The SSWC comprises data from six different reanalyses (Table 1): the National Aeronautics and Space Administration (NASA) Modern-Era Retrospective-analysis for Research and Applications version 2 (MERRA2), Japanese 55-year Reanalysis (JRA-55), European Centre for Medium-Range Weather Forecasts (ECMWF) 40-year Reanalysis (ERA-40), ECMWF Interim Reanalysis (ERA-interim), NOAA 20th Century Reanalysis version 2c (NOAA20CRv2c), and NOAA's National Centers for Environmental Prediction/National Center for Atmospheric Research (NCEP-NCAR I) reanalysis.

Reanalyses are derived from observations from multiple sources (including surface observations, aircraft, radiosondes, rocketsondes, and satellites) that are assimilated by global coupled land–atmosphere–ocean models to create spatially and temporally complete observational records. There are advantages and disadvantages of using reanalysis products for this database, as opposed to individual measurement sources or various stratospheric analyses. These analyses include that from the Freie Universitat Berlin, which produces a database of continuous daily gridded synoptic-scale analyses based largely on radiosonde measurements, but only for three stratospheric levels for a 35-year period (Labitzke and Collaborators, 2002), and from the NOAA Climate Prediction Center (CPC), which offers analyzed stratospheric temperatures at eight stratospheric levels based on satellite retrievals of the advanced microwave sounding unit (AMSU). The major advantage of reanalysis is that it allows consideration of the evolution of SSWs and their impacts throughout the entire atmosphere with a spatial and temporal extent that is not feasible using individual measurements or stratospheric analyses alone. A major disadvantage of using reanalysis is that due to sparse observations, particularly in the pre-satellite era, stratospheric reanalysis is poorly constrained, especially above 10 hPa (Manney et al., 2003), and tropospheric reanalysis may be poorly constrained over oceans and remote regions (e.g., Bosilovich et al., 2008). Reanalyses can also suffer from upper-boundary effects and discontinuities due to model streams or changes in the observations being assimilated (Fujiwara et al., 2016; Labitzke and Kunze, 2005). These issues should not have a strong effect on the daily-to-seasonal timescales documented in the SSWC, but should be kept in mind, especially for data above 10 hPa where the discontinuities are conspicuous.

Some biases and uncertainties in individual reanalysis products have been documented (see references in Table 1), and an evaluation of their stratospheric processes is currently the focus of an international effort by the Stratosphere-troposphere Processes And their Role in Climate (SPARC) Reanalysis Intercomparison Project (S-RIP; Fujiwara et al., 2016). While initial studies have shown that stratospheric dynamics and variability of and coupling to the surface are reasonably simulated in reanalyses (Martineau and Son, 2010), particularly in the latest generation products (Martineau et al., 2016), the SSWC enables quick comparison between reanalyses of sudden stratospheric warming events and their evolution on daily timescales. This capability is important when considering the substantial volume of data needed to calculate the daily climatology and anomalies for each grid point and pressure level in each reanalysis.

Certain reanalysis output provided in the SSWC should be used with caution. For example, we provide the reanalysis ozone mass mixing ratio and total column ozone output (where available) since there are interesting changes in ozone following a SSW event (e.g., Fig. 3). However, users should be aware that most reanalysis ozone fields are based on assimilated satellite measurements that utilize backscattered sunlight and cannot measure ozone during polar night. Reanalysis systems thus rely heavily on the model, which typically parameterizes heterogeneous chemistry, to simulate ozone at high latitudes, leading to potentially high errors (Dethof and Hólm, 2004; Dragani, 2011).

In addition, the evolution of SSW events prior to 1964, when concentrated efforts to observe the upper atmosphere using radiosondes and rocketsondes were begun in association with the International Years of the Quiet Sun (IQSY), should be viewed with skepticism. Even radiosonde measurements of the stratosphere were very limited during that time period, and so reanalysis fields may be almost entirely model-driven.

The NOAA20CRv2c is unique among the reanalyses, because it assimilates only surface pressure observations. Thus, the stratosphere is not constrained by any stratospheric observations, and the reanalysis winds are not realistic (Compo et al., 2011). However, because surface pressure observations do a reasonable job of constraining the model throughout the northern hemispheric troposphere (Compo et al., 2011), we include the NOAA20CRv2c to examine the tropospheric impacts of SSWs, using SSW event dates given by the JRA-55 reanalysis (Table 2). The NOAA20CRv2c reanalysis provides the unique opportunity to examine tropospheric and stratospheric interaction prior to and following SSWs, when only the surface is constrained by observations.

Major SSWs occur when the winter polar stratospheric westerlies reverse to easterlies. In minor warmings, the polar temperature gradient reverses but the circulation does not, and in final warmings, the vortex breaks down and remains easterly until the following boreal autumn. Because no unambiguous standard definition for major, minor, and final warmings yet exists (Butler et al., 2015), selecting SSW events to include in the Compendium is not straightforward.

The primary goal of the SSWC is to provide data for major SSWs, which have been found to have the largest surface impacts (Palmeiro et al., 2015). We recognize that any criteria we use may also select marginal events or miss events that perhaps should be considered major in terms of surface influences. We employ the following simple, commonly used definition for major warmings (Charlton and Polvani 2007; hereafter CP07): the central date or event date of a SSW occurs when the daily-mean zonal-mean zonal winds at 10 hPa and 60∘ N first change from westerly to easterly between November and March. The winds must return to westerly for 20 consecutive days between events (to avoid counting the same event twice, roughly equivalent to the thermal damping timescale at 10 hPa; Newman and Rosenfield, 1997). If the winds do not return to westerly for at least 10 consecutive days before 30 April, the warming is a final warming and is not included. The central dates for major NH SSWs in each reanalysis are provided in Table 2. We include in the SSW Compendium, for each reanalysis, every event detected in any reanalysis and shown in Table 2 (for example, we include data for the 30 November 1958 event for all reanalyses extending back to 1958, even though it was only detected in NCEP-NCAR). This includes the NOAA20CRv2c, even though that reanalysis detects only a single event.

There are two main types of SSW: displacement events in which the stratospheric polar vortex is displaced from the pole and split events in which the vortex splits into two or more vortices (Fig. 1). Some SSWs are a combination of both types. There are a number of methods for determining the type of SSW. We do not attempt to classify event types here; however, we do provide the filtered (and unfiltered) absolute vorticity field at 10 hPa (see Sect. 2.3), which may enable classification of split-type SSWs according to the CP07 definition, in which the edges of the vortex are identified by the location of the maximum absolute vorticity gradient. We also provide potential vorticity (PV) interpolated onto isentropic surfaces, and geopotential heights at 10 hPa, both of which can be used to assess vortex moment diagnostics and determine the SSW type (Mitchell et al., 2011; Seviour et al., 2013; Waugh, 1997). We note that the vortex moment diagnostics detect some different dates of SSWs compared to CP07 (and these events are not included in the Compendium), but the provided data would allow classification of the included events.

While almost all SSWs occur in the NH, we did examine their occurrence in the SH in the reanalyses (Table 3). The relevant dates for zonal-mean zonal wind reversals at 10 hPa and 60∘ S were between July and October, and the winds must return to westerly for at least 10 consecutive days before 30 November. Keeping in mind that prior to 1979 there were hardly any observations of the SH polar stratosphere, making reanalyses highly unconstrained, the only event detected occurred in September 2002. This event is included in the SSWC.

The production flowchart for the SSWC is shown in Fig. 2. We obtained the native horizontal and vertical pressure-level data for each reanalysis from various research data archives: NOAA20CRv2c and NCEP/NCAR I from the NOAA Earth System Research Laboratory, Physical Sciences Division (http://www.esrl.noaa.gov/psd/data/gridded/); JRA-55, ERA-interim, and ERA-40 from the University Corporation for Atmospheric Research (UCAR) Research Data Archive (http://rda.ucar.edu/); and MERRA-2 from the Modeling and Assimilation Data and Information Services Center (MDISC, http://disc.sci.gsfc.nasa.gov/mdisc/).

We extracted the following fields (when available): vertically integrated total column ozone; zonal winds, meridional winds, temperatures, geopotential heights, Ertel's potential vorticity (PV), and ozone mixing ratio, on provided pressure levels; and at the surface, mean daily temperature, minimum daily temperature, maximum daily temperature, mean sea level pressure, surface pressure, total precipitation liquid water equivalent, and total snowfall liquid water equivalent. Most raw reanalysis output is available every 6 h (for pressure-level fields) and sometimes up to every 3 h (for surface-level fields), but we computed daily means of all fields for the SSWC. We interpolated pressure-level fields onto a 2.5∘ × 2.5∘ latitude–longitude grid, while the surface-level fields are maintained at native horizontal resolution. We retained data on provided pressure levels, but we interpolated certain fields (PV and ozone mixing ratio) onto isentropic surfaces. Unless isentropic-level data are provided, we calculated potential temperature (θ) from temperature data on pressure levels using Eq. (1): Θ=Tp0pR/Cp, where T and p are atmospheric temperature and pressure, respectively, p0 is a reference pressure defined as 1000 hPa, R is the molar gas constant (287 J deg-1 kg-1), and cp is the specific heat capacity at constant pressure (1004 J deg-1 kg-1). The data, either on pressure or isentropic levels, are linearly interpolated at each time step onto 10 common isentropes (330, 350, 400, 450, 500, 550, 600, 700, 850, and 1000 K). Note that in JRA-55, isentropic-level data are provided but not at the 1000 K surface; therefore, in the SSWC missing values are indicated for this theta level.

There are two types of output provided by the SSWC: climatological statistics and event-based data. Climatological statistic files include the mean and standard deviations of all output fields and percentiles from the climatological distribution for a selection of surface fields: minimum and maximum surface temperature and precipitation. The climatological statistics are defined at each spatial point for 366 days spanning 1 July–30 June. The climatological mean is based on the entire time period of each reanalysis (Table 1). To calculate the climatological mean, we first calculate the mean of each day of the year over the full record. Then we calculate the Fourier transform of this daily mean climatology and retain the first four harmonics of the Fourier series (e.g., Wilks, 2006). This methodology smooths out the raw daily climatology while preserving low-frequency variability. The standard deviation is then calculated by taking the square root of the squared deviations in the raw daily data from this smoothed climatological mean. Percentiles are calculated following a method described in Zhang et al. (2005; see Eq. 1). Chosen percentiles are 5, 10, 90, and 95 %. These statistics are calculated using the entire data record.

Event-based files contain full field, anomaly, and derived fields for the 60 days prior to and following each SSW event in Tables 2 and 3. Anomalies are calculated using the smoothed climatology for each field, using the entire data record for each reanalysis. We caution that, while the climatologies for different time periods are generally quite similar, using different periods for the climatology for each reanalysis means that differences in reanalysis anomaly fields may partially be a result of the climatology chosen. In addition to full fields and anomalies, we derive a number of useful diagnostics for understanding dynamic processes and surface climate surrounding SSW events, as described below:

We provide the maximum and minimum daily temperatures. NCEP-NCAR I provides this output; we calculate these values for the other reanalyses. Note that no interpolation is used – just the minimum and maximum values of the 3 or 6 hourly data – so these values may underestimate the true maximum and minimum daily temperatures.

Assessing the composite response to SSWs is useful for separating the signals from internal noise and identifying where the signal is robust. Figure 3 shows, as a function of pressure level and time before and after the event, (a) zonal-mean zonal winds at 60∘ N and zonal-mean temperature anomalies averaged from 50 to 90∘ N, (b) the Northern Annular Mode index at each pressure level, and (c) ozone mixing ratios from 60 to 90∘ N, composited over all 41 northern hemispheric SSW events (Table 2), using the JRA-55 reanalysis. Figure 4 shows the surface response composited over the 60 days following the central date of all SSWs, including (a) mean sea level pressure anomalies, (b) surface temperature anomalies, and (c) precipitation anomalies.

These two figures illustrate several important and well-known features of SSWs and their impacts on circulation and surface climate (e.g., Baldwin and Dunkerton, 2001). In the stratosphere, the zonal-mean zonal winds change from westerly to easterly at 10 hPa and 60∘ N at lag zero (the central date), as constructed by the SSW definition (Fig. 3a). The zonal wind reversal is strongest near ∼ 3 hPa. In the composite, a complete wind reversal extends from 1 hPa down to ∼ 10 hPa, but a deceleration of the zonal winds extends throughout the whole stratosphere. The peak warming of the stratosphere occurs ∼ 1 day before the peak zonal wind reversal, and its location at ∼ 7 hPa is consistent with peak zonal wind decreases at higher altitudes, per the thermal wind relationship. At 10 hPa and higher, the zonal winds and temperatures rebound quickly after the SSW, reforming a colder westerly vortex above 10 hPa after 10–15 days. In the lower stratosphere, warmer, weaker vortex conditions persist 60 days following the SSW due to slow radiative timescales (Newman and Rosenfield, 1997). These changes near the tropopause may increase the persistence of the negative NAM phase in the troposphere (Fig. 3b), potentially providing a source of predictive skill for up to 60 days after the occurrence of the SSW (Maycock and Hitchcock, 2015). Following the SSW, the stratospheric ozone over the polar cap is greatly enhanced (Fig. 3c), both due to the increased transport of ozone-rich air into the stratosphere via the residual mean circulation and the horizontal mixing of high-ozone air into the region as the low-ozone region of the polar vortex is moved off the pole (either in one or two lobes, depending on whether a split- or displacement-type event has occurred).

At the surface, the composite response in mean sea level pressure anomalies comprises an anomalous high over the polar cap and Greenland and an anomalous low over the North Atlantic, a pattern that projects well onto the negative phase of the NAO, the regional equivalent of the NAM (Fig. 4a). The associated surface temperature anomalies include significant warming over western Greenland and eastern Canada and strong cold air outbreaks over much of northern Europe, Asia, and the eastern United States (Fig. 4b). Conditions are also anomalously wet over western and central Europe and dry over Scandinavia (Fig. 4c).

Composite analysis could also be used to consider differences in SSW evolution and impacts in relation to other factors, such as the differences between split- and displacement- type events, the differences between events that occur in El Niño or La Niña winters, or the different phases of the MJO. Figure 5 highlights the differences in the evolution of the 500 hPa geopotential height anomalies prior to and after a SSW during La Niña versus El Niño winters. Here we use the December–January–February ONI index to classify El Niño and La Niña years, with winters with ONI exceeding +0.5 ∘C defined as El Niño years and winters with ONI below -0.5 ∘C defined as La Niña years. While the sample size for these composites is small (13 events during El Niño years, 9 events during La Niña years), some major features are apparent; for example, the trough during El Niño and the ridge during La Niña in the North Pacific are evident throughout the evolution of the SSW. Note, however, the intensification of low-pressure anomalies in the northwest Pacific in the 60 days prior to SSWs in both El Niño and La Niña winters, a feature theorized in Garfinkel et al. (2012) to amplify planetary-scale waves from the troposphere into the stratosphere and weaken the stratospheric polar vortex. During El Niño winters, the tropospheric circulation pattern is strongest over North America in the days prior to a SSW, but strongest over the North Atlantic after a SSW. During La Niña winters, the anomalies over Greenland and Europe change sign before and after a SSW event, demonstrating the role of SSWs in winter climate over the North Atlantic–European region.

While compositing is useful for highlighting robust features of SSWs, the dynamic evolution and surface climate anomalies before and after each individual SSW can vary widely. The SSWC can be used to demonstrate this range of variability. Figure 6 illustrates the differences in the tropospheric climate following two similar split-type SSWs, one in January 1985 and the other in January 2009. In both events, the polar vortex split into two lobes: the one associated with the greatest warming anomalies centered over Canada and the other centered over northern Europe and Asia (Fig. 6a, b). The 2009 split SSW had a larger lobe that extended over most of Eurasia, but otherwise the stratospheric evolution was quite similar.

However, the subsequent surface and tropospheric responses in the weeks following the events differed in several ways. The 500 hPa height anomaly pattern following the 1985 event projects strongly onto the negative NAO pattern (Fig. 6c), with positive height anomalies over Greenland and negative height anomalies over the North Atlantic. This pattern is associated with much lower surface temperature anomalies over much of Europe and Asia. However, the height anomalies in the 2 months following the 2009 split-type event do not look like the negative NAO phase, though there are weakly positive height anomalies over the Arctic and two centers of low height anomalies over Europe and Asia (Fig. 6d). Temperature advection associated with these anomalous low-pressure centers may explain the regional cold air experienced over Asia and central Europe. Comparison of these two events shows how different modes of climate variability can impact the tropospheric climate during the period after a substantial SSW event. While 1985 and 2009 were both (essentially) La Niña winters (2009 misses official La Niña classification by the NOAA Climate Prediction Center by 0.1 ∘C), the location and strength of the North Pacific ridge during these 2 years was quite different. Other aspects of climate variability, such as the QBO, sea ice, or the MJO, may have played a role in the tropospheric climate during these time periods.

The SSWC allows easy evaluation of the spread among individual events for different features of SSWs. Figure 7 shows time series of the (a) amplitude and (b) latitude of the maximum temperature anomaly (that occurs within the range of 30–90∘ latitude and 300 to 1 hPa) and (c) the 200 hPa 40–70∘ N eddy heat flux anomaly. On average, the maximum temperature anomaly of ∼ 50 K peaks 1–2 days prior to the zonal wind reversal (Fig. 7a, bold black line), but the amplitude and timing vary substantially among the individual events (colored lines), with values from 10 to almost 100 K. Likewise, the mean latitude where the temperature maximizes tends to fall between 60 and 70∘ N (Fig. 7b) but ranges from ∼ 45∘ N to the pole. The 200 hPa heat flux anomaly represents the incoming heat fluxes from the troposphere via vertically propagating waves, which amplify and peak prior to the SSW (Polvani and Waugh, 2004; Sjoberg and Birner, 2014); however, during any individual year, there may be pulses of large heat fluxes that do not result in a SSW (Fig. 7c).

Finally, the SSWC includes data from six different reanalyses, both to aid in reanalysis intercomparison projects such as S-RIP and to allow users the ability to assess the robustness of SSW features in different products. Figure 8 demonstrates how these differences manifest during the January 2013 SSW event for (a) a modern reanalysis product (MERRA2), (b) an older reanalysis product with low model top (NCEP1), and (c) a reanalysis that only assimilates observations at the surface and has a strong bias in the stratosphere (NOAA20CR). In MERRA2, there is strong weakening of the zonal wind anomalies at 60∘ N, which starts near 1 hPa around the event date and descends over time to the tropopause (Fig. 8a, left panel). These anomalies are also evident in NCEP1, but output is only available up to 10 hPa, and the anomalies at 10 hPa tend to be slightly smaller than those in MERRA2 (Fig. 8b). The NOAA20CRv2c makes an interesting comparison because the model stratospheric winds are too strong but the surface is constrained by assimilated observations (Fig. 8c). This means that although NOAA20CRv2c does not capture the SSW event, the surface and tropospheric response contains information about the impact of this stratospheric event. Conversely, the mid- to upper-tropospheric zonal wind anomalies after the SSW event in NOAA20CRv2c are smaller (more positive) than in either NCEP1 or MERRA2, suggesting that the lack of stratospheric processes limits the ability of this reanalysis to capture the tropospheric climate response following major breakdowns of the polar vortex.

The surface temperature anomalies and the 200 hPa geopotential height anomalies for days 30–60 after the 2013 SSW are shown in the right-hand panels of Fig. 8. In the SSWC, surface variables are provided at their native horizontal resolution, which is reflected in these panels in the surface temperature anomalies. MERRA2 has the highest horizontal resolution, making more regional structure and detail apparent. The cold anomalies over Asia and parts of the Arctic, and the tropospheric circulation anomalies at 200 hPa (particularly in regions impacted by stratosphere–troposphere coupling, such as the North Atlantic), are weaker in the NOAA20CR relative to MERRA2 and NCEP1. Regional differences between all three reanalyses can be seen, particularly in the polar cap region where observations may not be available to constrain the reanalysis system.