• Allen, M. R., and S. F. B. Tett, 1999: Checking for model consistency in optimal fingerprinting. Climate Dyn., 15 , 419434.

  • Allen, M. R., and P. A. Stott, 2002: Estimating signal amplitudes in optimal fingerprinting: Part I. Theory. Climate Dyn., in press.

  • Bengtsson, L., E. Roeckner, and M. Stendel, 1999: Why is the global warming proceeding much slower than expected. J. Geophys. Res., 104 , 38653876.

    • Search Google Scholar
    • Export Citation
  • Brown, S. J., D. E. Parker, C. K. Folland, and I. Macadam, 2000a: Decadal variability in the lower-tropospheric lapse rate. Geophys. Res. Lett., 27 , 9971000.

    • Search Google Scholar
    • Export Citation
  • Brown, S. J., and D. Sexton, 2000b: Differential changes in observed surface and atmospheric temperature since 1979. Hadley Centre Tech. Note HCTN12, 41 pp.

    • Search Google Scholar
    • Export Citation
  • Chang, P., R. Saravanan, L. Ji, and G. C. Hegerl, 2000: The effect of local sea surface temperature on atmospheric circulation over the tropical Atlantic sector. J. Climate, 13 , 21952216.

    • Search Google Scholar
    • Export Citation
  • Chelliah, M., and C. F. Ropelewski, 2000: Reanalysis based tropospheric temperature estimates: Uncertainties in the context of global climate change detection. J. Climate, 13 , 31873205.

    • Search Google Scholar
    • Export Citation
  • Christy, J. R., and R. T. McNider, 1994: Satellite greenhouse warming. Nature,367, p.325.

  • Christy, J. R., R. W. Spencer, and W. D. Braswell, 2000: MSU tropospheric temperatures: Dataset construction and radiosonde comparisons. J. Atmos. Oceanic Technol., 17 , 11531170.

    • Search Google Scholar
    • Export Citation
  • Crowley, T. J., 2000: Causes of climate change over the past 1000 yrs. Science, 289 , 270277.

  • Gaffen, D. J., B. D. Santer, J. S. Boyle, J. R. Christy, N. E. Graham, and R. J. Ross, 2000: Multidecadal changes in vertical temperature structure of the tropical troposphere. Science, 287 , 12421245.

    • Search Google Scholar
    • Export Citation
  • Gillett, N., M. R. Allen, and S. F. B. Tett, 2000: Modeled and observed variability in atmospheric vertical temperature structure. Climate Dyn., 16 , 4961.

    • Search Google Scholar
    • Export Citation
  • Hartmann, D. L., 2002: Tropical surprises. Science, 295 , 811813.

  • Hasselmann, K., 1976: Stochastic climate models. Tellus, 28 , 473485.

  • Hill, D. C., M. R. Allen, and P. A. Stott, 2001: Allowing for solar forcing in the detection of human influence on tropospheric temperatures. Geophys. Res. Lett., 28 , 15551558.

    • Search Google Scholar
    • Export Citation
  • Houghton, J. T., L. G. Meiro Filho, B. A. Callander, N. Harris, A. Kattenberg, and K. Maskell, Eds. 1996: Climate Change 1995: The Science of Climate Change. Cambridge University Press, 572 pp.

    • Search Google Scholar
    • Export Citation
  • Houghton, J. T., Y. Ding, D. J. Griggs, M. Noguer, P. J. van der Linden, X. Dai, K. Maskell, and C. A. Johnson, Eds.,. 2001: Climate Change 2000: The Scientific Basis. Cambridge University Press, 892 pp.

    • Search Google Scholar
    • Export Citation
  • Hurrell, J. W., and K. E. Trenberth, 1998: Difficulties in obtaining reliable temperature trends: Reconciling the surface and satellite MSU records. J. Climate, 11 , 945967.

    • Search Google Scholar
    • Export Citation
  • Jones, P. D., 1994: Hemispheric surface air temperature variations: A reanalysis and an update to 1993. J. Climate, 7 , 17941802.

  • Jones, P. D., M. New, D. E. Parker, S. Martin, and I. G. Rigor, 1999: Surface air temperature and its changes over the past 150 years. Rev. Geophys., 37 , 173199.

    • Search Google Scholar
    • Export Citation
  • Kalnay, E., and Coauthors. 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77 , 437471.

  • Lambert, S. J., 1990: Discontinuities in the long-term Northern Hemisphere 500-mbar dataset. J. Climate, 3 , 14791484.

  • Lean, J., J. Beer, and R. Bradley, 1995: Reconstruction of solar irradiance since 1600: Implications for climate change. Geophys. Res. Lett., 22 , 31953198.

    • Search Google Scholar
    • Export Citation
  • Levitus, S., J. I. Boyer, and C. Stephens, 2000: Warming of the world ocean. Science, 287 , 22252229.

  • Mantua, J. N., R. S. Hare, Y. Zhang, J. M. Wallace, and R. C. Francis, 1997: A Pacific interdecadal climate oscillation with impacts on salmon production. Bull. Amer. Meteor. Soc., 78 , 10691080.

    • Search Google Scholar
    • Export Citation
  • Mitchell, T., and J. M. Wallace, 1996: ENSO seasonality: 1950–78 versus 1979–92. J. Climate, 9 , 31493161.

  • Parker, D. E., C. K. Folland, and M. Jackson, 1995: Marine surface temperature: Observed variations and data requirements. Climate Change, 31 , 559600.

    • Search Google Scholar
    • Export Citation
  • Parker, D. E., M. Gordon, D. P. N. Cullum, D. M. H. Sexton, C. K. Folland, and N. Rayner, 1997: A new global gridded radiosonde temperature data base and recent temperature trends. Geophys. Res. Lett., 24 , 14991502.

    • Search Google Scholar
    • Export Citation
  • Roeckner, E., L. Bengtsson, J. Feichter, J. Lelieveld, and H. Rodhe, 1999: Transient climate change simulations with a coupled atmosphere–ocean GCM including the tropospheric sulfur cycle. J. Climate, 12 , 30043032.

    • Search Google Scholar
    • Export Citation
  • Santer, B. D., J. J. Hnilo, T. M. L. Wigley, J. S. Boyle, C. Doutriaux, M. Fiorino, D. E. Parker, and K. E. Taylor, 1999: Uncertainties in “observational” estimates of temperature change in the free atmosphere. J. Geophys. Res., 104 , 63056333.

    • Search Google Scholar
    • Export Citation
  • Santer, B. D., and Coauthors. 2000a: Interpreting differential temperature trends at the surface and in the lower troposphere. Science, 287 , 12271231.

    • Search Google Scholar
    • Export Citation
  • Santer, B. D., and Coauthors. 2000b: Statistical significance of trends and trend differences in layer average atmospheric temperature time series. J. Geophys. Res., 105 , 73377356.

    • Search Google Scholar
    • Export Citation
  • Santer, B. D., and Coauthors. 2001: Accounting for the effects of volcanoes and ENSO in comparisons of modeled and observed temperature trends. J. Geophys. Res., 106 , 2803328059.

    • Search Google Scholar
    • Export Citation
  • Spencer, R. W., and J. R. Christy, 1992: Precision and radiosonde validation of satellite gridpoint temperature anomalies. Part II: A tropospheric retrieval and trends during 1979–90. J. Climate, 5 , 858866.

    • Search Google Scholar
    • Export Citation
  • Stott, P. A., S. F. B. Tett, G. S. Jones, M. R. Allen, J. F. B. Mitchell, and G. J. Jenkins, 2000: External control of 20th century temperature by natural and anthropogenic forcings. Science, 290 , 21332137.

    • Search Google Scholar
    • Export Citation
  • Tett, S. F. B., J. F. B. Mitchell, D. E. Parker, and M. R. Allen, 1996: Human influence on the atmospheric vertical temperature structure: Detection and observations. Science, 274 , 11701173.

    • Search Google Scholar
    • Export Citation
  • Thompson, D. W. J., and J. M. Wallace, 1998: The Arctic Oscillation signature in wintertime geopotential height and temperature fields. Geophys. Res. Lett., 25 , 12971300.

    • Search Google Scholar
    • Export Citation
  • Thompson, D. W. J., and G. C. Hegerl, 2000: Annular modes in the extratropical circulation. Part II: Trends. J. Climate, 13 , 10181036.

    • Search Google Scholar
    • Export Citation
  • Venzke, S., M. R. Allen, R. T. Sutton, and D. P. Rowell, 1999: The atmospheric response over the North Atlantic to decadal changes in sea surface temperature. J. Climate, 12 , 25622584.

    • Search Google Scholar
    • Export Citation
  • von Storch, H., and F. W. Zwiers, 1999: Statistical Analysis in Climate Research. Cambridge University Press, 484 pp.

  • Wallace, J. M., Y. Zhang, and J. A. Renwick, 1995: Dynamic contribution to hemispheric mean temperature trends. Science, 270 , 780783.

    • Search Google Scholar
    • Export Citation
  • Wallace, J. M., and Coeditors,. 2000: Reconciling Observations of Global Temperature Change. National Academy Press, 85 pp.

  • Wentz, F. J., and M. Schabel, 1998: Effects of orbital decay on satellite-derived lower-tropospheric trends. Nature, 344 , 661664.

  • Wentz, F. J., C. Mears, and D. Seidel, 2001: Lower tropospheric air temperature derived from a blended analysis of MSU, SSM/T1, SSM/I, NCEP/NCAR reanalysis, and Reynolds SST. RSS Tech. Rep. 062801, Remote Sensing Systems, Santa Rosa, CA, 46 pp.

    • Search Google Scholar
    • Export Citation
  • Yulaeva, E., and J. M. Wallace, 1994: The signature of ENSO in global temperature and precipitation fields derived from the microwave sounding unit. J. Climate, 7 , 17191736.

    • Search Google Scholar
    • Export Citation
  • Zhang, Y., J. M. Wallace, and D. S. Battisti, 1997: ENSO-like interdecadal variability: 1900–93. J. Climate, 10 , 10041020.

  • View in gallery

    (top) Comparison between annual and global (90°N–40°S) mean lower-tropospehric temperature (LTT; MSU2lt from satellite measurements and a comparable average from radiosonde data), (middle) surface air temperature (SAT; based on Jones et al. 1999), and (bottom) the difference time series SAT − LTT lapse rate, difference averaged only over grid points where both SAT and LTT data were available. LTT and lapse rate averages based on radiosonde data (black; Parker et al. 1997) are compared to those based on satellite data (gray; Christy et al. 2000), radiosonde data before 1964 are not shown due to homogeneity concerns. Ticks are offset by 0.2 K. All time series are centered on their mean over the period 1979–97

  • View in gallery

    Spatial pattern of the trend 1979–Jan 2000 in (a) LTT (based on satellite data), (b) SAT, and (c) lapse rate (K decade−1). The trend pattern for each field is based on the least square fit of a linear trend to the time series of each grid point

  • View in gallery

    Evolution of the average lapse rate (SAT − LTT) for NH, SH, and the Tropics from radiosonde (black) and satellite data (gray). Note the positive trend over the satellite period in all sections of the globe (see also Table 1)

  • View in gallery

    Trend pattern of zonally averaged atmospheric temperature in pressure coordinates (hPa) based on (a) radiosonde temperature data (1979–Apr 1998), (b) reanalysis data (1979–Jan 2000), (K decade−1). (c) Trend in zonal wind from reanalysis data (m s−1 decade−1). Each trend is least square fitted to zonally averaged temperature/wind at each latitude and pressure grid point. Note that the zonal temperature trend pattern from reanalysis data resembles that from radiosonde data more closely (particularly in the NH) if both trends are calculated to Apr 1998 only

  • View in gallery

    Cold ocean warm land (COWL) pattern from (a) SAT compared to a regression of the COWL index on (b) LTT, using satellite data, and (c) lapse rate. Positive COWL anomalies over the continents are weakened aloft. Therefore positive lapse rate anomalies over the continents coincide with positive COWL anomalies. Units relate to a 1 std dev change in the COWL index

  • View in gallery

    Comparison of the cold season (Nov–Apr) NH mean lapse rate north of 30°N with the COWL index. (top) The cold season lapse rate average for radiosonde (black) and satellite data (gray), (middle) COWL index, normalized as regressed on lapse rate, and (bottom) residual NH lapse rate time series after removing the influence of the COWL index and El Niño–related variability. The removal is based on spatial multiregression patterns (section 3c). The dashed curve in the middle panel shows the Arctic Oscillation index, which explains part of the variability of the COWL index

  • View in gallery

    Comparison of tropical and subtropical mean (30°N–30°S) LTT, SAT, CTI* (cold tongue index for Tropicswide surface temperature response on El Niño, scaled by its regression onto LTT), and lapse rate. The bottom time series (res. l. r.) is the lapse rate after regressing out the effects of El Niño and COWL based on spatial fields (same as in Fig. 6). The time of major volcanic eruptions is marked by vertical bars in the bottom two panels.

  • View in gallery

    Lapse rate response to El Niño based on regressions of surface and satellite data for (a) Nov–Apr and (b) May–Oct on CTI* [K (std dev)−1]

  • View in gallery

    Scatter diagram illustrating the covarying anomalies of tropical and subtropical mean surface temperature and lower-tropospheric temperature (averaged from 30°N to 30°S; only collocated grid points are used). Gray crosses compare satellite-based LTT and SAT (trends over the satellite era are removed), black crosses compare radiosonde-based LTT and SAT (variability beyond 5 yr removed). The slanted line indicates a regression line based on a (ordinary least square) regression from 11-month running mean data (1.5 enhancement of surface anomalies in the lower troposphere), the dotted line a (total least square) best fit using monthly data (1.8 enhancement). The fat circles show for comparison long-term trends (K per trend length) from satellite (gray) and radiosonde (black) data

  • View in gallery

    Comparison of tropical and subtropical lower-tropospheric temperature from radiosonde and satellite data. (top) Raw LTT and (bottom) LTT after the surface temperature influence has been subtracted (LTT*, subtraction based on enhancement of 1.5). (middle) Negative lapse rate (LTT − SAT) for comparison. Note that the influence of El Niño is reduced in LTT* compared to LTT − SAT. Alternative LTT* time series based on a differently estimated enhancement factor (estimates ranging between 1.3 and 1.8) are qualitatively very similar and also show negative trends over the satellite period. The time of major volcanic eruptions is given by vertical bars

  • View in gallery

    Regression of the LTT* time series (Fig. 10) based on surface and satellite data on (a) lower-tropospheric temperature and (b) surface temperature patterns to illustrate the origin of LTT* anomalies [K (std dev)−1]. Months 6–24 following volcanic eruptions (see Fig. 10) have been omitted and the long-term trend of LTT* over the satellite period has been subtracted prior to computing regression patterns. Correlation patterns with LTT* (not shown) are similar in the Tropics and subtropics, with peak correlations in the eastern half of the Tropics and subtropics (30°E–180°) up to 0.4, and very weak correlations in the extratropics. Correlation patterns with SAT are weak; structures are not robust

  • View in gallery

    (a) Regression of LTT* (tropical plus subtropical LTT minus surface influence; Fig. 10) on zonal mean surface (dotted) and lower-tropospheric (solid) temperature. Fat lines show the regression of satellite-derived LTT* on zonal mean SAT and satellite LTT, thin lines the regression of radiosonde-based LTT* (derived from radiosonde data from 1964 onward) on radiosonde data. A similar regression using Tropics only (20°N–20°S, enhancement 1.4) yielded a similar zonal pattern, however, with only very little enhanced warming in the subtropics. (b) Similar to (a), but for correlations instead of regressions. (c) Zonal average of LTT and SAT trend patterns between 1979 and Jan 2000, thin lines are results from trends based on raw fields, thick lines from LTT and SAT fields after removing ENSO- and COWL-related variability (section 3c)

  • View in gallery

    Regression of LTT* from satellite and surface data on (a) zonal mean temperature from radiosondes and from (b) reanalysis data. (c) Zonal mean zonal wind from reanalysis data. Correlation patterns (not shown) are similar in Tropics and subtropics, with peak correlations of 0.5–0.6 in the troposphere and correlations extending into the stratosphere, but weak in the extratropics.

  • View in gallery

    Comparison of (top) LTT* (Tropics and subtropics) and (bottom) an index of the variability characterized by Figs. 11–13. The index is the expansion time series of the spatial pattern of zonal wind at 200 mb between 50°N and 50°S (not shown); it is scaled to its regression on LTT*

  • View in gallery

    (a) Comparison between SAT and satellite-based LTT in Tropics and subtropics similar to Fig. 9 (crosses: monthly variability, trend removed; green circle: long-term observed trend in K per trend length) and variability in a coupled ocean–atmosphere general circulation model (Roeckner et al. 1999). The gray crosses show monthly variability in the “control” simulation of the model (same missing data mask and trend removal as in the observation; only every 15th monthly value displayed for clarity). The blue circles show trends from the control simulation over a period of 21.1 yr (same data mask as observations 1979–Jan 2001), the red circles from an anthropogenic climate change simulation (forced with increases in greenhouse gases, tropospheric ozone, and direct and indirect effects of sulfate aerosol forcing, 21.1-yr running trends from the model period 1950–2010). (b) Same as (a) but focusing on the observed period 1964–81, using the radiosonde data mask. Note that the model simulates monthly variability very well, but couples the surface and lower-tropospheric trends much more tightly along the regression line than observed. The comparison is limited to grid boxes covered in both LTT and SAT data for each panel

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 235 75 7
PDF Downloads 63 33 2

Influence of Patterns of Climate Variability on the Difference between Satellite and Surface Temperature Trends

View More View Less
  • 1 Department of Earth and Ocean Sciences, Nicholas School of the Environment, Duke University, Durham, North Carolina
  • | 2 Department of Atmospheric Sciences, University of Washington, Seattle, Washington
Full access

Abstract

During the past 20 years, satellite measurements of tropospheric temperature have shown a slower rate of global temperature increase than surface air temperature, yielding an increase in the surface to lower-troposphere lapse rate of 0.12 K decade−1 from 1979 to August 2001. This increase in lapse rate was preceded by a decrease over the previous 15-yr interval.

The influence of patterns of climate variability on the global- and hemispheric-scale lapse rate was investigated, based on observations of surface and tropospheric temperature from satellite, radiosonde, surface air, and sea surface data. It was found that a substantial fraction of winter-to-winter lapse rate variability in the Northern Hemisphere mid- to high latitudes is dynamically induced. In the Tropics and subtropics, a distinctive signature of El Niño is apparent in the interannual variations in lapse rate. A small additional amount of month-to-month variability can be attributed to zonally symmetric circulation changes at lower latitudes that are linearly independent of ENSO. Trends in these patterns can account only for a small fraction of the observed trend in lapse rate. The combination of strong surface warming and very small tropospheric warming in the Tropics and subtropics over the recent 21 yr is extremely unusual in the context of data from a coupled climate model. The same can be said of the observed trends of the previous 15 yr. Thus, it is concluded that structured patterns of climate variability account for much of the variability in lapse rate on monthly and interannual timescales, but not on interdecadal timescales.

Corresponding author address: Gabriele Hegerl, Dept. of Earth and Ocean Sciences, Nicholas School of the Environment, Duke University, Durham, NC 27708. Email: hegerl@duke.edu

Abstract

During the past 20 years, satellite measurements of tropospheric temperature have shown a slower rate of global temperature increase than surface air temperature, yielding an increase in the surface to lower-troposphere lapse rate of 0.12 K decade−1 from 1979 to August 2001. This increase in lapse rate was preceded by a decrease over the previous 15-yr interval.

The influence of patterns of climate variability on the global- and hemispheric-scale lapse rate was investigated, based on observations of surface and tropospheric temperature from satellite, radiosonde, surface air, and sea surface data. It was found that a substantial fraction of winter-to-winter lapse rate variability in the Northern Hemisphere mid- to high latitudes is dynamically induced. In the Tropics and subtropics, a distinctive signature of El Niño is apparent in the interannual variations in lapse rate. A small additional amount of month-to-month variability can be attributed to zonally symmetric circulation changes at lower latitudes that are linearly independent of ENSO. Trends in these patterns can account only for a small fraction of the observed trend in lapse rate. The combination of strong surface warming and very small tropospheric warming in the Tropics and subtropics over the recent 21 yr is extremely unusual in the context of data from a coupled climate model. The same can be said of the observed trends of the previous 15 yr. Thus, it is concluded that structured patterns of climate variability account for much of the variability in lapse rate on monthly and interannual timescales, but not on interdecadal timescales.

Corresponding author address: Gabriele Hegerl, Dept. of Earth and Ocean Sciences, Nicholas School of the Environment, Duke University, Durham, NC 27708. Email: hegerl@duke.edu

1. Introduction

The introduction of satellite measurements of lower-tropospheric, tropospheric, and stratospheric temperature in 1979 provided the first comprehensive global measurements of upper-air temperature. It also offered the possibility of verifying the reported strong surface warming measured by station data, upon which a large part of the claims of detection of an anthropogenic influence on climate have been based. However, the satellite measurements show only about half of the warming recorded by the surface data from 1979 to January 2000 and only a third if the analysis is extended to mid-2001 (Table 1, average from 40°S to 90°N due to limited surface coverage in the Southern Hemisphere south of 40°S, see below). Recently, a panel of the National Research Council (Wallace et al. 2000) concluded that errors in surface temperature measurements (e.g., due to the urban heat island effect, and corrections to systematic errors of sea surface temperature) are likely small over the last few decades in comparison to the observed warming (see also Jones et al. 1999). Warming has also been observed in the subsurface ocean (Levitus et al. 2000). The relatively close agreement between satellite and radiosonde data, particularly if differences in spatial coverage are explicitly taken into account (discussed below, see also Santer et al. 1999; Brown et al. 2000a), increases the degree of confidence that can be attributed to the satellite data. However, there is still observational uncertainty, for example, in the processing of satellite data (Wentz and Schabel 1998; Wentz et al. 2001). The National Academy panel concluded that the strong warming of surface temperature over the satellite era is “undoubtedly real,” and that the evidence suggests that the troposphere warmed less rapidly over the same time period. However, it is important to keep in mind that the exact magnitude of the difference between surface and lower-tropospheric warming is subject to observational uncertainty in both satellite and radiosonde data (see, e.g., Santer et al. 1999) as well as surface data.

Brown et al. (2000a) and Gaffen et al. (2000) both show that the disparity between the surface and tropospheric temperature trends is concentrated close to the earth's surface. Trends throughout the troposphere are rather uniform up to about 300 hPa (above which they decrease with height and eventually become negative). We also found that the evolution of microwave sounding unit lower-tropospheric (MSU2lt) global temperature anomalies is very similar to that from 1000- to 500-mb thickness in reanalysis data, also suggesting that the disparity in the trends is concentrated at low levels (not shown).

The present paper focuses on the influence of modes of climate variability on the difference between surface and lower-tropospheric temperature anomalies on monthly and longer timescales (a detailed definition of lower-tropospheric temperature is given below). We refer to this difference as the “lapse rate,” since it is a surrogate for variations in the average lapse rate in the troposphere. We attempt to determine whether trends in recognizable atmospheric modes of variability account for all or part of observed trends in the lapse rate. We find that a considerable fraction of month-to-month variability in the lapse rate can be explained by well-defined modes of variability, particularly in the mid- to high-latitude Northern Hemisphere. In the Tropics and subtropics, where the observed lapse rate tends to be close to moist adiabatic, the lapse rate has been shown to vary with surface temperature in accord with the flattening and steepening of the moist adiabatic lapse rate as surface temperature rises and falls (e.g., Hurrell and Trenberth 1998; Gillett et al. 2000). We also find evidence of this behavior. However, the overall trend in the global mean lapse rate since 1979 is nearly unaffected if this effect or the effect of modes of variability is subtracted.

This paper is structured as follows: In the next section, we describe the data used in this study and review the spatial pattern of trends over the satellite era. In the third section, dynamically induced lapse rate variability is discussed. The fourth section discusses the coupling between surface and lower-tropospheric averages and the temperature variability within the free troposphere that affects lapse rate averages. The fifth section illustrates coupling between the surface and lower troposphere in a coupled climate model and discusses a final attempt to distinguish a spatial pattern responsible for the observed lapse rate trend using model data.

2. Data and observed trends

Tropospheric temperature measurements have been compiled from satellite measurements by Spencer and Christy (1992) and Christy et al. (2000; see also Christy and McNider 1994). The MSU2lt (earlier versions were referred to as MSU-2r) are a weighted integral over tropospheric temperature from the surface into the stratosphere, with peak weights at about 650 hPa in pressure coordinates (which we also use for figures; note that the peak weights are lower if volume weighting is applied; J. R. Christy 2001, personal communication), and small, partly negative, weights in the stratosphere. We have employed a recent version of the data (“cycle D”), which has been corrected for known systematic biases, for example, orbital decay effects (Christy et al. 2000). Both surface and satellite data are continuously updated. Our analysis is based on data through early 2000; however, we have updated Fig. 1 and Table 1 using satellite and surface temperature data through August 2001.

Radiosonde data have been compiled by Parker and coworkers at the Hadley Centre. We have used the second version of the radiosonde dataset, HadRT2.1 (Parker et al. 1997), where systematic biases and inhomogeneities, for example, in the Australian radiosonde stations, have been removed using version c of the MSU4 and MSU2lt satellite record; we use data from 1958 to April 1998. Due to a reported inhomogeneity in radiosonde measurements (Lambert 1990) and very limited coverage in the Southern Hemisphere prior to about 1964 (Houghton et al. 1996), we restrict our analysis to the years after that. Other versions of the data (HadRT2.2, HadRT2.3) use spatial infilling techniques or are blended with reanalysis data to improve coverage. We have not applied these data in order to keep the satellite and radiosonde datasets as independent from each other as possible (note that bias corrections based on satellite data in HadRT2.1 have only been applied where the station history indicated an inhomogeneity). We apply a weighted average of temperature measurements over four layers of radiosonde data (850, 700, 500, and 300 hPa; if the highest level is missing the weighting is adjusted) with weights chosen to approximate the MSU2lt satellite average (Parker et al. 1997, we refer to this in the following as R-MSU2lt). The spatial coverage of the R-MSU2lt gridded dataset is rather sparse, its grid boxes cover 30%–45% of the globe between 90°N and 40°S, with coverage highest (∼60%) in the NH north of 30°N and lowest in the deep Tropics (20°N–20°S) with coverage ranging between 15% and 30%; the area most strongly affected by El Niño is virtually not covered. Coverage generally is highest around 1980, declining sharply back in time and also declining after 1985. Following Brown et al. (2000a,b) we disregard radiosonde data over the Indian subcontinent (5°–30°N, 60°–90°E). Additionally, a grid box in the Gulf of Guinea has been disregarded, since regressions of differences between satellite MSU2lt and radiosonde R-MSU2lt averages showed high weights there.

We also made limited use of data from the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis (Kalnay et al. 1996) in some of the figures. Note that the reanalysis provides a dynamically complete dataset of atmospheric variables, which allowed us to relate temperature fluctuations to zonal wind fluctuations (see below). Note that the reanalysis is not independent from the other datasets since it assimilates data related to them. Also, the reanalysis data are not homogeneous in time and therefore have to be treated with caution (Chelliah and Ropelewski 2000).

For surface temperature, we employ a recently updated dataset (Jones et al. 1999), which is based on land surface temperature observations (Jones 1994) blended with sea surface temperature observations (Parker et al. 1995). The data are compiled as surface temperature anomalies with respect to 1961–90 on a 5° latitude by 5° longitude gridbox basis. To study the lapse rate, the higher-resolution satellite and reanalysis data have been transformed to the 5° × 5° grid to provide a spatial field of the lapse rate.

Surface temperature data, and even more so radiosonde data, are very sparse south of 40°S. Therefore we used only data equatorward of 40°S in “global” and Southern Hemispheric (SH) averages. Spatial patterns of the lapse rate have only been computed at grid points where both surface and lower-tropospheric data are available. This procedure, referred to as “collocation of grid boxes” avoids spurious influences of variability only sampled in one dataset (as discussed in Santer et al. 1999). Note that the lapse rate based on surface minus radiosonde data therefore samples a much smaller fraction (on average 35%) of the full spatial field 90°N–40°S than the lapse rate based on surface minus satellite data (85%–94%; area of grid box taken into account). Since radiosonde data are most complete over the Northern Hemisphere, a simple area average would be biased to the NH average. Therefore, for all data, global averages are computed from an area-weighted average of SH (20°–40°S), NH (90°–20°N), and tropical (20°N–20°S) data. We also make use of averages of the mid- to high-latitude Northern Hemisphere (30°–90°N) and the Tropics and subtropics (30°N–30°S), since these appear to show coherent signatures of climate variability. All data are computed as anomalies with respect to the reference period 1979–97.

Figure 1 shows that the evolution of global mean surface air temperature and lower-tropospheric temperature is quite similar on monthly timescales. However, the low-frequency evolution of both is qualitatively different: while surface air temperature (SAT) exhibits a long-term upward trend, the average of lower-tropospheric temperature (LTT) is dominated by a pronounced temperature increase between 1975 and 1980, with little increase after that (see also Table 1). Therefore, the global mean lapse rate exhibits an overall decrease between 1970 and 1980, and an increase since the late 1980s, interrupted by the strong El Niño of 1997/98 (Fig. 1; see also Gaffen et al. 2000). Linear trends and their significance are given in Table 1. The significance test is based on a t test adjusting the sample size for autocorrelation (see, e.g., von Storch and Zwiers 1999). Note that such a test will only establish if a trend describes the data significantly better than a flat time series, it cannot establish if the evolution of the time series is unusual relative to low-frequency climate variability, which should have a more complicated time evolution than red noise on monthly timescales (for a more thorough discussion of trend significance testing see Santer et al. 2000b).

In contrast to the trends observed during the satellite era, the trends in the surface and lower troposphere from 1964 onward are quite similar with a global mean lapse rate trend of virtually zero between 1964 and 1998 (Table 1; note, however, that there are trends of opposite sign in lapse rate in the deep Tropics and the NH). The observed warming is approximately consistent with climate model simulations of anthropogenic climate change as has been shown by comparisons in large-scale-averaged data (Santer et al. 2000a,b) and using fingerprint detection approaches (e.g., Tett et al. 1996; Allen and Tett 1999; Hill et al. 2001).

Note the similarity of satellite and radiosonde data for the lower-tropospheric temperature and the lapse rate based on the two datasets. Correlations between global averages of radiosonde and satellite data in the time of overlap are 0.85 if the satellite data are sampled only at radiosonde locations and higher for hemispheric scale averages except in the SH (see Table 1).

Figure 2 shows the spatial pattern of the trend starting in 1979 for satellite and surface data (see also Wallace et al. 2000). The strongest difference in trends is observed in the Tropics, where the satellite data show weak and mostly negative trends, while surface temperature warms almost everywhere. This visual impression is confirmed in the time series of the NH, tropical, and SH lapse rate (Fig. 3), in which the tropical lapse rate increased at an average rate of 0.16 K decade−1 between 1979 and January 2000, while the NH north of 20°N has an upward trend of 0.03 K decade−1 and the SH of 0.03 K decade−1 (see also Table 1).

Figure 4 shows the trend in zonal cross sections of free-tropospheric temperature derived from radiosonde and reanalysis data, and compares it to trends in zonal wind fields. Temperatures show warming in the NH troposphere, and sections of the SH troposphere (note, however, that reanalysis trends particularly in the SH are questionable due to inhomogeneities of the record; Chelliah and Ropelewski 2000). The westerlies have strengthened in the subpolar latitudes and weakened throughout most of the lower latitudes. We will come back to these patterns later in the paper.

3. Influence of modes of variability connected to surface temperature

In the following, we discuss the influence of dynamically induced climate variability upon NH and tropical averages of the lapse rate. Relevant patterns of variability have been identified by regressing the hemispheric-scale lapse rate averages on spatial fields of SAT, LTT, and the lapse rate. The resulting pattern shows the strength of the local lapse rate variations that tend to coincide with a hemispheric-scale average lapse rate variation of 1 K, thereby showing which areas have the strongest influence on the lapse rate averages and identifying structured patterns of variability that covary with the lapse rate averages. To ensure a description of monthly variability independent of the decadal trends that we are trying to interpret, the trends in surface and lower-tropospheric data over the satellite era have been removed prior to regression throughout the paper. Similarly, low-frequency variability with periods longer than 5 yr during the radiosonde period has been removed from surface- and radiosonde-based lower-troposphere averages by applying a simple low-pass filter (application of a 61-month running combined with a 41-month running mean, the latter improves the filter characteristics and removes residual variability below 5 yr; see, e.g., von Storch and Zwiers 1999). Results are generally insensitive to this removal of low-frequency variability.

We define 30°N as a cutoff point between NH averages and the Tropics/subtropics since this proved a useful separation point particularly for the free atmosphere. Results where 20°N was used as the separation point were very similar.

a. Northern Hemisphere

Regression of NH lapse rate variations onto surface air temperature, lower-tropospheric temperature, and the lapse rate fields shows that anomalously strong lapse rates occur when the continents are relatively warmer than the oceans at the surface. This pattern is virtually identical to the “cold ocean warm land” (COWL) pattern (Wallace et al. 1995), which is shown in Fig. 5a. The COWL pattern is not a mode of variability in its own right, but is useful in isolating dynamically induced mean surface temperature fluctuations caused by warm air masses being preferentially situated over continents and cold air masses over oceans or vice versa. The difference in the heat capacity of continents and ocean leads to stronger anomalies over land, and hence, to warm NH mean surface temperature anomalies.

The COWL index is constructed by regressing hemispherically averaged temperatures (poleward of 20°N) onto spatial patterns of monthly cold season temperatures from which the spatial mean has been subtracted. The expansion coefficient of the so-obtained pattern is referred to as COWL index. If the COWL index is regressed onto the spatial field (without subtracting the spatial mean) the COWL pattern is obtained.1 The lower-troposphere temperature pattern that occurs in linear association with fluctuations in the COWL index (Fig. 5b) is similar, but less pronounced over the continents. The disparity in the strengths of the surface and lower-troposphere patterns leads to anomalously strong lapse rates over the continents (Fig. 5c), and hence, to a positive lapse rate anomaly for the NH as a whole.

The COWL index accounts for 40% of monthly cold season (November–April) NH lapse rate variations (based on satellite data) for the area north of 30°N (43% of the lapse rate variations north of 40°N). For the cold seasonal average, the COWL pattern explains even more of the variance (59% north of 30°N). Its systematic influence upon the lapse rate appears to be largely restricted to the colder six months of the year; it only explains 34% of the annual NH (north of 30°N) mean lapse rate. Lapse rate averages south of 30°N are virtually uncorrelated with the COWL index.

Figure 6 shows the time series of cold season averages of the lapse rate based on satellite and radiosonde data for the NH together with the COWL index, also averaged over the cold season. Fluctuations in the COWL index are induced by modes of NH circulation. For example, the Arctic Oscillation (AO; an annular mode of NH circulation; Thompson and Wallace 1998) explains about a third of the monthly COWL fluctuations in the cold season poleward of 40°N. Furthermore, the trend in the AO is responsible in part for the recent increase in the COWL index (Thompson et al. 2000). A further part of this increase might be due to an emerging anthropogenic warming signal that is expected to exhibit more warming over land (see Houghton et al. 1996, 2001).

b. Tropics and subtropics

The deep Tropics (20°N–20°S) account for 75% of the global mean lower-tropospheric temperature variability on monthly timescales and also show the most prominent trend in the lapse rate over the satellite era (Fig. 3; Table 1; see also Gaffen et al. 2000). Figure 7 shows that the monthly tropical (30°N–30°S) LTT strongly resembles the SAT average; both time series are dominated by El Niño variability.

Yulaeva and Wallace (1994) found that a simple thermodynamic model for the atmospheric response to El Niño–related warming explains the midtroposphere (MSU channel 2) temperature observations quite well. The model assumes that tropical tropospheric temperature exhibits a damped response to the local anomalies in the surface energy fluxes that occur in response to El Niño–induced sea surface temperature (SST) anomalies in the equatorial eastern Pacific. The mathematical formalism is based on Hasselmann's (1976) model of stochastically forced SST fluctuations, but in this case forcing by local SST fluctuations is prescribed in accordance with observations and the atmospheric response is inferred from the model. We have applied the same simple model to derive a Tropicswide index of El Niño variability that predicts LTT response (see the appendix). As a forcing term, a slightly modified cold tongue index (CTI; e.g., Mitchell and Wallace 1996; here an SST average from 5°N to 5°S and 180° to 90°W, derived from the same gridded surface temperature data as used in the rest of the paper) is used as a measure of El Niño variability. The thermodynamic model calculation yields an El Niño index CTI* that lags the original CTI (and also the Southern Oscillation index) by about 5 months. This lag is similar to the lag of average tropospheric temperatures relative to El Niño events (see, e.g., Brown et al. 2000b). Brown et al. (2000b) and Santer et al. (2001) discuss how to account for El Niño–related variability on lower-tropospheric temperature and the lapse rate time series, using differently lagged time series for El Niño variability and sophisticated statistical approaches to derive a best estimate of the lapse rate response to El Niño. We found that the CTI* index yielded very similar (in most cases slightly higher) correlations with the lapse rate and lower-tropospheric temperatures as lagged CTI. We prefer CTI* to lagged CTI since it has a stronger physical motivation.

Figure 7 shows that while the correlation of the CTI* time series with lower-tropospheric and surface averages is very high, the corresponding negative fluctuations in the lapse rate, though clearly visible, are small. The regression pattern of the index CTI* on lapse rate is shown in Fig. 8. Note that in order to reduce spurious influences of volcanism on CTI* response patterns we have disregarded months 6–24 after major volcanic eruptions in the regressions, since these show the most visible change in the lapse rate. This choice is a compromise between reducing the influence of climate response to volcanism on CTI* (and later other) response patterns and the need to retain enough data for calculating robust regression patterns (for more thorough discussions of the influence of volcanic eruptions, see Santer et al. 2001; Brown et al. 2000b; Wigley and Santer 2002, manuscript submitted to J. Climate). The regression pattern shows that anomalously high lapse variations occur at times of a high CTI* index (and therefore shortly after El Niño events) directly over the cold tongue and low anomalies in an equatorially symmetric pattern. The widespread negative anomalies reflect the overall enhanced warming of the tropical troposphere that occurs in association with El Niño (more discussed below). Differences of the regression pattern between the months November–April and May–October (Figs. 8a and 8b) probably reflect the stronger tropical heating anomalies and the stronger NH extratropical response during the NH cold season months.

We have also attempted to use other indices for the El Niño–related variability in the Tropics: the expansion coefficient for the dominant spatial pattern of decadally filtered global SST, in combination with sea level pressure or by itself, the Pacific Decadal Oscillation index (Zhang et al. 1997; Mantua et al. 1997) and the Southern Oscillation index show generally smaller correlations with the tropical mean lapse rate than CTI*. None of these indices can account for the decadal-scale fluctuations in the tropical lapse rate.

c. Removing El Niño and COWL influences

We have removed from the spatial patterns of the lapse rate the influence of El Niño variability as represented by the CTI* time series and Northern Hemispheric wintertime dynamics as represented by the COWL index. Since our results are consistent with findings of Brown et al. (2000a,b) and Santer et al. (2001), we do not discuss them in detail. In the present work, a multiple regression of the detrended monthly CTI* and COWL indices on satellite- and surface-based lapse rates was performed for the NH cold season months (November–April) to derive the lapse rate response pattern. These spatial patterns have then been removed from both satellite- and radiosonde-based lapse rate fields for every cold season month. For the NH warm season months, only the May–October response to CTI* was removed (Fig. 8), since the COWL response appears to be incoherent in boreal summer. The residuals for the NH mid- to high latitudes and the Tropics and subtropics are shown in the bottom panels of Figs. 6 and 7 (and their zonal trends in Fig. 12c).

Although the residual time series are noticeably simpler than the raw time series from which they were derived, the positive lapse rate trend (particularly in the Tropics and subtropics) is largely unaffected (even slightly increased) by the removal of the COWL and CTI* signatures. Note that a clearer volcanic signal is visible in the residual lapse rate in Fig. 7, with an anomalously high lapse rate particularly after the volcanic eruptions of El Chichon in 1983 and Mt. Pinatubo in 1991. These anomalies are caused by the cooling after major volcanic eruptions being stronger aloft than at the surface (see, e.g., Santer et al. 2001).

A regression of the time series of the residual lapse rate on surface temperature fields yielded featureless patterns, suggesting that we have succeeded in capturing the most important spatially structured modes of variability. Hence, we consider it unlikely that structured modes related to surface temperature variability can explain the observed lapse rate trend. The variability explained by the circulation indices is largest on interannual timescales with periods of less than 3 yr. Removing the component of the variability that is linearly proportional to the COWL and CTI indices actually increases the trend slightly.

4. What drives lower-tropospheric temperature anomalies in the Tropics and subtropics?

The previous section demonstrated that the modes of variability that affect surface temperature cannot explain trends in the observed lapse rate averages, particularly those in the Tropics and subtropics. We now focus on variability within the troposphere that affects the tropical and subtropical lapse rate averages (from 30°N to 30°S).

First, we derive an index of tropospheric variability that is linearly independent of the fluctuations in area-averaged SAT. It is apparent from Fig. 7 that monthly temperature anomalies at the surface and in the lower troposphere are highly correlated. However, the amplitude of variations in LTT is somewhat stronger. This enhancement of SAT variability has been also reported in the literature, for example, Hurrell and Trenberth (1998). Figure 9 illustrates the high degree of coherence between monthly tropical and subtropical averages of surface and lower-tropospheric temperatures. Note that in the figure only collocated surface and lower-troposphere grid boxes are used (yielding quite a small number of grid boxes for surface–radiosonde comparisons); also, the long-term trend in data over the satellite period and the variability at periods longer than 5 yr over the radiosonde period have been removed.

Both LTT and SAT averages align approximately along a regression line. If an 11-month running mean is applied to both SAT and LTT time series, a standard least square fit yields that anomalies of SAT averages are enhanced by a factor of 1.5 aloft (see Fig. 9) in both radiosonde- and satellite-based averages. The exact value of the enhancement factor is subject to uncertainty in how it is estimated; for example, if monthly data are used, the slope of a standard least square fit regression line is 1.4 (1.3 for Tropics only), while a total least square fit yields a slope of 1.8. A standard least square fit allows for noise in LTT averages (due to random instrumental error and variability disconnected from the surface), but not in surface averages. This asymmetry tends to bias slope estimates from a standard least square fit downward (see Allen and Stott 2002), while the total least square fit requires an estimate of the unknown ratio of noise in LTT relative to that in SAT (an enhancement of 1.8 based on a ratio of 1). Smoothing the time series will reduce noise and make this problem less severe. The results described below are not sensitive to the exact value of the enhancement ratio and are very similar if 1.8 is used.

Figure 9 also shows that in the month-to-month variability over the past 21 yr, SAT anomalies as large as the cumulative SAT rise of the past 21 yr have never occurred in combination with LTT anomalies as small as the observed cumulative LTT rise of the same period. We will come back to this point later.

We now define an index LTT* of (30°N–30°S) lower-tropospheric temperature anomalies linearly independent of the SAT anomalies from the relationship LTT* = LTT − α SAT, where α is the enhancement ratio, as defined in the previous paragraph (i.e., LTT* records anomalies away from the regression line in Fig. 9). The resulting time series, shown in Fig. 10, is similar to the inverted lapse rate time series LTT − SAT, but the residual El Niño signature apparent in the LTT − SAT time series is diminished (a complete removal cannot be expected, given the simplifications and linearizations applied). CTI* and LTT* (disregarding months following the Mt. Pinatubo eruption) show almost no correlation. Note also that the trend of LTT* is 1.6 times as large as that in LTT − SAT (−0.21 K decade−1).

The effect of volcanic eruptions appears also slightly reduced in LTT* compared to LTT − SAT, and more so compared to that in the residual lapse rate time series (Fig. 7, bottom). However, attempts to estimate the strength of the volcanic signal in LTT − SAT and LTT* [using an energy balance model simulation of surface temperature response to volcanism; Crowley (2000), or a response model similar to that used in the appendix] yielded suggestive, but not robust results. For a more thorough discussion of the effect of volcanism on the lapse rate see Brown et al. (2000b) and Santer et al. (2001).

The next question is where do the LTT* variations originate from. To shed some light on the cause of the variations in LTT*, the LTT* time series has been regressed on temperature patterns at the surface and aloft to obtain the patterns shown in Fig. 11. Positive LTT* anomalies coincide with positive anomalies of LTT throughout lower latitudes (Fig. 11a), with pronounced, zonally elongated maxima in the subtropics, which are roughly symmetric about the equator. Spatial correlation maps (not shown) confirm the existence of these features, which exhibit correlations of up to ∼0.4 with the LTT* index. The corresponding regression pattern for surface air temperature shows rather little structure, except for a faint pattern reminiscent of decadal ENSO-like variability in the Pacific (Fig. 11b; cf. Zhang et al. 1997), which is not robust (e.g., it is sensitive to whether satellite or radiosonde data are used in computing LTT*). Regression coefficients and correlations between zonally averaged LTT and LTT* shown in Fig. 12 confirm the equatorially symmetric appearance of the regression patterns and the enhanced values in the subtropics. The corresponding meridional profiles of zonal mean SAT are quite flat with slightly negative values in the Tropics and subtropics. The connection between the lapse rate and this zonally symmetric pattern in the lower troposphere was found robust to changes in the analysis procedure (i.e., not omitting months following volcanic eruptions, and basing LTT* on radiosonde data from 1964 onward, rather than satellite data from 1979 onward).

Figure 13 illustrates the vertical structure of the subtropical temperature anomalies connected to the LTT* fluctuations (analysis based on 1979 onward). Variations in LTT* coincide with warming in the troposphere that peaks near 350 hPa and 35° latitude, and the surface signature is again weak by construction. Figure 13c shows the regression of LTT* on zonal wind anomalies. Note that the vertical gradient of the zonal wind pattern is in thermal agreement with the horizontal temperature gradient, maximum wind anomalies occur around 200 hPa. Correlation maps (not shown) exhibit similar patterns with peak correlations above 0.55 for temperature. The higher correlations are largely restricted to low latitudes. A dynamical interpretation of the mechanisms responsible for the variability shown in Figs. 11–13 is beyond the scope of this paper.

We tried to represent the patterns in Figs. 11–13 in terms of a single time index (Fig. 14) that could be used to document their time history and to estimate more quantitatively their influence on LTT*. That the most successful of our trial indices (based on the meridional profile of 200-hPa zonal wind, not shown) is correlated with LTT* at a level of only 0.3 clearly indicates that the factors that cause LTT* to vary cannot be described in terms of a single coherent mode of variability. It is also notable that the index exhibits only a very small trend during the period of the satellite record and it is not of the right sign to account for the observed trend in LTT*. Nonetheless, the patterns in Figs. 10–12 appear to be robust, dynamically plausible, and accountable for a modest fraction of the month-to-month variance of LTT*.

Meridional profiles of observed trends in LTT and SAT (with and without the removal of the COWL and El Niño–related variability), shown in Fig. 12c, exhibit a hint of equatorial symmetry analogous to the regression patterns in Figs. 11–13 but the meridional scale is broader (see also Figs. 4 and 13). These profiles clearly show the strong contribution of the lower latitudes to the observed trend toward a stronger positive global lapse rate. Note also that there is very little difference in trends between raw zonal averages and averages with CTI* and COWL taken out, demonstrating that the climate variability captured by those indices does not explain the observed lapse rate trend.

5. Can data from a coupled ocean–atmosphere general circulation model help understand the lapse rate trend?

Consistent with the results presented in the previous two sections, Gaffen et al. (2000) and Santer et al. (2000a,b) found that the observed trend in the difference between surface and lower-tropospheric temperature is significantly larger than trends occurring due to dynamically induced climate variability, as estimated from coupled ocean–atmosphere general circulation models. In this section we will compare lapse rate variations in one of the climate models used in those studies with the observed variations considered in the previous section.

We use data from a state-of-the-art coupled climate model, ECHAM4/OPYC, which was developed at the Max Planck Institute for Meteorology and has been run for approximately 300 yr with constant interannual forcing and for several climate change simulations driven by transient anthropogenic forcing (Roeckner et al. 1999). Data from both the control simulation and a climate change simulation forced with anthropogenic greenhouse gas, sulfate aerosol (direct and indirect effect), and tropospheric ozone (the experiment is referred to as “GSDIO” in the literature) are used.

For comparing model-simulated LTT and SAT fluctuations with the observed, a series of ninety 21.1-yr samples, separated by 1-yr intervals have been extracted from the model data. The time–space fields for each sample were masked with the observed missing data mask (since collocated data are used, both MSU and surface fields have data only where surface monthly values are available) and processed exactly the same as the observations, including subtraction of the long-term trend for studying the patterns of month-to-month variability. Figure 15 shows a comparison of monthly variability between observations and the model. The simulated LTT/SAT enhancement ratio of 1.56 is very close to that from the observations (1.53, which is within 0.4 standard deviations). The relatively close agreement between the model and observations is illustrated by the similar area covered by simulated (gray) and observed (black) crosses in Fig. 15a. The standard deviation of LTT* is ∼30% larger in the observations than in the model, which is not surprising given that instrumental uncertainties and external forcing not included in the control simulation are likely to increase the observed LTT* spread. Figure 15b shows a comparison focusing on the time span 1964–81 in radiosonde data, model data have been masked in accordance with the (much sparser) radiosonde observation mask. The general impression is similar, although the spread of monthly variability is larger in both due to the poorer coverage inducing more sampling errors. We conclude that the model is realistic in its simulation of monthly coherence between the surface and lower troposphere, in agreement with a finding of B. D. Santer (2001, personal communication; Santer et al. 2000a).

Next, we compare the long-term observed trend with trends in the model control simulation. The trends from the control simulation in Fig. 15 align very closely along the regression line, showing that on long timescales, surface and upper-air temperature trends are much more closely coupled than they are on a month-to-month basis. Nowhere in the 110 yr from the control simulation does there occur a trend as far away from the regression line as the observed ones (the observed LTT* trend is 8 times stronger than the strongest LTT* deviation of the same sign in the model, and 15 times the estimated standard deviations of LTT* trends). If trends from the climate change simulation are processed in the same way as the control simulation (using the simulation of the time between 1950 and 2010 to collect 21.1-yr trends) a very similar picture emerges, except that the covarying SAT and LTT trends tend to be more positive due to greenhouse warming in the model. Furthermore, a comparison between the observed trends from the radiosonde period illustrates that the observed period 1964–81, in which the lapse rate and LTT* trends were of opposite sign to the most recent trend (Figs. 8 and 10), is also quite unusual with respect to model simulated trends (Fig. 15b).

We conclude that while the model simulates quite realistically the month-to-month variations in the coherence between surface and lower-tropospheric temperature, on the decadal timescale, it maintains a much tighter coupling between the surface and lower troposphere than is apparent in the observations. This problem appears not to be limited to the connection of LTT to the surface: Gillett et al. (2000) investigated vertical coherence within the troposphere (using temperature averages of different layers from radiosonde data) and found that on timescales beyond 6 yr a (different) coupled climate model was coupled significantly tighter than the observations. Studies have shown that other models also fail to explain the observed lapse rate trend by internal variability (Santer et al. 2000a; S. Tett et al. 2002, personal communication). Therefore, the difference between surface–troposphere coupling in the model and the observations does not appear to be limited to this particular climate model.

As discussed above, an analysis based on observations only failed to detect a pattern of variability responsible for the observed trend in the lapse rate, and the model appears not to simulate the mechanism for it. It is conceivable that the variability responsible for the trend is hidden among the stronger and more obvious short-term variability, which appears to be simulated realistically in the model. The large amount of model data for interannual lapse rate variability can be used to identify such a hidden “signal” from the observations (the observed lapse rate data are transformed so that variability occurring in the model is spatially white, and then an empirical orthogonal function analysis detects the pattern that is not simulated in the model; Venzke et al. 1999; Chang et al. 2000). However, the resulting pattern (not shown) is very close to the lapse rate trend pattern, and its expansion coefficient is very similar to the observed tropical and subtropical average lapse rate time series (correlation 0.79 for interannual data and 0.95 on timescales beyond 4 yr, not shown). We conclude that the most unusual feature of the observations compared to the climate model is the trend in the low-latitude lapse rate over the satellite period, and again no mechanism with clear spatial or time structure can be found that accounts for that trend.

6. Discussion and conclusions

We have investigated whether there exist spatially coherent “modes of variability” that drive monthly variations in the difference between surface and lower-tropospheric temperature (such as derived as “channel 2lt” from microwave satellite measurements). We find that the Northern Hemispheric variability that projects on the cold ocean warm land (COWL) index explains a large fraction of the lapse rate variations over the Northern Hemisphere extratropics during the cold season, but has little influence in the warm season and outside the Northern Hemisphere extratropics.

The lapse rate variations in the Tropics and subtropics, where the trend is strongest, are related to El Niño events. We have removed variability in the lapse rate associated with both the COWL index and an El Niño index tailored to the lapse rate and find that this reduces month-to-month and interannual variability, but leaves the observed trend toward a higher global mean lapse rate over the satellite era virtually unaltered. If the variations in the low-latitude lapse rate that are related to an enhancement of surface temperature anomalies in the troposphere are removed by a linear fit, an index for the residual variations in temperature within the lower troposphere can be derived. Such temperature anomalies are weakly organized in terms of zonally elongated bands, with the largest temperature amplitudes in the subtropics near the 350-hPa level and related to the zonal wind anomalies peaking at ∼200 hPa. However, this pattern accounts for only a small fraction of the month-to-month variability of the observed lapse rate and none of the trend.

Therefore, all attempts to explain all or a significant part of the observed lapse rate trend by modes of climate variability with structured patterns from observations have failed. An approach applying model data to isolate such a pattern has also failed. It is conceivable that the processes internal to the climate system could cause low-frequency variability of the lapse rate on long timescales without a structured pattern. Such processes, if they exist, are not simulated in coupled climate models. Surface and upper-air trends over 21 yr in an extended control run and in a transient anthropogenic climate change simulation are much more tightly coupled than both the observed trends during the satellite era (where the lapse rate has been increasing) and over the period 1964–81 (where the lapse rate has been decreasing). It is conceivable that our understanding of tropical climate variability is still incomplete (cf. Hartmann 2002).

Other possible causes for the observed lapse rate evolution are external forcings: anthropogenic forcing, volcanism, and variations in solar irradiance. Volcanism causes stronger cooling aloft than at the surface, and hence, positive anomalies of the lapse rate for a period after volcanic eruptions. However, volcanism alone is unlikely to explain the whole differential warming over the satellite period (see Bengtsson et al. 1999; Brown et al. 2000b; Santer et al. 2001). In our analysis, changes in the lapse rate immediately after volcanic eruptions are visible, particularly after modes of climate variability have been subtracted, but the magnitude of the volcanic signal and its extent in time are difficult to assess. We have not directly discussed the effect of variations in solar irradiance; however, its time series (e.g., Lean et al. 1995) is dissimilar to that of the lapse rate variations.

A simulation of anthropogenic climate change in a state-of-the-art coupled climate model was no closer to the observed lapse rate trend than the control simulation. This simulation does not incorporate stratospheric ozone forcing, which has been hypothesized to play a role in the observed lapse rate fluctuations since it cools the stratosphere and could cause some cooling of the lower troposphere relative to the surface (e.g., Santer et al. 2000a). However, recent anthropogenic climate change simulations including stratospheric ozone forcing, solar forcing, and volcanism (S. Tett 2002, personal communication) also fail to explain the observed lapse rate trend, while they impressively reproduce the twentieth-century surface temperature evolution (Stott et al. 2000). Therefore it seems unlikely that anthropogenic forcing alone or in combination with natural forcing can entirely explain the observed evolution of the lapse rate.

The observed trends in the global (or tropical) mean lapse rate are subtle, yet nonetheless require an explanation because they are so much larger than those recovered in the extended simulations of climate models. We have shown that they cannot be explained in terms of trends in recognizable patterns of climate variability. Nor does it seem to us that interdecadal variations in radiative forcing, such as might be caused by volcanic eruptions, variations in solar output, or stratospheric ozone depletion alone, offer a compelling explanation. It is conceivable that these factors in combination with observational error might explain the lapse rate trend. Nevertheless, at this stage we conclude that there remains a gap in our fundamental understanding of the processes that cause the lapse rate to vary on interdecadal timescales.

Acknowledgments

We thank Hank Seidel for support with graphics. Ben Santer helped in numerous aspects of the work and provided very thorough and useful comments as a reviewer. We also thank him for sharing preprocessed model data with us. We would also like to thank the following people: two further anonymous reviewers, John Christy, Nathan Gillett, Simon Brown, Myles Allen, David Thompson, and Jim Hurrell for discussions; and Phil Jones, John Christy, and David Parker for observational data. Todd Mitchell provided indices for variability and was very helpful in many aspects of the work. GCH was supported by a fellowship from the Alexander von Humboldt foundation, by NSF Grants ATM-9707069, ATM-0096017, and ATM-0296007, by JISAO, NOAAs Office of Global programs, and DOE in conjunction with the Climate Change Data and Detection element and by Duke University. Any opinions, findings, and conclusions expressed in this paper are those of the authors and do not necessarily reflect the views of the funding agencies.

REFERENCES

  • Allen, M. R., and S. F. B. Tett, 1999: Checking for model consistency in optimal fingerprinting. Climate Dyn., 15 , 419434.

  • Allen, M. R., and P. A. Stott, 2002: Estimating signal amplitudes in optimal fingerprinting: Part I. Theory. Climate Dyn., in press.

  • Bengtsson, L., E. Roeckner, and M. Stendel, 1999: Why is the global warming proceeding much slower than expected. J. Geophys. Res., 104 , 38653876.

    • Search Google Scholar
    • Export Citation
  • Brown, S. J., D. E. Parker, C. K. Folland, and I. Macadam, 2000a: Decadal variability in the lower-tropospheric lapse rate. Geophys. Res. Lett., 27 , 9971000.

    • Search Google Scholar
    • Export Citation
  • Brown, S. J., and D. Sexton, 2000b: Differential changes in observed surface and atmospheric temperature since 1979. Hadley Centre Tech. Note HCTN12, 41 pp.

    • Search Google Scholar
    • Export Citation
  • Chang, P., R. Saravanan, L. Ji, and G. C. Hegerl, 2000: The effect of local sea surface temperature on atmospheric circulation over the tropical Atlantic sector. J. Climate, 13 , 21952216.

    • Search Google Scholar
    • Export Citation
  • Chelliah, M., and C. F. Ropelewski, 2000: Reanalysis based tropospheric temperature estimates: Uncertainties in the context of global climate change detection. J. Climate, 13 , 31873205.

    • Search Google Scholar
    • Export Citation
  • Christy, J. R., and R. T. McNider, 1994: Satellite greenhouse warming. Nature,367, p.325.

  • Christy, J. R., R. W. Spencer, and W. D. Braswell, 2000: MSU tropospheric temperatures: Dataset construction and radiosonde comparisons. J. Atmos. Oceanic Technol., 17 , 11531170.

    • Search Google Scholar
    • Export Citation
  • Crowley, T. J., 2000: Causes of climate change over the past 1000 yrs. Science, 289 , 270277.

  • Gaffen, D. J., B. D. Santer, J. S. Boyle, J. R. Christy, N. E. Graham, and R. J. Ross, 2000: Multidecadal changes in vertical temperature structure of the tropical troposphere. Science, 287 , 12421245.

    • Search Google Scholar
    • Export Citation
  • Gillett, N., M. R. Allen, and S. F. B. Tett, 2000: Modeled and observed variability in atmospheric vertical temperature structure. Climate Dyn., 16 , 4961.

    • Search Google Scholar
    • Export Citation
  • Hartmann, D. L., 2002: Tropical surprises. Science, 295 , 811813.

  • Hasselmann, K., 1976: Stochastic climate models. Tellus, 28 , 473485.

  • Hill, D. C., M. R. Allen, and P. A. Stott, 2001: Allowing for solar forcing in the detection of human influence on tropospheric temperatures. Geophys. Res. Lett., 28 , 15551558.

    • Search Google Scholar
    • Export Citation
  • Houghton, J. T., L. G. Meiro Filho, B. A. Callander, N. Harris, A. Kattenberg, and K. Maskell, Eds. 1996: Climate Change 1995: The Science of Climate Change. Cambridge University Press, 572 pp.

    • Search Google Scholar
    • Export Citation
  • Houghton, J. T., Y. Ding, D. J. Griggs, M. Noguer, P. J. van der Linden, X. Dai, K. Maskell, and C. A. Johnson, Eds.,. 2001: Climate Change 2000: The Scientific Basis. Cambridge University Press, 892 pp.

    • Search Google Scholar
    • Export Citation
  • Hurrell, J. W., and K. E. Trenberth, 1998: Difficulties in obtaining reliable temperature trends: Reconciling the surface and satellite MSU records. J. Climate, 11 , 945967.

    • Search Google Scholar
    • Export Citation
  • Jones, P. D., 1994: Hemispheric surface air temperature variations: A reanalysis and an update to 1993. J. Climate, 7 , 17941802.

  • Jones, P. D., M. New, D. E. Parker, S. Martin, and I. G. Rigor, 1999: Surface air temperature and its changes over the past 150 years. Rev. Geophys., 37 , 173199.

    • Search Google Scholar
    • Export Citation
  • Kalnay, E., and Coauthors. 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77 , 437471.

  • Lambert, S. J., 1990: Discontinuities in the long-term Northern Hemisphere 500-mbar dataset. J. Climate, 3 , 14791484.

  • Lean, J., J. Beer, and R. Bradley, 1995: Reconstruction of solar irradiance since 1600: Implications for climate change. Geophys. Res. Lett., 22 , 31953198.

    • Search Google Scholar
    • Export Citation
  • Levitus, S., J. I. Boyer, and C. Stephens, 2000: Warming of the world ocean. Science, 287 , 22252229.

  • Mantua, J. N., R. S. Hare, Y. Zhang, J. M. Wallace, and R. C. Francis, 1997: A Pacific interdecadal climate oscillation with impacts on salmon production. Bull. Amer. Meteor. Soc., 78 , 10691080.

    • Search Google Scholar
    • Export Citation
  • Mitchell, T., and J. M. Wallace, 1996: ENSO seasonality: 1950–78 versus 1979–92. J. Climate, 9 , 31493161.

  • Parker, D. E., C. K. Folland, and M. Jackson, 1995: Marine surface temperature: Observed variations and data requirements. Climate Change, 31 , 559600.

    • Search Google Scholar
    • Export Citation
  • Parker, D. E., M. Gordon, D. P. N. Cullum, D. M. H. Sexton, C. K. Folland, and N. Rayner, 1997: A new global gridded radiosonde temperature data base and recent temperature trends. Geophys. Res. Lett., 24 , 14991502.

    • Search Google Scholar
    • Export Citation
  • Roeckner, E., L. Bengtsson, J. Feichter, J. Lelieveld, and H. Rodhe, 1999: Transient climate change simulations with a coupled atmosphere–ocean GCM including the tropospheric sulfur cycle. J. Climate, 12 , 30043032.

    • Search Google Scholar
    • Export Citation
  • Santer, B. D., J. J. Hnilo, T. M. L. Wigley, J. S. Boyle, C. Doutriaux, M. Fiorino, D. E. Parker, and K. E. Taylor, 1999: Uncertainties in “observational” estimates of temperature change in the free atmosphere. J. Geophys. Res., 104 , 63056333.

    • Search Google Scholar
    • Export Citation
  • Santer, B. D., and Coauthors. 2000a: Interpreting differential temperature trends at the surface and in the lower troposphere. Science, 287 , 12271231.

    • Search Google Scholar
    • Export Citation
  • Santer, B. D., and Coauthors. 2000b: Statistical significance of trends and trend differences in layer average atmospheric temperature time series. J. Geophys. Res., 105 , 73377356.

    • Search Google Scholar
    • Export Citation
  • Santer, B. D., and Coauthors. 2001: Accounting for the effects of volcanoes and ENSO in comparisons of modeled and observed temperature trends. J. Geophys. Res., 106 , 2803328059.

    • Search Google Scholar
    • Export Citation
  • Spencer, R. W., and J. R. Christy, 1992: Precision and radiosonde validation of satellite gridpoint temperature anomalies. Part II: A tropospheric retrieval and trends during 1979–90. J. Climate, 5 , 858866.

    • Search Google Scholar
    • Export Citation
  • Stott, P. A., S. F. B. Tett, G. S. Jones, M. R. Allen, J. F. B. Mitchell, and G. J. Jenkins, 2000: External control of 20th century temperature by natural and anthropogenic forcings. Science, 290 , 21332137.

    • Search Google Scholar
    • Export Citation
  • Tett, S. F. B., J. F. B. Mitchell, D. E. Parker, and M. R. Allen, 1996: Human influence on the atmospheric vertical temperature structure: Detection and observations. Science, 274 , 11701173.

    • Search Google Scholar
    • Export Citation
  • Thompson, D. W. J., and J. M. Wallace, 1998: The Arctic Oscillation signature in wintertime geopotential height and temperature fields. Geophys. Res. Lett., 25 , 12971300.

    • Search Google Scholar
    • Export Citation
  • Thompson, D. W. J., and G. C. Hegerl, 2000: Annular modes in the extratropical circulation. Part II: Trends. J. Climate, 13 , 10181036.

    • Search Google Scholar
    • Export Citation
  • Venzke, S., M. R. Allen, R. T. Sutton, and D. P. Rowell, 1999: The atmospheric response over the North Atlantic to decadal changes in sea surface temperature. J. Climate, 12 , 25622584.

    • Search Google Scholar
    • Export Citation
  • von Storch, H., and F. W. Zwiers, 1999: Statistical Analysis in Climate Research. Cambridge University Press, 484 pp.

  • Wallace, J. M., Y. Zhang, and J. A. Renwick, 1995: Dynamic contribution to hemispheric mean temperature trends. Science, 270 , 780783.

    • Search Google Scholar
    • Export Citation
  • Wallace, J. M., and Coeditors,. 2000: Reconciling Observations of Global Temperature Change. National Academy Press, 85 pp.

  • Wentz, F. J., and M. Schabel, 1998: Effects of orbital decay on satellite-derived lower-tropospheric trends. Nature, 344 , 661664.

  • Wentz, F. J., C. Mears, and D. Seidel, 2001: Lower tropospheric air temperature derived from a blended analysis of MSU, SSM/T1, SSM/I, NCEP/NCAR reanalysis, and Reynolds SST. RSS Tech. Rep. 062801, Remote Sensing Systems, Santa Rosa, CA, 46 pp.

    • Search Google Scholar
    • Export Citation
  • Yulaeva, E., and J. M. Wallace, 1994: The signature of ENSO in global temperature and precipitation fields derived from the microwave sounding unit. J. Climate, 7 , 17191736.

    • Search Google Scholar
    • Export Citation
  • Zhang, Y., J. M. Wallace, and D. S. Battisti, 1997: ENSO-like interdecadal variability: 1900–93. J. Climate, 10 , 10041020.

APPENDIX

An Index for El Niño Variability in the Lower Troposphere

Yulaeva and Wallace (1994) suggested a simple thermodynamic model for the atmospheric response to El Niño. They applied this model to simulate the temperature response of the midtroposphere (MSU channel 2). We use a very similar approach in the present paper and reiterate it here for completeness.

The model assumes that the atmospheric response to El Niño shows a delayed response to heating due to an uptake of the heat released by El Niño mainly by the tropical oceans, according to the mechanism for red climate variability by Hasselmann (1976). It yields the simple model for atmospheric temperature T:
i1520-0442-15-17-2412-ea1
where F is the external forcing, μ a sensitivity, and α a memory term. As a forcing term, we applied a variation of the cold tongue index for ENSO-related variability. The memory term is related to the heat capacity of the tropical ocean and land surface. We have fitted both parameters to simulate the tropical mean surface temperature anomalies T by a least square fit to estimate the values of α and μ. Solving (A1) numerically produces a time series for the Tropicswide SAT response to El Niño that is called “CTI*” in the following.

We fit surface air temperature instead of atmospheric temperature, which was used by Yulaeva and Wallace. Although lower-tropospheric temperature is the physically more plausible choice, the use of surface air temperature for the fit ensures that the lower-tropospheric temperature time series can then be used as a validation of the model without bias caused by fitting. The coefficients of the simple model are similar to those given in Yulaeva and Wallace; small differences in parameters to those in their work yielded nearly identical time series. The estimated memory term is consistent with an e-folding time of about 6.6 months. Similar as in Yulaeva and Wallace, this causes a lag of CTI* compared to CTI of about 5 months, which is close to the lag of LTT to El Niño (see, e.g., Brown et al. 2000b; Santer et al. 2001).

Fig. 1.
Fig. 1.

(top) Comparison between annual and global (90°N–40°S) mean lower-tropospehric temperature (LTT; MSU2lt from satellite measurements and a comparable average from radiosonde data), (middle) surface air temperature (SAT; based on Jones et al. 1999), and (bottom) the difference time series SAT − LTT lapse rate, difference averaged only over grid points where both SAT and LTT data were available. LTT and lapse rate averages based on radiosonde data (black; Parker et al. 1997) are compared to those based on satellite data (gray; Christy et al. 2000), radiosonde data before 1964 are not shown due to homogeneity concerns. Ticks are offset by 0.2 K. All time series are centered on their mean over the period 1979–97

Citation: Journal of Climate 15, 17; 10.1175/1520-0442(2002)015<2412:IOPOCV>2.0.CO;2

Fig. 2.
Fig. 2.

Spatial pattern of the trend 1979–Jan 2000 in (a) LTT (based on satellite data), (b) SAT, and (c) lapse rate (K decade−1). The trend pattern for each field is based on the least square fit of a linear trend to the time series of each grid point

Citation: Journal of Climate 15, 17; 10.1175/1520-0442(2002)015<2412:IOPOCV>2.0.CO;2

Fig. 3.
Fig. 3.

Evolution of the average lapse rate (SAT − LTT) for NH, SH, and the Tropics from radiosonde (black) and satellite data (gray). Note the positive trend over the satellite period in all sections of the globe (see also Table 1)

Citation: Journal of Climate 15, 17; 10.1175/1520-0442(2002)015<2412:IOPOCV>2.0.CO;2

Fig. 4.
Fig. 4.

Trend pattern of zonally averaged atmospheric temperature in pressure coordinates (hPa) based on (a) radiosonde temperature data (1979–Apr 1998), (b) reanalysis data (1979–Jan 2000), (K decade−1). (c) Trend in zonal wind from reanalysis data (m s−1 decade−1). Each trend is least square fitted to zonally averaged temperature/wind at each latitude and pressure grid point. Note that the zonal temperature trend pattern from reanalysis data resembles that from radiosonde data more closely (particularly in the NH) if both trends are calculated to Apr 1998 only

Citation: Journal of Climate 15, 17; 10.1175/1520-0442(2002)015<2412:IOPOCV>2.0.CO;2

Fig. 5.
Fig. 5.

Cold ocean warm land (COWL) pattern from (a) SAT compared to a regression of the COWL index on (b) LTT, using satellite data, and (c) lapse rate. Positive COWL anomalies over the continents are weakened aloft. Therefore positive lapse rate anomalies over the continents coincide with positive COWL anomalies. Units relate to a 1 std dev change in the COWL index

Citation: Journal of Climate 15, 17; 10.1175/1520-0442(2002)015<2412:IOPOCV>2.0.CO;2

Fig. 6.
Fig. 6.

Comparison of the cold season (Nov–Apr) NH mean lapse rate north of 30°N with the COWL index. (top) The cold season lapse rate average for radiosonde (black) and satellite data (gray), (middle) COWL index, normalized as regressed on lapse rate, and (bottom) residual NH lapse rate time series after removing the influence of the COWL index and El Niño–related variability. The removal is based on spatial multiregression patterns (section 3c). The dashed curve in the middle panel shows the Arctic Oscillation index, which explains part of the variability of the COWL index

Citation: Journal of Climate 15, 17; 10.1175/1520-0442(2002)015<2412:IOPOCV>2.0.CO;2

Fig. 7.
Fig. 7.

Comparison of tropical and subtropical mean (30°N–30°S) LTT, SAT, CTI* (cold tongue index for Tropicswide surface temperature response on El Niño, scaled by its regression onto LTT), and lapse rate. The bottom time series (res. l. r.) is the lapse rate after regressing out the effects of El Niño and COWL based on spatial fields (same as in Fig. 6). The time of major volcanic eruptions is marked by vertical bars in the bottom two panels.

Citation: Journal of Climate 15, 17; 10.1175/1520-0442(2002)015<2412:IOPOCV>2.0.CO;2

Fig. 8.
Fig. 8.

Lapse rate response to El Niño based on regressions of surface and satellite data for (a) Nov–Apr and (b) May–Oct on CTI* [K (std dev)−1]

Citation: Journal of Climate 15, 17; 10.1175/1520-0442(2002)015<2412:IOPOCV>2.0.CO;2

Fig. 9.
Fig. 9.

Scatter diagram illustrating the covarying anomalies of tropical and subtropical mean surface temperature and lower-tropospheric temperature (averaged from 30°N to 30°S; only collocated grid points are used). Gray crosses compare satellite-based LTT and SAT (trends over the satellite era are removed), black crosses compare radiosonde-based LTT and SAT (variability beyond 5 yr removed). The slanted line indicates a regression line based on a (ordinary least square) regression from 11-month running mean data (1.5 enhancement of surface anomalies in the lower troposphere), the dotted line a (total least square) best fit using monthly data (1.8 enhancement). The fat circles show for comparison long-term trends (K per trend length) from satellite (gray) and radiosonde (black) data

Citation: Journal of Climate 15, 17; 10.1175/1520-0442(2002)015<2412:IOPOCV>2.0.CO;2

Fig. 10.
Fig. 10.

Comparison of tropical and subtropical lower-tropospheric temperature from radiosonde and satellite data. (top) Raw LTT and (bottom) LTT after the surface temperature influence has been subtracted (LTT*, subtraction based on enhancement of 1.5). (middle) Negative lapse rate (LTT − SAT) for comparison. Note that the influence of El Niño is reduced in LTT* compared to LTT − SAT. Alternative LTT* time series based on a differently estimated enhancement factor (estimates ranging between 1.3 and 1.8) are qualitatively very similar and also show negative trends over the satellite period. The time of major volcanic eruptions is given by vertical bars

Citation: Journal of Climate 15, 17; 10.1175/1520-0442(2002)015<2412:IOPOCV>2.0.CO;2

Fig. 11.
Fig. 11.

Regression of the LTT* time series (Fig. 10) based on surface and satellite data on (a) lower-tropospheric temperature and (b) surface temperature patterns to illustrate the origin of LTT* anomalies [K (std dev)−1]. Months 6–24 following volcanic eruptions (see Fig. 10) have been omitted and the long-term trend of LTT* over the satellite period has been subtracted prior to computing regression patterns. Correlation patterns with LTT* (not shown) are similar in the Tropics and subtropics, with peak correlations in the eastern half of the Tropics and subtropics (30°E–180°) up to 0.4, and very weak correlations in the extratropics. Correlation patterns with SAT are weak; structures are not robust

Citation: Journal of Climate 15, 17; 10.1175/1520-0442(2002)015<2412:IOPOCV>2.0.CO;2

Fig. 12.
Fig. 12.

(a) Regression of LTT* (tropical plus subtropical LTT minus surface influence; Fig. 10) on zonal mean surface (dotted) and lower-tropospheric (solid) temperature. Fat lines show the regression of satellite-derived LTT* on zonal mean SAT and satellite LTT, thin lines the regression of radiosonde-based LTT* (derived from radiosonde data from 1964 onward) on radiosonde data. A similar regression using Tropics only (20°N–20°S, enhancement 1.4) yielded a similar zonal pattern, however, with only very little enhanced warming in the subtropics. (b) Similar to (a), but for correlations instead of regressions. (c) Zonal average of LTT and SAT trend patterns between 1979 and Jan 2000, thin lines are results from trends based on raw fields, thick lines from LTT and SAT fields after removing ENSO- and COWL-related variability (section 3c)

Citation: Journal of Climate 15, 17; 10.1175/1520-0442(2002)015<2412:IOPOCV>2.0.CO;2

Fig. 13.
Fig. 13.

Regression of LTT* from satellite and surface data on (a) zonal mean temperature from radiosondes and from (b) reanalysis data. (c) Zonal mean zonal wind from reanalysis data. Correlation patterns (not shown) are similar in Tropics and subtropics, with peak correlations of 0.5–0.6 in the troposphere and correlations extending into the stratosphere, but weak in the extratropics.

Citation: Journal of Climate 15, 17; 10.1175/1520-0442(2002)015<2412:IOPOCV>2.0.CO;2

Fig. 14.
Fig. 14.

Comparison of (top) LTT* (Tropics and subtropics) and (bottom) an index of the variability characterized by Figs. 11–13. The index is the expansion time series of the spatial pattern of zonal wind at 200 mb between 50°N and 50°S (not shown); it is scaled to its regression on LTT*

Citation: Journal of Climate 15, 17; 10.1175/1520-0442(2002)015<2412:IOPOCV>2.0.CO;2

Fig. 15.
Fig. 15.

(a) Comparison between SAT and satellite-based LTT in Tropics and subtropics similar to Fig. 9 (crosses: monthly variability, trend removed; green circle: long-term observed trend in K per trend length) and variability in a coupled ocean–atmosphere general circulation model (Roeckner et al. 1999). The gray crosses show monthly variability in the “control” simulation of the model (same missing data mask and trend removal as in the observation; only every 15th monthly value displayed for clarity). The blue circles show trends from the control simulation over a period of 21.1 yr (same data mask as observations 1979–Jan 2001), the red circles from an anthropogenic climate change simulation (forced with increases in greenhouse gases, tropospheric ozone, and direct and indirect effects of sulfate aerosol forcing, 21.1-yr running trends from the model period 1950–2010). (b) Same as (a) but focusing on the observed period 1964–81, using the radiosonde data mask. Note that the model simulates monthly variability very well, but couples the surface and lower-tropospheric trends much more tightly along the regression line than observed. The comparison is limited to grid boxes covered in both LTT and SAT data for each panel

Citation: Journal of Climate 15, 17; 10.1175/1520-0442(2002)015<2412:IOPOCV>2.0.CO;2

Table 1. 

Trends of hemispheric-scale temperature averages of surface air temperature (SAT) and lower-troposphere temperature (LTT) data (MSU based on the satellite MSU2lt record, R-MSU on a vertical average of radiosonde data that approximates the satellite vertical average) and from lapse rate between both (K decade\su\−1\r\). The trend period is given in the left column, along with the month if not at a full year. Trends over individual datasets (SAT, MSU) are based on all available data, while the lapse rate is based on the average of collocated local lapse rate data (this coverage difference can cause small differences between the lapse rate trend and SAT−LTT trends). The two bottom rows give correlations r between radiosonde- and satellite-based averages, first using averages based on all satellite data, and second based only on radiosonde grid boxes

Table 1. 

1

Note that the process of regressing indices onto a field and then computing the expansion coefficient of the so-obtained pattern converges to the first empirical orthogonal function that the original time series is not uncorrelated with, but this does not affect the validity of the COWL pattern.

Save