien et al. 1998). Hence, species diversity as such is not a measure of a region's biological importance the number of species in some regions would simply be low in accordance with their natural environment. However, relative changes in species diversity give in- sight into the vulnerability of a region with respect to climate change. Assessing the change of species diversity in a changing climate on a regional scale can identify vulnerable areas. This may, in turn, stimulate the arising awareness of the impact of climate change and, further- more, allow specific mitigation measures to be planned and implemented. Materials and methods Climate data We derived the calculated present and future climate data from Alcamo's image 2 model (1994). Image 2 pre- sent climate data are derived from an updated version of the IIASA climate database on a global terrestrial grid (Leemans & Cramer 1991). This climate database in- cludes measured values at an array of weather stations for the climatic-normal period of 1931��60. It is interpol- ated onto a grid with a resolution of 0.5 longitude by 0.5 latitude. Image 2 calculated future climate data for 2050 based on a simulation of the Conventional Wisdom Scenario (Alcamo 1994 pages 39��68) by calculating annual temperature and precipitation changes in latitu- dinal belts. These changes are interpolated towards the terrestrial grid by overlaying the normal climate with a scaled climate change based on the longitudinal patterns of a 3-dimensional climate model. This is a common approach in climate change impact assessments (Carter et al. 1994). The Conventional Wisdom Scenario makes conventional assumptions about future demographic, economic, and technical driving forces. It is a reference scenario, and makes no assumptions about climate- related policies. Input data for the main driving forces are based partly on the assumptions of the IS92a scenario for the Intergovernmental Panel on Climate Change (IPCC 1992). The climate sensitivity (global mean tem- perature increase for doubled CO2 conditions) of the model is 2.3 C, which is at the lower end of the IPCC range of 1.5��4.5 C (Houghton et al. 1996). The actual temperature increase used for this study is 1.8 C in 2050, with greenhouse gas concentrations of 513 ppmv, 2.8 ppbv, and 376ppbv for carbon dioxide (CO2), me- thane (CH4), and nitrous oxide (N2O), respectively. The resulting CO2-equivalent concentration is 575 ppmv. Climate can be described by numerous climate vari- ables. Woodward (1987) and Prentice et al. (1992) used various approaches to select ecologically relevant cli- mate variables. The variables should, at least, reflect summer and winter temperatures, and a measure for the available moisture, which are regarded as the main controlling factors for plant distribution (Holdridge 1947 �� 2002 Blackwell Science Ltd, Global Change Biology, 8, 390��407 E F F E C T S O F C L I M A T E C H A N G E O N E U R O P E A N H I G H E R P L A N T S 391

Holten 1990 Leemans & Cramer 1991). In various species-based studies, growing degree-days or effective temperature sums (Flannigan & Woodward 1994 Beerling et al. 1995 Huntley 1995 Sykes & Prentice 1995) or mean temperature of the warmest month (Malanson et al. 1992 Jefree & Jefree 1996) indicates warmth. Cold is reflected by the mean temperature of the coldest month (Malanson et al. 1992 Beerling et al. 1995 Huntley 1995 Sykes & Prentice 1995 Jefree & Jefree 1996), and moisture is indicated by annual precipitation (Fernandez-Palacios 1992 Malanson et al. 1992 Flannigan & Woodward 1994) or the ratio of actual to potential evapotranspiration (`Priestley & Taylor 1972 equation', used in Beerling et al. 1995 Huntley 1995 Sykes & Prentice 1995). The Image 2 dataset (Alcamo 1994) includes 13 climate variables as discussed below (see also Table 1), which we considered for regression analyses. The mean tempera- tures of the coldest (TCM) and warmest (TWM) months are directly selected from the array of monthly tempera- ture values. The effective temperature sums (ETS0 and ETS5) are the summations of quasi-daily values above the respective temperature thresholds of 0 C and 5 C, respectively. These quasi-daily values are obtained by adopting a spline��interpolation approach using the monthly mean temperature values. Annual precipitation (AP) is the sum of monthly precipitation values. The annual potential evapotranspiration (APE) is the summa- tion of quasi-daily values for temperature and cloudiness used together with a few parameters for vegetation prop- erties and latitude (Priestley & Taylor 1972). Actual eva- potranspiration is calculated using a simple bucket model for soil moisture dynamics. The annual actual evapotranspiration (AAE) is the sum of daily values, and is always less than the potential evapotranspir- ation. The alpha moisture index (AMI) is simply the ratio between actual and potential evapotranspiration. A value of 1 means a soil profile that is always moist. Lower values denote climates with a more distinct dry season. The details and rationales of the algorithms used to calculate the moisture balance are described by Prentice et al. (1992). The annual runoff (AR) is defined by the sum of the daily precipitation surplus values that cannot percolate into the soil bucket. The start of the growing season (SGS) is the (derived) Julian day on which the temperature increases above 5 C and when precipitation equals half the potential evapotranspira- tion. The growing season is defined according to the FAO definition that is, the period with adequate mois- ture and temperatures for plant growth based on the daily sequence of precipitation, temperature, and soil moisture (Leemans & van den Born 1994). The growing season ends when soil moisture drops below wilting point or temperatures become too low. The length of the growing season (LGS) is the number of days in between. Mean growing season temperatures (MGST0 and MGST5) are the average temperatures during the grow- ing season above 0 C and 5 C, respectively. Data on plant species distribution and data connection In the Atlas Florae Europaeae (afe) (Jalas & Suominen 1972��94), presence��absence data for (utm) grid cells of, on average, approximately 50sq. km depending on their longitude and latitude coordinates are given for 2432 relatively early evolutionary higher plant (sub) species (up to Cruciferae). These distribution maps have been digitised at the Botanical Museum of Helsinki (Lahti pers. comm. Ascroft pers. comm.). The distribution of all afe (Fig. 1) species shows plant diversity to vary geographically due to not only climatic stresses, such as drought and cold, but apparently also poor data col- lections (eastern Russia). Nevertheless, we assumed the afe data, collected by different individuals throughout Table 1 Climate variables with units and abbreviations as used in this study Climate variable Units Abbreviation Used in study 1. Temperature of the coldest month C TCM X 2. Effective temperature sum above 0 C degree-days ETS0 3. Effective temperature sum above 5 C degree-days ETS5 X 4. Temperature of the warmest month C TWM 5. Alpha moisture index �� AMI X 6. Annual precipitation mm AP X 7. Annual potential evapotranspiration mm APE 8. Annual actual evapotranspiration mm AAE 9. Length of growing season # days LGS X 10. Start of growing season Julian day SGS 11. Mean growing season temperature above 0 C 0 C MGST0 12. Mean growing season temperature above 5 C 5 C MGST5 X 13. Annual runoff mm AR �� 2002 Blackwell Science Ltd, Global Change Biology, 8, 390��407 392 M . B A K K E N E S et al.