Northern Hemisphere circumpolar vortex trends and climate change implications



[1] Trends in the Northern Hemisphere circumpolar vortex at 700, 500, and 300 hPa are examined to assess the relationship between circulation variability and air temperature. A vortex climatology is developed for the period 1949–2000. At each pressure level, three geopotential height contours are used to quantify the size and position of the vortex at 5° longitude resolution within and both north and south of the primary hemispheric baroclinic zone. This combination of spatial specificity and the long temporal record makes this the most comprehensive vortex climatology to date. The overall and seasonal vortex time series for the Northern Hemisphere are created for northern, middle, and southern contours at each of the three levels in the atmosphere. From the beginning of the record until 1970, the vortex exhibits a statistically significant expansion, but the vortex has been contracting significantly since then at all levels. The pre-1970 expansion and subsequent contraction is strongest in the lower latitudes and weakest in the higher latitudes. The trends are also stronger in the upper troposphere than in the lower troposphere. Spatial examination of the vortex indicates that the pre-1970 expansion and post-1970 contraction were driven primarily by expansion/contraction over Asia, Europe, and North America with little change over the Northern Hemisphere oceans. Although significant climate change debate focuses on the discrepancy between positive trends in surface air temperature and little or no trends in Microwave Sounding Unit (MSU) satellite temperatures, contraction of the circumpolar vortex at every level of the atmosphere implies that the atmosphere is warming at depth since 1970. Comparisons with the MSU temperature history indicate that the Northern Hemisphere circulation as a whole, as represented by the circumpolar vortex, accounts for almost two thirds of the interannual variability in midlatitude MSU temperature, indicating that vortex size and position are coupled strongly to atmospheric temperature and could be a good indicator of climate change. On a latitude-by-latitude and level-by-level basis, the lower latitudes are associated most strongly with MSU temperature in the midtroposphere while the middle and higher latitudes are more closely associated with MSU temperature in the upper troposphere. The vortex trends are also similar to observed surface warming trends.

1. Introduction

[2] An ongoing aspect of the anthropogenic climate change debate is the discrepancy between the observed surface warming and the lack of warming in satellite-based temperature measurements. While the surface has warmed at a rate of ∼0.1°–0.2°C/decade for the past 20 years since satellite observations commenced, satellite-based measurements of global temperature show little or no trend [Gaffen et al., 2000]. As there is a close link between temperature and atmospheric circulation at synoptic/global scales, careful examination of atmospheric circulation can add valuable information to the debate.

[3] The observed changes in surface warming have coincided with variations in atmospheric circulation such as changes in the North Atlantic Oscillation (NAO), the Pacific/North America (PNA) pattern, and the Arctic Oscillation (AO) [Hurrell, 1996; Thompson and Wallace, 1998]. The NAO shifted into its positive phase around 1980, indicating stronger than normal westerlies over the midlatitude Atlantic Ocean in conjunction with anomalous low pressure near the Icelandic Low and high-pressure anomalies in the subtropical Atlantic. This shift toward the NAO's positive phase has been correlated with the wintertime increases of surface air temperatures north of 20°N [Hurrell, 1996]. Changes in the PNA pattern occurred in the mid-1970s such that the Aleutian Low is deepened and shifted eastward, resulting in warm and moist air advection over western North America and cooler drier conditions over the central North Pacific [Hurrell, 1996]. These changes are related to the “Pacific Climate Shift,” a regime shift that occurred in the mid-1970s related to North Pacific Ocean variability [Miller et al., 1994]. Retrospective analyses indicate that spatial patterns of Pacific sea-surface temperature (SST) anomalies exhibited a significant reversal over a short period of time. The Pacific Climate Shift also coincided with abrupt changes in many climate records, such as tropical Pacific SSTs related to the El Niño-Southern Oscillation (ENSO) [Guilderson and Schrag, 1998], North Pacific SSTs [Miller et al., 1994], oceanic heat content [Levitus et al., 2000], the radiosonde record of Angell [1999], atmospheric sea level pressure, and surface air temperature [Minobe, 1997, 1999], etc. The AO, which is the hemispheric manifestation of the NAO, is a wintertime seesaw in atmospheric mass between the polar region and a zonal ring centered near 45°N. It has been significantly trending toward its positive phase, or a decrease in pressure over the polar region. This change in the AO is also characterized by a strengthening in the subpolar westerlies and a weakening of the jet stream in lower latitudes. Over 50% of the observed warming trend over Eurasia, and 30% of the total extratropical wintertime Northern Hemisphere warming, has been attributed to these changes in the AO [Thompson and Wallace, 2001].

[4] These studies, and others, establish the linkage between temperature trends and circulation features at climatic scales. Many of these analyses are based on circulation fields and indices developed from gridded data sets. An atmospheric circulation variable that has not been thoroughly evaluated with respect to climate change is the circumpolar vortex itself. The circumpolar vortex has been used for decades in numerous studies to quantify hemispheric circulation variability [Angell and Korshover, 1977, 1978, 1985; Markham, 1985; Angell, 1992, 1998; Davis and Benkovic, 1992, 1994; Burnett, 1993; Frauenfeld and Davis, 2000, 2002; Burnett and McNicoll, 2000]. LaSeur [1954] originally noted the usefulness of the vortex in weather studies, but as early as the late 1940s the expansion and contraction of the vortex had been considered in terms of atmospheric general circulation [Willett, 1949]. The circumpolar vortex is a useful and parsimonious measure of circulation because it captures multiple aspects of midlatitude circulation variability, the size, shape, strength, and Rossby wave pattern, in one variable. Vortex time series incorporate long-wave circulation trends such as changes in standing waves and regional circulation features such as strengthening or weakening of troughs and ridges. In addition, these vortex climatologies implicitly incorporate nontropical atmospheric teleconnection patterns, e.g., PNA and NAO, and changes thereof. Furthermore, variations in the amplitude of the circumpolar vortex are related to air mass advection, surface and midtropospheric temperatures, and precipitation [Angell and Korshover, 1977; Burnett, 1993]. Vortex variability previously has been related to climatological phenomena such as El Niño, the Southern Oscillation, the Quasi-Biennial Oscillation, volcanism, and the Pacific Climate Shift [Angell, 1992, 1998; Frauenfeld and Davis, 2000, 2002]. It has been argued that the circumpolar vortex could be useful in detecting a greenhouse warming signal, with fewer site biases and urbanization effects that can contaminate global temperature and precipitation records [Davis and Benkovic, 1992; Burnett, 1993]. The vortex is therefore a potentially useful and efficacious indicator of surface and tropospheric features and of climate change and variability.

[5] Atmospheric circulation trends represented by the Northern Hemisphere circumpolar vortex mainly have been investigated through the year 1990. Davis and Benkovic [1992, 1994] studied the January 500 hPa vortex for the years 1947–1990 and found an overall expansion of the vortex from 1966 to 1990. This expansion was driven by regional troughing over the North Pacific Ocean and eastern North America and was linked to more frequent warm air mass advection into Alaska and western Canada. Burnett [1993] also investigated size variations in the 1946–1989 January 500 hPa circumpolar vortex and similarly found a trend toward an expanded vortex after the mid-1960s dominated by changes over the Pacific and eastern North America/Atlantic. However, the last few years of that record indicated that the January vortex was contracting once again. Recently, Angell [1998] found that the 1963–1997 300 hPa vortex area was decreasing overall, but primarily in the Western Hemisphere. Although Angell [1998] investigated the 300 hPa vortex through 1997, his period of record only began in 1963.

[6] Angell [1998] indicated that around 1990, the circumpolar vortex switched from an expanded to a more contracted configuration. However, only this study has investigated the circumpolar vortex with the inclusion of more recent data, at only one level in the atmosphere, and therefore the implications of the vortex contraction in terms of hemispheric climate change are uncertain. The vortex data set first introduced by Burnett [1993], which was based on the National Meteorological Center (NMC) Northern Hemisphere octagonal grid data [Jenne, 1975], has recently been recalculated and updated using the National Center for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) reanalysis data through the year 2000 and expanded for various levels of the atmosphere. The addition of the last 10 years of data can be important for illustrating the Northern Hemisphere vortex response to the recent climate events, such as the prolonged El Niño during the early 1990s, as well as the strong 1997–1998 warm event. More importantly, the inclusion of 10 more years of data, to a total of 52 years, provides a considerable increase in the length of the data record for long-term assessments of climate change during a period of rapidly increasing concentrations of anthropogenic greenhouse gases.

[7] The goals of this analysis are to determine the trends and changes in the Northern Hemisphere circulation using the 52-year vortex data set for the high, middle, and low latitudes at the 300, 500, and 700 levels, respectively, and to relate those changes to atmospheric temperature variability. The important contribution of this research is not only the use of an extended period of record, but also the greatly increased spatial and temporal resolution of the data. Most of the aforementioned vortex studies only used one geopotential height contour at one level of the atmosphere, and only 1 month was analyzed over time [Davis and Benkovic, 1992, 1994; Burnett, 1993]. Further, while the methodology of Angell [1992, 1998] did include all months of the year, the Northern Hemisphere vortex was represented by four 90° longitude slices that might not characterize regional circulation features. Although analyzing one geopotential height contour captures well the overall variability of the circulation's primary baroclinic zone, certain circulation features such as split-flow regimes, as well as geopotential height gradient changes cannot be represented.

[8] To examine climate change impacts on the circumpolar vortex, vortex variability is compared to the trends in midlatitude microwave sounding unit (MSU) satellite temperature. Because the MSU data provide an average atmospheric temperature, Northern Hemisphere MSU trends should be similar to the Northern Hemisphere circumpolar vortex trends. However, the MSU record only provides a measure of average vertically integrated atmospheric temperature, whereas different layers of the atmosphere may exhibit different temperature trends depending on the vertical temperature structure of the atmosphere. Therefore relating the vortex at various levels in the atmosphere to the MSU temperature history is useful for determining how the different layers of the atmosphere behave over time and how strongly they are linked to the MSU record.

2. Data and Methods

[9] The circumpolar vortex data used consist of mean monthly Northern Hemisphere 700, 500, and 300 hPa geopotential heights for the period of January 1949 to December 2000 (624 months). The geopotential height data, from the NCEP/NCAR reanalysis data set, consists mainly of rawinsonde observations and are “type A variables,” the most reliable product of the reanalysis [Kalnay et al., 1996]. The definition of the circumpolar vortex is based on the assumption that the vortex is best described by the region of strongest meridional height gradient, within the main belt of the westerly flow at each of the three levels. Using the 52-year data record, we thus selected the vortex to be the geopotential height contour that consistently falls within the primary baroclinic zone of the 700, 500, and 300 hPa circumpolar vortex for each month of the year, in accordance with the established “definition” of the vortex [Angell and Korshover, 1977, 1978, 1985; Angell, 1992, 1998; Davis and Benkovic, 1992, 1994; Burnett, 1993; Frauenfeld and Davis, 2000, 2002; Burnett and McNicoll, 2000]. Specifically, geopotential height changes were examined along each 5° meridian to determine the contour that best describes the core of the westerly flow, i.e., the zone of greatest meridional height change. This representative contour is permitted to vary from month to month as the primary baroclinic zone expands and contracts in conjunction with seasonal temperature variations (Table 1). Since significant smoothing occurs when daily weather maps are averaged into mean monthly fields, height-change gradients are not readily discernable on mean monthly maps. Therefore daily weather maps for each month were also used in determining that month's contours to ensure proper contour selection. The representative geopotential height contour for each month was then interpolated at each 5° meridian, thus converting it into a series of 72 latitudinal and longitudinal intersections. The vortex is thereby represented as a series of 72 straight-line segments connecting the intersections [Burnett, 1993; Frauenfeld and Davis, 2000, 2002; Burnett and McNicoll, 2000]. This contour is called the “center contour.” To capture more completely the geometry of the vortex, a “southern contour” and a “northern contour” were also selected at each pressure level. These additional contours were chosen to be 2–3 standard contours south and north of the center contour, thereby capturing the circulation at lower and higher latitudes (Figure 1). Because of the weak height gradient during the summer months, the contours representing the higher and lower latitudes were sometimes the standard contours immediately adjacent to the center contour. However, because of the weaker gradient, the adjacent contours are spaced far enough apart to still capture the circulation of the middle, high, and low (subtropical) latitudes. Although contour selection could be argued to be subjective, because the contours within the westerly flow parallel each other in mean monthly fields, adjacent contours yield very similar changes in the vortex over time [Angell, 1992]. Representative contours and the latitudinal contour position vary from month to month, and a significant portion of that variability is a function of the seasonal cycle in hemispheric temperature gradients. To remove this seasonality, the data were standardized with respect to each month's mean and standard deviation by converting each vortex latitude to a z-score. For the seasonal analyses, the data were first averaged into seasons and then standardized with respect to each season's mean and standard deviation.

Figure 1.

Sample 500 hPa mean monthly geopotential height map depicting the “northern,” “center,” and “southern” contours for March 2001. The contour spacing at 500 hPa is 60 m.

Table 1. Geopotential Height Contours Used to Define the Northern Hemisphere Circumpolar Vortex at Higher (“North”), Middle (“Center”), and Lower (“South”) Latitudes for Each Montha
Month300 hPa500 hPa700 hPa
  • a

    Geopotential height contours are measured in units of m.


[10] The time series of the circumpolar vortex exhibit marked nonlinear trends, and preliminary analysis indicated that the vortex was expanding prior to 1970 and contracting thereafter (Figure 2). This analysis consisted of performing a two-phase regression model to determine the most significant trend change-point [e.g., Solow, 1987; DeGaetano and Allen, 2002]. If we consider the vortex time series Vi, i = 1, …, n, for n months, this model can be written as:

equation image

where a is the intercept, b is the slope, e represents the residuals, and r is the potential trend change-point. To find the optimal r in the time series, the point where both phases of the regression model exhibit the best fit (i.e., where the p value of the F statistic is the smallest for both regressions) was determined. This procedure objectively finds a time step at which to split the time series such that each of the two segments can be described by the best fit regression line. This analysis was performed on each of the nine contours' monthly time series and each of the 36 seasonal time series. As it is not likely that each of the 45 time series are characterized by the identical inflection point, we chose the mean inflection point of all 45 time series, which is 1969. It should be noted that in general the most significant trend change-point was between 1968 and 1972, and any of these points would have yielded very similar results. (Throughout the remainder of this paper, the 1949–1969 period is referred to as “pre-1970” and the 1970–2000 period as “post-1970.”) Fitting a linear trend through a nonlinear time series with an inflection point would give a misleading representation of the long-term vortex behavior. Therefore to quantify the pre-1970 expansion and post-1970 contraction, the trends were determined separately before and after 1970. During the pre- and post-1970 periods, the time series are approximately linear.

Figure 2.

Time series of the (top) 300 hPa center contour and (bottom) southern contour with linear least squares trend lines before and after 1970.

[11] For the overall vortex, the latitudinal position of each of the 72 standardized vortex slices was averaged across the hemisphere to obtain one mean vortex position for each of the 624 months. A least squares regression line was then fit to each contour at each level before and after 1970. The same analysis was repeated on the seasonal averages of the four seasons separately to determine the trends for winter (December–February), spring (March–May), summer (June–August), and fall (September–November). To assess potential changes in the strength of the geopotential height gradient of the Northern Hemisphere and thus changes in the strength of the westerly flow, the difference between the southern and northern contour (south minus north) is calculated for the monthly and seasonal time series. Differences between the pre-1970 and post-1970 gradient are then assessed statistically with independent-samples t-tests.

[12] A trend of overall vortex expansion, for example, could arise from large-scale changes across much of the vortex, very strong localized changes (like enhanced regional troughing), or a combination of the two. To establish the spatial vortex trends for each contour at each level, the slope of the least squares trend line was determined for the 72 time series representing each 5° longitude-wide vortex sector before and after 1970. These slopes were then plotted to assess where the circumpolar vortex is expanding and contracting with time.

[13] To compare the circumpolar vortex patterns with the Northern Hemisphere temperature history, the MSU temperature data for the Northern Hemisphere were used for 1979–2000 [Christy et al., 2000]. The MSU is a polar orbiting satellite-borne microwave radiometer that observes the Earth's upwelling radiation, the intensity of which is directly proportional to the temperature of the air. Two MSU records are available to represent tropospheric temperature: MSU 2 for the middle to upper troposphere and MSU 2LT for the lower troposphere. MSU 2LT was used here because the air temperature calculated from MSU 2 contains a significant signal from the lower stratosphere. Because the temperature of the lower stratosphere is thought to decline with a tropospheric temperature increase, detecting greenhouse warming changes becomes problematic when using MSU 2 [National Research Council, 2000] in any comparison with a troposphere-only vortex. Half of the Northern Hemisphere comprises the tropics, so 50% of the Northern Hemisphere MSU 2LT temperature is influenced by the tropical temperature variability. Since the vortex at all three levels examined here predominantly falls within the 30°N–60°N latitude band, the 2.5° by 2.5° gridded MSU temperatures were weighted by the cosine of the latitude to account for varying grid box size and averaged for the 30°N–60°N band only. Initially, the time series of each hemispherically averaged contour at each level of the vortex is correlated with the midlatitude MSU 2LT time series. To determine which seasons contribute most to the midlatitude MSU 2LT temperature-vortex relationship, the vortex is also correlated with MSU temperature for each season separately. Given the notion that the MSU data represent the vertically integrated average atmospheric temperature, a multiple regression model was built using all nine contours as dependent variables, thereby providing a three-dimensional characterization of the atmospheric circulation. To determine which level of the atmosphere is most closely linked to the MSU 2LT record, multiple regression models were also built separately for each of the three levels.

[14] As with most atmospheric data, the statistical assumption of independence is not satisfied for the circumpolar vortex time series, i.e., there is serial autocorrelation. To address the issue of autocorrelation and account for its effects, an effective sample size which represents an equivalent number of independent samples is calculated for each dependent time series [Wilks, 1995]. The effective sample size, N′, is calculated using the formula:

equation image

where N is the number of observations and ρ is the lag-1 autocorrelation of the time series [Mitchell, 1966]. It was also determined that the vortex time series' autocorrelations are greatest at lag-1 and are indeed characteristic of AR(1) processes (not shown); therefore, using the lag-1 autocorrelation is appropriate. For each time series, N′ is used for determining the degrees of freedom needed for establishing the significance of the various statistical tests performed in this investigation. This method of addressing serial correlation has been used in other vortex studies [e.g., Angell, 1992].

[15] As the circumpolar vortex contours in these analyses are derived from the NCEP/NCAR reanalysis, it should be noted that some controversy exists as to the reliability of using NCEP/NCAR reanalysis data in evaluation of long-term trends. While the reanalysis data assimilation system has been constant and is therefore essentially free of inhomogeneities because of changes in model resolution and physics, changes in the observing system have occurred [Kistler et al., 2001]. The reanalysis period can be divided into three eras: the period from 1948 to 1957, during which the upper-air observing network was established and continually improved, the subsequent period during which the modern observational network was fully in operation, and the satellite era, which began in November 1978. The era beginning in 1957, specifically June 1957, was also accompanied by a shift in observing time: before June 1957 upper-air observations were observed 3 hours earlier than the current synoptic times of 0000, 0600, 1200, and 1800 UT. While the consistent assimilation system of the reanalysis efforts produce consistent analyses which reduce the existence of spurious climate shifts, changes in the observational network such as the introduction of satellite data beginning around 1979 still produce small but spurious shifts. For instance, in an analysis of the NCEP/NCAR reanalysis for temporal stability, Pawson and Fiorino [1999] note such a spurious shift via an increase in temperature near the tropical tropopause around 1979. As tropical tropopause temperatures can have a significant impact on low-latitude geopotential height data, any shifts observed in 1979 must be interpreted cautiously. Similarly, any observed shifts in 1957 may be related to the change in observing time.

3. Average Vortex Trends

[16] The trends of the standardized 624-month time series of the hemispherically averaged vortex show statistically significant overall vortex expansion at every level in the troposphere before 1970 and statistically significant contraction thereafter (Table 2). At the 300 hPa level, the southern contour exhibits the strongest pre-1970 expansion and post-1970 contraction, and the center contour has a greater trend than does the northern contour. In addition to these pre- and post-1970 trends, the time series for the three 300 hPa contours also exhibit some other interesting features (Figure 2). A large positive anomaly and possible vortex regime shift during March 1977, coinciding with the Pacific Climate Shift, is evident in the time series of the 300 hPa southern contour. This shift is also somewhat evident in the 300 hPa center contour but not in the northern contour. Furthermore, all three time series seem to switch from mainly decadal variability before the late 1980s to much higher frequency variability in the last 10 years of the record. From 1949 until the early 1970s, the vortex was steadily expanding. Throughout the 1970s, the vortex remained in an expanded position and has been contracting since then. Beginning in the early 1990s, however, the vortex has been undergoing periods of expansion and contraction approximately every 2–3 years. The vortex expansion around 1992 coincides with the months following the eruption of Mt. Pinatubo in June 1991. The northern and southern 300 hPa contours also underwent an abrupt contraction during April of 1998, coinciding with the 1997–1998 El Niño event as well as one of the warmest years on record. The largest positive vortex contraction in the 300 hPa center contour occurred in March of 1990, coinciding with the largest positive global land surface mean temperature anomaly.

Table 2. Regression Slopes for the Linear Trends in the Vortex Before and After 1970 in z-Score Units Per Decadea
  • a

    Negative slopes indicate vortex expansion and positive slopes indicate vortex contraction. Bold values indicate statistical significance at the 0.05α level based on an F test.

300 hPa Level
500 hPa Level
700 hPa Level

[17] The patterns observed at 500 and 700 hPa are very similar to those at 300 hPa. The overall trend is again a statistically significant vortex expansion before 1970 and contraction since then for all three contours at both levels, and in each case the pre-1970 expansions and post-1970 contractions are strongest for the southern contour and weakest for the northern (Table 2). The same features observed at 300 hPa are also evident for each contour at the 500 and 700 hPa levels. The Pacific Climate Shift is again evident via a large positive anomaly in the southern contour at 500 hPa, but not in the center and northern contours (not shown). The warm March in 1990 again shows up as one of the greatest vortex contractions in the 500 hPa center contour, as does the April 1998 contraction in the southern contour. The 700 hPa vortex is very similar; however, while there is a large positive vortex anomaly in 1976–1977, no shift is observed in the southern contour (Figure 3). Similarly, the 1998 vortex contraction is not as evident as it was at 300 and 500 hPa, suggesting that the mid- and upper-tropospheric circulation is more responsive to surface temperature changes.

Figure 3.

Same as Figure 2 but for 700 hPa southern contour.

[18] The patterns of variability throughout the troposphere are very similar. They are most amplified in the lower latitudes (southern contour), less amplified in the midlatitudes (center contour), and most dampened in the higher latitudes (northern contour). Therefore the circumpolar vortex seems to vary in unison at the lower, middle, and upper troposphere as well as at the lower and middle latitudes. The overall pre-1970 expansion and post-1970 contraction of the vortex is strongest at 300 hPa and weakest at 700 hPa (Table 2). Since the vortex is contracting most in the lower latitudes and least in the higher latitudes, this could also indicate a trend toward an increased meridional height and pressure gradient and stronger westerly flow within the core of the Northern Hemisphere circumpolar vortex since 1970. To test this notion statistically, the difference between the standardized southern and northern contours was calculated at each atmospheric pressure level. If there has been a significant change in the meridional height gradient of the Northern Hemisphere, this south-north contour difference should reflect this change. Performing an independent-samples t-test between the north and south difference before and after 1970 indicates that indeed there is a significant change in the meridional height gradient at each of the three atmospheric levels (Table 3, “monthly” column). The differences before 1970 are always negative whereas the differences since 1970 are always positive. A contracted contour is characterized by a positive vortex departure (z-score) and an expanded contour is characterized by a negative departure (z-score). Therefore for the south minus north difference in the vortex departures (z-scores) to be negative, the gradient must be weak, as is observed before 1970. Conversely, for the south minus north vortex difference to be positive, the gradient must be stronger, as is the case since 1970. The t-tests between the differences of the two periods is statistically significant (α = 0.05); therefore there has been a statistically significant change toward a stronger geopotential height gradient since 1970.

Table 3. Mean Differences Between the Southern and Northern Vortex Contours as a Measure of Geopotential Height Gradienta
Time FrameMonthlyWinterSpringSummerFall
  • a

    Measurements are in z-score units. Bold values indicate a statistically significant difference (α = 0.05) before and after 1970, according to a t-test.

300 hPa Level
500 hPa Level
700 hPa Level

4. Seasonal Vortex Trends

[19] To assess the seasonal behavior of the circumpolar vortex, hemispheric mean monthly vortex latitudes were averaged for each season. At the 300 hPa level, the seasonal analysis shows that during winter, summer, and fall the vortex was expanding significantly prior to 1970 and contracting significantly thereafter (Table 4 and Figure 4). With the exception of the spring post-1970 trend in the northern contour, no significant relationships are observed during spring at 300 hPa. The Pacific Climate Shift is somewhat evident in the summer center and southern vortex time series only. The 300 hPa summer southern contour also exhibits 3–4 distinct periods of vortex variability (Figure 5). Before the late 1950s, the vortex was mostly contracted, at which point it switched to an almost exclusively expanded period for the next 20 years, until the time of the Pacific Climate Shift. After this, the vortex was again predominantly contracted. During the early 1990s, there was a shift toward a more variable period, as was observed in the overall time series from section 3. The 500 hPa level trends are very similar to those at 300 hPa, however, the winter northern contour trends are not significant before or after 1970 and the summer pre-1970 trends are also not significant for the northern and southern contours (Table 4). The 1976–1977 Climate Shift is again evident in the southern summer contour and to a lesser degree in the summer center and fall southern contours. The 700 hPa seasonal analysis indicates very similar results, but fewer of the pre- and post-1970 trends are significant (Table 4). The Pacific Climate Shift is somewhat evident at the 700 hPa level in the summer southern contour only.

Figure 4.

Time series of the winter 300 hPa center vortex contour. The straight lines correspond to the 1949–1969 and 1970–2000 linear least squares trends.

Figure 5.

Same as Figure 4 but for 300 hPa summer contour.

Table 4. Regression Slopes for the Linear Trends in the Vortex Before and After 1970 in z-Score Units Per Decadea
  • a

    Negative slopes indicate vortex expansion and positive slopes indicate vortex contraction; bold values indicate statistical significance at the 0.05α level.

300 hPa Level
500 hPa Level
700 hPa Level

[20] As was the case for the overall trends (Section 3), the southern contours generally exhibit the strongest pre-1970 and post-1970 trends during all seasons except spring. Throughout the entire troposphere, the trends tend to be strongest during winter. Virtually no significant vortex trends are observed during the spring season. The Pacific Climate Shift is generally evident at all levels, but only in the southern summer contours. Again, testing the changes in geopotential height gradient for the individual seasons indicates that during winter, there was a weaker gradient before 1970 and a stronger gradient since 1970 at every level in the atmosphere (Table 3). During spring, only the 700 hPa level has experienced this same change in geopotential height gradient. The 300 and 700 hPa levels during summer also experienced a weaker gradient pre-1970 and a stronger gradient post-1970, as does every level during the fall season. Overall, these changes in geopotential height gradient are strongest during winter, when the greatest changes in the vortex contours are observed.

5. Spatial Vortex Trends

[21] Thus far, the vortex has been evaluated as a single entity via examination of the mean hemispheric vortex latitude only. In this portion of the analysis, trends are determined for the 624-month time series of each 5° vortex slice for each contour at each level, thereby depicting where the vortex is expanding and contracting with time. Because of the nonlinear nature of the vortex time series, this spatial analysis was again performed on the 1949–1969 and 1970–2000 subperiods.

[22] Before 1970, the three contours at the 300 hPa level are significantly expanded over Eurasia and eastern North America (Figure 6, left). The strongest expansions are observed over eastern Asia and eastern Europe. Over Eurasia, the southern contour exhibits the most expansion, while the northern contour shows the least. The southern contour is significantly expanded over central North America, while all three contours are expanded over eastern North America, the center and northern contours more so than the southern. The southern contour is also expanded over the Pacific before 1970. After 1970, the regions with significant vortex trends are again Eurasia and North America (Figure 6, right). The center and southern contours are contracting significantly from Europe to western Asia, while all three 300 hPa contours are contracting significantly from central to eastern Asia. In fact, the southern contour is contracting significantly across the entire hemisphere since 1970. Over North America, all three contours are contracted and a large-scale contraction in the southern contour is evident from the western Atlantic into Europe.

Figure 6.

Spatial (5° longitude resolution) (left) pre-1970 and (right) post-1970 300 hPa vortex trends in z-score units. Trends outside of the shaded box are statistically significant at that longitude (α = 0.05). A base map is provided for longitudinal reference, illustrating where the vortex contours are expanding/contracting.

[23] Similar patterns are evident in the 500 and 700 hPa vortex (Figures 7 and 8). Before 1970, the vortex is expanding over Eurasia and eastern North America and contracting in these same regions since then. While the expansions over eastern Europe are greatest at 700 hPa, the expansions over eastern Asia are strongest at 300 hPa. The post-1970 contractions are strongest over Asia at 700 hPa and weakest over Asia at 300 hPa. Over North America, the contractions are of similar magnitude across all levels of the troposphere.

Figure 7.

Same as Figure 6 but for 500 hPa.

Figure 8.

Same as Figure 6 but for 700 hPa.

[24] The overall vortex trends in the hemispheric time series from Section 3 are therefore driven by strong regional expansions and contractions over Eurasia and North America. The trends over Eurasia are much stronger than the trends over North America. Only at the 300 hPa level are the post-1970 vortex contractions more hemispheric in nature. Before 1970, the southern contours always exhibit the strongest expansion and the northern contours the weakest. This spatial analysis was performed at the seasonal level as well, but the regional patterns are very similar to the monthly results (above) and therefore are not presented. As was the case for the seasonal hemispheric time series, the winter trends were strongest.

6. MSU Temperature Comparison

[25] To compare the Northern Hemisphere circumpolar vortex trends with the trends in hemispheric temperature and evaluate the degree to which the vortex is sensitive to climate change, the relationship between the midlatitude MSU temperature history and each Northern Hemisphere average vortex contour time series was established. Each monthly and seasonal vortex contour at each level was correlated with the midlatitude MSU temperature time series for the period 1979–2000 (Table 5 and Figure 9). The strongest correlation between the vortex and MSU temperatures is at the 300 hPa level and vortex variations account for from 25 to 50% of the variance in temperatures, depending on the contour (and from 26 to 57% seasonally). The correlations between the 500 hPa vortex and MSU temperature are slightly weaker, and the center and southern contours account for 33–38% (42–49% seasonal) of the variance in MSU temperature. At the 700 hPa level, again the center and southern contours are significant, accounting for 16–26% of the variance (18–36% seasonal).

Figure 9.

Time series of the standardized 300 hPa center contour (black line) and midlatitude MSU 2LT temperature departures (shaded line). The two time series are correlated at R = 0.71 (p < 0.01).

Table 5. Pearson's Correlation Coefficients for the Monthly, Residual, Seasonal, and Individual Season Vortex Contours and Midlatitude (30°N–60°N) MSU 2LT Temperaturea
  • a

    Significant coefficients are in bold (taking into consideration multiple comparison effects by adjusting the confidence levels, correlations are statistically significant at the 0.006 level).

300 hPa Level
500 hPa Level
700 hPa Level

[26] Given that two time series could be significantly correlated merely because they both exhibit a similar trend, these vortex-MSU correlations may not be indicative of their true relationship. Therefore, using least squares regression, the linear trend was removed from the respective vortex contours and the midlatitude MSU temperature time series. If the correlations between the vortex contours and MSU temperature are significant only because they both exhibit a positive trend, then these residuals should be uncorrelated. However, the correlation between the vortex residuals and MSU temperature residuals indicates that there is still the same strong correlation (Table 5, “residual” column). It can therefore be concluded confidently that the Northern Hemisphere circumpolar vortex is indeed correlated strongly with the midlatitude MSU temperatures.

[27] On a seasonal basis, the relationship is strongest between the vortex and midlatitude MSU 2LT during summer, where the strongest relationships are observed at 300 hPa and the weakest at 700 hPa (Table 5). The strongest correlation is evident for the 300 hPa southern contour, accounting for 70% of the variance between the vortex and atmospheric temperature. The weakest of the seasonal relationships, though still statistically significant and fairly strong, are observed during winter and fall. During winter, the correlations (R2 × 100) are between 19% (300 hPa northern contour) and 46% (500 hPa southern contour), and during fall the correlations are between 21% (300 hPa northern contour) and 59% (300 hPa southern contour). Correlations during spring are also strong, ranging from 19% (300 hPa northern contour) to 64% (300 hPa center contour). In general, the 300 hPa vortex is again related the strongest to midlatitude MSU temperature for each season.

[28] The discovery that upper tropospheric circulation is more strongly related to MSU 2LT temperature than is lower tropospheric circulation is perhaps unexpected. Since the MSU 2LT data are more heavily weighted by lower tropospheric temperatures, the strongest relationships between MSU 2LT and circulation might have been expected to occur at the 700 or 500 hPa levels. However, the MSU 2LT temperatures are most strongly correlated with the 300 hPa vortex and more poorly related to the 700 hPa vortex. This suggests that the circulation of the upper troposphere is most responsive to the trends in the MSU temperature, or vice versa.

[29] Multiple regressions are calculated separately using monthly (and seasonal) MSU 2LT temperature as the dependent variable and three vortex contours at each level as the independent variables. The three 300 hPa contours account for more of the MSU variability than the 500 hPa contours (55 versus 44% monthly; 63 versus 56% seasonally). At 700 hPa, only the southern contour is significant in the model and it accounts for 26% of MSU variance (37% seasonal). Since the MSU data represent an integrated measure of the temperature of the atmospheric column, a multiple regression model was calculated using the hemispherically averaged time series of all vortex contours. Inspection of the correlation matrix and the diagonal of the inverse correlation matrix indicated that a few of the vortex contours are highly intercorrelated and therefore have high variance inflation factors (VIFs). After removing these vortex contours as well as the nonsignificant ones, a stable model with four of the nine vortex contours was obtained that accounts for approximately 58% of the variance in midlatitude MSU temperature (Tables 6a and 6b). The 300 hPa level again seems to be most representative of MSU temperature variability, since all three 300 hPa contours are significant in the model (Tables 6a and 6b). The 700 hPa northern contour only accounts for approximately 3% of the MSU temperature variability, however, it is significant and is therefore included in the model. As was the case in the correlation analysis, the 700 hPa northern contour exhibits an inverse relationship with MSU 2LT temperature. A multiple regression model was also computed for the seasonally averaged vortex and MSU time series and after removing the intercorrelated variables that produced VIFs as well as the nonsignificant variables five vortex contours remained in the model. This stable model accounts for 69% of the MSU variance and again all three 300 hPa contours, as well as both the 500 and 700 hPa northern contours are significant in the model (not shown). The significant 700 hPa contour has an inverse relationship with MSU temperature.

Table 6a. ANOVA Table and Regression Summary for the Multiple Regression Between Midlatitude (30°N–60°N) MSU 2LT (Dependent Variable) and the Significant Noncollinear Vortex Time Series
SourceSum of SquaresDFMean SquareFSignificanceSummary
  • a


Regression12.18243.04690.8220.000R = 0.764
Residual8.6852590.034  R2 = 0.584
Total20.867263   R2 = 0.577a
Table 6b. Regression Coefficients for the Multiple Regression in Table 6aa
  • a

    IV, independent variable; UC, unstandardized coefficients; SC, standardized coefficients.

Constant−0.0470.012 −3.740.000
300 hPa north0.3500.0530.3676.620.000
300 hPa center0.3100.0500.3906.180.000
300 hPa south0.1240.0340.1943.630.000
700 hPa north−0.1860.042−0.207−4.410.000

[30] Almost two thirds of the variability in midlatitude MSU temperatures can therefore be accounted for by changes in the Northern Hemisphere circumpolar vortex. Although the northern contours have the weakest linear correlations with MSU temperature, these contours are always significant in the regression models. When considering circulation as a whole, the higher latitudes therefore still contribute significantly to the temperature variability of the Northern Hemisphere. As was the case in the correlation analysis, the 700 hPa level is related the least with MSU 2LT temperature.

7. Summary and Conclusions

[31] With the addition of 10 years of data to extend previous vortex climatologies, it is evident that the circumpolar vortex is becoming more contracted with time since 1970 at every level in the atmosphere. Previous analyses had indicated that the vortex at 500 hPa was expanding over time [Davis and Benkovic, 1992, 1994]; however, this trend has apparently changed with the inclusion of the last 10 years of data and now supports the findings of Angell [1998] for the 300 hPa level. The Northern Hemisphere circumpolar vortex can be characterized by two different regimes. The vortex was expanding significantly at 700, 500, and 300 hPa for the first 20 years of the analyzed record until 1970, at which time it underwent a trend reversal and has been contracting significantly since then. The pre-1970 expansion was strongest in the lower latitudes and upper troposphere and weakest in the higher latitudes and lower troposphere. Similarly, the contraction since 1970 has been strongest in the lower latitudes and the upper troposphere and weakest in the higher latitudes. It must be noted that the objectively determined trend reversal of 1970 is valid assuming that there are indeed two regimes. The 1976/1977 Pacific Climate Shift can be argued to represent a natural break in the time series as well, and perhaps drives both the change-point analysis as well as the post-1970 trends. However, performing the same analysis using 1976/1977 as the breakpoint in the time series yields very similar results (not shown).

[32] Since the lower latitude contours expanded more than the middle latitude contours and the higher latitude contours expanded the least, these results also suggest that before 1970, the meridional height gradient was weakening. Similarly, since the vortex in the lower latitudes has been contracting more than in the middle and high latitudes since 1970, the meridional height gradient has been strengthening. During the early period when the vortex was expanding, the westerly flow was therefore also weaker, and during the later period when the vortex was contracting, the westerly flow across the Northern Hemisphere has been stronger.

[33] The same pre- and post-1970 trends are evident for the individual seasons. The strongest temporal trends in the Northern Hemisphere circumpolar vortex are observed during winter, and virtually no significant trends are evident during spring before or after 1970. Since approximately 1990, vortex variability has also shifted from roughly decadal variability to more high frequency variability. Before 1990, the vortex was expanding or contracting for periods of approximately 20 years. However, since 1990, it has undergone much more frequent changes, with periods of expansion and contraction lasting only 2–3 years. This change in variability is evident in virtually all of the vortex time series.

[34] On a regional basis, the vortex expansions and contractions observed for every layer in the atmosphere are dominated by very strong pre-1970 expansion and post-1970 contraction over Asia as well as over Europe. Few changes are observed over the Northern Hemisphere oceans. The vortex over North America is also expanding and contracting before and after 1970, respectively, but to a lesser degree than over Eurasia. The lower latitude contours were expanding more than the higher latitude contours over Eurasia before 1970, again indicating that before 1970 the meridional height gradient was weaker. Similarly, the southern contours are contracting the strongest over eastern Europe and western Asia in the post-1970 period at all three levels, indicating a potential strengthening of the height gradient and westerly flow in these regions.

[35] Comparison of the temporal vortex trends to the trends in the MSU 2LT temperature history indicates that between 58 and 69% of the midlatitude atmospheric temperature variability can be accounted for by the combined vortex variations at all three pressure levels, although the upper troposphere accounts for a majority of that variability. This suggests that the upper tropospheric vortex is most indicative of large-scale climate change and variability. That the strongest relationships are found in the upper troposphere can be explained in terms of a well-mixed warming. For a surface warming to have the strongest relationship with circulation at 300 hPa, the warming cannot be confined to only a shallow surface layer, but instead must be well mixed throughout the atmosphere, which then results in the greatest height anomalies observed at the top of the troposphere. Investigating the vortex contours and atmospheric levels separately indicates that in the middle and upper troposphere, the center contours are more strongly linked with midlatitude atmospheric temperature; in the lower troposphere, the lower latitude contours are more strongly linked with atmospheric temperature. Although the higher latitudes are always related the weakest with MSU temperature, when atmospheric circulation as a whole is considered, the high latitudes (northern contours) are significant. This could be attributable to the fact that atmospheric flow is stronger at the higher latitudes where the atmosphere is more baroclinic and where the meridional exchange of energy and moisture takes place. The strong relationship between extratropical Northern Hemisphere circulation and midlatitude MSU temperature indicates that the MSU observations are linked to hemispheric climate variability.

[36] The fact that the circumpolar vortex is contracting with time since 1970 at 300, 500, and 700 hPa has important climate change implications. Following the argument that, hydrostatically, a contracted vortex means that the pressure surfaces are higher and therefore the atmosphere “thicker,” the atmosphere as a whole must also be warming at each of the three layers investigated here. These results are supported by Pielke et al. [1998] who investigated atmospheric thickness based on the NCEP reanalysis data and found a significant Northern Hemisphere warming throughout the troposphere in the period 1973–1996, which coincides closely with our post-1970 period. Pielke et al. also found their thickness trends agree closely with MSU temperature trends but not with surface data.

[37] The trends in the Northern Hemisphere circumpolar vortex are also qualitatively comparable to trends in surface warming. The periods of expansion, contraction, and subsequent higher variability observed in the trends of the hemispherically averaged vortex time series agree with observations of surface temperatures [Jones et al., 2001]. Northern Hemisphere temperatures were decreasing until the mid-1970s, increasing steadily thereafter, and highly variable in the 1990s. The anomalously cool years from 1992–1993 are most likely related to the dust veil produced by Mt. Pinatubo's eruption in June 1991, a feature clearly discernable in the circumpolar vortex time series as a period of strong vortex expansion. Similarly, 1998 was one of the warmest years on record as well as an El Niño year, a feature which is clearly evident as a strong vortex contraction. The Pacific Climate Shift is also apparent in some of the vortex contours, generally more so in the lower latitudes than the middle latitudes, but not at all in the higher latitudes. The seasonal analysis reveals that the Shift is evident mainly during summer in the circumpolar vortex.

[38] The vortex contraction and the analogous warming are strongest over Eurasia and weakest over North America. These spatial warming patterns closely match the regions of warming related to changes in the AO which account for 50% of the extratropical Eurasian wintertime warming and 30% of the wintertime Northern Hemisphere warming [Thompson and Wallace, 2001]. The spatial contraction trends in the vortex over North America can also be argued to represent an enhancement of the PNA pattern. Trends toward significant contraction at every level in the atmosphere in the vicinity of the Alberta High, one of the PNA centers of action, result in a strengthening of the ridge in that region. Given that the strongest contractions are observed over Asia during summer, it seems reasonable that this change is related to changes in the monsoon circulation and to the general warming of the Eurasian landmass.

[39] The strong link between the vortex and MSU temperature demonstrates that the integrated temperature measure provided by the MSU is not only an indicator of atmospheric temperature variability but also of atmospheric circulation. The size and positions of the circumpolar vortex are circulation variables that are not very widely used in circulation studies, despite their effectiveness in quantifying atmospheric temperature variability in the midlatitudes, surface temperature variability, and therefore climate change. The vortex responds not only to gradual decadal climate variability such as the cooling and warming periods in the observational surface and satellite temperature records, but also to interannual changes such as perturbations related to El Niño and volcanism and seasonal variability such as monthly surface air temperature anomalies.

[40] Regarding potential inhomogeneities in analyses based on NCEP/NCAR reanalysis products, we see no evidence of spurious nonclimatic shifts in the circumpolar vortex data. While the Pacific Climate Shift is most evident in the southern contours of the vortex, this shift occurs during the winter of 1976/1977, 2 years before the incorporation of satellite data in the NCEP/NCAR reanalysis. Furthermore, no shift is evident in any of the vortex time series around 1979 and it can therefore be concluded that the 1979 discontinuity in tropical tropopause [Pawson and Fiorino, 1999] temperatures is not affecting the circumpolar vortex data. Similarly, the June 1957 three-hour shift in observing time does not seem to introduce any bias in the vortex data (e.g., Figure 5). We conclude that the Northern Hemisphere circumpolar vortex data, though derived from the NCEP/NCAR reanalysis data, is relatively immune to spurious shifts resulting from nonclimatic events. It could also be argued that, due to sparse observational data, the NCEP/NCAR reanalysis data may not be reliable prior to the late 1960s. In light of this fact, the pre-1970 trends discussed here may be over or underestimated. However, we are confident in the post-1970 trends of vortex contraction throughout the midlatitude troposphere and the resultant climate change implications of tropospheric warming.

[41] MSU temperatures have been criticized as to their ability to accurately reflect temperature trends in the atmosphere, and significant controversy exists as to the discrepancy between surface warming trends and those measured by satellite-borne instruments [e.g., Hurrell and Trenberth, 1998; Santer et al., 1999, 2000a, 2000b]. Given the very close correspondence between the circumpolar vortex and MSU temperatures, the trends in MSU temperature have in essence been verified by the trends in the circumpolar vortex (and vice versa). Since one would expect a close correspondence between atmospheric temperature and atmospheric circulation, if the atmospheric temperature measurements made by satellite-borne instruments were not reflective of the true atmospheric temperature trends, the strong relationships presented here would not be evident. However, as a caveat it must be noted that the TIROS operational vertical sounder data incorporated into the NCEP/NCAR reanalysis consist of infrared measurements from the High-Resolution Infrared Sounder and microwave measurements from MSU. Therefore the circumpolar vortex and MSU temperatures are not strictly independent.

[42] An important consideration for future work is exploring the influence of external forcings on atmospheric circulation and, hence, atmospheric temperature variability. What role do the oceans, in particular the Pacific, play in driving vortex variability, and vice versa? Evidence exists for a link between the circumpolar vortex and interannual tropical Pacific SST variability, such as ENSO [Angell, 1992; Frauenfeld and Davis, 2000, 2002]. Distinctly multidecadal associations between the circumpolar vortex over the Eurasian landmass, where the greatest vortex changes have been observed, and tropical Pacific non-ENSO SST variability have also been observed [Frauenfeld and Davis, 2002]. What role does the Eurasian landmass play in driving the observed changes in the vortex and the tropical Pacific, or vice versa? It is the goal of our ongoing research to address these issues.


[43] We thank Adam Burnett for providing the circumpolar vortex data. Thanks also to Michael E. Mann and P. Chip Knappenberger for their insights and suggestions. Three anonymous reviewers provided substantive comments which greatly improved this research.