1. Introduction
Changes in Arctic climate extremes driven by anthropogenic emissions are of great societal relevance (e.g., Walsh et al. 2020). Several important time scales can be distinguished in defining climate extremes. For example, at annual-to-monthly scales, Arctic (poleward of 60°N) surface air temperature (SAT) has increased in recent decades as a result of global warming, with 2016, 2019, and 2020 ranking as the three warmest years in the more than 100-yr instrumental record of measurements (IPCC 2022). Regionally, record-breaking hot months, seasons, and years were observed over northern Europe, Siberia, Greenland, and North America in recent decades (e.g., Walsh et al. 2020).
There is widespread agreement in the literature that the magnitude, frequency, and duration of daily-scale warm (cold) extremes increase (decrease) because of background warming and sea ice loss (e.g., Sui et al. 2017; Bieniek and Walsh 2017; Thoman and Walsh 2019; Walsh et al. 2020; Overland 2021). For example, Graham et al. (2017) analyzed a combination of floating buoys and ERA-Interim and found a positive trend in the duration of winter warming events with air temperatures above −5°C near the North Pole (4.25 days decade−1) and in the Pacific sector (1.16 days decade−1) of the Arctic. They also discovered that the North Pole’s warmest midwinter temperatures have been rising at a rate that is twice as fast as the warming of the region’s mean midwinter temperatures. Moore (2016) evaluated multiple reanalyses and argued that midwinter warming occurrences (SAT > 0°C) near the North Pole, which occur once or twice per decade, are linked to strong surface cyclones and a highly disturbed polar vortex that shifted the jet stream toward the high Arctic. The rate of increase for the warmest midwinter temperatures at the North Pole has been twice as fast as the pace for mean midwinter temperatures there, a finding corroborated by Graham et al. (2017). According to Matthes et al. (2015), who examined weather station and ERA-Interim 1979–2013 data, there has been a general drop in Arctic cold spells in winter and summer of up to 4 days decade−1. Changes in warm spells have largely been small (less than 1 day decade−1) and statistically insignificant throughout the Arctic. Sui et al. (2017) showed particularly large and statistically significant decreases in cold nights and cold days in Canada while particularly large and statistically significant increases in warm nights and warm days in northwest Eurasia using the Climate Data Extremes (CLIMDEX) metrics for climate extremes over the 1979–2015 period. Significant heat waves affected Siberia (2020), northern Europe (2018, 2019, 2021), Alaska (2016), and other high-latitude locations in recent decades (e.g., Walsh et al. 2020).
However, there are complicating factors that make evaluating climate extremes difficult. From the statistical standpoint, it is more difficult to discern significant trends in extremes because they are rare. Physically, there are a host of factors that control climate extremes. Extreme temperatures, for instance, are influenced by intricate linkages and feedback mechanisms in the high-latitude system. A mechanism by which the planetary boundary layer modulates the SAT response to changes in the surface energy balance may explain the observed diurnal asymmetry in the global warming trend, where nighttime temperatures have increased more quickly than daytime temperatures (Davy et al. 2017). This is because nighttime temperatures are intrinsically more sensitive to changes in the radiation balance and will rise faster in the presence of a uniform forcing (such as that caused by the accumulation of greenhouse gases). The most remarkable is how sensitive subseasonal-to-seasonal temperature extremes are to sea ice loss in the Arctic (Screen et al. 2015; Hartmuth et al. 2022). Climatic linkages across multiple time scales, complicated by evolving Arctic boundary conditions (i.e., diminishing and thinning sea ice conditions), make it challenging to identify and characterize the nature of climate extremes. Note also that climate extremes can be defined in a variety of ways, further complicating their analysis. Through this work we analyze ERA5 to update monthly, seasonal, and annual trends of pan-Arctic SAT extremes and pinpoint periods, as well as provide analysis, of the most extreme daily temperature events since 1979.
2. Data and methods
In this study, we use atmospheric reanalysis data from the European Centre for Medium-Range Weather Forecasts fifth-generation atmospheric reanalysis (ERA5) covering 1979–2021 (Hersbach et al. 2020). This period provides sufficient temporal coverage to resolve long-term trends and several modes of the observed atmospheric variability and overlaps with the period of satellite sea ice observations. Ballinger et al. (2023) and references therein justify the usage of this dataset over other atmospheric reanalyses.
a. A brief description of ERA5 atmospheric reanalysis
ERA5 hourly surface air temperatures (SAT), sea level pressure (SLP), 10-m winds, and radiative and turbulent surface heat fluxes data were downloaded from the Copernicus Climate Data Store and used in this study. The horizontal resolution of the data is 0.25°. The area for the analysis was limited by 60°N latitude. Pan-Arctic (>60°N) time series were created by averaging gridded data, taking into account the area of grid cells.
b. Estimating quality of SAT from ERA5
Monthly SAT data spanning 1875–2017 from terrestrial weather stations (referred to in Fig. 1 and Table 1 as “Meteo”) were used to create pan-Arctic (>60°N) annual and seasonal time series using the Bekryaev et al. (2010) archive of SATs updated by recent data. Spatiotemporal Meteo data coverage is shown in Figs. 1b and 1c. The pan-Arctic Meteo-based anomalous SAT time series were composed using the box method (Bekryaev et al. 2010). This method involved dividing the entire area > 60°N into boxes (10° latitude × 20° longitude), and averaging anomaly (relative to 1991–2020 means) time series from individual meteorological stations inside each box. To create a single pan-Arctic time series, the average time series that resulted for each grid box were again averaged taking into account the area of each box. The sensitivity of the box method to the choice of parameters was discussed in Bekryaev et al. (2010). The trends of the annual and seasonal SAT time series, which were computed using boxes with each side halved to reduce the box area by a factor of 4, are compared here. The results are statistically similar to those in Fig. 1. Additionally, a large number of stations have records that start in the 1960s and 1970s (Fig. 1c). As such, our trend analyses do not show discernable sensitivity when we exclude stations with records of <50 years.
Time series of surface air temperature (SAT) (a) annual and (b)–(e) seasonal (DJF, MAM, JJA, SON) anomalies (relative to 1991–2020 means) from meteorological stations (black) and ERA5 (blue) for the region poleward of 60°N. In (a) and (d)–(g), black dotted and solid lines are used for unsmoothed and 7-yr running means. In (b), the map shows locations of meteorological stations. In (c), the time series shows the number of active stations for each year since 1875 with >700 stations in the 1990s to 2000s.
Citation: Journal of Climate 37, 8; 10.1175/JCLI-D-23-0266.1
Comparison of ERA5 and Meteo Arctic (>60°N) SATs for 1979–2017.
The Meteo dataset has been only partially assimilated into ERA5 (R. Bekryaev 2023, personal communication), and therefore is helpful in validating the quality of the ERA5 SATs over land areas. For example, Fig. 1 demonstrates a good match between the annual and seasonal time series of SAT anomalies derived from the two products, attesting to the robust quality of the ERA5 data. Table 1 shows further evaluation and validation of ERA5 quality, showing comparable means, standard deviations (SD), and trends as with the Meteo dataset for overlapping 1979–2017 years. Moreover, all seasonal and annual ERA5 time series have very similar SD and trends and are highly correlated (R > 0.89) with the Meteo time series. The high quality of the trends derived from ERA5 data is critically important for our analysis of the time evolution of the Arctic SAT and its extremes. Warm biases in ERA5 at annual and seasonal scales have been noted. However, Arctic terrestrial and marine lower-tropospheric air temperature biases in ERA5 have been shown to be lower than other atmospheric reanalyses (Graham et al. 2019; Avila-Diaz et al. 2021). Note also that the SAT trends estimated using ERA5 for 1979–2017 (0.62° ± 0.15°C decade−1) and 1979–2021 (0.61° ± 0.11°C decade−1) are statistically indistinguishable.
c. Defining extremes
There are numerous publications on the topic of Arctic meteorological extremes (e.g., Frich et al. 2002; Zhang et al. 2011; Matthes et al. 2015; Messori et al. 2018; Walsh et al. 2020). Perhaps the most popular datasets analyzed on this theme are the Expert Team on Climate Change Detection and Indices (ETCCDI), also referred to as Climate Data Extremes (CLIMDEX). These indices are described in Sillman et al. (2013) and have been used in many past studies and climate assessments, perhaps most notably the Fifth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC 2013).
Following these publications, several climate extreme indices have been developed and analyzed in this study, including daily maximum temperature (DMAT) and daily minimum temperature (DMIT) indices derived from hourly ERA5 SAT for each grid point northward of 60°N. The DMIT and DMAT daily time series were also used to compute various percentiles to define climate extremes’ thresholds.
For our analysis, we selected definitions of annual and seasonal (monthly) extreme indices that form continuous annual or seasonal time series that are particularly helpful in the examination of the time evolution of SAT extremes. Examples of such indices used here are annual and monthly indices of highest DMAT (TXx; this and other index notations are standard; e.g., Zhang et al. 2011), lowest DMIT (TNn), and counts of days when DMAT is greater than the 90th percentile (TX90p) and DMIT is less than the 10th percentile (TN10p). Note that our definition of the TX90p and TN10p differs slightly from what was used in Zhang et al. (2011) in that we used the count of days (which is more explicit), whereas they used a percentage of time. The analysis of the time evolution of the warm extreme TXx and TX90p indices and the cold extreme TNn and TN10p indices represented by continuous annual and/or monthly time series is straightforward via, for example, the calculation of their linear trends using the least squares best-fit method. Pan-Arctic (>60°N) time series of the climate extremes were obtained by averaging gridded data, taking into account the area of grid cells.
d. Defining seasonal cycle
Detection of change in time of seasonal SAT variations was critically important for this study. Wavelet analysis was used to define the amplitudes and phases of the time-variable seasonal signal, which were estimated using wavelet transformations of the time series. The standard package of wavelet programs is used for the calculation of wavelet transforms based on the DOG Mother function (Torrence and Compo 1998). The 95% confidence intervals and cones of influence shown in wavelet presentations are provided by the same package. Since wavelet analysis generally does not produce an estimate of the spectral function precisely at an annual period, we generated the annual series by interpolating the series of spectral densities with periods closest to the annual one (0.929 and 1.105 years for ERA5 hourly SAT). Time series of wavelet power spectra at the annual period regressed on the original time series defined the sought-after seasonal signal. An example of using the wavelet method in defining the seasonal signal and deseasonalizing for pan-Arctic (>60°N) SAT time series is shown in Fig. S1 in the online supplemental material. This approach has been used in previous studies (e.g., Polyakov et al. 2020). When needed, we produced a daily-resolution time-constant seasonal cycle by averaging daily data across all years of the 1979–2021 record.
Removing a time-constant seasonal cycle (for that, all available values for each day of the year throughout the record were averaged), the residuals inform how the index is changing independent of the time of year. Thus, the residuals show us how the index is evolving independently of the season using a time-constant seasonal cycle. We investigate how these extremes are altering in light of modifications to the seasonal cycle brought on by climate change using a time-variable seasonal cycle.
e. Defining warm and cold spells
In addition to continuous SAT extreme indices (time series), discrete temporal warm and cold spells were analyzed. They were defined as events with at least 5 consecutive days when DMAT > 99.9th percentile (DMIT < 0.01th percentile) of 1979–2021 for annual time series and 99.0th and 1.0th percentiles for monthly time series. To obtain a sufficient number of spells for shorter monthly time series, different percentiles for annual and monthly time series are utilized. Monthly DMIT and DMAT time series were transformed into anomalies by subtraction of the time-constant seasonal cycle in order to avoid a significant seasonal trend of DMIT and DMAT within each month (especially during transition seasons, such as April or May, when the highest temperatures are found at the end of the month). Annual counts of at least several consecutive days of maximum temperature > 90th percentile or minimum temperature < 10th percentile are often used for analysis of warm and cold spells (e.g., Zhang et al. 2011). However, the higher percentiles used in our study for annual and seasonal time series ensured that outstanding Arctic annual or seasonal events were identified and analyzed. Warm spells of this magnitude and duration are often called heat waves (IPCC 2022).
In addition, the impact of the SAT seasonal signal reduced by the increasing moderating role of the Arctic Ocean in structuring cold and warm spells was investigated. To do this, all identified spells were examined to see if they remained identifiable when the modulation of the SAT seasonal amplitude was removed from the time series. This was accomplished by subtracting the time-variable seasonal signal from each time series and adding the time-constant (averaged in time) seasonal signal and repeating the spell identification procedure. In Figs. 12 and 13, red circles show spells that were identified for time series with both retained and removed modulation of the seasonal SAT amplitudes, whereas green circles represent spells that were not identified when the modulation of the seasonal amplitude was eliminated (and therefore they were caused by this modulation).
f. Computing probability density functions and cumulative distribution functions
The available data range was separated into bins, and the number of values falling into each bin was counted. The histogram was defined by the distribution of the number of counts within each bin. A series of sensitivity experiments determined the number of bins (and their width) (Wilks 2011). The probability density function (PDF) underlying the same set of discrete observations was then calculated using the kernel density estimator method based on the Epanechnikov kernel (Ouyang 2005) and compared with the histogram. We found a very close match between the histogram and the nonparametric technique utilizing the nonparametric kernel estimator. By integrating PDF values less than or equal to the provided bin value, cumulative distribution functions (CDFs) were generated.
g. Computing and plotting statistical significance of trends
The least squares best-fit method was used to assess trends in the SAT and climatic extreme time series. Student’s t statistic with a 95% confidence interval is used to assess the statistical significance of trends (Brooks and Carruthers 1953). Furthermore, multiple hypothesis testing was assessed by correcting for the false discovery rate for monthly trends of SAT extremes with 0.05 as the global significance threshold (e.g., Wilks 2016). When this technique was used, it resulted in additional rejections of locally (at every grid box) statistically significant (e.g., having a p value less than 0.05) trends in climate extremes. For example, the percentage of rejected local statistically significant trends ranged from 3% in November TXx to more than 15% in February TXx, with an average of 8.6% of statistically significant local TXx trends discarded.
3. Arctic surface air temperature extremes
The selected 1979–2021 extreme Arctic indices were dominated by linear statistically significant trends showing increasing highest and lowest daily temperatures (TXx, TNn indices) and the number of days when temperature exceeded the 90th percentile (TX90p) and decreasing the number of days when the temperature was lower than the 10th percentile (TN10p) (Fig. 2). It is important for our further discussion that the trend of the highest summer temperatures (TXx) of 0.46° ± 0.21°C decade−1 (Fig. 2), although strong, was lower than the annual SAT trend of 0.62° ± 0.15°C decade−1 (∼25% reduction) whereas the trend of the lowest winter temperatures (TNn) of 1.13° ± 0.22°C decade−1 was very strong, well exceeding (almost doubling) annual SAT trends (Fig. 2, Table 1). Consistently, the magnitude of the trend in the TN10p index is 20% larger than the magnitude of the trend in the TX90p index. This is strongly compatible with the dampening of the SAT seasonal cycle modulated by warming as illustrated in Fig. 3.
Pan-Arctic 1979–2021 annual time series of the highest and lowest daily (a) maximum and (b) minimum SAT (TXx and TNn, respectively), and counts of days when (c) daily maximum SAT is greater than the 90th percentile (TX90p) and (d) daily minimum SAT is less than the 10th percentile (TN10p). Anomalies are relative to 1991–2020 means. Trends (also shown by green lines) and means are for 1979–2021.
Citation: Journal of Climate 37, 8; 10.1175/JCLI-D-23-0266.1
Idealized schematic illustrating different rates of changes of SAT and maximum and minimum SAT extremes. SAT, represented by a combination of the warming trend and seasonal signal, shows dampened amplitude with time. This is the result of strong amplification of the minimum temperature trend (dash–dotted red line) relative to annual SAT trend (red solid line) and weakened trend of maximum temperature (dashed red line).
Citation: Journal of Climate 37, 8; 10.1175/JCLI-D-23-0266.1
Probability distribution functions (PDFs) and cumulative distribution functions (CDFs) for the TNn and TXx indices are shown in Fig. 4. There is a clear distinction between PDFs and CDFs composed for the winter, summer, and transition (i.e., spring and autumn) months. For example, the CDF has the steepest slope in summer [June–September (JJAS); e.g., the transition from the 10th to the 90th percentile does not exceed 1°C in July]. This is a signature of the least variable, most stable summer weather (see Figs. 5 and 6, inserts). The winter [December–March (DJFM)] CDF slope is not as steep than that in summer (∼5°C transition from the 10th to the 90th percentile). The smallest slope (∼6.0°/7.5°C transition, signifying a wide spread of values due to very unstable weather conditions) is during the transitional seasons of spring [April–May (AM)] and fall [October–November (ON)]. A reflection of these unstable conditions is a multipeak PDF in AM and ON (Fig. 4).
Probability distribution functions (PDFs) and cumulative distribution functions (CDFs) for the pan-Arctic monthly time series of the highest (TXx; red) and lowest (TNn; blue) daily maximum and minimum SAT averaged over 1979–99 (solid lines) and 2000–21 (broken lines). The 10th and 90th percentiles are indicated by broken horizontal lines.
Citation: Journal of Climate 37, 8; 10.1175/JCLI-D-23-0266.1
Maps of 1979–2021 linear trends (°C decade−1) of the monthly highest maximum daily temperature (TXx) index. Monthly mean TXx is indicated in the bottom-right corner of each panel. Areas with statistically insignificant trends—typically, those with weak trends—do not display color. Pan-Arctic time series of anomalous TXx (anomalies are relative to the 1991–2020 mean; blue lines) and their trends (red lines and digits) are shown in inserts (the vertical axis is in °C).
Citation: Journal of Climate 37, 8; 10.1175/JCLI-D-23-0266.1
Maps of 1979–2021 linear trends (°C decade−1) of the monthly lowest daily minimum temperature (TNn) index. Monthly mean TNn is indicated in the bottom-right corner of each panel. Areas with statistically insignificant trends—typically, those with weak trends—do not display color. Pan-Arctic time series of anomalous TNn (anomalies are relative to the 1991–2020 mean; blue lines) and their trends (red lines and digits) are shown in inserts (the vertical axis is in °C).
Citation: Journal of Climate 37, 8; 10.1175/JCLI-D-23-0266.1
A transition of PDF and CDF in time from 1979–99 to 2000–21 is very similar between TXx and TNn, which is probably a reflection of the overall similar impact of multidecadal warming on the gradual increase of minimum and maximum temperatures (Fig. 4). The lack of asymmetry in Arctic summer and winter daily temperatures prevents realization of the mechanism by which the planetary boundary layer modulates the SATs as proposed by Davy et al. (2017). Therefore, further analysis is required to identify the processes associated with the absence of asymmetry in transition seasons. A parallel shift of 1979–99 versus 2000–21 CDF is used for an analysis of temporal changes by considering changes at a selected percentile. Here, we use the 50th percentile. The most pronounced temporal changes are found for the cold season (JFMA and OND), with subtle changes for the warm [May–September (MJJAS)] season. However, due to the steepest summer CDF slope, even the relatively small shift of the summer CDF over time results in significant changes in statistical percentiles. For example, the 90th percentile TNn from 1979 to 1999 becomes a 50th–60th-percentile TNn for all months in 2000–21 (Fig. 4). Thus, the hottest (>90th percentile) days in the 1980s to 1990s would be days with normal high daily temperatures in the 2000s to 2010s; the coldest (<10th percentile) days since 2000 would be considered days with close to normal low daily temperatures in the 1980s to 1990s.
The spatial distributions of TXx, TNn, TX90p, and TN10p trends are shown in Figs. 5–8 (see also Fig. 9 showing pan-Arctic monthly trends). The trends peak in September–November over the Arctic Ocean, with a hint of the fastest increase of the indices at the basin’s periphery, where sea ice is more dynamic and less compact. Accelerated loss of Arctic sea ice (e.g., Serreze and Barry 2011; Taylor et al. 2022; Stroeve and Notz 2018), particularly losses in September and October, results in the release of additional oceanic heat into the atmosphere through thin ice and leads (Fig. 9), causing regional atmospheric warming and an impact on air temperature extremes. Positive TX90p and negative TN10p trends over the Arctic Ocean are noteworthy (Figs. 7 and 8). Also note a clear winter maximum in TNn in the Barents Sea, emphasizing the role of the Barents Sea as an ongoing Arctic hotspot (Isaksen et al. 2022). In contrast, in January and February, TXx and TNn indices showed negative trends over the Eurasian continent, consistent with the warm Arctic–cold Eurasia pattern (e.g., Kim and Son 2020). In summer (JJA) in the central Arctic Ocean, the TXx and TNn trends are negative, albeit weak. This pattern of changes in the Arctic SAT extremes is consistent with the reduction of sea ice (Fig. 9) and emerging role of the Arctic Ocean as a moderator of SAT anomalies.
Maps of 1979–2021 linear trends of the monthly TX90p index (the day count when daily maximum temperature exceeded 90% threshold). Areas with statistically insignificant trends—typically, those with weak trends—do not display color. Pan-Arctic monthly time series and their trends are shown in inserts (the vertical axis is in days).
Citation: Journal of Climate 37, 8; 10.1175/JCLI-D-23-0266.1
Maps of 1979–2021 linear trends of the monthly TN10p index (the day count when daily minimum temperature was lower than 10% threshold). Areas with statistically insignificant trends—typically, those with weak trends—do not display color. Pan-Arctic monthly time series and their trends are shown in inserts (the vertical axis is in days).
Citation: Journal of Climate 37, 8; 10.1175/JCLI-D-23-0266.1
Monthly trends of pan-Arctic (>60°N) (top) mean sea ice concentration and net heat fluxes, (middle) highest daily maximum temperature (TXx) and lowest daily minimum temperature (TNn) indices, and (bottom) TX90p (the day count when daily maximum temperature exceeded the 90% threshold) and TN10p (the day count when daily minimum temperature was lower than the 10% threshold) indices. Shaded areas bounded by dotted lines show uncertainty evaluated by one standard error.
Citation: Journal of Climate 37, 8; 10.1175/JCLI-D-23-0266.1
4. Climate change modulation of the seasonal signal in shaping SAT extremes
Global climate models predict that rising greenhouse gas concentrations will induce a reduction of sea ice driving shift in the seasonal cycle of surface temperature, which will result in later minimum and maximum yearly temperatures and more warming in the winter than the summer (Dwyer et al. 2012). In this section, we provide evidence that the modern Arctic Ocean becomes more connected with the atmosphere as sea ice cover retreats in all seasons, which more effectively moderates Arctic seasonal SAT variations in recent decades than in the past. This phenomenon explains why the annual SAT trend for the period of 1979 to 2021 is greater than the trends of the annual “warm” TXx index and weaker than the trends of the annual “cold” TNn index.
According to Fig. 10, the seasonal pan-Arctic SAT fluctuation range between 1979 and 1999 was 2°C greater than that between 2000 and 2021. The seasonal ranges of daily highest (DMAT) and lowest (DMIT) Arctic temperature, as expected, exhibit nearly identical change in time compared with the SAT range between 1979–99 and 2000–21 (Fig. 10). Given that the overall SAT trend from both meteorological stations (i.e., Meteo) and ERA5 was 0.62°C for the period of 1979 to 2021 (Table 1), this is a striking, recent reduction in temperature range. Additionally, the 7% decline in seasonal SAT amplitude is very consistent with the 5% rate predicted for the twenty-first century by 24 models from phase 3 of the Coupled Model Intercomparison Project (CMIP3) (Dwyer et al. 2012). The seasonal SAT amplitude decrease during the twenty-first century, according to analysis of CMIP-3 model output, was primarily caused by sea ice loss. The authors showed that as sea ice melts, the previously unexposed open ocean increases the surface layer’s effective heat capacity and dampens the air temperature response via altering surface heat fluxes. This is consistent with our observational findings (see Fig. 9).
(left) Seasonal cycles and (right) their differences of the (top) SAT and (middle) maximum and (bottom) minimum daily pan-Arctic (>60°N) temperatures in 1979–99 (blue) and 2000–21 (red).
Citation: Journal of Climate 37, 8; 10.1175/JCLI-D-23-0266.1
That mechanism explains winter and summer SAT peak suppression over the Arctic Ocean which is clearly visible in the spatial patterns of monthly differences in the seasonal fluctuations of the SAT in 2000–21 compared to 1979–99 (e.g., positive anomalies in winter and negative anomalies in summer; Fig. 11). This is additional proof of the Arctic Ocean’s new function as a moderator, via enhanced air–sea interactions, of seasonal SAT variations, trends, and extremes consistent with the findings of Screen and Simmonds (2010) and Papritz (2020) and CMIP-3 modeling results (Dwyer et al. 2012). Moreover, there often appears a weak tendency for the ocean and land to have opposing signs in Fig. 11 (e.g., January). Increased ocean–atmosphere coupling also aids in the explanation of the monthly TXx and TNn trend patterns in Figs. 5 and 6 discussed in the previous section.
Monthly changes in time (2000–21 relative to 1979–99) of the seasonal SAT cycle (°C; SAT detrended). Regions with decreasing monthly SAT due to modulated seasonal signal amplitude are indicated by the blue color, whereas regions with increasing monthly SAT due to modulated seasonal signal amplitude are indicated by the red color.
Citation: Journal of Climate 37, 8; 10.1175/JCLI-D-23-0266.1
A comparison of monthly maps of changes in seasonal SAT amplitudes (Fig. 11) and the portion of the SAT extremes’ linear trends induced by seasonal amplitude fluctuations (Figs. S2–S5) adds to the evidence that the modulated seasonal cycle dominates the patterns of the TXx, TNn, TX90p, and TN10p trends. Pattern correlations between monthly fields provide a quantitative measure for these tight linkages. For example, the correlation varies from R > 0.9 for the peak summer [July–August (JA)] and winter [January–February (JF)] months to lesser but still high correlations (R > 0.7) in March–June and September–December for the TXx trend defined by seasonal cycle modulation. Similar correlations are obtained for the other climate extreme indices as well.
The patterns of the Arctic SAT extreme indices TXx and TNn, shown in Figs. 5 and 6, show modest correlation with the pattern of the modulated SAT seasonal signal (Fig. 11). For example, the cold extreme index TNn correlates at R = 0.52/0.73 in winter months from October to March, with lower correlations of R ∼ 0.22/0.27 in the summer months. Estimates of trends using monthly time series of the TNn and TXx indices are highly consistent with the spatial pattern, a finding one would expect from a warming trend in conjunction with the shrinking SAT seasonal cycle (e.g., Fig. 3), with generally damped trends of “warm” TXx index and amplified “cold” TNn index trend (cf. Fig. 5, inserts; see also Table 1). Thus, this analysis provides further confirmation of the importance of modulation of the seasonal signal for shaping climate extreme indices in the Arctic.
Finally, we note that the 1979–2021 changes in the SAT extremes driven by climate change are substantial. The reduction of the seasonal SAT signal likely due to the emerging moderating role of the Arctic Ocean has an important impact on the SAT extremes by doubling trends of the cold SAT extremes and dampening trends of the warm SAT extremes by ∼25% (see the first paragraph of section 3). This important finding is consistent with the early study by Sulikowska et al. (2019), who found that winter cold extremes in Alaska are decreasing faster than summer warm extremes are increasing although our study shows this phenomenon at the pan-Arctic scale.
5. Warm and cold spells
Warm (cold) spell events of exceptional strength and lasting at least 5 consecutive days when DMAT > 99.9th percentile (DMIT < 0.01th percentile) of 1979–2021 within the annual cycle and 99.0th and 1.0th percentiles within the seasonal time series were identified (Figs. 12 and 13). As expected, the two most extreme warm events (heat waves) were found in recent years since 2000, while the two most extreme cold events happened much earlier, in the 1970s and 1980s, reflecting the tendency of warm (cold) spells to be more likely in recent (earlier) decades. We briefly document and examine the synoptic settings of these most extreme events, and focus attention on days when these most warm and cold extremes impacted the largest respective areas in the Arctic.
14–21 July 2016: According to the NOAA National Centers for Environmental Information (2023), this year was the warmest on record across Earth. The Arctic region was not an exception, hitting the record-highest annual temperature (Ballinger et al. 2022), a finding corroborated by our analysis (Fig. 1a), which contributed to the development of a heat wave of exceptional strength in the Barents–Kara Seas region in July (Fig. 14). This heat wave intensified across northwestern Siberia and was subsequently advected northward over Barents-Kara waters by southerly winds. While over Siberia, the magnitude of the SAT anomalies were not exceptional. But once the continental air mass propagated over the Kara and Barents Seas, the associated SAT anomalies exceeded the 99.9th percentile aided by strong, downward net radiative fluxes collocated with high pressure and clear skies. This is a signature of the new Arctic when the loss of sea ice precedes heat waves over areas where in the past it was not possible due to sea ice constraining the near-surface air temperature to around 0°C. This finding is consistent with the results of Papritz (2020), who found that poleward advection of warm air masses and subsidence are primary contributors that shape exceptional summertime warm extremes in the Arctic.
23–27 July 2020: This summer is known for a scorching heat wave in June when the temperature in Siberia reached 38°C (Overland and Wang 2021). Just one month later, as it follows from our analysis, close-to-record-setting temperatures were found in the Canadian Arctic and in several areas surrounding the Barents Sea and in eastern Siberia (Fig. 15). The high pressure system associated with anticyclonic circulation led to clear skies and amplified positive (downward) radiative heat fluxes over the Canadian Archipelago. Anomalously high SATs in this area were then advected into the central Arctic, reminiscent of the synoptic situation of the July 2016 heat wave.
4–11 February 1979 and 21–25 January 1989: These two cold spells were associated with the coldest decades of the second half of the twentieth century (Fig. 1). During both spells, the cold air mass originated and was modified in different regions but followed a very similar synoptic pattern. Winter high pressure systems over central Siberia and the Canadian Archipelago (1979 cold spell) and eastern Siberia and Alaska (1989 cold spell) resulted in exceptionally strong heat loss over continents (Figs. 16 and 17). The centers of low SATs were formed there, and the cold polar air was then transported from these areas into the high Arctic to produce exceptionally cold weather conditions.
Warm spell events. Pan-Arctic (>60°N) time series of (top) annual daily maximum SAT and (bottom rows) their monthly anomalies relative to the time-constant seasonal cycle. Red dots in the top panel show two exceptional warm spells (>99.9th percentile) of 14–21 Jul 2016 and 23–27 Jul 2020; in addition, one event (12–18 Jul 1991) is identified for the first (colder) half of the time series using 99.9th percentile for this half (red-yellow dot). Monthly warm spells of exceptional strength (>99th percentile) are represented by red and green dots in the bottom panels. Green dots show spells that were not identified when the seasonal amplitude modulation was removed (for more information, see section 2). Threshold values are indicated in red font.
Citation: Journal of Climate 37, 8; 10.1175/JCLI-D-23-0266.1
Cold spell events. Pan-Arctic (>60°N) time series of (top) annual daily minimum SAT and (bottom rows) their monthly anomalies relative to the time-constant seasonal cycle. Red dots in the top panel show two exceptional cold spells (<0.1th percentile) of 4–11 Feb 1979 and 21–25 Jan 1989; in addition, one event (9–13 Feb 2002) is identified for the second (warmer) half of the time series using 0.1th percentile for this half (red-yellow dot). Monthly cold spells of exceptional strength (<1st percentile) are represented by red and green dots in the bottom panels. Green dots show spells that were not identified when the seasonal amplitude modulation was removed (for more information, see section 2). Threshold values are indicated in red font.
Citation: Journal of Climate 37, 8; 10.1175/JCLI-D-23-0266.1
Arctic heat wave on 14–21 Jul 2016. (top four rows, from left to right) The daily mean area of the warm spell (red) and sea ice edge (blue line), SAT, SLP, and geostrophic winds, net radiative [shortwave (SW) and longwave (LW)] heat fluxes, and turbulent heat fluxes [sensible heat (SH) and latent heat (LH)]. Heat fluxes are positive downward. (bottom two rows) The same five parameters, but zoomed in on the area of spell in the Barents–Kara Seas region.
Citation: Journal of Climate 37, 8; 10.1175/JCLI-D-23-0266.1
Arctic heat wave on 23–27 Jul 2020. (top four rows, from left to right) The daily mean area of the warm spell (red) and sea ice edge (blue line), SAT, SLP and geostrophic winds, net radiative (SW and LW) heat fluxes, and turbulent (SH and LH) heat fluxes. Heat fluxes are positive downward. (bottom two rows) The same five parameters, but zoomed in on the area of spell in the Canadian Arctic region.
Citation: Journal of Climate 37, 8; 10.1175/JCLI-D-23-0266.1
Arctic cold spell on 4–11 Feb 1979. (top four rows, from left to right) The daily mean area of the warm spell (red) and sea ice edge (blue line), SAT, SLP and geostrophic winds, net radiative (SW and LW) heat fluxes, and turbulent (SH and LH) heat fluxes. Heat fluxes are positive downward. (bottom two rows) The same five parameters, but zoomed in on the area of spell in the Kara Sea and neighboring Siberia region.
Citation: Journal of Climate 37, 8; 10.1175/JCLI-D-23-0266.1
Arctic cold spell on 21–25 Jan 1989. (top four rows, from left to right) The daily mean area of the warm spell (red) and sea ice edge (blue line), SAT, SLP and geostrophic winds, net radiative (SW and LW) heat fluxes, and turbulent (SH and LH) heat fluxes. Heat fluxes are positive downward. (bottom two rows) The same five parameters, but zoomed in on the area of spell in the Alaska–Beaufort Sea region.
Citation: Journal of Climate 37, 8; 10.1175/JCLI-D-23-0266.1
Placing these analyses in a longer-term perspective, we additionally evaluated one warm spell from the first half of the record (July 1991) and one cold spell from the second half of the record (February 2002), which were identified using the 99.9th and 0.1th percentiles for the corresponding periods (Figs. S6 and S7).
-
12–18 July 1991: The factors that were driving this warm spell are reminiscent of those of the aforementioned 2000s warm spells, when a high pressure system developed in conjunction with strong downward radiative fluxes dominated by shortwave radiation. This setting created anomalous warming across eastern Siberia, and this warm air mass was transported over adjacent marginal seas by offshore winds.
-
9–13 February 2002: With the establishment of a high pressure system and abnormally large heat losses by longwave radiation resulting to exceptionally low temperatures, the overall course of the 2002 cold spell is similar to that of the 1979 cold spell (including the region) (cf. Fig. 16 and Fig. S7). Although the area of the cold spell in 2002 is smaller than it was in 1979 and the absolute SAT is not as low, the atmospheric causes of the event appear to be very similar.
These findings are consistent with the analysis of Papritz (2020), who found no systematic changes in the relative importance of various processes driving the high Arctic (>80°N) lower-tropospheric temperature extremes in 1979–2017. Thus, following Papritz (2020), one can say that the mechanisms causing the Arctic’s warm and cold spells are consistent through time, but that the intensity of the warm and cold spells has changed due to amplified, background thermodynamic forcing (e.g., warmer near-surface temperatures, less sea ice, and more open water).
Finally, we note that monthly spells were significantly affected by the modulated seasonal signal. For example, 51 exceptional monthly warm spells (99th percentile) were identified, 57% of which would not have been identified as exceptional if the variable seasonal signal had been eliminated (Fig. 12). Similarly, 78 monthly cold spells were observed, with 39% of them not being identified as extraordinary if the variable seasonal signal was eliminated (Fig. 13). Thus, the modulated seasonal signal of Arctic SAT played an important role in shaping exceptionally strong SAT anomalies.
6. Discussion and conclusions
Both ERA5- and Meteo-based SAT time series are characterized by a strong upward trend of +0.62°C decade−1 in 1979–2017 (Fig. 1). Due to this Arctic warming, the hottest days in the 1980s to 1990s would be considered “normal” days now, and the coldest days of recent years would be considered normal days in the 1980s–90s.
Yet, several modes of Arctic climate variability on broad spatiotemporal scales complicate the warming. Multidecadal fluctuations are particularly noteworthy (with positive anomalies in the 1930s to 1940s and recent decades and negative anomalies in the early part of the twentieth century and in the 1960s to 1970s; Fig. 1). Large-amplitude multidecadal variability hinders the analysis of recent trends. Seasonal oscillation is yet another statistically significant factor. Our analysis demonstrates that the range of pan-Arctic SAT seasonal variations decreased by ∼2°C, which constitutes a 7% reduction of the total SAT range over 1979–2021 (Fig. 10). We provide an observational perspective on the CMIP-3 model-based finding (Dwyer et al. 2012), which indicates that the new Arctic Ocean is acting as a moderator in reducing seasonal SAT fluctuations caused by sea ice loss and oceanic warming via dampening summer and amplifying winter warming near the surface, thereby reducing the seasonal signal. This effect of the new Arctic Ocean, characterized by diminished ice area and less thick ice cover, on seasonal SAT fluctuations grows stronger over time.
As a consequence, the dampened seasonal SAT signal has a considerable impact on the SAT extremes, amplifying the upward trends of the cold extremes by a factor of 2 compared to the annual SAT warming trend and dampening the upward trends of the warm SAT extremes by around 25%. Related to the same moderating effect of the Arctic Ocean, the strongest 1979–2021 TNn and TXx trends occur during fall; trends in summer over the Arctic Ocean are weak and mostly negative. The strongest TX90p and TN10p trends are also in fall consistent with temporal changes in sea ice freeze-up and melt onset.
Our analysis showed that the modulated seasonal signal in Arctic SAT played a significant role in shaping extraordinarily intense monthly cold and warm spells. Furthermore, we corroborated Papritz’s (2020) earlier conclusion that the mechanisms underlying the Arctic’s warm and cold spells have not changed consistently over time, but that the intensity of the warm and cold spells has varied significantly in association with the strong moderation of the Arctic SAT seasonal cycle with time.
The Arctic Ocean’s moderating impact on the seasonal fluctuations of air temperatures is likely to increase in a warming climate. Crawford et al. (2021) analyzed climate models comprising phase six of the Coupled Model Intercomparison Project (CMIP6) and found that 2°C warming above the 1850–1900 mean coincident with sea ice melt out to 2100 will likely give way to larger areas and longer interannual periods (e.g., ∼2–3 months yr−1) when the Arctic Ocean surface is exposed to the atmosphere (Crawford et al. 2021). As a result, the rate of growth of the Arctic winter SAT extremes may continue to accelerate, whereas the rate of increase of the summer SAT extremes may be rather modest. A recent study by Hartmuth et al. (2023) using the Community Earth System model large ensemble (CESM-LE) similarly found future SAT reductions due to sea ice loss, most notably in the Barents Sea hotspot. However, the summer SAT extreme increase may accelerate in the future since the seasonal signal likely will not continue to decline indefinitely, posing significant challenges to projecting temperature extremes in a warming climate.
Acknowledgments.
This study was supported by NSF Grant 1724523 (IP) and ONR Grant N00014-21-1-2577 (TB, XZ, IP).
Data availability statement.
The ERA5 data are available from https://cds.climate.copernicus.eu/cdsapp#!/home.
REFERENCES
Avila-Diaz, A., D. H. Bromwich, A. R. Wilson, F. Justino, and S.-H. Wang, 2021: Climate extremes across the North American Arctic in modern reanalyses. J. Climate, 34, 2385–2410, https://doi.org/10.1175/JCLI-D-20-0093.1.
Ballinger, T. J., and Coauthors, 2022: Surface air temperature [in “State of the Climate in 2021”]. Bull. Amer. Meteor. Soc., 103, S264–S268, https://doi:10.1175/BAMS-D-22-0082.1.
Ballinger, T. J., and Coauthors, 2023: Alaska terrestrial and marine climate trends, 1957–2021. J. Climate, 36, 4375–4391, https://doi.org/10.1175/JCLI-D-22-0434.1.
Bekryaev, R. V., I. V. Polyakov, and V. A. Alexeev, 2010: Role of polar amplification in long-term surface air temperature variations and modern Arctic warming. J. Climate, 23, 3888–3906, https://doi.org/10.1175/2010JCLI3297.1.
Bieniek, P. A., and J. E. Walsh, 2017: Atmospheric circulation patterns associated with monthly and daily temperature and precipitation extremes in Alaska. Int. J. Climatol., 37, 208–217, https://doi.org/10.1002/joc.4994.
Brooks, C. E. P., and N. Carruthers, 1953: Handbooks of Statistical Methods in Meteorology. Meteorological Office, 412 pp.
Crawford, A., J. Stroeve, A. Smith, and A. Jahn, 2021: Arctic open-water periods are projected to lengthen dramatically by 2100. Commun. Earth Environ., 2, 109, https://doi.org/10.1038/s43247-021-00183-x.
Davy, R., I. Esau, A. Chernokulsky, S. Outten, and S. Zilitinkevich, 2017: Diurnal asymmetry to the observed global warming. Int. J. Climatol., 37, 79–93, https://doi.org/10.1002/joc.4688.
Dwyer, J. G., M. Biasutti, and A. H. Sober, 2012: Projected changes in the seasonal cycle of surface temperature. J. Climate, 25, 6359–6374, https://doi.org/10.1175/JCLI-D-11-00741.1.
Frich, P., L. V. Alexander, P. Della-Marta, B. Gleason, M. Haylock, A. M. G. Klein Tank, and T. Peterson, 2002: Observed coherent changes in climatic extremes during the second half of the twentieth century. Climate Res., 19, 193–212, https://doi.org/10.3354/cr019193.
Graham, R. M., L. Cohen, A. A. Petty, L. N. Boisvert, A. Rinke, S. R. Hudson, M. Nicolaus, and M. A. Granskog, 2017: Increasing frequency and duration of Arctic winter warming events. Geophys. Res. Lett., 44, 6974–6983, https://doi.org/10.1002/2017GL073395.
Graham, R. M., S. R. Hudson, and M. Matarilli, 2019: Improved performance of ERA5 in Arctic gateway relative to four global atmospheric reanalyses. Geophys. Res. Lett., 46, 6138–6147, https://doi.org/10.1029/2019GL082781.
Hartmuth, K., M. Boettcher, H. Wernli, and L. Papritz, 2022: Identification, characteristics and dynamics of Arctic extreme seasons. Wea. Climate Dyn., 3, 89–111, https://doi.org/10.5194/wcd-3-89-2022.
Hartmuth, K., L. Papritz, M. Boettcher, and H. Wernli, 2023: Arctic seasonal variability and extremes, and the role of weather systems in a changing climate. Geophys. Res. Lett., 50, e2022GL102349, https://doi.org/10.1029/2022GL102349.
Hersbach, H., and Coauthors, 2020: The ERA5 global reanalysis. Quart. J. Roy. Meteor. Soc., 146, 1999–2049, https://doi.org/10.1002/qj.3803.
IPCC, 2013: Climate Change 2013: The Physical Science Basis. Cambridge University Press, 1535 pp.
IPCC, 2022: Climate Change 2022: Impacts, Adaptation, and Vulnerability. Cambridge University Press, 3056 pp.
Isaksen, K., and Coauthors, 2022: Exceptional warming over the Barents area. Sci. Rep., 12, 9371, https://doi.org/10.1038/s41598-022-13568-5.
Kim, H.-J., and S.-W. Son, 2020: Eurasian winter temperature change in recent decades and its association with Arctic sea ice loss. Polar Res., 39, 3363, https://doi.org/10.33265/polar.v39.3363.
Matthes, H., A. Rinke, and K. Dethloff, 2015: Recent changes in Arctic temperature extremes: Warm and cold spells during winter and summer. Environ. Res. Lett., 10, 114020, https://doi.org/10.1088/1748-9326/10/11/114020.
Messori, G., C. Woods, and R. Caballero, 2018: On the drivers of wintertime temperature extremes in the high Arctic. J. Climate, 31, 1597–1618, https://doi.org/10.1175/JCLI-D-17-0386.1.
Moore, G. W. K., 2016: The December 2015 North Pole warming event and the increasing occurrence of such events. Sci. Rep., 6, 39084, https://doi.org/10.1038/srep39084.
NOAA National Centers for Environmental Information, 2023: State of the climate: Global climate report for 2022. NOAA, accessed 18 January 2023, https://www.ncei.noaa.gov/access/monitoring/monthly-report/global/202213.
Ouyang, Z., 2005: Univariate kernel density estimation. Duke University, accessed 26 September 2023, http://www.stat.duke.edu/∼zo2/shared/research/readings/kernelsmoothing.pdf.
Overland, J. E., 2021: Rare events in the Arctic. Climatic Change, 168, 27, https://doi.org/10.1007/s10584-021-03238-2.
Overland, J. E., and M. Wang, 2021: The 2020 Siberian heat wave. Int. J. Climatol., 41, E2341–E2346, https://doi.org/10.1002/joc.6850.
Papritz, L., 2020: Arctic lower-tropospheric warm and cold extremes: Horizontal and vertical transport, diabatic processes, and linkage to synoptic circulation features. J. Climate, 33, 993–1016, https://doi.org/10.1175/JCLI-D-19-0638.1.
Polyakov, I. V., and Coauthors, 2020: Weakening of cold halocline layer exposes sea ice to oceanic heat in the eastern Arctic Ocean. J. Climate, 33, 8107–8123, https://doi.org/10.1175/JCLI-D-19-0976.1.
Screen, J. A., and I. Simmonds, 2010: The central role of diminishing sea ice in recent Arctic temperature amplification. Nature, 464, 1334–1337, https://doi.org/10.1038/nature09051.
Screen, J. A., C. Deser, and L. Sun, 2015: Projected changes in regional climate extremes arising from Arctic sea ice loss. Environ. Res. Lett., 10, 084006, https://doi.org/10.1088/1748-9326/10/8/084006.
Serreze, M. C., and R. G. Barry, 2011: Processes and impacts of Arctic amplification: A research synthesis. Global Planet. Change, 77, 85–96, https://doi.org/10.1016/j.gloplacha.2011.03.004.
Sillman, J., V. V. Kharin, X. Zhang, F. W. Zwiers, and D. Bronaugh, 2013: Climate extremes indices in the CMIP5 multimodel ensemble: Part 1. Model evaluation in the present climate. J. Geophys. Res. Atmos., 118, 1716–1733, https://doi.org/10.1002/jgrd.50203.
Stroeve, J., and D. Notz, 2018: Changing state of Arctic sea ice across all seasons. Environ. Res. Lett., 13, 103001, https://doi.org/10.1088/1748-9326/aade56.
Sui, C., Z. Zhang, L. Yu, Y. Li, and M. Song, 2017: Investigation of Arctic air temperature extremes at north of 60°N in winter. Acta Oceanol. Sin., 36, 51–60, https://doi.org/10.1007/s13131-017-1137-5.
Sulikowska, A., J. P. Walawender, and E. Walawender, 2019: Temperature extremes in Alaska: Temporal variability and circulation background. Theor. Appl. Climatol., 136, 955–970, https://doi.org/10.1007/s00704-018-2528-z.
Taylor, P. C., and Coauthors, 2022: Process drivers, inter-model spread, and the path forward: A review of amplified Arctic warming. Front. Earth Sci., 9, 758361, https://doi.org/10.3389/feart.2021.758361.
Thoman, R., and J. E. Walsh, 2019: Alaska’s Changing Environment: Documenting Alaska’s Physical and Biological Changes through Observations. H. R. McFarland, Ed., International Arctic Research Center, University of Alaska, 16 pp.
Torrence, C., and G. P. Compo, 1998: A practical guide to wavelet analysis. Bull. Amer. Meteor. Soc., 79, 61–78, https://doi.org/10.1175/1520-0477(1998)079<0061:APGTWA>2.0.CO;2.
Walsh, J. E., T. J. Ballinger, E. S. Euskirchen, E. Hanna, J. Mård, J. E. Overland, H. Tangen, and T. Vihma, 2020: Extreme weather and climate events in northern areas: A review. Earth-Sci. Rev., 209, 103324, https://doi.org/10.1016/j.earscirev.2020.103324.
Wilks, D. S., 2011: Statistical Methods in the Atmospheric Sciences. International Geophysical Series, Vol. 100, Academic Press, 676 pp.
Wilks, D. S., 2016: “The stippling shows statistically significant grid points”: How research results are routinely overstated and overinterpreted, and what to do about it. Bull. Amer. Meteor. Soc., 97, 2263–2273, https://doi.org/10.1175/BAMS-D-15-00267.1.
Zhang, X., L. Alexander, G. C. Hegerl, P. Jones, A. K. Tank, T. C. Peterson, B. Trewin, and F. W. Zwiers, 2011: Indices for monitoring changes in extremes based on daily temperature and precipitation data. Wiley Interdiscip. Rev.: Climate Change, 2, 851–870, https://doi.org/10.1002/wcc.147.
