Climate change and projections for the Barents region: what is expected to change and what will stay the same?

The polar regions are expected to be the place where the most pronounced climate changes will take place in connection with a global warming. The stronger polar response is caused by a 'polar amplification' [1, 2] and will have potentially serious consequences for both ecosystems and communities. These concerns, in addition to scientific curiosity, have triggered a number of initiatives and scientific programs with the objective to enhance our understanding of the current situation and future outlooks for the polar regions. The environmental changes are being felt especially in the Arctic, both because of communities of people living there and because a warming is associated with a retreat in the sea-ice cover [3, 4]. A diminishing sea-ice extent combined with coastal erosion and storm surges affect the Arctic communities dramatically, e.g. likely need for costly relocation of some settlements [5–7]. Furthermore, more severe winter storms and blizzards alter the deposition pattern of snow that may lead to fatal avalanches (e.g. Longyearbyen, Svalbard, 19 December, 2015).

There have been comprehensive studies, such as the Arctic Monitoring and Assessment Programme's (AMAP) project known as the Arctic Climate Impact Assessment in 2004 [6], followed by the fourth International Polar Year in 2007–2008 [8], an assessment of changes in the Arctic through the initiative known as Snow Water Ice Permafrost (SWIPA) in 2011 [9], and the European project FP7-ACCESS [10] (2011–2015). The rapid climate changes in the Arctic, the accumulation of new observations, progress in new analytical capabilities, and new knowledge have prompted the Arctic Council to commission a new set of reports through AMAP; one SWIPA update and another on Adaptive Actions in a Changing Arctic (AACA). Furthermore, a new set of climate projections is available since the previous assessments, based on the fifth coupled model inter-comparison experiment (CMIP5) experiment [11] which is described in 'AR5', the Intergovernmental Panel on Climate Change's (IPCC) fifth assessment report [12]. It was established in AR5 that the models' ability to capture the downward trend in Arctic summer sea-ice extent had improved compared to the previous assessment report [13].

1.1. Motivation

2.1. Method

2.2. Data

Only stations above the 65°N were used in this analysis (figure 1), and the downscaling used 316 daily rain gauge and 297 temperature station records from the European Climate Assessment & Dataset (ECA&D) project [32] for the predictands (downloaded through the esd package). For precipitation data, annual wet-day mean values exceeding 20 mm day⁻¹ were ignored and set to missing value, as such high values were not physically credible (this only applied to a couple of instances). The downscaling was carried out for aggregated temperature parameters for four seasons (Dec–Feb, Mar–May, Jun–Aug, Sep-Nov) but only for two seasons for precipitation ('cold': Jan–Mar + Oct–Dec; 'warm': Apr–Sep). Precipitation is intermittent both in time and space, and the real size of the data sample depends on the number of wet days (defined as days with more than 1 mm day⁻¹ precipitation).

Zoom In Zoom Out Reset image size

The predictor used in the calibration was the NCEP/NCAR reanalysis [33] (1949–2015), and for projection of future outlooks, the predictors were taken from the CMIP5 [34] ensembles following the RCP2.6, RCP4.5, and RCP8.5 respectively [35]. The RCP2.6 ensemble included 65 simulations, RCP4.5 had 108 runs, and RCP8.5 involved 81 ensemble members. The reanalysis and the CMIP5 results were combined and common EOFs were used to represent the predictors in the regression analysis [28]. The downscaling of storm statistics used NCEP/NCAR SLP reanalysis [33] over the North Atlantic and European region (90°W–90°E/30°–80°N) as a predictor and the aggregated storm track count over the Barents Sea as predictand. A set of five leading EOFs calculated from the annual mean SLP was used to describe the large scales, and the predictand was derived from published storm identification and track analysis [36, 37] to estimate number of deep cyclones (n_s with central pressure < 980hPa) per year over the Barents region (10°–70°E/65°–80°N).

The elevation information used in the gridding of the station data was based on the etopo5 data set with a 5 min latitude/longitude grid [38], however, the elevation taken from the station meta-data was used to calibrate the gridding model to the topography.

Figure 2 presents the results of a sensitivity assessment for the mean temperature (T_2m), the day-to-day temperature variability (${\sigma }_{{\rm{T}}}$), the temperature persistence ($\alpha (\tau )$), the wet-day mean precipitation (μ), wet-day frequency (f_w), and the mean number of consecutive dry days (n_cdd). The mean seasonal variations in T_2m was included as a well-known benchmark to verify the methods and provide a level against which the other parameters could be assessed (figure 2(a)). It followed a clear seasonal cycle with cold winters and warm summers, as expected, and the estimates for ${\sigma }_{{\rm{T}}}$ were more prominent during winter than summer (figure 2(b)). The results for $\alpha (\tau )$, on the other hand, was not sensitive to variations in the ambient conditions, and the tight clustering of the curves in figure 2(c) suggested that it was insensitive to different geographical conditions too. The precipitation-based parameters, however, revealed a stronger geographical dependency. Estimates of μ exhibited a seasonal response, albeit with a character depending on the site (figure 2(d)), whereas f_w and n_cdd showed a less pronounced systematic response to the seasonal variations (figures 2(e) and (f)). This sensitivity analysis was supported by trend analyses (SM), and there have been clear past trends in the seasonal mean temperature $\bar{{T}_{2{\rm{m}}}}$, but increases in μ were less pronounced. There was no historic trend in n_cdd, and f_w has increased in some regions and decreased in others. Regional differences in these trends may be associated with shifts in circulation patterns or a displacement of storm tracks. Assuming that these results provide some indication of the general sensitivity to changing conditions in the Barents region, then $\alpha (\tau )$ and the spell length statistics are not expected to change significantly in the future. Hence, projections of future climate parameters were limited to $\bar{{T}_{2{\rm{m}}}},\mu ,{f}_{{\rm{w}}}$, and n_s.

Zoom In Zoom Out Reset image size

A validation of the GCMs was based on a comparison of the standard deviation of the principal components (PCs) from the common EOF analysis over the 1949–2015 interval common to both reanalysis and GCMs. The GCMs simulated spatial patterns of variability for seasonal mean temperatures over the predictor region with similar characteristics as those found in the reanalysis. There were no model that distingushed itself as being consistently poorer or superior to the rest of the ensemble, but the evaluation suggested that different models had different systematic biases (SM). Cross-validation correlation scores and R² statistics were used to evaluate the step-wise multiple regression used in the ESD, and the results for temperature indicated that the downscaling model could account for about 90% of the seasonal temperature variability. The validation also involved an assessment of the degree that the downscaled ensembles of GCM projections reproduced observed magnitudes of interannual variability and past trends. The trends were in general agreement with observations over common time intervals, and the magnitude of interannual variability was slightly exaggerated by the downscaled ensembles. Overall, the ESD for seasonal mean temperature provided credible results that were consistent with changes in the past.

Projected changes in future temperature and precipitation were based on ESD of entire CMIP5 ensembles and subject to gridding. A 'typical warm winter' in 2099 over Fennoscandia and the high Arctic was estimated to be 7 °C warmer than a corresponding warm winter in 2010 (see SM), given the RCP4.5 emission scenario (figure 3(a)). Generally higher $\bar{{T}_{2{\rm{m}}}}$ in the future was a robust feature in all emission scenarios, seasons, and for near, mid, and distant future. These results are presented in the atlas in the SM, showing changes since 2010 in the 95-percentile of the downscaled ensemble. The most extreme warming was associated with emission scenario RCP8.5 and exceeded 18 °C.

Zoom In Zoom Out Reset image size

The evaluation of the ESD method indicated moderate skills for downscaling f_w (the occurrence of precipitation) when SLP was used as predictor. The downscaled ensemble results slightly under-estimated the interannual range of variability, and a histogram of past proportional trends in f_w exhibited a bimodal structure, which was consistent with increased precipitation frequency in some regions and a decrease elsewhere. One explanation for such regional differences is a shift in the location of storm tracks. The projected changes in f_w differed with season, and included regions with both increases and decreases in the range −5–+10% (RCP4.5).

The picture for μ was more complex than for f_w, and the evaluation of the ESD method used for projection gave poor skill scores. Nevertheless, the sensitivity analysis suggested a dependency between past trends in μ and $\bar{{T}_{2{\rm{m}}}}$ of 8.75% °C⁻¹ which implies an increased winter-time precipitation intensity of 20%–70% by 2099 (RCP4.5), with strongest increase in the region with greatest warming (figure 3(b)). Diagnostics for ESD analysis applied to μ and reanalysis indicated a connection between North Atlantic cyclone activity and intense rainfall along the west coast of northern Norway during both warm and cold seasons. A typical feature in the large-scale temperature over the North Atlantic was a cooler ocean region west off the British Isles and south of Greenland, and higher temperatures further south and over the North Sea. Some of the storm systems from the North Atlantic reach the Barents Sea, and the downscaled results for deep cyclone storm tracks in the Barents region suggested higher frequency in the future and was a robust result with respect to RCP (figure 4).

Zoom In Zoom Out Reset image size

The results presented here reflected changes in the upper range of the ensembles, taken as the 95-percentile. This upper range was sensitive to both a shift in simulated change in the magnitude interannual variability and the models' different climate sensitivity. There is a complex link between circulation patterns, storm statistics, precipitation and temperature [39] that was emphasised here, and is a small part of a larger information content embedded in large GCM ensembles and historical observations. The use of large ensembles is key, and even if a GCM and an RCM were 'perfect', they would not provide a good picture of weather related risks based on just one simulation because the objective is to predict the longer term statistics for temperature and precipitation.

The lowest confidence of the downscaled results concerned μ and its spatial-temporal intermittency may be one factor explaining its wide range of uncertainty and small effective statistical sample. A small number of very extreme cases can have a strong influence, and the sampling issue is exacerbated by the rain gauges' small cross-section area of capture. In addition, a number of different phenomena are known to produce precipitation such as cyclones, fronts, orographic effects and local cumulonimbus, and precipitation has geographically distributed sources of moisture. The downscaling in this study was designed to capture precipitation connected with storms and remote moisture sources, but not local cumulonimbus and the effect of increased area of open ocean on μ [40, 41]. Nevertheless, there were multiple indications suggesting that a higher temperature is associated with an increase in μ, and these effects can to some degree be accounted for in terms of the ratio of local trends in μ and $\bar{{T}_{2{\rm{m}}}}$ respectively. The dependency between these local trends was considered to be more reliable than the results from ESD based on North Atlantic temperatures, due to the poor ESD skill scores.

Another caveat was that the recorded increase in the precipitation may be due to a changing portion of the precipitation falling as snow and rain and systematic under-catch of snow [42]. Evaporation E takes place over a wide area A_E but the precipitation P involves a more limited area A_P (cloud-covered area where ${A}_{{\rm{E}}}\gt {A}_{{\rm{P}}}$). A constant atmospheric water vapour content implies that the global rate of evaporation equals the global rate of precipitation averaged over time and space: $\bar{{\int }_{{A}_{{\rm{E}}}}E(a,t){\rm{d}}a}=\bar{{\int }_{{A}_{{\rm{P}}}}P(a,t){\rm{d}}a}$. Since these involve different areas, an increase in the precipitation is expected to be proportionally higher than the evaporation by a factor of γ, as in $\delta \langle P\rangle =\gamma \delta \langle E\rangle$, where $\gamma =({A}_{{\rm{E}}}/{A}_{{\rm{P}}})$ and $\langle .\rangle$ is the spatial average. In other words, we can deduce that the precipitation will increase faster than the atmospheric moisture because E is a smooth variable in space but P is intermittent in time and space.

Another caveat concerns the variable station network density, and there were fewer data points in Russia which emphasises Fennoscandia with a dense network (figure 1). This choice influenced the selection of predictor domain and the ESD in terms of favouring the heavy precipitation over northern Norway and its association with low-pressure systems. A sparse Russian network of stations also has an effect of the gridding of the results.

The present study was limited to a small set of climate parameters and there is a need for future work addressing the question of downscaling precipitation and additional statistics, such as the number of days with above- or below-freezing temperatures, the length of the winter season, and the future prospect for the snow cover. Such analyses may make use of dependencies between the seasonal mean temperature and number of warm/cold days for some sample stations [14], in addition to probabilities of heavy precipitation associated with μ and f_w [25, 26]. There is also a risk for abrupt changes and a triggering of tipping points, which may or may not be captured by GCMs, e.g. a collapse of ice sheets, release of methane, and rapid changes to the landscape and vegetation. Nevertheless, the projections presented here show the best current information there is on future climate outcomes in the Arctic.

Projected changes in climate parameters for the Barents region indicate a temperature increase for 'typical warm' winters with ∼7 °C over northern Fennoscandia by 2099 and the high Arctic under the RCP4.5 emission scenario. The scenarios for the wet-day frequency and mean precipitation were not as robust with respect to greenhouse gas emissions and season as for temperature, however, the inferred changes in the wet-day mean were most severe in winter (up to 70% for 2099 and RCP4.5). The wet-day frequency pointed to greatest increases ($\delta {f}_{{\rm{w}}}\quad \sim$ 5%–10%) along the northern coast of Norway but reduced frequencies in northern Sweden and the vicinity of the Bothnian Bay. The number of deep synoptic cyclones can also be expected to increase in a warmer Arctic, but there was little indication that persistence in the temperature and precipitation will change in the future.

This work was supported by the Norwegian Meteorological Institute, and carried out for AMAP/AACA. We acknowledge the efforts by the IMILAST team for the cyclone analysis, ECA&D for the station data, the team behind the NCEP/NCAR reanalysis, and the people behind the CMIP5 results.