Journal of Geophysical Research: Atmospheres

A comparison of the Hadley circulation in modern reanalyses



[1] Previous studies using reanalysis data suggest an intensification and poleward expansion of the tropical Hadley circulation (HC) throughout the twentieth century, yet the HC climatology and trends remain undocumented for many of the newest reanalyses. An intercomparison of eight reanalyses is presented to better elucidate the mean state variability and trends concerning HC intensity and width. Significant variability between reanalyses was found in the mean HC intensity with less variability in HC width. Certain reanalyses (e.g., ERA40 and the Climate Forecast System Reanalysis) tend to produce stronger meridional overturning, while others (National Centers for Environmental Prediction–National Center for Atmospheric Research and Modern-Era Retrospective-Analysis for Research and Applications) are constantly weaker. The NOAA–Cooperative Institute for Research in Environmental Sciences Twentieth Century Reanalysis best matched the ensemble averages with the exception of a poleward shift in the subtropical terminus. Ensemble trends regarding HC intensity and width are broadly consistent with previous work, indicating a 0.40 (0.07) × 1010 kg s−1 decade−1 intensification in the northern (southern) cell and a 1.1° decade−1 widening in the past 30 years, although some uncertainty remains regarding the intensity of the southern cell. Longer-term ensemble trends (i.e., 1958–2008) containing fewer ensemble members suggest a weaker northern cell intensification but stronger southern cell intensification and a more modest widening of the HC (i.e., 0.53° decade−1) compared to the last 30 years. Separation of the seasonally averaged stream function magnitudes by the El Niño–Southern Oscillation (ENSO) phase revealed a weak clustering and statistically significant strengthening of the mean circulation for El Niño compared to ENSO neutral and La Niña events for the winter cell with little difference in the summer cell intensity.

1. Introduction

[2] The global mean meridional circulation is traditionally divided into three zones comprising a thermally direct polar cell, thermally indirect Ferrel cell, driven by midlatitude eddies, and a thermally forced cell at low latitudes. The last of these phenomena is commonly referred to as the Hadley circulation (HC) and consists of an idealized zone of tropospheric ascent near the equator, poleward flow aloft, subsidence in the subtropics, and a return flow at low levels in each hemisphere. Deep convection in the tropics fuels the HC [e.g., Riehl and Malkus, 1958; Riehl and Simpson, 1979; Fierro et al., 2009] with the ascending branch following the seasonal migrations of the Intertropical Convergence Zone (ITCZ) and a broader, weaker area of descent usually located between 20°–30° latitude in each hemisphere. The HC accounts for the largest portion of global overturning in the meridional-vertical plane (stream function values often peak in excess of 1 × 1011 kg s−1) and is responsible for a major redistribution of energy and heat from the equator to higher latitudes.

[3] Researchers have long studied the HC given its importance in both determining local weather and climate (e.g., tropical rainfall patterns and suppression of precipitation in the subtropics) and influences on weather patterns at higher latitudes due to impacts on the general circulation. Although the existence of the HC has been well documented for several centuries, questions remain as to how the HC has evolved over the period of record and how future global changes may affect the HC and resulting weather patterns and climate. Studies of the HC require global observations (e.g., satellites or other large-scale, upper-air arrays), numerical reanalyses, or atmospheric general circulation models (GCMs) given the comprehensive nature of the phenomenon.

[4] Comparison of observational metrics, reanalyses, and GCMs often reveal large differences in the mean representation of the HC [e.g., Mitas and Clement, 2005; Johanson and Fu, 2009], with more significant discrepancies in the observed and forecast trends of HC activity throughout the twentieth [e.g., Hu and Fu, 2007; Seidel and Randel, 2007; Mantsis and Clement, 2009] and twenty-first [Lu et al., 2008] centuries. Substantial variability may also exist in the products of those data sets considered alike, with HC trends derived from different reanalyses producing opposite results [e.g., Song and Zhang, 2007]. Furthermore, opinions differ on whether recent trends in increased equatorial rainfall and decreased subtropical humidity and cloudiness can be viewed as a strengthening of the tropical circulation [Chen et al., 2002; Fu et al., 2006; Sohn and Park, 2010] or are better attributed to instrument error and data matching across multiple satellites or sampling during prolonged El Niño–Southern Oscillation (ENSO) periods [Trenberth, 2002]. The absence of a proper consensus regarding the observed fluctuations of the HC during the previous decades thereby makes verification of GCMs and reanalyses a difficult task.

[5] Although there are numerous studies that compare precipitation and sea surface temperature (SST) from different reanalysis data sets [e.g., Quartly et al., 2007; Bosilovich et al., 2008; Ma et al., 2009; Higgins et al., 2010], few studies appear in the formal literature with a specific intercomparison of the tropical Hadley cell. Previous investigations were limited to a small number of data sets; at least eight global reanalyses are now available for study (Table 1, and section 2.1), often with increased resolution and improved model physics and data assimilation schemes relative to their earlier counterparts. Those recent studies using HC metrics derived from next-generation reanalyses (e.g., tropopause height statistics related to the HC width [Birner, 2010]) find divergent trends among older and newer data sets. Discrepancy among HC trends from reanalyses studies [e.g., Mitas and Clement, 2005; Hu and Fu, 2007] and disagreement with observations and GCMs [e.g., Mitas and Clement, 2006; Seidel and Randel 2007; Johanson and Fu, 2009] suggest the need for additional intercomparison studies using a multireanalysis ensemble to better elucidate decadal trends and potential biases.

Table 1. Reanalysis Data Sets Used in This Study
Data SetSourceData RangeResolutionAnalysis Output Resolutiona
  • a

    The analysis output resolution refers to the highest resolution available for each data set.

  • b

    Some surface variables (e.g., precipitation) are instead output on a Gaussian grid of ∼1.875° × 1.904°.

  • c

    Most data output on pressure levels is only available at a reduced resolution of 1.25° × 1.25°.

  • d

    Some variables (e.g., surface diagnostics) are available as 1-hourly output.

JRAJMA1979–2007T106L401.125° × 1.125°23 levels6-hourly
ERAINTECMWF1989–presentT255L601.5° × 1.5°37 levels6-hourly
ERA40ECMWF1957–2002T159L602.5° × 2.5°23 levels6-hourly
NNRPNCEP/NCAR1958–presentT62L282.5° × 2.5°b17 levels6-hourly
NDRPNCEP/DOE1979–2008T62L282.5° × 2.5°b17 levels6-hourly
CFSRNCEP1979–presentT382L640.5° × 0.5°37 levels1-hourly
MERRANASA1979–present2/3° × 1/2°, L602/3° × 1/2°c42 levels3-hourlyd
20CRNOAA/CIRES1871–2008T62L282.0° × 2.0°b24 levels6-hourly

[6] While the HC demonstrates a well-known annual mode [e.g., Dima and Wallace, 2003], the interannual variability is less well understood. Oort and Yienger [1996] were among the first to investigate the correlation between SSTs in the eastern equatorial Pacific and the maximum (minimum) values of the meridional mass stream function in the northern (southern) hemisphere. Oort and Yienger [1996] found that the absolute value of the stream function anomaly was generally maximized during warm ENSO events, with weakening usually observed during La Niña years. Studies since continue to attribute a large portion of the HC interannual variability to ocean-atmospheric perturbations induced by ENSO cycles [e.g., Quan et al., 2004; Ma and Li, 2008], while others find significant non-ENSO variability [e.g., Caballero, 2007].

[7] Additional considerations related to HC intensity may include changes to the oceanic mean state with warming in the tropical Indian and western Pacific oceans [Quan et al., 2004], links to monsoon activity [Trenberth et al., 2000; Dima and Wallace, 2003], and the influence of subtropical stability and midlatitude baroclinic eddy stresses on the descending branch of the HC [e.g., Held, 2000; Walker and Schneider, 2006; Caballero, 2007; Frierson et al., 2007; Lu et al., 2007; Korty and Schneider, 2008; Lu et al., 2008]. Moreover, Johanson and Fu [2009] were unable to reproduce the observed trends in HC width [e.g., Hu and Fu, 2007] when using GCMs forced with prescribed SSTs, concluding that there must be some other influence for HC width beyond SST alone. Examination of a multireanalysis ensemble provides the opportunity to identify whether SST anomalies associated with a particular ENSO phase (using prescribed or predicted SSTs) are able to sufficiently explain HC variability in reanalyses or determine if other controlling factors are present in these data sets.

2. Data and Methods

2.1. Reanalysis Data

[8] As identified in section 1, atmospheric reanalyses are used herein to examine the structure and properties of the large-scale circulation. Multiple reanalysis data sets have become publically available over the last few years and eight reanalyses (comprising both older and more recent data sets) were identified for the purposes of this study (Table 1). Selected reanalyses include the Japan Meteorological Agency (JMA) 25 year Reanalysis Project (JRA) [Onogi et al., 2007], the European Centre for Medium-Range Weather Forecasts (ECMWF) Interim Reanalysis (ERAINT) [Dee and Uppala, 2009], the ECMWF 40 year Reanalysis (ERA40) [Uppala et al., 2005], the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP/NCAR) 40 year Reanalysis Project (NNRP) [Kalnay et al., 1996], the National Centers for Environmental Prediction–Department of Energy (NCEP/DOE) Reanalysis Project (NDRP) [Kanamitsu et al., 2002], the National Centers for Environmental Prediction (NCEP) Climate Forecast System Reanalysis (CFSR) [Saha et al., 2010], the National Aeronautics and Space Administration (NASA) Modern-Era Retrospective Analysis for Research and Applications (MERRA) [Rienecker et al., 2011], and the National Oceanic and Atmospheric Administration–Cooperative Institute for Research in Environmental Sciences (NOAA/CIRES) 20th Century Reanalysis Version 2 (20CR) [Compo et al., 2011].

[9] Whereas each reanalysis was developed to meet specific goals with distinct model physics and resolutions, nearly all of the reanalyses ingest a variety of surface, upper-air, and satellite observations (when available) using either a 3D or 4D variational assimilation technique. The 20CR does not include any upper-air or satellite observations, however, and only assimilates surface pressure, SSTs, and sea ice coverage using an ensemble Kalman filter. Several of the newest reanalyses contain adaptive schemes for changing concentrations of atmospheric aerosols, CO2, and other trace gases and may thus be useful in discovering multidecadal HC trends in the recent climate. All of the reanalyses are forced with specified SSTs with the notable exception of the CFSR, which is a fully coupled land-ocean-atmosphere reanalysis. Other technical details related to the differences in the reanalysis frameworks can be found in the references above.

[10] As a first-order approach to eliminate resolution dependency in the reanalysis solutions, all relevant surface and upper-air variables were regridded to a common horizontal fixed grid of 2.5° × 2.5° using either spherical harmonics or bilinear interpolation. Upper-air variables were also regridded in the vertical with a common 10 hPa pressure increment (ranging from 1000 to 10 hPa) using linear interpolation. Unphysical solutions resulting from the interpolation (e.g., negative precipitation rates) were corrected by specifying appropriate upper and lower boundaries for each affected variable. Additional tests were performed to document the sensitivity of selected variables to interpolation functions and the mean variable quantities (and to a lesser extent, maximum and minimum values) only changed by a small amount. Monthly averages (if not already available) were calculated for all variables.

[11] The meridional stream function, Ψ, satisfying the zonal mean continuity equation in spherical coordinates can be calculated at each pressure, p, and latitude, φ, as a function of the downward integrated meridional wind, v, and is expressed as

equation image

where a is the planetary radius, g is the gravitational acceleration, and bracketed terms denote a zonal average. Using this notation, v is by definition positive (i.e., northward) in regions where −equation image > 0. Stream function values are set to zero at the top of the atmosphere, and the lowest level is modified such that Ψ equals zero at the lower boundary to ensure mass conservation and a steady state solution to the continuity equation.

[12] Several quantities of interest were determined from the latitude-pressure cross sections of Ψ. The maximum stream function value centered in the northern hemisphere, ΨN*, is a common index to measure the overturning strength [e.g., Oort and Yienger, 1996; Quan et al., 2004; Caballero, 2007] and corresponds to a critical pressure and latitude, pN* and φN*, respectively. Similar coordinates (pS* and φS*) can be defined for the minimum stream function value, ΨS*, in the southern hemisphere. Stream function maxima-minima were limited to coordinates above 800 hPa to avoid contamination by low-level features near the cell edges. The subtropical HC termini, φN and φS, are defined as the first latitudes poleward of the cell centers (φN* and φS*) in which the 700–400 hPa average value of Ψ equals zero in each hemisphere, using linear interpolation. Previous studies have used either the value of Ψ at 500 hPa [Frierson et al., 2007; Lu et al., 2007, 2008] or the 600–400 hPa average [Hu and Fu, 2007; Johanson and Fu, 2009]; the width results are generally not sensitive to using either a single level or a vertical average [e.g., Johanson and Fu, 2009]. Finally, the HC width (Δφ) can be determined as the difference between φN and φS. These metrics were calculated for each data set, in addition to those retrieved from an equally weighted ensemble average of the zonally averaged meridional stream function (rather than a simple arithmetic mean) using all the reanalyses available at any given time.

2.2. Other Data

[13] Monthly average precipitation estimates were taken from the Global Precipitation Climatology Project (GPCP) data set, Version 2.1 [Adler et al., 2003]. The GPCP data combine numerous precipitation measurements from satellite observations (low Earth orbiting microwave radiances, IR values from geostationary sensors, etc.) with surface rain gauge observations to create a blended multisource precipitation estimate on a global grid that is independent of any numerical models or physical parameterizations. Monthly precipitation data are available at a 2.5° × 2.5° horizontal resolution and were interpolated to match the coordinates of the reanalysis products used in this study.

[14] Finally, the Climate Prediction Center (CPC) Oceanic Niño Index (ONI) is used as source of SST anomalies and identification of ENSO phase. The ONI classifies an event as El Niño (La Niña) if the average SST anomaly in the Niño 3.4 region (5°N–5°S, 120°W–170°W) is +0.5°C (−0.5°C) during a 3 month running average for at least 5 consecutive months, compared to the 1971–2000 base period. The CPC ONI uses the NOAA Extended Reconstructed SST (ERSST) version 2 data set, details of which can be found in the work of Smith and Reynolds [2004]. The ENSO classifications for individual months were extended to seasonal or yearly periods if more than half of the months comprising the period of interest were identified as a particular ENSO phase.

3. Results

[15] The following analysis presents an intercomparison of reanalyses for January 1979–December 2008. The majority of the reanalyses are limited to the satellite era (most data sets begin at or after 1979; see Table 1), with only three data sets (ERA40, NNRP, and 20CR) initiating prior to the satellite epoch. Some discussion is included regarding the long-term trends (1958–2008) in these reanalyses.

3.1. HC Climatology and Ensemble Variability

[16] The annual average, zonal mean meridional mass stream function for each reanalysis data set is shown in Figure 1. All of the reanalyses reveal the expected structure of a roughly symmetric two-cell pattern with the mutual ascending branch located slightly north of the equator, in agreement with the annual average position of the ITCZ. The corresponding circulations for JJA and DJF are shown in Figures 2 and 3, respectively; the MAM and SON stream function values are similar to the annual average. Each data set demonstrates a transition toward a dominant one-cell pattern with ascent in the summer hemisphere and descent in the winter hemisphere during the solsticial seasons, as previously identified in reanalyses [e.g., Dima and Wallace, 2003] and observations [e.g., Oort and Rasmusson, 1970].

Figure 1.

Annual average zonal mean meridional mass stream function values for each reanalysis data set during the 1979–2008 period. Positive (negative) values are indicated with solid (dashed) contours and warm (cold) colors representing counterclockwise (clockwise) circulations. The thick solid contours correspond to values where Ψ = 0. Contour interval for Ψ is 2 × 1010 kg s−1.

Figure 2.

Same as Figure 1 but for JJA.

Figure 3.

Same as Figure 1 but for DJF.

[17] Though the solutions appear similar, notable differences exist among the reanalysis ensemble. Low-level (below 800 hPa) eddies appear in the subtropics for all data sets, with the exception of the ERA40. Despite having identical native vertical resolution and data output on similar pressure levels as the next-generation ECMWF reanalysis (ERAINT), the subtropical eddy feature is absent from the ERA40 solutions during all seasons and for the annual average. The 20CR data have a poleward displacement of the Ψ = 0 boundary throughout the 1000–600 hPa layer which is not seen in the annual average for the other reanalyses (Figure 1). The 20CR boundary anomalies are more pronounced for the southern hemispheric cell during JJA (Figure 2) than the northern cell in DJF (Figure 3). These edge effects are likely attributed to the difference in the data assimilation strategy between the 20CR and the remaining reanalyses; the 20CR does not assimilate upper-air or satellite data and thus should be more error prone in the southern hemisphere where the number of observations over land is greatly reduced. Finally, the NNRP meridional overturning is significantly weaker than the remaining data sets, particularly for the southern hemispheric cell during the boreal summer (Figure 2).

[18] The multireanalysis ensemble average stream function for the annual average HC is shown in Figure 4. With the exception of the previously identified edge effects, the ensemble average (ENS) is most similar in both magnitude and structure to the 20CR. Whereas the 20CR data may not be expected to verify for individual events (given the reduced number of available observations used in the data assimilation scheme), the ability to best simulate the ENS stream function potentially justifies its use in tropical and subtropical climate studies.

Figure 4.

Same as Figure 1 but for the multireanalysis ensemble average.

[19] The annual mean stream function values, level of maximum overturning, and HC width for each data set are listed in Table 2. The annual average ENS stream function has a long-term mean of 10.46 × 1010 kg s−1 and −13.10 × 1010 kg s−1 for the northern and southern hemispheric cells, respectively. These values are smaller in magnitude than the arithmetic mean of the individual members because the ENS quantities are derived from the equally weighted, zonally averaged meridional stream function, which accounts for differences in the vertical structure among data sets. The individual values of ΨN* range from 9.39 × 1010 kg s−1 (NNRP) to 12.84 × 1010 kg s−1 (ERA40), while the corresponding values of ΨS* vary from −10.38 × 1010 kg s−1 (NNRP) to −15.45 × 1010 kg s−1 (CFSR), indicating an ensemble variability of 33.2% and 38.7% for the northern and southern cells relative to the ENS mean. The ensemble variability is amplified (e.g., in excess of 40%) during the solsticial seasons with larger percent differences relative to the ensemble mean (not shown). The ERA40 has the strongest overturning for the northern cell, while the MERRA and NNRP contain the weakest circulations. The ERA40 is again among the strongest circulations for the southern hemisphere, with the CFSR producing similarly large values. The NNRP is a more obvious weak outlier for the southern cell.

Table 2. Annual Average Values of ΨN*, ΨS*, pN*, pS*, and Δφ for Each Reanalysis During the 1979–2008 Perioda
Data SetΨN*pN*ΨS*pS*Δφ
  • a

    Units for stream function magnitude, pressure level, and width are × 1010 kg s−1, hPa, and degrees of latitude, respectively.


[20] The climatological ensemble variability in HC stream function magnitude is corroborated by the annual average, zonal mean total precipitation rate produced by each reanalysis (Figure 5). All of the data sets overpredict the average total precipitation rate relative to GPCP throughout the entire tropics and most of the subtropics with the most significant differences centered near 7.5°N and 5°S. Zonal mean precipitation rates along the ITCZ vary from 5.7 mm d−1 (NNRP) to 8.9 mm d−1 (ERA40). The ERA40 produces the most tropical precipitation, whereas the MERRA and NNRP have the least rainfall, in agreement with those data sets identified in the previous analysis containing the largest and smallest stream function values. Although the coupled reanalysis (CFSR) produces the strongest annual average ΨS*, it falls in the middle to lower portion of the distribution for tropical rainfall and resides in the upper portion for subtropical precipitation rates. The CFSR also overpredicts precipitation in excess of 2.0 mm d−1 (relative to GPCP) along the tropical peak, despite previous work indicating that the CFSR is more skilled than former NCEP reanalyses in the midlatitudes [e.g., Higgins et al., 2010].

Figure 5.

Annual average zonal mean total precipitation rate (mm d−1) for the GPCP and reanalysis data sets during the 1979–2008 period.

[21] Table 2 also shows that the JRA and NNRP have higher circulation centers, pN* and pS*, than the remaining reanalyses. Although the total amount of dry air mass in the HC is identical for similar values of the stream function regardless of the central pressure level, the height of the circulation center could have significant effects upon the resulting estimates of cross-equatorial water vapor transport, as done for the NNRP by Cohen et al. [2000] and Sohn and Park [2010]. Whereas a higher circulation center may be explained by the presence of more organized convection and elevated latent heat release, it is not possible to determine these properties from the reanalyses. A common region of anomalous northward velocities (∼1.5–2.0 m s−1) was identified for the JRA and CFSR during JJA and for the annual average (not shown), located across much of the eastern equatorial Pacific at 400 hPa. The northward anomalies contribute to a weakening of the zonally averaged mass flux and stream function values at these heights and higher circulation centers in these data sets (although the circulation center for the CFSR is within reasonable agreement of the remaining data sets, it experiences a local minimum at ∼450 hPa during JJA; see Figure 2). A similar region (and similar magnitude) of northward anomalies was present in the NNRP data at 700 hPa, resulting in weaker low-level overturning and a higher circulation center. A common region of anomalous southward velocities was not identified for the northern cell during DJF, however, and further investigation is necessary to identify the root causes of the anomalous meridional winds.

[22] The annual average HC width (Table 2) has a long-term ENS average of 65.3°, varying from 62.4° (CFSR) to 67.5° (JRA). The range of the width estimates is 7.8% relative to the ensemble mean, which is significantly smaller than the relative ensemble spread for HC intensity. Curiously, there is no direct correspondence between intensity and width in the reanalysis data sets. The JRA (a relatively strong HC) might be expected to have the narrowest circulation based upon conservation of mass; the MERRA (a relatively weak HC) might likewise have a larger width. The JRA has the widest circulation of all the reanalyses, however, with the MERRA producing a relatively narrow HC. Clearly, HC intensity in the reanalyses is controlled by factors other than the width alone.

3.2. Trends in HC Intensity and Width

[23] Time series of the annual average HC intensity (ΨN* and ΨS*) and width (Δφ) are shown for each reanalysis data set and the ensemble average in Figures 6 and 7. Trends of the regression lines fit to the annual average quantities from 1979 to 2008 for each of the above variables are documented in Table 3. Longer trend values for 1958–2008 are shown in parentheses when available. All of the reanalyses show a strengthening of the northern cell with time (Figure 6, top), with significant trends at 95% ranging from 0.37 × 1010 kg s−1 decade−1 to 1.43 × 1010 kg s−1 decade−1 (Table 3). Three of the reanalyses (ERAINT, NDRP, and 20CR) show weak and nonsignificant trends. The intensification trends for the ERA40 and NNRP during 1958–2008 are somewhat smaller than those for 1979–2008, suggesting a more rapid intensification during the later period. The 20CR, which should minimize any artificial long-term trends potentially introduced by the evolution and assimilation of different satellite data streams, demonstrates a weak (though statistically significant) intensification rate of 0.09 × 1010 kg s−1 decade−1 for 1958–2008, noticeably smaller than estimates from the previous data sets. All of the reanalyses demonstrate larger intensification trends during the DJF season (with the exception of the NDRP), with the ERA40 as high as 2.68 × 1010 kg s−1 decade−1 (not shown). Overall, the decadal trends of all the reanalyses fall within the range documented in previous research. Mitas and Clement [2005] found significant intensification in the ERA40 and NNRP, no significant trend with the NDRP (or reconstructed atmospheres from global radiosonde networks), and only a slight increase with GCMs, consistent with the estimates above.

Figure 6.

Time series of the annual average (top) maximum and (bottom) minimum zonally averaged meridional mass stream function, ΨN* and ΨS*, for each reanalysis during the 1958–2008 period.

Figure 7.

Same as Figure 6 but for the Hadley cell width, Δφ.

Table 3. Yearly Trends of Annual Average Quantities From Each Reanalysis Data Set for the 1979–2008 (1958–2008) Perioda
Data SetΨN*ΨS*Δφ
  • a

    The seasonal cycle is removed from the monthly data prior to determining the long-term trend. Slopes are calculated using all available data in the period of interest except when limited by the data ranges identified in Table 1. Significant (95%) values are denoted with an asterisk. Units for HC intensity and width trends are × 1010 kg s−1 decade−1 and degrees of latitude decade−1, respectively.

  • b

    Data for the JRA are only available from 1979 to 2007.

  • c

    Data for the ERAINT are only available from 1989 to 2008.

  • d

    Data for the ERA40 are not available after August 2002.

JRAb0.93* 0.54* 1.48* 
ERAINTc0.03 −0.21 0.78* 
NDRP0.04 −0.05 1.40* 
CFSR0.37* 0.46* 0.29 
MERRA0.62* −0.20* 0.33 

[24] Figure 6 (bottom) shows more variations in trends for ΨS*. Table 3 indicates that only three reanalyses have a statistically significant intensification of the southern hemispheric cell from 1979–2008 (i.e., ERA40, NNRP, and MERRA). The JRA and CFSR illustrate a statistically significant weakening trend, with a more pronounced weakening of 1.69 × 1010 kg s−1 decade−1 and 1.30 × 1010 kg s−1 decade−1 during JJA (not shown). These two reanalyses employ chemical models for ozone production (in addition to the 20CR, which has a weak strengthening of the southern cell that is not statistically significant) instead of using directly assimilated observations or climatological profiles, suggesting that the different representation of stratospheric ozone (and associated radiative and dynamic feedbacks) may play a role in HC trends. Polvani and Kushner [2002] showed that polar stratospheric cooling (a consequence of long-term ozone reduction in the southern hemisphere) may significantly alter the meridional temperature gradient, with a resulting poleward displacement of the upper-tropospheric jets and widening (and weakening) of the tropical circulation. The NDRP and MERRA also indicate a statistically significant weakening trend in excess of 0.50 × 1010 kg s−1 decade−1 during JJA (not shown). These results cast doubt on the actual qualitative trends (i.e., strengthening or weakening) of the southern hemispheric cell during this period. The ERA40 and NNRP demonstrate much smaller intensification rates for the 1958–2008 period, suggesting again a more rapid intensification during recent decades. The corresponding 20CR long-term trend indicates only a slight (though statistically significant) intensification.

[25] Trends in the HC width, Δφ, are shown in Figure 7 and indicate a general widening with time of the annual average HC. Five of the reanalyses suggest a statistically significant widening (Table 3). The widening trends for 1979–2008 (significant values ranging from 0.78° decade−1 to 1.48° decade−1) are broadly consistent with those determined by Hu and Fu [2007] for the ERA40, NNRP, and NDRP of ∼1.1° decade−1 over the 1979–2002/2005 period. The long-term trend (1958–2008) for the NNRP is similar, while the 20CR exhibits a more modest rate of cell expansion. The ERA40 actually suggests a weak narrowing during the 1958–2008 period, casting some doubt on the certainty of the long-term trends. Moreover, the amplitude of the annual average widening for the ERA40 during 1979–2008 is also smaller than previously reported (0.41° decade−1, not statistically significant). The trend becomes slightly more comparable to the work of Hu and Fu [2007] when using their criteria for HC width (averaging the stream function throughout the 600–400 hPa layer in place of the 700–400 hPa as done here), producing a statistically significant widening of 0.54° decade−1. Differences in the exact values of the ERA40 widening trends might arise from the use of different data resolutions and the regridding methods described in section 2, though these hypotheses require further investigation.

[26] Comparison of the HC trends reveals no clear relationship between intensity and width. The JRA, which has the second largest intensification rate for the annual average ΨN*, also has the greatest widening trend (Table 3). Consequently, there is no simple correspondence of either mean state intensity and width or related HC trends (i.e., conservation of mass alone would predict a narrowing trend with HC intensification) in some of the reanalysis data sets. Finally, it is worth noting that although the range in estimates of the HC width increase among the reanalysis ensemble during the second half of the 1979–2008 period (Figure 7), five of the reanalyses converge near the ENS value by 2008. The increase in ensemble variability is therefore attributed to just a few data sets which become more pronounced outliers near the end of the period.

3.3. Interannual Variability and Connections to ENSO

[27] As previously identified, it remains a topic of debate as to how ocean-atmospheric interactions might modulate the long-term and interannual variability of the HC through connections with ENSO anomalies. Although Oort and Yienger [1996] found a significant correlation between their time series of HC stream function and eastern equatorial Pacific SST, equal amplitude SST perturbations did not generate equal stream function anomalies and several points appear anticorrelated in their data, suggesting something other than ENSO must be contributing to the variability on yearly time scales. Caballero [2007] found that non-ENSO variability accounted for more than 70% of the detrended stream function variance in the ERA40, with the increased cell strength balanced by greater extratropical wave fluxes impinging upon the tropics. The subtropical jets are thought to shift equatorward in response to an increased eddy stress, resulting in a narrower and thus more intense (given the constraint of mass conservation) overturning circulation.

[28] To better determine the contributions to interannual HC intensity, yearly and seasonally averaged values were calculated for ΨN* and ΨS* with the long-term trend removed and further categorized by ENSO phase (either as a warm, neutral, or cold event) for each data set using the ONI (see section 2.2). The interannual variability among the detrended data sets (defined as the difference of the stream function value for the maximum and minimum seasons divided by the long-term average) before ENSO subsetting ranged from 21% to 29% (ENS value of 27%) for the northern cell during DJF, with the notable exception of the ERA40. The ERA40 contained anomalously large values of ΨN* during strong overturning events, with a corresponding interannual variability of 43% relative to the data set mean. Interannual variability estimates for the southern cell during JJA ranged from 16% to 27% (ENS 12%).

[29] Detrended stream function values are shown for the northern hemispheric cell during DJF in Figure 8 (i.e., when ENSO is most active), with each ENSO phase color coded. The corresponding sample means for each phase are summarized in Table 4. Overall, the distributions show a weak clustering of the stream function values by ENSO phase for the northern (winter) cell, with El Niño events generally accounting for most of the occurrences above the 75th percentile and neutral and La Niña events representing weaker stream function values. Although the variances for individual ENSO categories may occasionally be large and contain overlap with other phases, a simple t test statistic revealed that the sample means were almost always significantly different (at 95%) for warm-neutral and/or warm-cold comparisons during DJF for most reanalyses (Table 4). The separation between neutral and cold events was more ambiguous, however, as substantial overlap in the distributions resulted in only one of the reanalyses (ERAINT) being able to identify a statistically significant difference in the sample means for the northern cell. Similar significance patterns were identified for a special long-term (1958–2008) data set (ENS50), comprising stream function values (and corresponding ENSO classifications) from an equally weighted, restricted ensemble average containing only those data sets with extended coverage (ERA40, NNRP, and 20CR). Considerable overlap exists for those ENSO neutral and La Niña events in the ENS50 data set (Figure 8), with identical phase means (18.30 × 1010 kg s−1, Table 4). A larger sample size is therefore considered not necessary in order to sufficiently determine statistical significance in the remaining data sets.

Figure 8.

Box whisker diagram of ΨN* for each reanalysis data set during DJF for the 1979–2008 period. Markers represent critical stream function values for individual seasons throughout the period and are categorized by the ENSO phase. Box plot boundaries indicate the sample median and 25th and 75th percentiles, with whiskers indicating maximum and minimum values. The long-term trend has been removed from the data to focus on interannual variability. The ENS50 data contain the long-term (1958–2008) ENSO classifications for the average of selected data sets.

Table 4. Mean Values and Number of Occurrences of ΨN* for the 1979–2008 Period for Each Reanalysis Categorized by ENSO Phase During DJFa
Data SetEl Niño(Number)Neutral(Number)La Niña(Number)All(Number)
  • a

    Stream function units are × 1010 kg s−1. Values significantly different (95%) from other ENSO phase means are noted for warm-neutral (WN), warm-cold (WC), and neutral-cold (NC) conditions.

  • b

    Data for the JRA are only available from 1979 to 2007.

  • c

    Data for the ERAINT are only available from 1989 to 2008.

  • d

    Data for the ERA40 are not available after August 2002.

  • e

    ENS50 is for 1958–2008 (see text for description).

JRAb21.96 (WN)(9)20.62(13)20.55(6)21.04(28)
ERAINTc22.08 (WN, WC)(6)20.93 (NC)(8)20.15(5)21.09(19)
ERA40d25.89 (WN)(6)23.03(11)23.28(6)23.84(23)
NNRP17.88 (WN, WC)(9)16.66(13)16.50(7)17.00(29)
NDRP22.67 (WN, WC)(9)20.05(13)20.02(7)20.85(29)
CFSR22.13 (WN, WC)(9)20.62(13)20.03(7)20.95(29)
MERRA18.92 (WN)(9)17.49(13)18.03(7)18.06(29)
20CR21.07 (WN, WC)(9)19.49(13)19.07(7)19.88(29)
ENS20.62 (WN, WC)(9)18.99(13)19.00(7)19.50(29)
ENS50e19.38 (WN, WC)(16)18.30(19)18.30(15)18.65(50)

[30] The ENSO clustering for the southern (winter) cell during JJA is less evident than the corresponding northern hemisphere winter cell (compare Tables 4 and 5). Only four of the reanalyses indicate a statistically significant difference in the mean values of ΨS* for warm-neutral and/or warm-cold conditions; all eight data sets demonstrated an ENSO dependency for ΨN* during the local hemispheric winter. Furthermore, the long-term ensemble average (ENS50) contains no statistically significant values (Table 5).

Table 5. Same as Table 4 but for ΨS* During JJA
Data SetEl Niño(Number)Neutral(Number)La Niña(Number)All(Number)
JRA−27.67 (WC)(8)−26.57 (NC)(17)−24.66(4)−26.61(29)
ERA40−27.69 (WC)(7)−27.29(13)−26.09(4)−27.21(24)
CFSR−28.98 (WN)(8)−27.91(18)−27.48(4)−28.14(30)
20CR−24.35 (WN, WC)(8)−23.42(18)−22.87(4)−23.59(30)
ENS−24.50 (WN, WC)(8)−23.82(18)−23.14(4)−23.91(30)

[31] Although the average ENSO SST anomalies are generally weaker in JJA than those observed during DJF, the different behavior between the northern and southern winter cells may be partially controlled by stratospheric ozone. Recent studies have suggested variability on interannual time scales linking ENSO and polar stratospheric temperatures in the southern hemisphere [Hitchman and Rogal, 2010; Hurwitz et al., 2011]. Polar stratospheric temperatures are generally warmer during El Niño events, resulting in a weaker polar vortex. The relaxation of the meridional temperature gradient results in a reduced jet intensity and a presumably narrower (and thus stronger) HC given the absence of any significant poleward jet contraction. The polar stratospheric feedbacks on HC intensity are thus the same sign as the expected ENSO effects inside the tropics, suggesting the overturning values should be more significant (i.e., compensating for the weaker SST anomalies during JJA) when categorized by ENSO phase. Those reanalyses using a chemical model and include ozone radiative feedbacks (JRA, CFSR, and 20CR) demonstrate a statistically significant difference for ΨS* for warm-neutral or warm-cold comparisons during JJA, suggesting these reanalyses contain the appropriate polar stratospheric feedbacks on HC intensity and width. Those reanalyses using directly assimilated observations or indirect climatological ozone profiles may have different levels of ozone forcing and stratospheric temperature response (e.g., the ERAINT and ERA40 data sets may overestimate polar stratospheric ozone in winter by up to 40% [Dragani, 2011]), thereby limiting the ability of some reanalyses to identify unique ENSO phase means for the southern hemisphere winter cell.

[32] The sensitivity to ENSO phase is nearly nonexistent during DJF or JJA for the corresponding summer hemisphere cells, with ENSO neutral or La Niña seasons responsible for the strongest overturning event in each the JRA, NNRP, NDRP, and CFSR data sets for the southern cell (not shown). Consequently, only a few of the mean values for a given phase demonstrate a statistically significant difference, telltale that other mechanisms beyond tropical SSTs must be in control of the interannual variability for the summer cell in the reanalysis data sets.

[33] Statistics regarding the detrended annual average HC width and ENSO phase are presented in Table 6. Whereas the width distributions display a similar weak ENSO phase clustering with occasional overlap (not shown) analogous to ΨN* in DJF, half of the reanalyses (including the ENS) show a statistically significant difference for warm-neutral and warm-cold events, irrespective of an overall small sample size. El Niño events account for the narrowest average HC, with La Niña years generally corresponding to the widest circulations. For example, values for the CFSR were 62.0° and 63.6° for warm and cold ENSO conditions, respectively, consistent with the expectations of Seager et al. [2003]. A smaller range was observed for the estimates of interannual variability, with HC width varying from 5% to 11% (ENS 7%) among data sets.

Table 6. Same as Table 4 but for the Annual Average Δφa
Data SetEl Niño(Number)Neutral(Number)La Niña(Number)All(Number)
  • a

    Units are degrees of latitude.

ERA4062.5 (WC)(7)63.3 (NC)(12)64.9(4)63.3(23)
NDRP64.0 (WC)(9)64.9 (NC)(17)66.5(4)64.8(30)
CFSR62.0 (WC)(9)62.5 (NC)(17)63.6(4)62.5(30)
MERRA63.0 (WC)(9)63.5 (NC)(17)64.7(4)63.5(30)
ENS64.7 (WC)(9)65.4(17)66.3(4)65.3(30)

[34] While the above results (i.e., narrower circulations during El Niño and wider cells during La Niña) are generally true for longer time periods, almost none of the reanalyses demonstrate a statistically significant difference from the other ENSO phase means for individual seasons (not shown). While the width calculations are often noisy during the solsticial seasons due to fluctuations in the poleward terminus of the summer (and weak amplitude) cell, similar null results were obtained during MAM and SON when the width retrievals become more stable. Calculations using the width of the individual cells (in place of the entire tropical width, Δφ) demonstrated some ENSO dependence, but not to the degree of significance identified for the cell intensities on seasonal time scales (e.g., DJF, Table 4). As such, it is possible that different mechanisms (and time scales) affect the HC intensity and width. At the onset of a warm ENSO event, the circulation intensity may increase (particularly in the dominant cell) as a result of warmer boundary conditions driving an increase in near-equatorial clouds and precipitation while maintaining a similar HC width. The effects of the SST anomalies may propagate to the upper tropical troposphere with a maximum correlated response lagging by 1–2 seasons [e.g., Newell and Weare, 1976; Pan and Oort, 1983; Yulaeva and Wallace, 1994]. These anomalies eventually alter the meridional temperature gradient and contract the subtropical jets (as outlined by Seager et al. [2003]), resulting in a narrower tropical circulation. The long-term narrowing (in response to SSTs) would eventually result in an additional intensification of the HC, attributed to conservation of mass, separate from the initial strengthening owing to the amplified convective mass flux. The ENSO metric used in this study, nevertheless, requires the presence of a persistent SST anomaly for 5 months prior to the period being classified as an ENSO event, thereby perhaps allowing a sufficient time for the subtropical jets to contract and result in a narrower HC for the seasonal periods in question.

4. Summary and Discussion

[35] Previous studies using observations and reanalyses suggest an intensification and poleward expansion of the tropical Hadley circulation throughout the twentieth century. Although the rates of intensification and expansion vary by study (or may occasionally be absent as in many GCMs), the climatological representation of the HC and decadal trends were previously undocumented for many of the newest reanalyses currently being produced by several meteorological centers worldwide.

[36] Significant ensemble variability was found in the mean state variables describing the HC intensity. Differences in the range of climatological mean values for the annual average meridional mass stream function among data sets was upwards of 33.2% and 38.7% of the mean ensemble average for the northern and southern cells, with higher relative differences observed for shorter (i.e., seasonal) periods. The ERA40, JRA, and CFSR produced the strongest meridional overturning, whereas the MERRA and NNRP were consistently the weakest. Mean state ensemble variability was consistent with the zonal average total precipitation rate and midlevel vertical velocity among data sets. Although differences of a few degrees were identified in the reanalysis ensemble for the annual average tropical width, the range of ensemble variability was only 7.8% of the mean ENS width.

[37] The NNRP, perhaps the most widely used atmospheric reanalysis, produced anomalously low stream function amplitudes for the southern hemispheric cell during JJA, in addition to the annual average. Moreover, the NNRP and JRA both produced a higher circulation center compared to other reanalyses, leading to possible biases in cross-equatorial vapor transport and other moisture quantities used to indirectly assess the strength of the tropical circulation in these data sets.

[38] The 20CR best matched the multireanalysis ensemble average HC with the exception of a poleward shift in the low-level subtropical terminus, likely attributed to the differences in data assimilation and lack of land-based observations over the subtropics, particularly in the southern hemisphere. While the 20CR may not validate for individual events, it nevertheless produced a realistic HC structure and intensity similar to the average of seven other reanalyses, all which include countless more observations in their data assimilation (including upper-air and satellite data), thereby justifying its use in potential future tropical and subtropical climate studies. Long-term (1958–2008) trends in the 20CR data set suggest a weak intensification (0.12 × 1010 kg s−1 decade−1 average for both cells) and a modest widening (0.62° decade−1) of the tropical circulation, with slope values smaller than that of previous studies using reanalysis data sets [e.g., Mitas and Clement, 2005; Hu and Fu, 2007].

[39] The latest reanalyses generally fall within the previous ensemble spread for mean state HC strength with larger uncertainty in the tropical width owing to a few outliers during the most recent years. Some discrepant trends emerged among newer data sets. The JRA and the CFSR both indicated a statistically significant weakening of the southern hemispheric cell, unlike the remainder of the reanalyses. The coupled reanalysis (CFSR) produced the smallest positive trend for HC widening in the annual average during the 1979–2008 period, though was not statistically significant. The JRA produced the strongest widening trend over the period of interest (∼1.5° decade−1), which was larger than previous trends using reanalysis data [Hu and Fu, 2007] but smaller than those widening estimates derived from observations [e.g., Hudson et al., 2006; Seidel and Randel, 2007]. Furthermore, the sign of the widening trend is discrepant with the work of Birner [2010], in which the JRA was the only data set to suggest a narrowing of the HC when using tropical tropopause statistics derived from multiple reanalyses.

[40] Large ranges were identified in the relative variability of HC intensity among the detrended data sets, with interannual variability estimates of 21%–43% and 16%–27% for the northern and southern hemispheric winter cells, respectively. Interannual variability of the annual average HC width varied from 5% to 11%. Previous HC interannual variability was thought to be highly influenced by ENSO cycles. Dynamical theories predicting the HC extent (and intensity) fall into two main categories: those that determine the tropical width as a function of (1) “interior” diabatic forcing under the assumption of upper-tropospheric angular momentum conservation [e.g., Held and Hou, 1980] or (2) “exterior” forcing mechanisms including the role of momentum fluxes by midlatitude baroclinic eddies [e.g., Held, 2000] and subtropical stability [Frierson et al., 2007; Lu et al., 2007; Korty and Schneider, 2008] or baroclinicity [Lu et al., 2008]. Certain regions of the HC are thought to be better represented by different theories. The dominant cross-hemispheric winter cell may be more controlled by diabatic forcing whereas the weaker summer cell is likely highly influenced by eddies [e.g., Caballero, 2007]. Separation of the seasonally averaged stream function magnitudes by ENSO phase revealed a weak clustering and statistically significant difference between the mean values for El Niño and ENSO neutral or La Niña events in almost all the reanalysis for the winter cell intensity, with little difference for the summer cell. The reanalysis results give credence to the above ideas, suggesting that ENSO cycles and diabatic forcing from SST anomalies dominate the variability of the winter cell, whereas other factors must exert an important influence on the summer cell. Clustering by ENSO phase was less evident for the southern hemispheric winter cell during JJA and may be related to other physical mechanisms including stratospheric ozone. The statistical significance of ENSO phase was only identified for the annual average HC width, despite exerting a significant influence on the HC intensity at seasonal time scales. A potential explanation focusing on the response times of convective mass fluxes versus the contraction of the subtropical jet was provided in section 3, though this hypothesis requires more evaluation.

[41] Although this research aims to provide an intercomparison of reanalysis solutions regarding the tropical Hadley circulation, it remains a topic of debate whether reanalysis data can be used to identify long-term dynamical and physical climate trends given the nature of discontinuous data assimilation and the uncertainty associated with analysis fields prior to the age of global satellite coverage [e.g., Thorne and Vose, 2010]. While reanalysis data provides some information about modeled clouds (generally restricted to fractional cloud coverage over three discrete atmospheric layers), reanalysis itself does not explicitly predict cloud type. Future work will use reanalysis data to simulate observed cloud regimes or “weather states,” as sometimes done for GCM output [e.g., Zhang et al., 2005; Williams and Tselioudis, 2007; Williams and Webb, 2008]. The ability (or lack thereof) to simulate specific cloud regimes among different reanalyses could be paramount to understanding differences in the mean state representation or long-term HC trends in the multireanalysis ensemble [e.g., Song and Zhang, 2007].


[42] The JRA data are generated by the cooperative research project of the JRA-25 long-term reanalysis by the Japan Meteorological Agency (JMA) and the Central Research Institute of Electric Power Industry (CRIEPI). ECMWF ERA-Interim data have been obtained from the ECMWF data server. The CFSR data were developed by NOAA's NCEP. The CFSR data used for this study are from NOAA's National Operational Model Archive and Distribution System (NOMADS), which is maintained at NOAA's National Climatic Data Center (NCDC). MERRA data were generated by the NASA Global Modeling and Assimilation Office (GMAO) and disseminated by the Goddard Earth Sciences (GES) Data and Information Services Center (DISC). The 20th Century Reanalysis V2 data are provided by the NOAA/OAR/ESRL PSD, Boulder, Colorado, from their Web site at The NNRP, NDRP, ERA40, and JRA data were obtained from the Research Data Archive (RDA), which is maintained by the Computational and Information Systems Laboratory (CISL) at the National Center for Atmospheric Research (NCAR). NCAR is sponsored by the National Science Foundation (NSF). The original data are available from the RDA ( in data sets ds090.0, ds091.0, ds120.0/ds120.1, and ds625.1. Two anonymous reviewers provided helpful comments. This study was funded by NSF grant ATM-0449782 and NASA grant NNX10AG89G.