OBSERVATIONS

Climate analysis in the Arctic is seriously hampered by deficiencies in the data sets. The observational base is quite limited with few long-term stations and a paucity of observations in general. However, both in the northern part of Fennoscandia, in Greenland, Iceland and the Faeroe Islands systematic weather observation were initiated at a few sites before 1900. In the Svalbard region the first permanent weather observations started in 1911, and around 1920 weather stations were established at Bjørnøya and Jan Mayen. In a joint Nordic effort, a consistent high-quality monthly data set (the NARP-dataset) for a selection of stations in the Nordic Arctic was established (Førland et al., 2002) covering the period 1890–2002 ( http://projects.met.no/∼narp). The data series were quality-checked and homogenized as far as possible at the national meteorological institutes.

When the World Meteorological Organisation (WMO) in 1935 agreed to calculate reference values (“normals”), it was decided to operate with averaging periods of 30 yr. The climatological standard normals are thus defined as “averages of climatological data computed for consecutive periods of 30 yr as follows: 1901–1930, 1931–1960, 1961–1990, etc” (WMO, 1989). As these climate normal values often are used as references for global, regional and local climate conditions, the present analyses are mainly based on mean values for the specific 30-yr periods. To illustrate the current conditions, mean values for two recent time periods (1976–2000 and 1990–2002) are also included.

DOWNSCALED SCENARIOS

The temperature scenarios in this study are based on the ECHAM4/OPYC3 AOGCM with the transient GSDIO integration (Roeckner et al., 1999), which includes greenhouse gases, tropospheric ozone, and direct as well as indirect sulphur aerosol forcing. In this integration, the concentrations of greenhouse gases have been specified according to the IPCC IS92a scenario. Compared to IPCCs new set of emission scenarios (SRES) (IPCC, 2001: 532), the projected increase in the global mean temperature up to 2050 for the IS92a scenario is similar to SRES B1, and lower than for the other SRES scenarios.

The present study focuses on empirically downscaled monthly mean temperature. Empirical downscaling consists of revealing empirical links between large-scale patterns of climate elements (e.g., air pressure, sea surface temperature, etc.) and local climate elements (e.g., temperature, precipitation, etc.), and applying them on output from global or regional models (Benestad, 2001). Hanssen-Bauer et al. (2003) showed that empirical downscaling has the potential for describing spatial climate features that are not resolved by the currently available regional climate models.

The empirical downscaling was based on an approach utilising common EOFs as described in Benestad (2001, 2002), using multiple regression for calibrating the empirical models. The predictor consisted of gridded 2-m temperature fields from the NCEP reanalysis (Kalnay et al., 1996) and the ECHAM4/OPYC3 GSDIO results. In order to carry out the common EOF analysis, the AOGCM results were interpolated onto the same grid as the NCEP reanalysis. The predictors for deriving the local climatic series were taken from three domains (Fig. 2): Greenland: Domain I (90°W30°W–52°N80°N); Fennoscandia and Iceland: Domain II (40°W40°E–52°N80°N); Svalbard (Bjørnøya and Svalbard Airport): Domain III (35°W40°E–67°N85°N). The downscaled results for Svalbard were sensitive to the choice of predictor domain, and are therefore subject to high degree of uncertainty. Part of this uncertainty may be related to the sparse observational network in the Arctic, affecting the quality of the predictor fields in this region.

The downscaled results may be sensitive to the choice of domain (Benestad, 2002), and it is therefore important to find an appropriate region that yields realistic results and has a solid physical justification. High R² values from the multiple regressions are a minimum criterion for an appropriate predictor domain. The downscaling was applied to the 12 calendar months separately for each location.

It is important to keep in mind the fact that one climate scenario represents one plausible description of a future climate, and should not be interpreted as a “forecast”. Benestad (2003) analyzed climate model reproductions of past temperature trends for 1880–1999 and concluded that one single model cannot provide a reliable description of these trends due to the presence of strong chaotic variability. On the other hand, different AOGCMs give different accounts of internal variability, and multimodel ensembles are able to span the past historic evolution. Some members of multimodel ensembles may give a less realistic description of the local climatic conditions than others, such as the HadCM3 and the NCAR-CSM models producing too cold climates in the Barents Sea or around Iceland (Benestad et al., 2002a). Hence, the use of multimodel ensembles is not straight forward, since some climate models are less realistic than others (Giorgi and Mearns, 2002).

ESTIMATION OF DAILY TEMPERATURES

Estimates of the length and degree-day-sum of growing, heating and freezing seasons are usually based on daily mean temperatures. For the Nordic Arctic, very few digitized long-term series of daily temperatures are available. On the other hand, the NARP-dataset contains monthly records back to 1890. In the present study, daily mean temperatures were interpolated from the mean monthly temperatures by fitting a spline curve through the 12 monthly mean temperatures (Press et al., 1992). A constraint was added to the spline equation to ensure that the deviation between the original mean monthly temperature, and the mean monthly temperature based on the estimated daily values did not exceed a tolerance criterion of 0.001°C. The amplitude of the spline curve was adjusted by shifting the positions of the monthly mean (default in the middle of the month). This was done iteratively until the tolerance criterion was fulfilled. The technique provides a simple, fast and robust method that can be applied everywhere where only mean monthly temperatures are available. Figure 3 shows examples of smoothed and averaged daily temperatures for Karasjok and Bjørnøya for the period 1961–1990.

AIR TEMPERATURE

The mean annual temperature in the Nordic Arctic has undergone large variations during the 20th century (Førland et al., 2002). Because of substantial differences in standard deviations, the variation in annual mean temperatures is more affected by the variation in winter temperatures than by summer temperatures. For most of the stations in the region there is a positive trend up to the late 1930s, a negative trend from the 1930s to the 1960s, and from the late 1960s the temperature has increased at all stations except for Nuuk. For all stations, the warmest two decades on annual basis were the 1930s and the 1950s. Despite the significant increase in global (Folland et al., 2001) and Fennoscandian (Tuomenvirta et al., 2000) temperatures during the 20th century, the trends in annual mean temperature in the Nordic Arctic are not statistically significant at any station (Førland et al., 2002). This is partly due to the large year-to-year variations, and the consequent low “signal-to-noise” ratio. Even though climate models indicate substantial greenhouse induced warming in the Arctic, the low “signal-to-noise” ratio implies that it is not evident that the first significant “greenhouse signal” will be found in this region.

Table 1 shows that the 1961–1990 normal annual temperature at the Nordic arctic stations varies between −7.3°C at Upernavik to +6.5°C at Torshavn. The temperature during the normal period 1901–1930 was lower than the present normals at all stations, except for Karasjok and Karesuando. For the western Greenland stations Upernavik and Illullisat the normal values were around 0.5°C lower than the present normals. The 1961–1990 normal period was colder than the 1931–1960 normal period at all arctic stations, with the largest deviation (>1°C) at Tasilaq, Svalbard Airport and Jan Mayen. The recent decades (1976–2000 and 1990–2002) have been warmer than the 1961–1990 normals at all stations, except for stations in western Greenland. At Svalbard Airport and Jan Mayen the mean annual temperature during 1990–2002 was 1.3 to 1.5°C higher than the 1961–1990 normals.

The downscaled scenarios for 2021–2050 suggest further warming over the entire region, with mean annual temperatures 3 to 5°C higher than the present normals in the Svalbard region and large parts of northern Fennoscandia. A more modest temperature increase (<2°C) is projected for the stations in Iceland, Greenland, and the Faeroes. The low temperature increase in this area, and particularly in southeastern Greenland, is consistent with the large-scale temperature pattern from several AOGCMs (e.g., ECHAM4, HadCM2, HadCM3 [Raïsänen, 2001]).

Smoothed daily temperatures for four locations are shown in Figure 4. For single months the downscaled scenarios may give unstable results, as, e.g., for Karasjok in December. For Karasjok the period 1961–1990 has lower winter temperatures than the other time-slices, while the scenario-period 2021–2050 has the highest values for most of the year. Also for Svalbard Airport the scenario indicates a warming during most of the year. For Torshavn a small warming is projected for parts of the year, while the projected temperatures for Tasilaq are similar to what has been experienced in the 20th century.

GROWING SEASON

Plant growth is in general influenced by air temperature, and is especially vulnerable during spring (Menzel, 2002). The air temperature is found to be a limiting factor for growth potential, thus the growing season is rather short at high latitudes. Different species respond differently to air temperature, some are sensitive to lower temperatures while others are more resistant to cold climate. It should however be emphasized that plant growth also depends on additional factors, both climatological (precipitation, snow cover, radiation) as well as soil, moisture, exposure, etc. Since the early 1960s, the length of the growing season has increased through advanced onset during spring in mid latitudes (Menzel and Fabian, 1999, Chmielewski and Rötzer, 2002). Groisman et al. (2003) found that during the past 50 yr the largest absolute and relative changes in duration of the growing season in the Arctic occurred over Alaska and western Canada (increase of 15 and 10 d, or 19 and 8% per 50 yr, respectively), but a significant increase of the growing season duration was also found for Russia (from 7 in the east to 10 d in the west, or 8% per 50 yr). On average over the entire land area north of 50°N, an increase of 6% per 50 yr was found in the length of the growing season (Groisman et al., 2003).

Based on a normalized vegetation index derived from satellite images, Høgda et al. (2001) found that the growing season had increased in large parts of Fennoscandia between 1982 and 1998, particularly in Denmark, southern Sweden, and in southernmost parts of Norway. For northern continental parts of Fennoscandia the growing season had been stable, or slightly shortened.

Different definitions of the thermal growing season exist (Brinkmann, 1979; Carter, 1998). The number of days with daily mean air temperatures (2 m) above a given threshold temperature is often used. Carter (1998) argues that the season for active plant development and growth in the Nordic countries should be defined as the period during which the mean daily air temperatures remain above 5°C, and that this threshold should also be adopted as a base temperature for computing the growing degree-days sum (GDD). The GDD provides a measure of the intensity of the growing season.

The thermal growing season is in this study defined as the period of the year when the smoothed daily mean temperature (T_i) is above 5°C, while the growing-degree-days (GDD) are the accumulated degree sum above the threshold temperature T̂ = 5°C:

By the above definitions, the thermal growing season in the Nordic arctic (Table 2) in the present normal period starts between 17 April at Torshavn and 2 August at Jan Mayen, and ends between 16 August (Upernavik and Svalbard Airport) and 9 November (Torshavn). The length of the growing season varies between zero days at Bjørnøya (cf. Fig. 3) to 207 d at Torshavn. In Copenhagen on the other hand, the thermal growing season for the normal period lasts for 322 d; from 4 April to 18 November.

During 1901–1930 the growing season was shorter than during 1961–1990 at most stations, while it during 1931–1960 was substantially longer at the Icelandic stations and at the Greenlandic stations Nuuk and Tasilaq. During the recent decades (1976–2000 and 1990–2002), the results indicate a substantial increase in the thermal growing season at Jan Mayen and Bjørnøya, but the low GDD-values imply that the thermal conditions for plant growth are still marginal.

The downscaled scenarios for 2021–2050 project a further increase in the length of the growing season at all stations, except for eastern Greenland (Tasilaq). For most of the locations in Table 2, the length of the growing season is projected to increase by more than 3 wk.

The degree-day-sum (GDD) during the growing season varied from zero at Bjørnøya, to nearly 800 at Jokkmokk. At most stations the GDD was lower during 1901–1930 and higher during 1931–1960 than for the present normal period. The GDD values for the latest decades are generally higher than for the normal period 1961–1990. However, at the Faeroes and some stations at Iceland and Greenland, the 1931–1960 values were higher than during the most recent decades. The projected values for 2021–2050 indicate a substantial increase in GDD-values at all stations, with the largest absolute increase at the Icelandic stations. Carter (1998) studied the projected change in growing season up to 2050 at nine locations in Fennoscandia, including Stykkisholmur, Helsinki, and Stockholm, and found estimates quite similar to the values in the present study.

HEATING SEASON

The heating season is the period of the year when buildings need to be heated. The sums of heating degree-days closely correlate to energy consumption for heating, and have numerous other practical implications (Quayle and Diaz, 1980; Guttman and Lehman, 1992). The amount of energy for heating of buildings is also depending on other climatological factors (wind speed, radiation), as well as factors related to demographic changes, living standards, and building instructions (e.g., volume of heated buildings, preferred indoor temperatures, thermal insulation, etc.) (Venäläinen et al., 2004). The heating season is in the present study defined as the period of the year when the smoothed daily mean temperature is below a threshold T̂, while heating degree-days (HDD) are the sum of the difference between a base temperature T_base and the daily mean temperature T_i (Taylor, 1981):

Groisman et al (2003) found a statistically significant decrease in annual heating degree-days during the past 50 yr of 6% over the entire Arctic, with a maximum absolute and relative reduction in heating degree-days over western Canada and Alaska (9 and 8% per 50 yr, respectively). For Eurasia, significant reductions were found for Russia (6–7% per 50 yr), indicating that there have been reduced heating costs in relative terms.

Table 3 indicates that the heating season lasts the whole year through at most of Greenland and in the Svalbard region (cf. Fig. 3). This is true for all normal periods, for the latest decades, and for most of the stations in this region it is also true for the scenario period 2021–2050. In Copenhagen on the other hand, the heating season in the present normal period lasts for 204 d, and is projected to decrease by 26 d up to 2021–2050.

Generally the length of the heating season and sum of heating-degree-days (HDD) were higher for 1901–30 and lower for 1931–60 than for the present normal period. During the recent decades the length and HDD–sum have been lower than for the normal period 1961–90. The highest HDD-sum (>8500 degrees) for the normal period 1961–90 are found for Upernavik and Svalbard Airport, while it in e.g. Oslo, Helsinki and Stockholm is ca. 3500–4000, and in Copenhagen ca 2700 degrees. The projected values for the scenario period 2021–2050 indicate a reduction of heating-degrees of 10 to 20% at most of the stations in Iceland and Fennoscandia.

FREEZING SEASON

The length of the frost-free period is among the most carefully monitored variables in the Arctic (Groisman et al., 2003). To characterize the severity of the cold season, the sum of negative temperatures is often used. The thermal freezing season is in this study defined as the period of the year when the smoothed daily mean temperature (T_i) is below 0°C, while the freezing-degree-days (FDD) are the accumulated degree sum below a threshold temperature T̂ = 0°C:

Time series of FDD indicate that the “severity” of the cold season has substantially decreased everywhere in the Arctic, except eastern Canada (Groisman et al., 2003). In terms of absolute values of FDD, the mean circumpolar decrease is 13% per 50 yr (Groisman et al., 2003).

The freezing season in the Nordic Arctic (Table 4) starts between 16 September (Svalbard Airport) and 10 December (Teigarhorn), and ends between 28 January (Teigarhorn) and 4 June (Upernavik and Svalbard Airport). In Torshavn there is no “freezing period” as the smoothed mean daily temperature is above zero for the whole year round (cf. Fig. 4). The length of the freezing period varies from zero days at Torshavn to 260 d at Svalbard Airport and Upernavik. For all Icelandic stations, the freezing period was shorter and the degree-day sums lower during 1931–1960 than for the latest decades.

The downscaled scenarios indicate a reduction of the freezing period at all stations except the eastern Greenlandic station Tasilaq. For several stations the scenarios imply a freezing season more than 50 days shorter than during 1961–1990. At Reykjavik and Stykkisholmur, there will be no “freezing period” during 2021–2050 according to the definition used in this study.