Assessment of climate change in Europe from an ensemble of regional climate models by the use of Köppen–Trewartha classification

Authors


Correspondence to: C. Gallardo, Instituto de Ciencias Ambientales, Universidad de Castilla-La Mancha, Avda. Carlos III s/n, 45071 Toledo, Spain. E-mail: clemente.gallardo@uclm.es

ABSTRACT

Through the use of the climatic classification of Köppen–Trewartha (K-T), the ability to reproduce the current climate of Europe has been shown for an ensemble of 15 regional climate model simulations nested in six global climate models. Depending on the simulation, between 55.4 and 81.3% of the grid points are in agreement with observations regarding the location of climate types in current climate simulations (1971–2000). In this respect, the result of the ensemble of 15 simulations is better than that of any individual model, with 83.5% of the grid points in agreement with observations. K-T classification has also been used to analyse the projected climate change over the 21st century under the SRES-A1B emissions scenario. It was found that 22.3% of the grid points in the domain change their climate by the period 2021–2050 compared to current climate and 48.1% change by 2061–2090. The climate shifts affecting the biggest extensions are projected in Central Europe and Fennoscandia, but other smaller areas suffer more intense changes which potentially are more dangerous to vegetation and ecosystems. Generally, these changes occur at a sustained rate throughout the century, reaching speeds of up to 90 × 103 km2 decade−1 in the retreat or expansion of some climates.

1. Introduction

A climate-vegetation scheme, like the Köppen climate classification (Köppen, 1936) or its improvements (Trewartha and Horn, 1980) is a complex system of climates, which is based on the two variables most frequently used in climate studies: precipitation and temperature. The categories or types of these classifications are not only related to the different climates that exist on the Earth, but they are structurally related to the potential vegetation of each zone, and are also indirectly related to the feasible crops and ecosystems. These relationships allow us not only to establish a projection of the future changes in the climate, but also to give a basic estimation of the possible effects on the natural vegetation, crops and ecosystems.

A Köppen-like climate classification has two additional advantages. First, it can be applied practically everywhere on the planet, as the temperature and precipitation data are available almost anywhere over the globe. Second, these variables are also part of the standard output of global climate models (GCMs) and regional climate models (RCMs). Owing to these properties, the Köppen methodology can be applied to track past and future changes in the climate, using the observations that have been gathered over the last 100 years or so, and the outputs of the climate models for past, present or future periods.

Within the field of the study of climate change, the Köppen climate classification and its variants have been used by several authors. Lohmann et al. (1993) used the Köppen classification to check whether a GCM was able to reproduce the present day climate and to analyse how the main climate regions could change as a result of global warming. Leemans et al. (1996) analysed the global biome distribution by applying the Köppen method to the output of four GCMs. Kleidon et al. (2000) estimated the effect of vegetation on the global climate by performing several climate model simulations and then applying the Köppen classification to illustrate the differences among them. Jylhä et al. (2010) used the traditional Köppen classification to study climate trends in Europe with a set of 19 GCMs. Feng et al. (2012) assessed current and future climate changes in the Arctic from the output of 16 GCMs.

The Köppen classification has also been applied to the output of RCMs in order to evaluate climatic refuge for the People's Republic of China (Baker et al., 2010), assess the possible increase of aridity caused by the late 21st century climate change in the Mediterranean region (Gao and Giorgi, 2008), quantify the potential impact of climate change on ecosystems of the Barents Sea Region (Roderfeld et al., 2008) and estimate the climate change effects in Europe (Castro et al., 2007).

The Köppen classification was also used to characterize the climate of certain regions (Baltas, 2007), or to detect the 20th century climate change in the Arctic region (Wang and Overland, 2004), in the United States (Diaz and Eischeid, 2007) or in Europe (Gerstengarbe and Werner, 2009).

In this work, the outputs of 11 high-resolution RCMs were used to reproduce the current climate in Europe and the Mediterranean area and to assess the possible magnitude of future climate change under SRES-A1B emission scenario. Regarding the concentrations of equivalent CO2 over the 21st century, the A1B scenario is intermediate for both the SRES scenarios group and the new RCP scenarios. The uncertainty associated with the emissions scenario has not been explored in this work, but the use of an extreme scenario has been avoided. The applied RCMs were driven by six GCMs, resulting in a total of 15 simulations. This makes a difference to other RCM-based works, where only one GCM is considered, by providing boundary conditions to only one or several RCMs. This means that the analysis presented in this study is more robust, because, in addition to using an ensemble of several RCMs, it also incorporates the uncertainty related to the GCMs. Unlike some previous works (Castro et al., 2007), in this study the whole period of 1961–2090 was considered, which made it possible to analyse the tendency of the changes throughout the 21st century.

The study is organized in the following way. In Section '2. The observed present day K-T climate distribution', a brief description of Köppen–Trewartha (K-T) climate classification scheme and its observed present day distribution is shown. In Section '3. Climate model simulations', the climate simulations are described. In Section '4. Results', the evolution of the climate in Europe and the Mediterranean area is analysed. Finally, some concluding remarks are presented in Section '5. Concluding remarks'.

2. The observed present day K-T climate distribution

The Köppen climate classification and its variants are the most widely used climate classification systems. In the present study, an improvement of the original system, the K-T (Trewartha and Horn, 1980) climate classification (Table 1) was used.

Table 1. The Köppen–Trewartha climate classification
Climate typeDescriptionClassification criteria
  1. a

    Dry month: < 6 cm monthly precipitation.

  2. b

    A = 2.3 T − 0.64 Pw + 41, being T, the mean annual temperature (° C) and Pw, the percentage of annual precipitation occurring in the coolest 6 months.

  3. c

    Dry summer: the driest summer month < 3 cm precipitation and less than one-third of the amount in the wettest winter month.

  4. d

    Dry winter: precipitation in the wettest summer month higher than 10 times that of the driest winter month.

ArTropical humidAll months above 18 °C and less than 3 dry monthsa
AwTropical wet-drySame as Ar but 3 or more dry months
BWDry aridAnnual precipitation P (cm) ≤0.5 Ab
BSDry semiaridAnnual precipitation P (cm) > 0.5 A but smaller or equal than A
CsSubtropical summer-dry8–12 months above 10 °C, annual rainfall < 89 cm and dry summerc
CwSubtropical summer wetSame thermal criteria as Cs, but dry winterd
CrSubtropical humidSame as Cw, with no dry season
DOTemperate oceanic4–7 months above 10 °C and coldest months above 0 °C
DCTemperate continental4–7 months above 10 °C and coldest months below 0 °C
EOSub-arctic oceanicUp to 3 months above 10 °C and temperature of the coldest month above − 10 °C
ECSub-arctic continentalUp to 3 months above 10 °C and temperature of the coldest month ≤− 10 °C
FTTundraAll months < 10 °C
FIIce capAll months below 0 °C

The K-T classification was applied to monthly mean temperature and precipitation data derived from the E-OBS data set (Haylock et al., 2008) of the European Climate Assessment & Dataset (ECA&D) project. In the present study, version 3.0 of E-OBS data, released in April 2010, was used on a 0.25° regular latitude/longitude grid for the period 1971–2000. All but four K-T subtypes (Ar, Aw, Cw and FI) were present in Europe and the Mediterranean area (Figure 1(a)). The climate types covering the largest part of Europe are DO (temperate oceanic) and DC (temperate continental). Subtropical climates (Cs and Cr) can be found mainly south of 45°N (except for the coastal areas of western France), while sub-arctic and polar climates (EO, EC and FT) are located approximately north of 60°N as well as in the Alps, as there is no separate alpine climate in the K-T classification.

Figure 1.

Köppen–Trewartha climate types distribution for (a) the E-OBS data set and (b) the ensemble mean of the RCM simulations. Both for the reference period (1971–2000). The thick black line shows the boundary of the domain DOM_1. The area of the map with colours other than white defines the domain DOM_2

When calculating the K-T climate types, a much localized feature was seen over the north of Romania (not shown). While in the surrounding areas the DC climate type was dominant, in about 50 grid points BW and BS types were detected. It was found that this behaviour was caused by anomalous data in the E-OBS data set for these few grid points. To further investigate this issue, the K-T classification was applied to monthly mean temperature and precipitation data from the CRU data set (on 0.5° and 10′ resolution as well; Mitchell et al., 2004). The above-mentioned feature did not appear in the K-T distribution obtained from the CRU data, and, therefore, it was decided to manually change the climate type of the grid points in question from BW/BS to DC (the climate subtype of these grid points obtained from the CRU data set) in the E-OBS-based K-T distribution, and use this modified data for any further analysis.

3. Climate model simulations

For the following analysis, data from the ENSEMBLES project (Hewitt and Griggs, 2004) were used. The applied RCMs were driven by a variety of GCMs under the SRES-A1B emission scenario (Nakićenović and Swart, 2000). A short overview of these models is given below, followed by a brief description of the experiments. The respective institutions, model names and acronyms are listed in Table 2.

Table 2. Regional climate models (RCMs) from which data have been analysed
AcronymsInstituteModelDriving GCM (see Table III)Source
C4IMet Eireann, IrelandRCA3HadCM3Q16Kjellström et al. (2005)
CNRMMétéo-FranceRM4.5ARPEGERadu et al. (2008)
DMIDanish Meteorological InstituteHIRHAM5ECHAM5-r3, BCM, ARPEGEChristensen et al. (2006)
ETHZSwiss Institute of TechnologyCLMHadCM3Q0Böhm et al. (2006)
ICTPThe Abdus Salam International Centre for Theoretical PhysicsRegCM3ECHAM5-r3Giorgi and Mearns (1999)
KNMIThe Royal Netherlands Meteorological InstituteRACMO2ECHAM5-r3van Meijgaard et al. (2008)
HC-Q0UK Met Office, Hadley CentreHadRM3Q0HadCM3Q0Collins et al. (2011)
HC-Q3 HadRM3Q3HadCM3Q3 
HC-Q16 HadRM3Q16HadCM3Q16 
MPIMax Planck Institute for MeteorologyREMOECHAM5-r3Jacob (2001)
SMHISwedish Meteorological and Hydrological InstituteRCA3.0ECHAM5-r3, BCM, HadCM3Q3Kjellström et al. (2005)

When selecting the RCM simulations to be used, two main issues had to be considered: spatial and time coverage. As each of the RCMs covers a slightly different area, the largest possible common domain (hereafter DOM_1) was defined. Only those points that are inside the common domain were taken into account. As the common domain excludes the north of Scandinavia, a larger additional area was also selected (hereafter DOM_2). In this case, the number of simulations available (11) is less than for the other domain (15). The reason why two different domains (Figure 1) are used is that by using the maximum common domain (DOM_1) in some analysis more RCMs could be included (15 simulations), with the consequent increase in the robustness of the results. The broader domain (DOM_2) allows us to analyse the evolution of climate in a region like northern Fennoscandia, which is very sensitive to climate change. The other criterion was the time period covered. As one of the main goals was to analyse the tendency of the changes in the K-T distribution throughout the 21st century, only those simulations covering the entire 1961–2090 time period were considered; the rest were excluded. The period 1961–2090 was a compromise solution, instead of the longer 1951–2100 interval, as some of the simulations started/finished a few years later/earlier.

The GCMs driving the RCMs simulations are listed in Table 3. The three HadCM3 simulations were based on the same model but with different parameter setting, in order to obtain different climate sensitivities (Murphy et al., 2007).

Table 3. GCMs that drive the simulations of the RCMs
AcronymsInstituteSource
  1. a

    For this GCM, three different versions (Q0, Q3 and Q16, see Table 2) were run with differing climate sensitivities.

HadCM3aUK Met Office, Hadley CentreGordon et al. (2000)
ARPEGEMétéo-FranceGibelin and Déqué (2003)
ECHAM5-r3Max Planck Institute for MeteorologyRoeckner et al. (2003)
BCMUniversity of Bergen, NorwayFurevik et al. (2003)

The K-T climate classification was applied to the above-mentioned climate simulations, both for the present day climate (1971–2000) and for the future over 30 year periods overlapped for 20 years. The K-T climate types were calculated for both domains (DOM_1 and DOM_2). All the tables presented in the article are based on the results of the common domain (DOM_1 with a 15 member ensemble), while the figures are composites of the results obtained for the two areas (except for Figures 4 and 2, which are based on the results of DOM_1).

Figure 2.

Transfers between the different K-T climate subtypes. The + or − signs (in parentheses) indicate a net gain or loss of area, respectively. The numbers on the arrows give the net area (103 km2) that undergoes the given transition. First number is for the period 2021–2050, while the second number is for 2061–2090

The modelled K-T distribution for the reference period was compared over land points to the one derived from the E-OBS data to check how the RCMs simulate present day climate. On the other hand results for the future were compared to that of the reference period to detect possible future changes in the climate distribution over Europe and the Mediterranean area. Results of this study are presented in Section '4. Results'.

4. Results

4.1. Evaluation of the RCM runs for the reference period

The monthly values of 2 m temperature and precipitation were averaged over the control period (1971–2000) and over the whole set of simulations. Then, the K-T climate subtypes corresponding to these mean fields were determined on each grid point (Figure 1(b)). These fields of average K-T subtypes are the ensemble mean of reference (EMR). K-T subtypes were also calculated for the observational database E-OBS for the same period (Figure 1(a)).

A grid to grid comparison of the individual simulations and the EMR with the E-OBS database has been done through the development of co-occurrence matrices. These matrices show the correspondences between K-T subtypes of the E-OBS climatology and every simulation (not shown) or the EMR (Table 4) for the period 1971–2000. A total of 15 847 grid points with data provided by E-OBS were analysed in the DOM_1 domain. For a better interpretation of the co-occurrence matrices the following points should be remembered:

  • The main diagonal of each matrix indicates the number of grid points where the K-T subtype according to the E-OBS database matches that generated from a simulation or the EMR.

  • A location outside the main diagonal of the matrix indicates lack of coincidence. A larger separation from the diagonal indicates larger differences between the climatology and the simulations.

  • Nonzero elements below the main diagonal indicate that the simulations are warmer or drier than E-OBS.

  • Conversely, nonzero elements above the main diagonal indicate that the simulations are cooler or wetter than E-OBS.

Table 4. Co-occurrence matrix between EMR and E-OBS for DOM_1 domain
  Ensemble mean (1971–2000)
  BWBSCsCrDODCEOECFT
  1. Columns contain the number of grid points of the land domain that correspond to each of the K-T subtypes according to the EMR. Rows contain the number of grid points for each K-T subtype following the climatology of the E-OBS database.

E-OBS (1971–2000)BW1321000000
 BS84695910620000
 Cs31097131272120000
 Cr01879630000
 DO02112234871944600
 DC00016475655251870
 EO000008878279107
 EC000000212038190
 FT00000060302

RCM simulations reproduce the K-T subtypes of E-OBS quite well in most of the cases. The percentage of coincidences of subtypes ranges from 55.4 to 81.3% and is over 70% in 10 of the 15 RCM simulations analysed (Table 5). The EMR has 83.5% of the grids in the main diagonal of the matrix of co-occurrence, more than any of the simulations used for its formation. The EMR has 10.7% of the points above the main diagonal and 5.8% below; this means that it has a slight cold bias. These figures are in line with those reported by Castro et al. (2007) and indicate that the EMR performs better than all the models that constitute it. The explanation for this result is not obvious and would need further research to clarify it.

Table 5. Percentage of the total studied grid points located on the main diagonal of the co-occurrence matrices for each analysed GCM-RCM pair for the domain DOM_1
 GCM
RCMECHAM5HadQ0HadQ3HadQ16BCMARPEGE
C4I   75.75  
CNRM     79.98
KNMI80.41     
SMHI80.29 68.26 69.38 
MPI73.61     
ETHZ 80.58    
HC 81.3274.8879.72  
DMI65.32   55.4172.52
ICTP69     

A large part of the differences in K-T subtypes between EMR and E-OBS is due to a significant number of grids that belong to the DC subtype for E-OBS and correspond to the subtype DO for the EMR (Table 4). This result was already observed by Castro et al. (2007), but this time the border between DC and DO from north to south across Central Europe is fairly well outlined by the RCMs (Figure 1) and differences focus on the eastern side of the Danube basin and the Crimean Peninsula. This tendency of the models to establish as DO some grids that E-OBS classify as DC is the most striking, but the relative error of other simulated subtypes is larger. For instance, the number of grid points with the subtype FT for the EMR is 94.5% more abundant than in E-OBS. This increase comes at the expense of the decrease in grid points with EC and EO subtypes in the mountainous areas of Norway and Sweden and the Kola Peninsula in Russia. This is caused by the tendency of models to produce cooler summers in this area which is the key factor to distinguish between types E and F of K-T. It is also noteworthy that EMR shows 32% less grid points for the subtype Cs, owing to the appearance of DO subtype in the central area of the Iberian Peninsula and Cr and BS subtypes in different Mediterranean coastal areas. This assignment of DO instead of Cs is due to mismatches in the simulation of the temperature in the transitional seasons, while Cr and BS cases are attributable to differences in precipitation.

4.2. Climate change scenario simulations

The K-T climate subtypes were calculated for consecutive and overlapping 30 year periods between 1961 and 2090 (1961–1990, 1971–2000, 1981–2010 and so on to 2061–2090) for the individual RCM simulations and for the ensemble mean. Calculations were performed using the monthly averages of 2 m temperature and precipitation over land points only. Because of the large number of models, only the results obtained with the ensemble mean are shown in this part of the analysis, for two intervals: 2021–2050 (Figure 3(a)) and 2061–2090 (Figure 3(b)). In order to facilitate the detection of areas where the climate simulations project changes with respect to the reference period (1971–2000), in Figure 4 only these grid points are plotted. Comparing these figures with that of the reference period (Figure 1(b)) the following important changes can be observed:

  • Northeastward shift of the boundary between DO and DC climate types. By the end of the 21st century the DC type withdraws drastically to the northeast. The DO type is extending in eastern Europe, but at the same time is loosing territory in western Europe.

  • Climate types EO, EC and FT in Fennoscandia withdraw drastically to the north. By 2021–2050 the area covered by EC and FT types is significantly reduced, and by 2061–2090 the EC and FT types almost completely disappear from northern Europe.

  • Cs and Cr gain more area in southern and western Europe, especially by the end of the 21st century.

  • The BS type gains area in southeast Spain, Italy, Greece, Turkey and the coastal zones of northern Africa.

Other figures similar to Figure 4, but for 1981–2010 and 1991–2020 (not shown), reveal that the first clear climate changes projected are the retreat of the EC climate in Finland, Sweden and western Russia and the DC subtype in central Europe. These projections fit reasonably well with observed climate transitions from the period 1950–1978 to the period 1979–2006 (Jylhä et al., 2010). In the following two periods (2001–2030 and 2011–2040) the most striking projected transitions occur to the east of the Kola Peninsula in Russia (FT-EC), western France (DO-Cr), in southern Finland (EO-DC) and north of Black Sea (DC-DO).

Figure 3.

Köppen–Trewartha climate type distribution for the ensemble mean of the RCM simulations for the period (a) 2021–2050 and (b) 2061–2090. The thick black line shows the domain DOM_1. The area with colours other than white defines the domain DOM_2

Figure 4.

Projected climate type transitions for the ensemble mean during (a) the 2021–2050 period and (b) 2061–2090 period. Both figures with respect to the reference period (1971–2000)

For a deeper analysis, Tables 6 and 7 contain the co-occurrence matrices (domain DOM_1) for the ensemble mean of the climate scenario runs. These matrices express the point-by-point agreement between the climate types of the reference period (1971–2000) and the periods 2021–2050 (Table 6) and 2061–2090 (Table 7).

Table 6. Co-occurence matrix for the ensemble mean of the RCM simulations for the time interval 2021–2050 (domain DOM_1) with respect to the reference period (1971–2000)
  Ensemble mean (2021–2050)
  BWBSCsCrDODCEOECFT
  1. Areas in 103 km2.

Ensemble mean (1971–2000)BW1170.900000000
 BS80.4276.90000000
 Cs077.0792.9000000
 Cr013.246.0204.000000
 DO029.0151.4222.52295.90000
 DC0000785.72119.3000
 EO000031.6200.4304.300
 EC00000117.1230.2457.40
 FT00000065.244.4122.8
Table 7. Co-occurence matrix for the ensemble mean of the RCM simulations for the time interval 2061–2090 (domain DOM_1) with respect to the reference period (1971–2000)
  Ensemble mean (2061–2090)
  BWBSCsCrDODCEOECFT
  1. Areas in 103 km2.

Ensemble mean (1971–2000)BW1170.900000000
 BS202.2155.00000000
 Cs0215.9654.0000000
 Cr030.478.3154.500000
 DO078.8376.9617.11625.90000
 DC00.6001478.81425.6000
 EO000085.2327.4123.700
 EC00000353.1424.826.70
 FT00000.32.1146.428.754.8

The portion of land area shifting from the current K-T climate type to a warmer or drier one is 21.3% for 2021–2050 and 45.2% for 2061–2090. This means that by the end of the 21st century almost half of the studied area will undergo a climate type change. It is important to notice that there are no entries above the main diagonal.

Some unexpected differences between the values in Tables 6 and 7 and in Table 4 arise because the E-OBS data do not cover the entire area simulated by the RCM. This is especially evident for BW subtype, as E-OBS does not provide data in 1 678 grid points in North Africa where such a climate is simulated by the models.

The largest changes can be observed in the current climate types DO, DC and EC.

  • The DO type is gaining territory mainly from DC, and to a much smaller extent from EO. On the other hand, DO is loosing area to BS, Cs and Cr types, among which DO-Cr is the most dominant shift, followed by DO-Cs.

  • The net results of the changes in the DC and EC covered areas are both negative. The DC subtype is loosing territory to DO, but also gaining (though considerable less) from EO and EC. The EC subtype gains some area from FT, but looses to DC and EO.

It is also important to mention that the BW and BS subtypes are both predicted to increase their area with respect to the control period. The BW subtype (dry arid or desert) appears over a much localized area in the southeast of Spain, while the BS subtype (dry semiarid or steppe) gains more area in Spain, Italy, Greece, Turkey and northern Africa.

Figure 2 shows a schematic plot of these projected transfers between the different K-T subtypes, also indicating the sign of the net change of area of the given subtype. It can be observed that FT, EC and DC are projected to undergo a net loss of area, while the net result of the changes in the remaining subtypes (EO, DO, Cr, Cs, BS and BW) is projected to be positive. This indicates that the projection points to a decrease in the diversity of climates in Europe, with potential consequences on natural ecosystems and crops.

It is also interesting to see how many grid points undergo a change from one main climate group (A, B, C, D, E and F) to another. Compared to the reference period of 1971–2000, 9.7% of the total area experiences such a change by 2021–2050. By 2061–2090 the area affected by this change increases to 23.0%. The grid points with main climate group change (i.e. the areas with potentially more dangerous ecosystem changes) can be found in western France, Alps, Scotland, Fennoscandia, northwest of Russia, Iceland, several regions of southern Europe and coastal areas of Wales, Ireland and southern England.

Figure 5 shows the evolution of K-T climates over time. It can be seen that, in general, the climates DC, EC and FT, that is three of the four colder climates, lose area progressively. BW, BS, Cs, Cr and DO (the warmest climates in Europe) increase their area. EO subtype has a special behaviour, as EO shows an area gain for the ensemble in the second half of the 21st century which is bigger than in any RCM. This is because the threshold criterion between EO and EC refers to a temperature of the coldest month above or below − 10 °C (see Table 1). As the minimum temperatures in some grid points do not coincide in the same month for all the RCM simulations, the ensemble could have its coldest month above − 10 °C although most RCMs do not. Therefore, in the border areas between these two subtypes there are some grid points in the ensemble that are EO, while in most simulations they are EC. As a result, at the end of studied period, EC climate in the ensemble has decreased more than in any of the RCMs.

Figure 5.

Evolution of K-T climatic subtypes through the increase/decrease over the period 1971–2000 of the area occupied by each subtype climate. In the x-axis the 11 time periods analysed are represented, while the y-axis shows the variation of the occupied area in units of 103 km2. The black line represents the evolution of the ensemble, while the grey band represents the spread of the area covered by the models. For obtaining this band the two models with lowest values and the two models with highest values were discarded. The 11 time periods are: (1) 1961–1990, (2) 1971–2000, (3) 1981–2010, (4) 1991–2020, (5) 2001–2030, (6) 2011–2040, (7) 2021–2050, (8) 2031–2060, (9) 2041–2070, (10) 2051–2080 and (11) 2061–2090

It is also remarkable that for the subtype Cr the ensemble shows more increases than most of the RCMs (Figure 5). This is mainly because, owing to the criteria that define these climates, in the peripheral regions of subtype Cs one or a few simulations that are excessively wet in summer or dry in winter can lead to an ensemble mean with subtype Cr, even while a majority of simulations show a Cs climate.

The two climates with the largest areas (DO and DC) are also those with the widest band in Figure 5 and therefore with the highest degree of spread in square kilometers between the different models. However, if we relate this spread in the period 2061–2090 to the extent of these climates in the EMR, it represents only 38% for DO and 22% for DC. The Cr subtype also has a large spread in square kilometers despite not occupying a large area; this implies a relative spread of 192%. The difference in square kilometers is moderate for intermediate extension climates (EC, EO and Cs) and small for the rest of the climates that are not widespread (FT, BS and BW). The relative difference for the subtype EO stands out from the rest of the last six subtypes as it reaches 72%, while for the others it is always below 40%. For all subtypes, the largest spread occurs in the last periods.

Looking at the temporal evolution of the ensemble (Figure 5) six out of nine existing subtypes show a roughly constant growth (BW, BS and Cs) or reduction (DC, EC and FT). On the other hand, Cr has a tendency to accelerate the increase in expanse. The general trend of the ensemble to a monotonous evolution is not always found in all simulations. Examples of this can be easily seen (Figure 5) in the graphs of the subtypes DO, DC and EO, but also exist for other climates.

Another important aspect is the rate at which the different subtypes of climate expand or contract. The two coldest subtypes in the studied domain (FT and EC) decrease their joint area, on average and in net value, at a rate slightly over 100 × 103 km2 decade−1. Subtropical climates (Cs and Cr) will expand with an average net rate of over 80 × 103 km2 decade−1. Dry climates (BW and BS) will increase its area at an average net rate of approximately 36 × 103 km2 decade−1. Among the two temperate and more widespread climates, the subtype DO will expand at a rate of about 55 × 103 km2 decade−1, while the DC will shrink at a rate of about 90 × 103 km2 decade−1. These rates of change in climate could be large enough that some types of vegetation and ecosystems would be unable to adapt or migrate to survive.

5. Concluding remarks

The projected climate change has been analysed for Europe and northern Africa by applying the K-T climate classification to a set of 15 RCM simulations nested in six GCMs for the A1B emissions scenario. The ensemble of simulations reproduces well the observed climate in the late 20th century (83.5% of matches with the observations) and better than any of the simulations used in its formation. The most significant climate spatial transitions observed in this period in Fennoscandia and Central Europe are simulated with good approximation.

The ensemble projects for the 21st century a climate type change for large areas of the domain. Taking as reference the subtypes of the period 1971–2000, 21.3% of the total area changes its K-T climate by 2021–2050 and 45.2% by 2061–2090. In addition, rates of net change are steadily high throughout the 21st century, ranging from a decrease of 90 × 103 km2 decade−1 of the area covered by DC climate, to an increase of 55 × 103 km2 decade−1 for Cr and DO climates. Climate-type changes are likely to occur at a fairly sustained rate throughout the 21st century, which could pose a threat to many ecosystems starting from the first decades of the century.

The largest changes from the point of view of spatial displacement of climates could occur in Central Europe, where the border between subtypes DO and DC is shifted far to the northeast; in Fennoscandia, where EC and FT climates almost disappear; in southern and western Europe, where Cs and Cr subtropical climates significantly advance; and in the Mediterranean basin, where BW and BS dry climates also extend. Meanwhile, the areas where changes are more intense, and potentially more dangerous, are western France, Alps, Scotland, Fennoscandia, northwest of Russia, Iceland, several regions of southern Europe and coastal areas of Wales, Ireland and southern England.

Acknowledgements

The ENSEMBLES model data used in this work was funded by the EU FP6 Integrated Project ENSEMBLES (Contract GOCE-CT-2003-505539) whose support is gratefully acknowledged. We also acknowledge the E-OBS data set from the EU-FP6 project ENSEMBLES (http://ensembles-eu.metoffice.com) and the data providers in the ECA&D project (http://eca.knmi.nl). The high-resolution climate data set available through the Climatic Research Unit and the Tyndall Centre was also very useful to complete this work.

Ancillary