While Fennoscandia is well accustomed to and prepared for winter conditions, very low temperatures have become less frequent in the recent decades and events such as this have the power to surprise and be particularly impactful. Direct impacts of the coldwave will have been felt most strongly by vulnerable groups notably people experiencing energy poverty, living in poor housing stock, and/or experiencing homelessness. Elderly people and children are also more vulnerable physiologically to extreme temperatures. The cold also caused traffic disruptions, school closures, power cuts and infrastructure damage.

Scientists from Norway, Sweden, Finland, France, the Netherlands, the UK and the US collaborated to assess whether and to what extent human-induced climate change has modified the likelihood and intensity of this cold wave.

Published peer-reviewed methods were used by the team to analyse the event, on twodifferent scales and locations: Firstly the annual (Jul-Jun) minimum of 5-day averaged minimum temperatures in a region impacted by the onset of the cold wave in Northern Scandinavia, as shown in Figure 1; and secondly the single-day (Jul-Jun) minimum temperature for the station of Oslo, to the south of the study region, impacted slightly later.

Main findings 

● Extreme cold such as experienced during this event can have severe direct impacts on health, and secondary impacts on roads, electricity grids, and services such as schools and public transport.
● Cold waves, like other extreme weather events, put significant pressure on energy, healthcare, and water systems. These systems must be designed and able to absorb additional (or different) needs, and early warnings, preparedness plans, and response capacity are all crucial to mitigate the impacts of these events.
● Homelessness in particular, but also energy poverty and bottlenecks with the associated price hikes and, in some cases, poor housing conditions, exacerbate the impacts of cold waves on particularly vulnerable socio-economic and demographic groups.
● Even though some very low temperatures were recorded, the dataset based on observations characterises the area average 5-day cold spell as a 1-in-15 year event in today’s climate, ranking as 12th coldest since 1950. For Oslo, the one day minimum temperature was rare, approximately a 1-in-200 year event in today’s climate.
● To estimate the influence of human-caused climate change on this extreme cold we use a combination of climate models and the observations. We find that because of human-induced climate change the area-averaged event would have been about 4 degrees colder in a 1.2°C cooler climate. This corresponds to such cold spells having become about 5 times less frequent.
● For the Oslo station the cold single day minimum would also have been about 4 degrees colder without human-induced climate change. This corresponds to such cold days having become about 12 times less frequent.
● At global mean temperatures of 2°C above pre-industrial levels, large-scale cold spells as rare as this one are projected to be about another 2.5 °C less cold for the area average and about another 2 °C less cold for Oslo. They will become even less frequent than today.
● Climate change does not mean that cold waves will no longer happen. In fact, less severe and less frequent cold waves may be more impactful than past ones if risk perception and preparedness decrease due to the less frequent event occurrences.

Main findings

• By analysing temperature observations and climate model simulations we found that without human-caused climate change, such temperatures in April would have been even lower by approx. 1.2°C.
• However, human-caused warming also affected the earlier occurrence of bud burst, characterized here by a growing-degree-day index value. This effect is stronger than the decrease in spring cold spells, thus exposing young leaves to more winter-like conditions with lower minimum temperatures and longer nights.
• The climate models used here simulate a cooling of growing-period annual temperature minima of about 0.5°C. This effect is larger in the observations based analysis than in climate model simulations, this number thus serves as a lower bound rather than an exact quantification.
• Overall, we conclude that human-caused climate change made the 2021 event 20% to 120% more likely.
• In a climate with global warming of 2°C (compared to the present day level of global warming of about 1.2°C) growing-period frost events such as observed in 2021 are projected to further intensify by about 0.2°C to 0.5°C.

Figure 1: Stations with March (left) high records broken and April (right) low records broken in 2021 in France. Source: Météo-France.

Background

Frost days and cold spells are decreasing in frequency in the northern latitudes. Yet, due to the natural chaotic variability of the atmosphere, severe cold spells continue to occur across the mid-latitudes. When occurring after the start of the growing period, the invasion of polar air into central and Southern Europe can create devastating frost damages in young leaves such as happened in early April 2021.

From 6 to 8 April 2021 exceptionally low daily minimum temperatures, below -5°C, were recorded in several places, leading to severe damages in grapevines and fruit trees in these places. The temperatures broke negative records at many weather stations (see Figure 1). To make matters worse, the cold event happened a week after an episode of record-breaking high March temperatures over France and Western Europe (Figure 1), leading to the growing season to start early and leaving new leaves exposed to the deep frost episode that followed.

The resulting damages affected “several hundreds of thousands of hectares” according to the French Ministry of Agriculture that also called the event “probably the biggest agricultural disaster in the beginning of the 21st century”.

To investigate the role of climate change in this event we focussed on central/northern France [-1°- 5°E; 46°-49°N] in order to investigate a relatively homogeneous, low-elevation area. It covers most of the grapevine agriculture areas of Champagne, Loire Valley and Burgundy which were identified as specifically vulnerable.

The event is further defined as the minimum temperature (TNn) conditioned on the so-called Growing Degree Day (GDD). The GDD is the cumulative temperature above 5°C  after the winter solstice that leads to buds bursting. This cumulative temperature threshold is species dependent and was between 150°C and 350°C the day before the frosts started. This is high for the season, but not a record, which was in fact set in 2020. Given we are not interested in a specific species but the overall frost damage to agriculturally important species we consider three different GDD thresholds: 150°C, 250°C and 350°C. We also considered minimal temperatures in a range of GDDs : between 250 and 350, in order to characterize the early growing season when leaves are vulnerable. For each of these three thresholds or ranges we calculate the minimum temperature in April-July after the GDD threshold has been reached, denoted as TNnGDD150, TNnGDD250, TNnGDD350 and TNnGDD250-350, averaged over the region of central France. We also looked at the change in intensity and frequency for the minimum April-July temperatures over the same region alone, finding that the event would have been even colder without human-caused climate change.

Conditioning the assessment on the four GDD-indices we found instead that frost intensities on bursted buds increased due to climate change. While the exact quantitative results differ between the indices, the results are qualitatively the same.

The differences between the changes in TNnGDD-indices simulated in the models and those calculated based on observations are large, we thus have confidence in the qualitative results but not the exact quantification.

We do not find any evidence for an intensification of these types of cold waves due to the Arctic warming faster than the mid-latitudes. On the contrary, they seem to be warming faster than the winter mean as the Arctic air coming south is less cold now.

Key points

• The two-week cold outbreak in the northeastern U.S. and southeastern Canada was highly unusual in the current climate; however, it is not unprecedented.
• Cold waves like this have decreased in intensity and frequency over the last century, but still occur.
• Temperatures like these are now about fifteen times rarer. This is equivalent to cold waves being about 4ºF (2ºC) warmer than they used to be.

Introduction

Over recent years (2014, 2016 and now), there have been a number of extreme cold events in North America, although in different regions and of different durations. Low temperatures have been observed over much of northeastern North America during the two weeks between December 26, 2017 through January 8, 2018 (Fig. 1, top), and these temperatures are unusually cold in a large region centered over the Great Lakes extending to the eastern seaboard (Fig. 1, bottom). In 2014, cold air was concentrated in the Great Lakes area (van Oldenborgh et al, 2015). December 2016 also saw a strong cold spell in North America, but over more western regions (see our analysis of that event). We here analyse this year’s cold using the same methodology as we applied to those other recent cold events.

Comparison with recent observations

First, to help us define formally this year’s event, we compare this year’s temperatures in the general area of the cold wave to what is used as the “normal” period of 1981-2010. We compare the coldest two weeks in the 2017/18 winter so far with the coldest two weeks in previous winters in North America. We chose this comparison, rather than using temperatures on the exact same dates in the past, because a cold wave can occur any time in winter and affects people similarly, regardless of the exact timing. We find that the coldest two weeks this winter have been 7ºF to 11ºF (4ºC to 6 ºC) colder than the coldest two weeks have been, on average, during the past “normal” period of 1981–2010 over most of the area affected by the cold wave (Fig. 2).

Event definition

The area with coldest two-week daily mean averaged temperatures is approximately 40ºN–50ºN, 65ºW–95ºW. This is indicated on the map in Figure 3. Note that only a few Canadian stations have up-to-date data in the GHCN-D v2 dataset. We use the spatial average over land points in this area as an index for this cold wave. This index is thus optimised to express these large anomalies. A time series of daily values of this temperature index since mid-October shows a clear and expected decrease through the end of 2017 — signaling the arrival of winter — but the temperatures over December 26, 2017 through January 7, 2018 are much colder than expected for this time of year, with deviations below the 1981–2010 mean close to 25ºF (14°C) (Fig. 3). The last dip was on Saturday, January 6, after which this cold wave came to an end.

Comparison with historical observations

To put this in a longer historical perspective and compute actual odds of such an event, we used the Berkeley Earth System Temperature daily temperature analyses 1880–2013, extended by ERA-interim up to 2016/17 and ECMWF analyses and forecasts for the last few months. Over the overlap period, this data agrees well after a small bias correction in the mean. The time series of the temperature of the coldest two-week period of the year, averaged over the area of the 2017–2018 cold wave, is shown in Figure 4. Note that although we have chosen the area to give the largest cold anomaly in 2017/18, there have been many similar or colder two-week periods in that region in earlier periods, notably in 1960–1994 and before 1920, but none since 1993/94. If not unprecedented in absolute values, this year’s cold stands out for occurring so early in the season. Only in 1886/87 there was an earlier colder two-week period. In 1887/88 a similarly cold period started one day later than this year.

Return time

To quantify the effect of warming on the coldest two-week period of the winter in the area, we fitted the series to a Generalised Extreme Value (GEV) distribution whose mean location is allowed to shift proportionally to the smoothed global mean temperature time series (Fig. 5). That factor is estimated to be about two, indicating that the temperature of the coldest two-week period has increased about two times faster than the global mean temperature rise. This has been explained as a consequence of stronger warming trends in the Arctic, where the cold air originates  (Screen, 2014; van Oldenborgh et al, 2015). The fit also shows that these two weeks are very unusual in the current climate, with a return time of very roughly 250 years in this region.

Change in probability

The significant dependence of the GEV location on the values of global mean temperature allows us to ask what the return time of such an event was in the climate of a century ago, i.e., for values of the global mean temperature at the beginning of the 20th century. Our estimates indicate that a cold wave like this occurred on average every seventeen years. This means that cold waves like this have become approximately a factor fifteen less frequent due to global warming, as shown in Figure 5.

Trends over North America

In fact, the temperature of two-week cold waves has increased over all of North America, as expected in a warming world (Fig. 6). We see no evidence of an increased intensity of these cold waves, but rather a faster warming than the mean winter temperature in northern North America. The same was shown for one-day cold extremes in previous articles (van Oldenborgh, et al, 2015), both in observations and climate models.

Conclusions

We conclude that this was an exceptional two-week cold wave in the area in the current climate. Cold outbreaks like this are getting warmer (less frequent) due to global warming, but cold waves still occur somewhere in North America almost every winter.

References

Barnes, E. A. and J. Screen. (2015) The impact of Arctic warming on the midlatitude jet-stream: Can it? Has it? Will it? WIREs Climate Change, 6: 277–286. doi: 10.1002/wcc.337

Van Oldenborgh, G.J., R. Haarsma, H. De Vries, en M.R. Allen. (2015) Cold extremes in North America vs. mild weather in Europe: the winter of 2013–14 in the context of a warming world. Bulletin of the American Meteorological Society, 96: 707–714, doi: 10.1175/BAMS-D-14-00036.1

Screen, J.A. (2014) Arctic amplification decreases temperature variance in northern mid- to high-latitudes. Nature Climate Change, 4: 577–582. doi: 10.1038/nclimate2268

Not only are the temperatures very low, they have also persisted for a long time. In Serbia maximum temperatures have been around -10ºC (14ºF) for five days in a row, with minimum temperatures around -15ºC (5ºF).

The forecast calls for less severe cold today and gone by tomorrow.

The largest deviations from normal are in the Balkans. Montenegro, Serbia, the Republic of Macedonia and Bulgaria had temperatures of more than 12ºC (21.6 ºF)  below normal averaged over the last five days (Figure 2). Around this core area it was also cold in eastern Italy, Greece, western Turkey and Romania with five-day mean temperatures 5ºC to 10ºC  (9ºF – 18ºF) below normal in January.

Weather situation

The weather that caused the cold outbreak was a developing high pressure system over western Europe. On January  1, 2017 this high was evident on the weather maps, although at that time it was still too far west to influence the Balkan peninsula. It gradually moved east, transferring cold air from Siberia south to southeastern Europe, following the clockwise circulation around this high pressure system (see Figure 3).

Snow showers develop easily over the relatively warm Mediterranean Sea when a compact low pressure system forms in the region, as it did from January 5 -8. From January 8 – 11 the high weakened and moved farther to the east. The low stayed in the region and caused more snow to fall. On January 12 the high all but disappeared and the inflow of cold air stopped. By January 13 the temperatures will almost be back to normal in southeastern Europe.

How exceptional is this?

The temperature of the last few days in the Balkans was very low, but not at record levels. The lowest daily mean temperature averaged over Serbia was around -15.6ºC (3.9ºF) on January 7. Since 1950 colder days were observed in the winters of 1953/54, 1955/56, 1962/63 and 2011/12. The lowest temperature in this series was measured on January 24, 1963 at -19.7 ºC (-3.5ºF). (Another dataset, the Berkeley Earth analysis, shows that the winters of 1928/29 and 1941/42 had even colder days.) A daily mean temperature as low as the one observed on January 7, or lower, occurs on average about once every 35 years.

The extended period of cold is also not very unusual. The lowest five-day mean temperature in the Serbia-wide series from E-OBS is from January 7 – 11, with a temperature of -12.8ºC (-9ºF). Colder five-day means in the series from 1950 to now were in 1953/54, 1955/56 and 1962/63, and before that again in 1928/29 and 1941/42. This analysis concludes that the five-day cold spell was about as exceptional as the one-day extreme, with a probability of once every 35 years on average (three percent every year).

In the north of the Balkans and in the mountains, snow is very common. Belgrade, Serbia has snow on the ground almost every year, although the amounts show a clear decrease since 1950. On the coasts snow is much more rare. Dubrovnik, Croatia only has a day with a few centimeters or inches of snow once every few years. This also seems to have become more rare over the years. The Greek islands also get snow occasionally according to satellite observations.

Trends in observations

The observational analysis shows a trend towards higher temperatures of the cold outbreaks, but the trend is not statistically significant, as the variability from year to year is much larger than the trend. Temperatures as low as the ones observed the last few days have become a factor of about two less common since 1950, but the uncertainty range stretches from more common to much less common. The equivalent trend in temperature is between -1.7ºC and 4.0ºC (19.4ºF – 39.2ºF) over the last 65 years, again probably a warming trend but no definitive conclusions can be drawn from the observations. An analysis of five-day cold spells shows very similar results

Trends in models

Climate models used for the 2013 Fifth IPCC Assessment Report show a trend of about 2ºC  (3.6ºF) for the temperature of the coldest day of the year in Serbia (Figure 4). This is completely compatible with the observed trend. The Balkans are in a transition zone between the very strong warming trends of cold days in Russia and more moderate trends around the Mediterranean

Conclusion

A high-pressure system moving slowly eastwards over Europe brought cold Siberian air to southeastern Europe along its eastern side. Montenegro, Serbia, the republic of Macedonia and Bulgaria were much colder than normal, with temperatures as low as -15 ºC (5ºF) over five consecutive days. The surrounding countries of Italy, Greece, Turkey and Romania were 5ºC to 10ºC (9ºF – 18ºF) colder than normal for the time of year. The freezing temperatures with a developing low over the Mediterranean brought large amounts of snow to many places.

This kind of cold outbreak is not unprecedented in this region. These cold snaps occur on average every 35 years or so. This means that every year there is a three percent chance of a cold event like this one, or colder, of happening. The temperature of these cold waves has increased since 1950. This increase is not statistically significant due to the variability of the weather, but climate models show the same increase. Before global warming, an extreme cold snap like the recent one in southeastern Europe would have been even colder.

This year, the North Pole and the surrounding Arctic region are seeing record-high temperatures in November and December (Figure 1a) and record-low ice extent (Figure 1b). Fall usually ushers in the season of sea ice growth, but November saw a brief retreat that was virtually unprecedented in nearly 40 years of satellite records, according to the National Snow and Ice Data Center. That dip helped November set a record low for sea ice area since 1850 (Walsh et al, 2016). By December, the area around the North Pole is typically 95 percent covered by sea ice. However, this year it is only about 80 percent (Figure 1b).

Mid-November saw an early winter “heatwave” with the temperature on November 11 reaching -7 ºC (19 ºF) – that is 15 ºC (27 ºF) above normal for the time of year. The monthly mean November temperature was 13 ºC (23 ºF) above normal on the pole. Temperatures in this region declined slightly after that but remained well above normal. The forecast for the next few days is again more than 15 ºC (27 ºF) above normal at the North Pole itself and 10 ºC (18 ºF) averaged over 80–90 ºC N (Fig. 2).

To quantify how rare the event was, we computed the November-December averaged temperature around the North Pole (80–90 ºN) in the ERA-interim reanalysis augmented with the ECMWF analysis and forecast up to December 25 and persistence up to December 31. The temperatures are slightly less extreme than at the pole itself (Figure 2), but unprecedented in the satellite era from 1979 onwards (Figure 4).

We first note some of the impacts of these high temperatures in the Arctic regions. Next we describe the physical mechanisms that are responsible for sea ice and temperature variability in the North Pole region, which is warming faster than anywhere else in the world. We then analyse the observed temperatures via a related time series encompassing the northern-most meteorological observations on land (70–80 ºN). This allows for a reconstruction of Arctic temperatures back to about 1900, which clearly shows how the current warmth is unprecedented over that period. It also gives a rough estimate of how rare it is in the current climate and how much the probability has changed over the last century.

Finally, we perform a multi-method analysis of North Pole temperatures with two sets of climate models: the CMIP5 multi-model ensemble that was used for the 2013 IPCC AR5 report, and a large ensemble of model runs in the Weather@Home project. Both sets of models give very comparable results, showing that the bulk of the increase is due to anthropogenic emissions. The results are combined in the synthesis and conclusions.

Impacts

Extreme warm events in the Arctic extend into biological and social implications. Unlike the Antarctic, Arctic lands are inhabited and their socio-economic systems are greatly affected by the impacts of extreme and unprecedented sea ice dynamics and temperatures. Recent work by Henri Huntington and colleagues reported on these effects on the timing of marine mammal migrations, their distribution and behaviour and the efficacy of certain hunting methods in the Beaufort Sea. Less than a month ago, Forbes and colleagues reported on the relationship between unseasonal sea ice decline in winter and the enhanced probability of rain-on-snow events in the Yamal Peninsula in north-Western Siberia. While rain on snow does not cause problems in spring, it can be catastrophic for reindeer in the autumn when rain turns to an ice crust as plummeting normal temperatures return. This crust, often several centimetres thick, prevents the reindeer from feeding on fodder beneath the snow throughout the winter months. Two extreme weather events in 2006 and 2013 caused mass starvation among the reindeer herds (one event alone in 2013 resulted in 61,000 reindeer deaths, about 22 percent out of 275,000 reindeer on the Yamal Peninsula), and were linked with sea ice loss in the adjoining Barents and Kara seas. These two examples only illustrate the variety of ways in which climate change in the Arctic directly affects its ecosystems and societies. Many others include effects on the timing of phytoplankton blooms (which are at the base of the food chain in the Arctic ocean) and terrestrial plants, their consequences for animals that depend on these resources, opportunities for invasive southern species to colonise the region, enhanced mobility around the Arctic region for marine mammals such as whales, or reduced and largely modified habitat for animals that directly depend on sea ice such as the polar bear.

Background

The Arctic is the fastest-warming region on Earth, with the largest temperature increases in winter. In summer, the temperature is constrained by all the ice melting and it never rises very far above zero. However, the extra heat absorbed in summer is released in autumn and winter as the water remains warmer for longer and re-freezing sea ice releases the latent energy of melting.

This Arctic amplification of the global warming trend has several causes (see e.g., IPCC WG1 AR5 Box 5.1, Screen & Simmonds, 2010; Bintanja et al, 2011). The best-known is the extra warming due to the change in color (albedo-feedback): when white snow and ice is replaced by dark land and ocean, a much larger fraction of the incoming sunshine is absorbed instead of reflected. This extra heat warms the surface even more. Another positive feedback is less well known and is based on the observation that the warming trend in the Arctic is mainly confined to the lower part of the atmosphere. This is still far below the levels where thermal radiation escapes from the atmosphere on average, which is around 5 km (3 miles) high. This implies that the radiative cooling of the warmer air near the surface is not as efficient as in lower latitudes where the trend at altitude is similar to (mid-latitudes) or higher than (tropics) near the surface. The reduction of radiative cooling due to the vertical structure of the warming trend is called the lapse-rate feedback.

Variability has also increased in the Arctic. On top of the positive feedbacks mentioned above, this is due to two other factors. First, the shape of the Arctic Ocean. Before global warming began in earnest, the basin was filled with ice for most of the year, which implies that except for a few summer months the edge of the ice was relatively short and confined to the Atlantic sector. Now, there are more months in which the ice edge is free of the continents. This allows larger changes from year to year (van der Linden et al, 2014). Another factor is the thickness of the ice. The ice has thinned dramatically, with most of it now single-year ice that is much thinner, and hence flows and melts more quickly (Tschudi et al, 2016). This type of ice changes much more quickly at the whim of the ever-capricious weather, especially combined with the waves and swell that now develop in the ice-free regions (Thomson and Rogers, 2014).

Finally, warm episodes near the pole are usually due to so-called ‘moist intrusions’: warm and moist air from the mid-latitudes that not only advects heat, but also clouds that block outgoing long-wave radiation and hence trap more heat. There is some evidence that these have become more frequent (Woods and Caballero, 2016).

Observations

Satellite observations from about 1979 onwards, coupled with modern weather forecast models, allow a good reconstruction of November–December temperatures on the North Pole over the last 38 years (Fig. 4). There is a clear upward trend since 1990, with November–December 2016 almost 5 ºC (9 ºF) above this trend. However, the trend is not simply up: some cooling occurred in the 1980s. We therefore need longer time series to pin down which part is due to climate change and which to natural variability.

Unfortunately, there are few thermometers near the North Pole, and they tend to drift away as the sea ice flows towards the Atlantic Ocean. The best alternative is to use the thermometers on land around the North Pole. These have been combined into area-averaged temperatures by several teams. We use the GISTEMP (1200 km) data between 70 and 80 ºN, but alternatives (e.g., NOAA/NCDC, Berkeley Earth) give similar results. This series is well correlated to the ERA-interim analysis for 80–90 ºN during the overlap period (r=0.84). The regression, however, is only about 0.6, meaning that the variability and trend of this analysis are lower than at the North Pole itself. The value for December 2016 is estimated from the ECMWF analysis and forecast for the same area, corrected by the local regression over 1979–2015. This gives a November–December temperature of almost 5 ºC (9 ºF) above the 1951–1980 average, far higher than any other values in the series (Fig. 5).

As is well known, this series shows a warm period in the 1930s and 1940s, followed by cooling up through the 1970s and 1980s. The record value of 1944 was not surpassed until 2001 in this series. It turns out that this decadal variability is well described by the Atlantic Multidecadal Oscillation (AMO), nowadays often referred to as Atlantic Multidecadal Variability (AMV) as it does not have a clear period. We use the AMO index of van Oldenborgh et al (2009), which is independent of the warming trend and emphasizes the northern part. Data before 1900 have been discarded. The AMO index was slightly negative in November 2016 with colder than usual sea surface temperatures in the North Atlantic. This decreased the probability of a warm extreme somewhat. We ignore this small effect in 2016 for the rest of the analysis and compute the return period and changes for neutral AMO/AMV, which is a slightly more conservative estimate.

The time series with the regression on the AMO index subtracted is shown in Fig. 6 as a function of the global mean temperature. The temperature increases 2.5±0.6 times faster than the global mean, as discussed in the previous section. The 2016 event is seen to be about two standard deviations above the trend line, which translates to a return period of about 50 years (95 percent CI: 20 to 200, the intersection of the purple and red lines in Fig. 7). Around 1900, when the AMO was also close to neutral, such a temperature would have been so rare that the return period is impossible to estimate (the fit formally gives a few hundred thousand years, but we do not trust such a large extrapolation).

Finally, we add the AMO back into the analysis. The highest value of the AMO index in November-December over the last century was almost 0.6 ºC. This translates to an increase of 0.8 ºC in the 70–80 ºN temperature, so with the highest observed temperatures in the North Atlantic this would now be a 1 in 10 year event (5 to 25), but still less than 1 in 3000 year in the climate of a century ago. Hence, even with a favourable phase of the AMO, 2016 temperatures would have been extremely unlikely at that time. The probability of temperatures reaching the value observed in 2016, including all variability would be between the two extremes of AMO neutral (one in 50 per year) and AMO maximum (one in 10), albeit with large uncertainties. Note that this is the return time of the 70–80 ºN anomaly in November–December 2016. The temperature anomaly in the study region 80–90 ºN is larger (Fig. 1a) and hence the return time in this region will also be larger.

Another limitation of the observational analysis is that the fit assumes that the standard deviation of the Gaussian is constant. There is simply is not enough data to also fit a change in variability. However, we have reason to believe, on physical grounds, that the variability has likely increased. This implies that the true return period of the current event in the current climate is somewhat smaller than the fitted value of 200 year for AMO-neutral conditions, and that the increase in probability is larger than estimated here. Only an analysis using climate models can show how large this effect is.

Coupled climate models

The state-of-the-art climate models from the fifth phase of the Coupled Model Intercomparison Project (CMIP5; Taylor et al. 2012) were used to examine the event with multiple models. This analysis is similar to that described in King et al. (2015).

Table 1. Climate model simulations used in this study. Models shown in bold passed the validation test and were used in the event attribution analysis.

Sixteen climate models with at least three historical simulations (natural and anthropogenic forcings) over 1861-2005, one historicalNat simulation (natural forcings only) for 1861-2005, and one RCP8.5 simulation (projected climate under high greenhouse gas emissions) for 2006-2100, were analysed (see Table 1). Two-month (November and December) temperature anomalies were extracted from all model simulations relative to their own historical 1979-2004 baseline periods. The historical simulations were then compared with ERA-Interim over the 1979-2005 period for their trends and year-to-year variability (Fig. 8). Models with at least one-third of historical simulations having trends greater than double the ERA-Interim linear trend were removed from the analysis leaving the 13 used for subsequent analysis (bold in Table 1). Since the variability in ERA-Interim is close to the average across the CMIP5 simulations, it was not used to remove more models for the analysis described below. However, as a sensitivity test, models with more than one-third of simulations failing either the trend test or with average variability of more than 0.5 ºC (0.9 ºF) above or below the ERA-Interim average were also removed. The results were insensitive to this further restriction on the models.

The 13 remaining models (Table 1) were then used to estimate the change in likelihood of events like November-December 2016 or hotter. The likelihoods of such events were estimated in our ensemble of natural runs (for 1901-2005) and the current world runs (2006-2026 in RCP8.5). The temperature anomalies are shown in Fig. 9. Events as hot as 2016 or hotter were not seen in our natural world ensemble. In contrast, events like 2016 or hotter occur in our current world model simulations but are rare, with a return interval of roughly 200 years. These results suggest that it is extremely unlikely this event would occur in the absence of human-induced climate change.

We also investigated when an event like 2016 could become “normal” (e.g., Lewis et al. 2016) in the future. Under the high emissions scenario RCP8.5, we calculated when extreme heat in the Arctic—like 2016—becomes at least a one-in-two year event in 21-year moving windows. We found that by around 2050 (best estimate 2047) heat events like this one would be commonplace, occurring about 50% of the time. This corresponds to a global mean temperature of 2 to 3 ºC (3.6 to 5.4 ºF) above pre-industrial. So even if global warming were limited to 2 ºC (3.6 ºF), such high Arctic temperatures as in 2016 would be warmer than average but not unusual.

Ensemble of SST-forced climate simulations

We also investigated the November-December 2016 Arctic temperatures using ensembles from the weather@home project. These consist of thousands of simulations of the HadAM3P atmospheric model (Massey et. al. 2015), computed by volunteers on their home computers. Three ensembles of simulations were considered. Firstly a climatological set of simulations representing 1986-2013 was used to determine the baseline climate of our model. These used observed sea-surface temperatures and sea-ice from the OSTIA dataset and were one-year simulations from December to November of the following year (climatology). Second was a set of simulations representing November–December 2016 conditions (‘actual’). To produce data for 2016 ahead of time, these used sea-surface temperatures from the UK Met Office seasonal forecast model GloSea5, and 2012 sea-ice from OSTIA, which is closest to 2016 as of November. The third set of simulations also represent 2016, but with human influences removed (‘natural’). This was done by using pre-industrial levels of greenhouse gases and subtracting estimates of anthropogenic SST warming patterns obtained from 11 different CMIP5 models from the GloSea5 sea-surface temperatures (see Haustein et al., 2016). OSTIA sea-ice from the year with greatest sea-ice extent in the dataset (1986) was used in the Arctic. For this analysis we used 3141 2016 ‘actual’ simulations, 6125 2016 ‘natural’ simulations and 8400 climatology simulations (300 per year).

In order to quantify the role of atmospheric circulation in the event, we also analysed subsets of the ensembles which most closely represented the observed atmospheric flow of November 2016 (flow analogues), following the method of Vautard and Yiou (2009). To represent the atmospheric flow we used the 500 hPa geopotential height between 60–90 ºN, and compared with the NCEP-NCAR reanalysis. We selected the 200 closest analogues from each of the ‘actual’ 2016 and ‘natural’ 2016 ensembles, based on the observed November 2016 NCEP-NCAR reanalysis. The distance for these analogues was calculated from the anomalies of the weather@home and NCEP-NCAR reanalysis climatologies respectively. Note that the analogues were chosen on the November 2016 flow and in Fig. 10, the temperature is shown for November-December, for the continuous simulations matching the November analogues.

When analysing the weather@home simulations, we are comparing against anomalies relative to the average over the 1986-2013 climatology period. As the climatology simulations run December to November, we calculate the November and December climatological averages separately rather than from continuous November-December simulations. We note that the weather@home November-December climatological average is -25.53ºC compared to -23.65ºC for ERA-Interim over the same period, which is a 1.88ºC cold bias.

Year-to-year variability of the model is relatively realistic in this region, with the climatological values across years and ensemble members within about 1σ of the ERA-interim variability. However, prescribing SST and sea ice lowers the variance by about 45 percent. In the year 2016, this variance suppression effect appears even more pronounced, 70 percent to 75 percent, due to the larger ice-free area over which SST is prescribed. Future analyses will determine how this influences the atmospheric circulation and maximum temperatures over the region. However, prescribing SST and sea ice and thus artificially reducing variability, means that return periods calculated below are not directly comparable to the return periods obtained in the observational analysis and coupled model framing.

When considering the anomaly for the November-December 2016 event, there is only a single ‘actual’ 2016 simulation as warm as the event, and no ‘natural’ 2016 simulations, see Fig. 10. As there are 3000 ‘actual’ simulations, this indicates that this event is on the order of a 1 in 1000 year event in the present day and extremely unlikely in a climate without human influence with prescribed SST and sea ice.

The mean of the ‘natural’ 2016 ensemble is over 4ºC colder than the ‘actual’ 2016 ensemble, showing a very large change when the anthropogenic signal is removed. This change is still present when considering the 200 simulations that display the closest analogues to the observed flow. The difference in temperature between the ‘actual’ and ‘natural’ analogues is just under 4ºC, so slightly less than the difference of the full ensemble. Hence the temperature trend over years that is characterised by this kind of warm intrusion does not appear to be significantly different from the overall trend in all years, confirming an assumption made in the observational analysis

Conclusions

We have investigated the rarity of the November-December 2016 average temperature around the North Pole and assessed how much November-December average temperatures have changed over the past century using observations over a wider region. We also attempted to quantify how much high Arctic temperatures have changed due to anthropogenic emissions (greenhouse gases and aerosols) in two climate model ensembles.

The observations and the bias-corrected CMIP5 ensemble point to a return period of about 50 to 200 years in the present climate, i.e., the probability of such an extreme is about 0.5 percent to two percent every year, with a large uncertainty. This is rare, but it should be kept in mind that we are focusing on this particular November–December period precisely because an unusual event has occurred. For a random two-month period it would be between six and 12 times more likely. The prescribed SST design of the HadAM3P simulations precludes estimating an absolute return period.

The observations show that November–December temperatures have risen on the North Pole, modulated by decadal North Atlantic variability. For all phases of this variability a warm event like the one of this year would have been extremely unlikely in the climate of a century ago. The probability was so small it is hard to estimate, but less than 0.1 percent per year. The model analyses show that the event would also have been extremely unlikely in a world without anthropogenic emissions of greenhouse gases and aerosols, attributing the cause of the change to human influences. This also holds for the warm extremes caused by the type of circulation of November 2016. If nothing is done to slow climate change, by the time global warming reaches 2 ºC (3.6 ºF) events like this winter would become common at the North Pole, happening every few years.

References

Bintanja, R., Graversen, R.G. and Hazeleger, W. (2011) Arctic winter warming amplified by the thermal inversion and consequent low infrared cooling to space. Nature Geoscience, B: 758–761.  doi: 10.1038/NGEO1285

Forbes, B.C., Kumpula, T., Meschtyb, N., Laptander, R., Macias-Fauria, M., Zetterberg, P., Verdonen, M., Skarin, A., Kim, K.-Y., Boisvert, L.N., Stroeve, J.C. and Bartsch, A. (2016) Sea ice, rain-on-snow and tundra reindeer nomadism in Arctic Russia. Biology Letters, 12(11): 20160466. doi: 10.1098/rsbl.2016.0466

Haustein, K., Otto, F.E.L., Uhe, P., Schaller, N., Allen, M.R., Hermanson, L., Christidis, N., McLean, P. and Cullen, H. (2016) Real-time extreme weather event attribution with forecast seasonal SSTs. Environmental Research Letters, 11(6):  doi: 10.1088/1748–9326/11/6/064 006

Huntington, H.P., Quakenbush, L.T. and Nelson, M. (2016) Effects of changing sea ice on marine mammals and subsistence hunters in northern Alaska from traditional knowledge interviews. Biology Letters, 12: 20160466. doi: 10.1098/rsbl.2016.0198

King, A. D., van Oldenborgh, G.J., Karoly, D.J., Lewis, S.C. and Cullen, H. (2015) Attribution of the record high central England temperature of 2014 to anthropogenic influences. Environmental Research Letters, 10(5): 054002. doi: 10.1088/1748-9326/10/5/054002

Lewis, S.C., King, A.D. and Perkins-Kirkpatrick, S.E. (2016) Defining a new normal for temperature extremes in a warming world. Bulletin of the American Meteorological Society, 98(6): 1139–1151. doi: 10.1175/BAMS-D-16-0183.1

Massey, N., Jones, R., Otto, F.E.L., Aina, T., Wilson, S., Murphy, J.M., Hassell, D., Yamazaki, Y.H. and Allen, M.R. (2015) weather@home— development and validation of a very large ensemble modelling system for probabilistic event attribution. Quarterly Journal of the Royal Meteorological Society, 141: 1528–1545. doi: 10.1002/qj.2455

Screen, J.A. and Simmonds, I. (2010) The central role of diminishing sea ice in recent Arctic temperature amplification. Nature, 464(7293): 1334-1337, doi: 10.1038/nature09051

Taylor, K.E., Stouffer, R.J. and Meehl, G.A. (2012) An overview of CMIP5 and the experiment 962 design. Bulletin of the American Meteorological Society, 93: 485–498. doi: 10.1175/BAMS-D-11-00094.1

Thomson, J. and Rogers, W.E. (2014) Swell and sea in the emerging Arctic Ocean. Geophysical Research Letters, 41: 3136–3140, doi: 10.1002/2014GL059983.

Tschudi, M., Fowler. C., Maslanik., J., Stewart, J.S. and Meier, M. (2016) EASE-Grid Sea Ice Age, Version 3. NASA National Snow and Ice Data Center Distributed Active Archive Center. doi: 10.5067/PFSVFZA9Y85G.

Vautard, R. and Yiou, P. (2009) Control of recent European surface climate change by atmospheric flow. Geophysical Research Letters, 36: L22702. doi: 10.1029/2009GL040480

van der Linden, E., Bintanja, R., Hazeleger, W. and Katsman, C. (2014) The role of the mean state of Arctic sea ice on near-surface temperature trends. J. Climate, 27: 2819–2841. doi: 10.1175/JCLI-D-12-00617.1.

van Oldenborgh, G.J., te Raa, L.A., Dijkstra, H.A. and Philip, S.Y. (2009) Frequency- or amplitude-dependent effects of the atlantic meridional overturning on the tropical Pacific Ocean. Ocean Science, 5(3): 293–301. doi: 10.5194/os-5-293-2009

Walsh, J.E., Fetterer, F., Scott Stewart, J. and Chapman, W. L. (2016) A database for depicting Arctic sea ice variations back to 1850. Geographical Review, 107(1): 89–107. doi: 10.1111/j.1931-0846.2016.12195.x

Woods, C. and Caballero, R. (2016) The role of moist intrusions in winter Arctic warming and sea ice decline. Journal of Climate, 29 (12): 4473–4485, doi: 10.1175/jcli-d-15-0773.1

First, what is going on? The map below (Figure 1a) shows the daily mean temperature forecast for December 8th compared to the 1981–2010 average. Cold air (blue) is seen to flow south from Alaska down to the Rocky Mountains. The map highlights the day the coldest air extends farthest south.

A few days later the cold air is forecast to reach all the way down to the Southeast as shown below (Figure 1b), but with less intensity. Western Canada remains cold throughout.

Such “cold air outbreaks” are a common occurrence in winter in North America. To determine how rare such an event is we consider two stations that are forecast to be in the center of the cold weather: Boulder, Colorado and New Orleans, Louisiana. Both these stations have daily observations going back to 1893, albeit with a few missing years.

The Boulder forecast for daily mean temperature on 8 December is 3.9ºF (-15.6 ºC). This is definitely a cold day, but not very unusual there: the coldest day of the winter is on average 3.4ºF (-15.9 ºC). Almost every other year has a day that is colder than today’s forecast. There is a positive trend in the temperature of the coldest day in the year (Figure 2), but because of the big differences in weather from year to year it is not significantly different from zero.

In New Orleans (Figure 3), the forecast for 10 December is a daily mean temperature of just 45ºF (7.2 ºC). This is nothing unusual, it gets colder than that every winter. The coldest day of the year there has a very significant upward trend: cold extremes are now on average 3.2ºF (1.8 ºC) warmer than they were a century ago.

Cold air outbreaks are influenced by two main factors (de Vries et al., 2012): the frequency and intensity of cold northerly winds, and the temperature of the cold air from the Arctic. There are no indications that we are getting more, stronger or more persistent northerlies above the large variability from winter to winter (Screen & Simmonds 2013), whereas observations show that the temperature of the source regions of the cold air is increasing rapidly (Screen, 2014; van Oldenborgh et al., 2015).

Climate models confirm this analysis, showing a strong warming trend in coldest day of the year in the midlatitudes. In these models this also reflects the strong warming of the Arctic. They also show little influence of the sea ice extent on the circulation over North America. While some research suggests that cold air outbreaks could have become more frequent as a result of melting Arctic sea ice (Francis and Vavrus, 2012), the majority of studies find no influence of sea ice at all (e.g., references in Jung et al., 2015). A recent paper by Screen et al (2015) finds that decreasing sea-ice cover results in fewer cold extremes over central and eastern North America. That said, there is some evidence for the sea ice decline strengthening the Siberian high somewhat (Mori et al, 2014) leading to colder extremes in that region (Screen et al, 2015).

In conclusion: the cold air outbreak of the next few days is nothing unusual, and neither inconsistent with an overall picture of a warming world, nor evidence that global warming is making cold weather more extreme. Such cold air outbreaks are, in fact, decreasing in intensity both in observations and climate models primarily because the source of the cold air, the Arctic, is warming strongly.

References

Francis, J.A. and S.J. Vavrus (2012) Evidence linking Arctic amplification to extreme weather in mid-latitudes. Geophysical Research Letters, 39(6): L06801. doi: 10.1029/2012GL051000

Jung, T., Doblas-Reyes, F., Goessling, H., Guemas, V., Bitz, C., Buontempo, C., Caballero, R., Jakobson, E., Jungclaus, J., Karcher, M., Koenigk, K., Matei, D., Overland, J., Spengler, T. and Yang, S. (2015) Polar lower-latitude linkages and their role in weather and climate prediction. Bulletin of the American Meteorological Society, 96: ES197–ES200. doi: 10.1175/BAMS-D-15-00121.1

Mori, M., Watanabe, M., Shiogama, H., Inoue, J. and Kimoto, M. (2014) Robust Arctic sea-ice influence on the frequent Eurasian cold winters in past decades. Nature Geoscience, 7: 869–873. doi: 10.1038/ngeo2277

Van Oldenborgh, G.J., R. Haarsma, H. De Vries, en M.R. Allen. (2015) Cold extremes in North America vs. mild weather in Europe: the winter of 2013–14 in the context of a warming world. Bulletin of the American Meteorological Society, 96: 707–714, doi: 10.1175/BAMS-D-14-00036.1

Screen, J.A. and AI. Simmonds. (2013) Exploring links between Arctic amplification and mid-latitude weather. Geophysical Research Letters, 40: 959–964. doi: 10.1002/grl.50174

Screen, J.A. (2014) Arctic amplification decreases temperature variance in northern mid- to high-latitudes. Nature Climate Change, 4: 577–582. doi: 10.1038/nclimate2268

Screen, J. A., Deser, C. and Sun, L. (2015) Projected changes in regional climate extremes arising from Arctic sea ice loss. Environmental Research Letters, 10: 084006. doi: 10.1088/1748-9326/10/8/084006

de Vries, H., Haarsma, R. and Hazeleger, W. (2012) Western European cold spells in current and future climate. Geophysical Research Letters, 39: L04706.