Here, we analyse Tarfala's continuous mean daily temperature record that dates back to 1 January 1965. Data prior to 1965 are restricted to the summer months (Grudd & Schneider 1996) and not considered here. From 1965 to 1989, air temperatures were measured with Pt100 platinum resistance thermometers (Vaisala, Vantaa, Finland) placed in a Stevenson-type radiation screen and recorded every minute by chart recorders. In 1988, an automatic weather station was installed in immediate vicinity to the Stevenson screen including an MP100 temperature/humidity sensor (Rotronic, Bassersdorf, Switzerland) placed into a multi-plate radiation shield (Young, Traverse City, MI, USA). Both systems were operated in parallel until 1998 to allow direct comparison. The new system sampled data every 10 s, and hourly (May–September) or three-hourly means (October–April), and daily means were stored on a Campbell Scientific Datalogger (Logan, UT, USA). Hourly means all year around became available in 1995. Since 1 July 1989, the data from the automatic station have been used in the Tarfala temperature series. After 1990, several additional Pt100 temperature sensors have been added at the same site for comparison and as backup for instrument failures. More details about the Tarfala weather station, including an analysis of the data prior to 1965, are given by Grudd & Schneider (1996).

In July 1995, the Swedish Meteorological and Hydrological Institute installed an automatic weather station approximately 20 m east of the Tarfala weather station as part of their national network. These data and those from the additional sensors at the Tarfala station were used to fill occasional gaps in the data series based on regression relationships between the data of these sensors and Tarfala's Rotronic sensor. When simultaneous measurements were not available, gaps (19 days between 1985 and 1994) were filled using downscaled ERA-40 (produced by the European Centre for Medium-Range Weather Forecasts) data of the grid cell containing Tarfala (Radić & Hock 2006). A month-long gap in January 2009 was filled by a monthly mean temperature derived from regression analysis of historic January temperatures between the Swedish Meteorological and Hydrological Institute for station Katterjokk (515 m a.s.l., approximately 60 km NNW of Tarfala) and Tarfala. Temperature data from this period are omitted in the analysis of sub-monthly timescales.

The daily mean air temperature series from Tarfala 1965–2011 used in this study is available through Stockholm University's Bolin Centre database (http://www.bolin.su.se/data/).

In March 2000, a shallow (15 m) and a deep borehole (100 m) were drilled at Tarfalaryggen at an elevation of 1540 m a.s.l., approximately 1.6 km north-east of Tarfala. The boreholes are part of a series of boreholes drilled and instrumented between 1998 and 2000 along a north–south transect from the Mediterranean to Svalbard as part of the Permafrost and Climate in Europe (PACE) project (under the umbrella of the European Union's Fourth Framework Programme), aiming at long-term monitoring of permafrost (Harris et al. 2001, 2003).

Bedrock at the drill site consists of massive amphibolite, which is overlain by an unconsolidated regolith of roughly 4 m. The surface is blocky and largely lacks vegetation (Isaksen et al. 2007). Permafrost thickness is estimated at 350 m±19 m (Isaksen et al. 2001). The depth of seasonal temperature variations defined as the depth where seasonal amplitudes are ≤0.1°C was approximately 20 m below the surface (Isaksen et al. 2007). During winter, the area is usually covered by a thin snow cover (<0.3 m).

The boreholes were instrumented with negative temperature coefficient YSI 44006 thermistors (Yellow Spring Instruments, Yellow Springs, OH, USA; Von der Muehll & Holub 1992), with an estimate absolute accuracy of ±0.05°C and a relative accuracy of ±0.02°C. The thermistor chain in the 100 m hole included 30 thermistors spaced apart with increasing distance between sensors with depth ranging from 0.2 m in the first metre and 10 m beyond 30 m below the surface, and denser spacing again beyond 90 m below the surface. The 15 m hole contained 17 thermistors with the same spacing as the 100 m hole and served as quality control. All thermistors recorded the temperature once every 24 h. More frequent measurements (every 6 h) are recorded at all thermistors in the shallow hole and the first 11 thermistors of the deep borehole (down to 5 m below the surface). More information about the borehole and its instrumentation has been presented by Sollid et al. (2000), Isaksen et al. (2001) and Isaksen et al. (2007). Due to disturbance from the drilling activities, temperatures at shallow depths initially showed a significant cooling and therefore only borehole data from 2001 onwards, when the temperature signal had stabilized, are used here.

At the borehole site, air temperature is measured at 2 m above the ground using a Vaisala HMP45D temperature–humidity sensor mounted in a solar radiation shield. Prior to installation, the sensor was installed at the Tarfala weather station for several weeks for calibration.

Part of the borehole data has been presented and discussed in previous publications (e.g., Harris et al. 2003; Isaksen et al. 2007; Christiansen et al. 2010). However, many of the previous analyses were limited to a short time period, hampering trend detection. Here, we present updates on ground temperatures and their trends based on the available time series 2001–11.

Snow and ice melt models often use temperature lapse rates, that is, linear temperature decreases with increasing elevation to extrapolate point temperature measurements to the domain of interest (e.g., Hock & Holmgren 2005; Gardner et al. 2009). Here, we use the mean daily temperature measurements at Tarfala and at the PACE borehole site, located 410 m higher than Tarfala (Fig. 1) to compute temperature lapse rates and to investigate their seasonal variation. We emphasize that our lapse rates are near-surface lapse rates referring to the temperature changes along the terrain slopes rather than free-air adiabatic lapse rates. Lapse rates are defined positive if the temperature decreases with increasing elevation.

The air temperature series at the PACE site had substantial gaps due to instrument failure. We did not fill the gaps but only used the existing data in the subsequent analysis. The available mean daily data generally correlate well with Tarfala's air temperatures (Fig. 6). As expected from their higher altitude, the borehole site temperatures are systematically lower. Using all available mean daily data, we find an average lapse rate of 4.5± 4.6°C km⁻¹ (±1σ). Mean monthly lapse rates (based on all available daily lapse rates in each month) vary between roughly 2 and 6°C km⁻¹ (Fig. 7). Lapse rates tend to be higher from May to October. The scatter is large in all months, though smaller for the summer months. Higher summer time lapse rates are consistent with the findings from Radić & Hock (2006), who derived lapse rates from Tarfala's temperature data and ERA-40 reanalysis data. Their values are somewhat higher (up to 8°C km⁻¹ in summer); however, their lapse rates are not directly comparable since they are affected by possible horizontal temperature differences between the grid cell and Tarfala's data and other re-analysis biases.

We correlate Tarfala's air temperatures with Storglaciären's mass-balance record. The mass balance is computed from detailed stake measurements, snow probings and snow density measurements (Holmlund et al. 2005). Results show that Tarfala's mean summer air temperature (JJA) is highly correlated with the summer mass balance and explains 76% of the variance (Fig. 8a). Melt models are often driven by positive-degree days (PDD) rather than air temperatures (Hock 2003). These so-called degree-day models relate melt to the positive degree-day sum (PDDS) (ϕ), defined as the integral, in Kd (Kelvin days), of the excess of temperature (T) above the melting point (T _m) over a stated span of time (t) in days (Cogley et al. 2011): ϕ = ∫ t 1 t z max [ 0 , T ( t ) - m ] d t

The integral is often approximated by summing positive mean daily temperatures (in °C) (e.g., de Woul & Hock 2005). Here, we calculate annual PDDS based on Tarfala's daily temperature record for the period 1965–2011. PDDS have been found to be key indicators not only for glacier/snow melt (Hock 2003), but also many other temperature-dependent variables, such as permafrost thaw and vegetation growth. Due to the observed temperature increases in Tarfala, PDDSs are expected to increase during the study period.

Annual PDDSs and the number of PDD per year are shown in Fig. 9. From 1965 to 1994, there were 129±12 days with positive mean temperatures in a given year (mean±1σ) with no temporal trend apparent in the time series. This number is significantly higher during the period 1995–2011, with an average of 139±10 days (Fig. 9a). Annual PDDSs fluctuated around 660 Kd but increased steadily since the mid-1990s and reached values not previously recorded (Fig. 9b). Correlating summer mass balance with annual PDDSs yields an explained variance of 77%, which is similar to the one found for summer air temperature.

The NAO index is a measure of the north-to-south pressure gradient over the North Atlantic region generally defined by the difference of atmospheric pressure at sea level between the Icelandic Low and the Azores High (Hurrell 1995). The NAO determines the strength and direction of westerly winds and storm tracks across the North Atlantic, especially in winter, and varies over time with no particular periodicity. Higher than usual pressure gradients (positive NAO index) are associated with the deflection of warmer Atlantic air masses to northern Scandinavia.

Previous studies have found correlations between NAO and seasonal mass balances of Scandinavian glaciers including Storglaciären (Pohjola & Rogers 1997; Nesje et al 2000; Linderholm & Jansson 2007). Here, we investigate the correlation between Tarfala's temperature record and the NAO. Fig. 10 indicates that at least for the first half of the time series variations in Tarfala, winter temperatures (DJF) may be driven by the air pressure variations described by the NAO index. As the results from the non-parametric Spearman rank tests show that the correlation varies with time (Table 2). It is largest during the period 1976–1995, when more than 50% of the variance is explained by the NAO. However, correlations are weaker during the last decades indicating a decoupling between Tarfala's winter temperatures and the NAO index.

Fig. 11 and 12 show ground temperature time series at different depths and temperature profiles for each year suggesting that the permafrost is warming at a considerable rate.

Fig. 12 indicates that temperatures have increased at all levels that are not subjected to the influence of seasonal fluctuations (below approximately the upper 15–20 m below the surface). The minimum temperature of the record below 20 m is −3.35°C (30 m below the surface at the beginning of the record). The temperature at the bottom of the 100 m borehole varies between −2.78 and −2.74°C during the investigated period. Based on earlier data from the time series, Harris et al. (2009) report that Tarfala's ground temperature gradients are negative in the upper 40–45 m. Below this depth, ground temperatures increase with depth, hence, gradients are positive. Fig. 12b reveals a steady lowering of the turning point where the gradient turns from negative to positive. The 2011 profile indicates that the first 60 m experience a negative gradient; hence a lowering of the turning point by 15–20 m has occurred in the recent years. This finding is consistent with the profile of linear warming trends (Fig. 13). Linear trends over the 11-year period from 2001 to 2011 range from 0.002°C yr⁻¹ at 100-m depth to 0.047°C yr⁻¹ at 20-m depth. Trends increase exponentially from the bottom of the borehole to the depth of 20 m, though temperatures at 20 m have decreased since 2010. The trends are similar to those found for an earlier period by Isaksen et al. (2007).

Christiansen et al. (2010) report an increase in active layer thickness (defined as the layer that experiences seasonal freezing and thawing) at the Tarfala drill site over the period 2000–08. Fig. 14 shows substantial warming in the upper layers over the period 2001–03, with the active layer thickness increasing from 1.19 to 1.66 m, but no apparent trend of active layer thickness in the following years. However, during the earlier years values are probably underestimated due to large data gaps. During the period 2003–11, the active layer thickness fluctuates between 1.52 and 1.66 m (mean: 1.58±0.04 m). The lack of a positive trend may be attributed to interannual variations in snow cover at the drill site.