• Anderson, G. P., S. A. Clough, F. X. Kneizys, J. H. Chetwynd, and E. P. Shettle, 1986: AFGL atmospheric constituent profiles (0–120 km). Rep. AFGL-TR-86-0110, 47 pp. [Available from AFGL, Hanscom AFB, MA 01731.].

  • Bailey, M. J., A. O’Neill, and V. D. Pope, 1992: Stratospheric analyses produced by the U.K. Meteorological Office. Climate Research Tech. Note 24, 50 pp. [Available from UKMO, London Rd., Bracknell, Berkshire RG12 2SY, United Kingdom.].

  • Barnett, T. P., 1986: Detection of changes in the global tropospheric temperature field induced by greenhouse gases. J. Geophys. Res.,91, 6659–6667.

  • Bates, J. J., X. Wu, and D. L. Jackson, 1996: Interannual variability of upper-troposphere water vapor band brightness temperature. J. Climate,9, 427–438.

  • Brownscombe, J. L., J. Nash, G. Vaughan, and C. F. Rogers, 1985: Solar tides in the middle atmosphere. I: Description of satellite observations and comparison with theoretical calculations at equinox. Quart. J. Roy. Meteor. Soc.,111, 677–689.

  • Charlock, T. P., 1984: CO2 induced climatic change and spectral variations in the outgoing terrestrial infrared radiation. Tellus,36B, 139–148.

  • Edwards, D. P., 1992: GENLN2 A general line-by-line atmospheric transmittance and radiance model. Version 3.0 description and users guide. NCAR Tech. Note NCAR/TN-367+STR, 147 pp. [Available from NCAR, P.O. Box 3000, Boulder, CO 80307-3000.].

  • ——, and L. L. Strow, 1991: Spectral line shape for limb temperature sounders. J. Geophys. Res.,96, 20 859–20 868.

  • Gaffen, D. J., 1994: Temporal inhomogeneities in radiosonde temperature records. J. Geophys. Res.,99, 3667–3676.

  • Geer, A. J., J. E. Harries, and H. E. Brindley, 1999: Spatial patterns of climate variability in upper-tropospheric water vapor radiances from satellite data and climate model simulations. J. Climate,12, 1940–1955.

  • Harries, J. E., H. E. Brindley, and A. J. Geer, 1998: Climate variability and trends from operational satellite spectral data. Geophys. Res. Lett.,25, 3975–3978.

  • Holton, J. R., 1992: An Introduction to Dynamic Meteorology. Academic Press, 511 pp.

  • Intergovernmental Panel on Climate Change, 1990: Climate Change:The IPCC Scientific Assessment. J. T. Houghton, G. J. Henkins, and J. J. Ephraums, Eds., Cambridge University Press, 365 pp.

  • ——, 1995: Climate Change 1995: The Science of Climate Change. J. T. Houghton et al., Eds., Cambridge University Press, 572 pp.

  • International TOVS Working Group, 1997: TOVS data in climate studies. A Report on the Ninth Int. TOVS Study Conf., Igls, Austria, ITWG, 9–13.

  • Johns, T. C., R. E. Carnell, J. F. Crossley, J. M. Gregory, J. F. B. Mitchell, C. A. Senior, S. F. B. Tett, and R. A. Wood, 1996: The second Hadley Centre coupled ocean–atmosphere GCM: Model description, spinup and validation. Climate Dyn.,13, 103–134.

  • Kidwell, K. B., 1995: NOAA Polar Orbiter data users’ guide. NOAA/National Environmental Satellite, Data, and Information Service, 250 pp. [Available from NOAA/NESDIS, Satelllite Data Services Division, Washington, DC 20233.].

  • Kiehl, J. T., 1983: Satellite detection of effects due to increased atmospheric carbon dioxide. Science,222, 504–506.

  • ——, 1986: Changes in the radiative balance of the atmosphere due to increases in CO2 and trace gases. Adv. Space Res.,6, 55–60.

  • Labitzke, K., 1994: Stratospheric temperature changes after the Pinatubo eruption. J. Atmos. Terr. Phys.,56, 1027–1034.

  • ——, B. Naujokat, and M. P. McCormick, 1983: Temperature effects on the stratosphere of the April 4, 1982 eruption of El Chichon, Mexico. Geophys. Res. Lett.,10, 24–26.

  • McCormick, M. P., R. E. Veiga, and W. P. Chu, 1992: Stratospheric ozone profile and total ozone trends derived from SAGE I and SAGE II data. Geophys. Res. Lett.,19, 269–272.

  • ——, L. W. Thomason, and C. R. Trepte, 1995: Atmospheric effects of the Mt Pinatubo eruption. Nature,373, 399–404.

  • Miller, D. E., J. L. Brownscombe, G. P. Carruthers, D. R. Pick, and K. H. Stewart, 1980: Operational temperature sounding of the stratosphere. Philos. Trans. Roy. Soc. London,296, 65–71.

  • Mitchell, J., C. Johns, J. M. Gregory, and S. Tett, 1995: Climate response to increasing amounts of greenhouse gases and sulfate aerosols. Nature,376, 501–504.

  • Nash, J., and J. L. Brownscombe, 1983: Validation of the stratospheric sounding unit. Adv. Space Res.,2, 59–62.

  • ——, and G. F. Forrester, 1986: Long-term monitoring of stratospheric temperature trends using radiance measurements obtained by the TIROS-N series of NOAA spacecraft. Adv. Space Res.,6, 37–44.

  • ——, and P. R. Edge, 1989: Temperature changes in the stratosphere and lower mesosphere 1979–1988 inferred from TOVS radiance observations. Adv. Space Res.,9, 333–341.

  • Oort, A. H., and H. Liu, 1993: Upper-air temperature trends over the globe, 1958–1989. J. Climate,6, 292–307.

  • Parker, D. E., 1986: Strategy for early detection of climate change. Adv. Space. Res.,6, 27–36.

  • ——, and J. L. Brownscombe, 1983: Stratospheric warming following the El Chichón volcanic eruption. Nature,301, 406–408.

  • Pick, D. R., and J. L. Brownscombe, 1981: Early results based on the stratospheric channels of TOVS on the TIROS-N series of operational satellites. Adv. Space Res.,1, 247–260.

  • Ramaswamy, V., M. D. Schwarzkopf, and W. J. Randel, 1996: Fingerprint of ozone depletion in the spatial and temporal pattern of recent lower stratospheric cooling. Nature,382, 616–618.

  • Reynolds, R. W., 1983: A comparison of sea surface temperature climatologies. J. Climate Appl. Meteor.,22, 447–459.

  • Salby, M. L., 1996: Fundamentals of Atmospheric Physics. Academic Press, 627 pp.

  • Santer, B. D., T. M. L. Wigley, and P. D. Jones, 1993: Correlation methods in fingerprint detection studies. Climate Dyn.,8, 265–276.

  • Smith, W. L., H. M. Woolf, C. M. Hayden, D. Q. Wark, and L. M. McMillin, 1979: The TIROS–N Operational Vertical Sounder. Bull. Amer. Meteor. Soc.,60, 1177–1187.

  • Spencer, R. W., and J. R. Christy, 1990: Precise monitoring of global temperature trends from satellites. Science,247, 1558–1562.

  • ——, and ——, 1993: Precision lower stratospheric temperature monitoring with the MSU: Technique, validation, and results 1979–1991. J. Climate,6, 1194–1204.

  • Spiegel, M. R., 1988: Theory and Problems of Statistics. 2d ed. Schaum’s Outline Series, McGraw-Hill, 504 pp.

  • Stolarski, R. S., P. Bloomfield, and R. D. McPeters, 1991: Total ozone trends deduced from NIMBUS 7 TOMS data. Geophys. Res. Lett.,18, 1015–1018.

  • Taylor, F. W., J. T. Houghton, G. D. Peskett, C. D. Rogers, and E. J. Williamson, 1972: Radiometer for remote sounding of the upper atmosphere. Appl. Opt.,11, 135–141.

  • 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 observation. Science,274, 1170–1173.

  • Wang, W.-C., Z. Zeng, and T. R. Karl, 1990: Urban heat islands in China. Geophys. Res. Lett.,17, 2377–2380.

  • ——, M. P. Dudek, X.-Z. Liang, and J. T. Kiehl, 1991: Inadequacy of effective CO2 as a proxy in simulating the greenhouse effect of other radiatively active gases. Nature,350, 573–577.

  • View in gallery

    Aspects of the SSU1 response function: (a) Wide-band (dashed line) and pressure-modulated (solid line) transmission filters; (b) weighting function showing effects of cell pressure, p0, changes. Mean cell pressure is given in Table 1, maximum/minimum values are derived by adding/subtracting the associated error.

  • View in gallery

    (a) Monthly mean measured SSU1 cell pressure. Vertical dotted lines indicate where a switch to a new satellite has been made. (b) Simulated effects of cell pressure, p0, history on SSU1 TB for two atmospheric cases. The effect is given in terms of a ratio between brightness temperatures calculated with p0 varying according to Fig. 1a, and those obtained when p0 was held constant over the measurement period.

  • View in gallery

    (a) Time series of SSU1 global monthly mean TB from Jan 1979 to Dec 1994. (b) Power spectrum of (a). Horizontal axis gives the frequency component, f, equal to t/N, where t is the time period, and N is the total number of months in the time series. Hence, for a period of 12 months, f = 0.083.

  • View in gallery

    Climatological monthly mean SSU1 TB fields constructed using the complete 1979–94 time series: (a) Jan; (b) Jul.

  • View in gallery

    (a) Temporal progression of monthly mean zonally averaged SSU1 brightness temperatures. (b) Deseasonalized monthly mean globally averaged SSU1 TB anomalies. (c) As in (b) but using data adjusted for the NOAA-9 radiometric error. Deseasonalized anomalies are constructed at each grid point by subtracting the relevant monthly climatological field from each corresponding monthly mean field over the time series. These gridpoint values are then averaged over the requisite spatial scale.

  • View in gallery

    (a) Temporal progression of monthly mean zonally averaged SSU1 TB anomalies (original data). Vertical lines indicate operative period of NOAA-9. (b) As in (a) using data adjusted for the NOAA-9 radiometric error. Vertical lines indicate times corresponding to El Chichón (EC) and Mount Pinatubo (MP) eruptions. (c) As in (a) for 20°N–20°S. (d) As in (b) for 20°N–20°S.

  • View in gallery

    Time series of SSU1 annual mean globally averaged TB. (a) Original data. (b) Adjusted for NOAA-9 radiometric error. (c) Adjusted for NOAA-9 radiometric error and cell pressure effects.

  • View in gallery

    Annual mean spatially resolved SSU1 TB fields for selected years: (a) 1986, (b) 1987, (c) 1992, (d) 1993.

  • View in gallery

    (a) Temporal standard deviations of annual mean SSU1 TB data in K. (b) Linear trends in SSU1 TB fields from 1979 to 1994 obtained using a least squares fit. Shaded areas indicate trends of greater than 90% significance.

  • View in gallery

    SSU1 TB fields simulated by GENLN2 using HADCM2 decadal mean atmospheres as input: (a) M1865, (b) M2045, (c) M2045 − M1865.

  • View in gallery

    Global mean temperature profiles for 1865 and 2045 from HADCM2. Dashed lines indicate the level corresponding to the peak of the SSU1 weighting function in each case.

  • View in gallery

    Temperature at the peak of the SSU1 weighting function: (a) M1865, (b) M2045, (c) M2045 − M1865.

  • View in gallery

    Selected annual mean spatially resolved SSU1 TB anomaly fields, ΔD(x, t) in K: (a) 1979, (b) 1984, (c) 1989, (d) 1994.

  • View in gallery

    (a) Pattern similarity statistics R(t) and C(t) with corresponding least squares linear trends. (b) Spatial mean [b〈ΔD(t)〉] and pattern anomaly [aR(t)] components of C(t).

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 168 15 1
PDF Downloads 16 7 0

Climate Variability and Trends in SSU Radiances: A Comparison of Model Predictions and Satellite Observations in the Middle Stratosphere

View More View Less
  • 1 Space and Atmospheric Physics Group, Imperial College, London, United Kingdom
Full access

Abstract

Several recent studies have highlighted the potential of utilizing statistical techniques to pattern match observations and model simulations in order to establish a causal relationship between anthropogenic activity and climate change. Up to now these have tended to concentrate upon the spatial or vertical patterns of temperature change. Given the availability of contiguous, global-scale satellite observations over the past two decades, in this paper the authors seek to employ an analogous technique to spatially match model predictions to directly measured radiances. As part of the initial investigations, the technique to channel 1 of the Stratospheric Sounding Unit, sensitive to stratospheric temperature and carbon dioxide concentrations, is applied. Over the majority of the globe the observations show a negative trend in brightness temperature, with significant decreases occurring throughout the Tropics. The influence of the volcanic eruptions of El Chichón and Mount Pinatubo can also be clearly identified. Simulated brightness temperature fields, against which the satellite data are compared, are calculated using atmospheric temperature profiles from a transient climate change run of the Hadley Centre GCM. The modeled change pattern also indicates a global reduction in brightness temperature but with an altered spatial distribution relative to the observations. This tendency is reflected in the trends seen in the correlation statistics. One, dominated by the spatial mean change, shows a significant positive trend; while the other, influenced by patterns around this mean, exhibits a reducing correlation with time. Possible reasons for this behavior are discussed, and the importance of both improving model parameterizations and performing additional“unforced” simulations to assess the role of natural variability is stressed.

Corresponding author address: Dr. H. Brindley, SPAT, Imperial College, Blackett Laboratory, Prince Consort Road, London SW7 2BZ, United Kingdom.

Email: h.brindley@ic.ac.uk

Abstract

Several recent studies have highlighted the potential of utilizing statistical techniques to pattern match observations and model simulations in order to establish a causal relationship between anthropogenic activity and climate change. Up to now these have tended to concentrate upon the spatial or vertical patterns of temperature change. Given the availability of contiguous, global-scale satellite observations over the past two decades, in this paper the authors seek to employ an analogous technique to spatially match model predictions to directly measured radiances. As part of the initial investigations, the technique to channel 1 of the Stratospheric Sounding Unit, sensitive to stratospheric temperature and carbon dioxide concentrations, is applied. Over the majority of the globe the observations show a negative trend in brightness temperature, with significant decreases occurring throughout the Tropics. The influence of the volcanic eruptions of El Chichón and Mount Pinatubo can also be clearly identified. Simulated brightness temperature fields, against which the satellite data are compared, are calculated using atmospheric temperature profiles from a transient climate change run of the Hadley Centre GCM. The modeled change pattern also indicates a global reduction in brightness temperature but with an altered spatial distribution relative to the observations. This tendency is reflected in the trends seen in the correlation statistics. One, dominated by the spatial mean change, shows a significant positive trend; while the other, influenced by patterns around this mean, exhibits a reducing correlation with time. Possible reasons for this behavior are discussed, and the importance of both improving model parameterizations and performing additional“unforced” simulations to assess the role of natural variability is stressed.

Corresponding author address: Dr. H. Brindley, SPAT, Imperial College, Blackett Laboratory, Prince Consort Road, London SW7 2BZ, United Kingdom.

Email: h.brindley@ic.ac.uk

1. Introduction

The problem of being able to ascribe a causal relationship between observed climatic trends and human activities has long occupied researchers and continues to provide a major challenge to our understanding of the climate system. Recent efforts to attribute climate change to an anthropogenic modification have tended to utilize highly statistical methods to correlate patterns of observed temperature trends with patterns predicted by GCMs given a particular forcing scenario (Intergovernmental Panel on Climate Change 1995 and references therein). The idea underlying much of this work is to define a model response pattern given a particular forcing and to see whether this pattern becomes more evident in the observed field with time through the use of a specifically defined pattern correlation coefficient (e.g., Barnett 1986; Santer at al. 1993). The rationale behind such an approach is that the model-predicted change in the spatial–vertical pattern of the temperature field should be unique for a given forcing; hence, detection of a similarly time-evolving pattern in the temperature observations will allow greater confidence in the attribution of the change to the given perturbation. Results from the most recent attempts to pattern match theoretical predictions from both equilibrium and transient GCMs to the observed radiosonde temperature records are encouraging, suggesting that when sulfate aerosol and ozone changes are included a significant trend in the correlation statistic can be obtained (Mitchell et al. 1995; Tett et al. 1996).

Despite these findings, the use of temperature alone as a detection variable is not ideal. Although the surface- and radiosonde-derived atmospheric temperature records provide the longest available time series against which to test climate change predictions, there still remain questions regarding the accuracy and internal consistency of the data (Parker 1986). Specific problems may arise because of changes in measurement techniques (e.g., Reynolds 1983), alterations in the spatial distribution of reporting stations over time, or as a result of the increase in urbanization seen over much of the globe during the past century (e.g., Wang et al. 1990). At mid- to upper-tropospheric levels the accuracy of radiosonde data is particularly uncertain; in addition, over the longer time periods, consistency between different instrumental designs is also open to question (Gaffen 1994). Uneven spatial coverage is a major problem when attempting to construct global or zonal averages of the measurements obtained, with a definite bias in station location toward the Northern Hemisphere continental regions at the expense of the majority of the Southern Hemisphere (Oort and Liu 1993).

Given the above considerations, an alternative method of obtaining measurements that exhibit continuity and consistency on a global scale becomes attractive. The most obvious candidate for providing such data are the measurement archives of the earth-observing satellites, some of which provide global coverage from 1979 to the present day in terms of outgoing radiances for channels of various frequency. To exploit this data, Harries et al. (1998) develop a “spectral-spatial signature” detection method. The approach involves modeling the expected response of selected spectral channels to a given climate perturbation before combining them to obtain a spatially varying, multivariate signal. This signal is then searched for in the corresponding data records in an analogous manner to pattern matching in temperature fields. Through the careful selection of available channels, one can obtain an alternative measure of changes in the spatial and vertical temperature structure of the atmosphere while retaining information concerning perturbations in the concentrations of the absorbing constituents, which, it is proposed, are primarily responsible for these changes (Kiehl 1983; Charlock 1984).

One benefit associated with the use of spectrally decomposed radiances would be the motivation such an approach would provide to improve the accessibility and documentation of satellite radiance records. Although there is a vast store of archived satellite climate data, up to now attempts to produce long-term, self-consistent, fully calibrated data time series have been fairly limited (e.g., Spencer and Christy 1990; Bates et al. 1996). Part of the problem lies in the original purposes for which the satellites were designed (i.e., short-term forecasting/model validation), since complexities may arise because of an absence of information concerning alterations in data-processing techniques (Spencer and Christy 1993). Adequate instrument intercalibration also presents a particular difficulty where periods of overlap between successive operational satellites are short or even nonexistent. The question of ensuring data compatibility will become even more pertinent as existing satellite series are phased out and replaced by alternative observational campaigns (International TOVS Working Group 1997).

This study is intended as a companion to the papers of Harries et al. (1998) and Geer et al. (1999). As an example of the techniques that could be employed to develop a spectral signature of climate change, the former paper focused on two spectral bands sounded respectively by the High Resolution Infrared Spectrometer (HIRS/2) and the Stratospheric Sounding Unit (SSU). Both instruments are component parts of the Television Infrared Observational Satellite Operational Vertical Sounder, which has been flying on the National Oceanic and Atmospheric Administration (NOAA) series of earth observation satellites, operational from 1979 onward (Kidwell 1995).

From considerations of data availability and consistency, the two channels selected for study were HIRS/2 channel 12 (HIRS12), centered at 6.7 μm, in the wing of the 6.3-μm water vapor band, and SSU channel 1 (SSU1), sounding the midstratosphere and located within the 15-μm carbon dioxide band (Smith et al. 1979). Geer et al. (1999) obtained a measure of the time-evolving spatial agreement between the observed HIRS12 brightness temperature fields and the predicted change from GCM simulations. The various factors influencing the degree of correlation between the observations and predictions were discussed in detail. A particular finding was the tendency of short-term quasiperiodic climate fluctuations such as El Niño to dominate variability within the channel and thus reduce the significance of trends within the data. Here we seek to provide a similar investigation in terms of the SSU1 observations and model simulations.

In order to achieve this aim, the study will be split into the following five main sections. Section 2 will provide a brief description of the SSU instrument, followed by a more in-depth discussion of the techniques used to simulate the instrument response. Particular emphasis will be placed on the sensitivity of the model simulations to variations in instrumental parameters and atmospheric composition. Section 3 will focus on the SSU1 data record from 1979 to 1994, outlining the temporal progression of both the global mean and spatially resolved observations. In section 4 we illustrate simulated global SSU1 fields derived using a combination of the techniques outlined in section 2 and the results from a transient GCM experiment. The evolving level of agreement between these model fields and the 16-yr data record is then tested in section 5, and the conclusions to be drawn from the investigations are presented in section 6.

2. SSU channel 1 characteristics

a. Instrument description

The stratospheric sounding unit is a three-channel infrared radiometer designed to measure the radiance emitted by stratospheric carbon dioxide within the 15-μm ν2 band (Miller et al. 1980). The instrument employs the pressure modulation principle, obtaining a signal in each channel by viewing the atmosphere through an absorption cell in which the pressure of the absorbing gas (carbon dioxide) is suitably modulated. This pressure modulation has the effect of varying the optical transmission of the gas at all wavenumbers corresponding to lines in the carbon dioxide absorption spectrum. An interference filter then confines the spectral response so that contributions from CO2 bands other than that at 15 μm are neglected. By selecting only the modulated part of the signal, any other radiation not originating from within the 15-μm band is rejected.

Viewing angles of the instrument range from ±35° to the nadir via eight steps of 10°. The field of view has a diameter of 10°, resulting in a ground resolution of approximately 200 km. Calibration is achieved via periodic views of cold space and an internal blackbody target within the instrument.

b. Simulation and sensitivity of instrument response

In order to realistically simulate the response of the SSU pressure modulated radiometer (PMR), the GENLN2 line-by-line radiative transfer code (Edwards 1992) was employed. Following the approach outlined by Edwards, if Iν,z is the radiation incident on the radiometer then R, the signal measured is
i1520-0442-12-11-3197-e1
where Fν is the total radiometer response function and can be written as the product of the wideband interference filter, Gν, and the high-frequency gas cell response, Hν,
FνGνHν
Assuming the interference filter varies very slowly with ν, then Gν can be averaged over a suitably selected“wide” wavenumber interval i so that R becomes
i1520-0442-12-11-3197-e3
where
i1520-0442-12-11-3197-e4
and N is the total number of wide wavenumber intervals. The pressure modulation cycle within the gas absorption cell can be modeled using a two-cell approximation,
Hντlowpcellντhighpcellν

The resulting high-resolution transmission filter is then convolved with the appropriate atmospheric radiances at each individual wavenumber point before being integrated over i and multiplied by the corresponding wideband transmissivities. Summation over all N frequency intervals follows to give the total channel radiance measured by the radiometer, which is then converted to an equivalent brightness temperature, TB. Plots of the wideband transmission function and the pressure-modulated transmission filter, Hν, calculated using the above approach are shown in Fig. 1a.

Details of the specifications of the relevant parameters needed to simulate the SSU radiometric response are given in Table 1. In each case these represent best estimates over the period of SSU1 measurements obtained from archived values at the Remote Sensing Instrumentation Branch, U. K. Meteorological Office (UKMO), Farnborough. All have varied over the lifetime of an individual satellite as well as between different satellites. Where available, root-mean-square errors associated with each parameter have been stated. In order to investigate the impact that realistic variations in these parameters could have upon measured brightness temperatures, sensitivity studies utilizing a representative global mean atmosphere constructed from the Air Force Geophysics Laboratory (AFGL) profiles (Anderson et al. 1986) were performed. A variation in cell temperature of 3 K induces an effect of less than 0.01% on SSU1 TB. Similarly, fluctuations of the error magnitude quoted in cell length and modulation amplitude produce a change in TB of less than 0.01% and 0.02%, respectively.

Of potentially more importance is the noted uncertainty in mean pressure within the absorption cell, p0. The positioning of the weighting function associated with a given SSU channel is determined principally by the amount of carbon dioxide and p0 (Pick and Brownscombe 1981). For channel 1, the peak of the weighting function is nominally located at 15 mb. Variations in mean cell pressure manifest themselves through shifts in the altitude of the peak of the radiometer weighting function. Equation (6) gives an analytical expression for the pressure at which the peak of the weighting function would be expected for a CO2 PMR based on the assumption of strong, independent Lorentz lines (Taylor et al. 1972),
pPEAKβLa1/2p0
where β is a factor accounting for carbon dioxide self-broadening, and L is the length of the cell in centimeters. The variable a is the thickness in centimeters that the total column of carbon dioxide would have if it were compressed to standard temperature and pressure. From Eq. (6), increases (decreases) in p0 lead to an increase (decrease) in the pressure level of the weighting function peak (Fig. 1b). Since the stratosphere displays an increase in temperature with height, increases (reductions) in pPEAK manifest themselves through decreases (increases) in the measured TB. The magnitude of these changes becomes greater for a given shift as the stratospheric lapse rate increases.

Figure 2a illustrates the mean monthly progression of p0 for SSU1 from 1979 to 1994. Values of p0 were determined routinely by the UKMO typically at 6-month intervals or upon the launch of a new spacecraft. The horizontal dashed line shows the nominal mean cell pressure. Departures from this value are clearly apparent over the complete observation period. Vertical dotted lines indicate the period for which each NOAA satellite provided the measurements used to construct the dataset employed here. The correspondence between a satellite switch and a distinct jump in the cell pressure is marked. Also noticeable is the drift in cell pressure over individual satellite lifetimes, particularly for NOAA-9 and to a lesser extent NOAA-11.

To assess the possible effects of the cell drifts upon measured brightness temperatures, response functions relevant to the annual mean cell pressure were calculated for each year of the measurement period. Each individual function was then convolved with a single global mean atmospheric radiance value, and the resulting channel TB obtained. These values were then used to scale the channel TB obtained using the nominal mean cell pressure to give a multiplying factor dependent purely on cell pressure drifts (Fig. 2b). The calculated effects on TB of the cell pressure shifts are less than 0.1%; however, the temporal progression of this multiplying factor can substantially reduce the trends seen in the observed data (section 3). Given the dependence of TB on stratospheric lapse rate, a worst-case scenario was also obtained by repeating the above calculation using the tropical AFGL atmosphere. As expected, the magnitude of the multiplying factor is seen to increase, albeit by less than 0.03% for all years.

Sensitivity studies were also carried out to assess the effects of spectral resolution, atmospheric gaseous absorption, and the impact of CO2 line mixing on the modeled channel radiances. Table 2 gives an indication of the importance of these various considerations relative to a reference calculation. In each case the Voigt line shape was used for non-CO2 gases. A Voigt shape was also employed for CO2 when line mixing was ignored, the modified Voigt shape accounting for sub-Lorentzian line wings being employed in the cases where line mixing was included (Edwards and Strow 1991).

In all cases, the table shows that the impact on channel TB is minimal. Particularly noteworthy is the insensitivity of channel 1 both to direct O3 absorption via the 701-cm−1 band and the explicit inclusion of a line-mixing model. On the basis of these results, the response functions used in section 4 utilize a spectral resolution of 0.01 cm−1, with CO2 as the only absorbing gas and with line mixing included.

3. Observational data

The dataset used here was provided by the UKMO in the form of daily global nadir radiance fields on a 5° × 5° grid. Full details of the processing applied to the raw radiances are given in Bailey et al. (1992). The procedure involves a quality control check for anomalous values, and linear spatial and temporal interpolation of the raw data to a common time and global grid. Fourier analysis is performed on the daily fields to provide smoothing on a spatial scale of approximately 30° latitude × 15° longitude. The processing procedure applied to the raw data has been consistent over the complete observation period.

Bailey et al. (1992) also provide a brief discussion of the accuracy of the stratospheric data. In terms of the radiance fields, the key influencing factor is the stability of the measuring instrument. Using a combination of laboratory studies and in-orbit comparisons, Nash and Brownscombe (1983) found that a continuous series of radiance observations with a relative error of less than 0.2 K in equivalent brightness temperature could be obtained from the SSU instruments mounted on successive satellites. Nash and Forrester (1986) extended these findings temporally but noted the dependence of the relative error upon intercalibration with an SSU stable in orbit. In addition, they indicated the importance of accounting for differences in observation times between satellites because of the influence of solar diurnal tides (Brownscombe et al. 1985). They also mention the presence of a radiometric error corresponding to a 0.7 K increase in brightness temperature in SSU1 measurements from NOAA-9. After correction for this error, intercomparisons of SSU1 measurements on the earlier NOAA satellites show good long-term accuracy (Nash and Edge 1989). However, a lack of temporal overlap between NOAA-9 and NOAA-11 means that intercalibration between these later instruments has not been performed. The dataset used here does not account for the effects of solar diurnal tides or any drift in sampling times.

For consistency with the HIRS12 analyses used by Geer et al. (1999), the SSU1 observations were confined to between 70°N and 70°S. The amount of missing data over the observation period composed 10% of the total measurement volume, the only significantly poor coverage occurring during September 1981; March, April, and August 1982; October 1984; April 1985; and April–July 1986. Data holes were filled using the results of linear interpolation in time between the nearest reliable daily observations.

Monthly mean SSU1 brightness temperatures were then constructed from the “filled” daily sets. These mean fields were then area weighted and globally averaged to produce the temporal distribution shown in Fig. 3a. A high degree of seasonality is apparent, and this is confirmed by applying a Fourier transform to the time series (Fig. 3b). The strong peak seen at a frequency component of 0.083 corresponds to a time period of 12 months in the data. Closer examination of the monthly record reveals that minimum brightness temperatures occur during austral winter (JJA), while maximum values are found during austral summer (DJF). To investigate this further, monthly mean climatological TB fields for the period 1979–94 were constructed. The climatological TB fields for January and July (Figs. 4a,b) indicate that the fluctuation within the global means is primarily controlled by the intensity of the respective hemispheric winter seasons. Orographically forced planetary waves penetrating into the stratosphere during the winter season act to perturb the polar vortex from radiative equilibrium. The more marked orography in the Northern Hemisphere results in stronger wave activity and hence a more disturbed winter stratosphere. The weaker planetary waves forced during austral winter result in the high latitudes of the Southern Hemisphere remaining closer to radiative equilibrium, with concurrent colder temperatures and a smoother zonal pattern (Salby 1996). This pattern is further emphasized by Fig. 5a, which shows the progression of zonally averaged, monthly mean SSU1 brightness temperatures with time. Within the Southern Hemisphere, TB maxima and minima tend to be confined to latitudes greater than 20°S and they tend to be more extreme than the corresponding Northern Hemisphere values. These latter features can penetrate well into equatorial regions and generally provide the only noticeable deviation from a constant TB background value of between 225 and 230 K throughout the Tropics.

In order to discern the presence of any trends within the SSU1 observations over time, deseasonalized globally averaged monthly mean TB anomalies from 1979 to 1994 are displayed in Fig. 5b. Apparent from this plot is the tendency toward a brightness temperature decrease with time, with three distinct features superimposed on the general trend. The second of these features, extending from April 1985 to November 1988, corresponds exactly to the period of operation of NOAA-9. As noted earlier, the SSU aboard this satellite exhibited a radiometric error with an effect equivalent to a 0.7 K increase in brightness temperature within channel 1. Reducing measurements by 0.7 K across the globe during the relevant period and recalculating the monthly mean anomalies result in the plot shown in Fig. 5c. Note the total removal of the NOAA-9 feature, with a generally smooth downward trend interrupted only by brightness temperature increases occurring in early 1982 and mid-1991. These increases coincide with the volcanic eruptions of El Chichón (April 1982) and Mount Pinatubo (June 1991). The injection of volcanic aerosols into the stratosphere allows local heating through enhanced absorption of terrestrial radiation (Parker and Brownscombe 1983). Radiosonde observations (Labitzke et al. 1983; Labitzke 1994) appear to confirm the link between increased stratospheric aerosol loading and local increases in lower stratospheric temperature, with significantly increased temperatures at Northern Hemisphere low latitudes following both the El Chichón and Pinatubo eruptions. For a given carbon dioxide concentration, within the 15-μm band such a warming would translate directly into enhanced emission and hence increased brightness temperatures (Kiehl 1986).

To investigate the spatially resolved picture, the temporal progression of the zonally meaned TB anomalies for each 5° latitude band is shown for the original (Fig. 6a) and NOAA-9 corrected (Fig. 6b) SSU1 data. Apparent from both plots is the general tendency for TB to decrease with time. By comparing Fig. 6a with Fig. 6b, the extra NOAA-9 induced warming across the globe is also clear. The largest zonal mean anomalies are located poleward of 30°N/S and are a consequence of small shifts in the position and intensity of the midlatitude maxima and minima identified in Fig. 5a relative to the climatological pattern. In comparison, TB anomalies within the Tropics are small (<|3 K|); nonetheless, positive anomalies are seen after both the El Chichón and Pinatubo eruptions. Figures 6c and 6d allow a more detailed study by confining the zonally meaned TB anomaly plots to ±20°. With the removal of the NOAA-9 feature, the warmest anomalies are found immediately after the two volcanic eruptions, three other noticeably warm periods also being present during late 1979, early 1987, and mid-1989. Comparing Fig. 6d to Fig. 6b, it is obvious that the influence of these last three features is reduced in the global mean due to strong negative anomalies at higher latitudes. In contrast, the positive tropical anomalies seen after El Chichón and Pinatubo are reinforced across all latitude bands. Notice also the prolonged period of negative tropical TB anomalies from January 1993 onward.

Given that the model predictions are in the form of decadal mean brightness temperatures, Fig. 7a illustrates the global mean annual average SSU1 brightness temperatures over the observation period. Also shown is the trend in the data calculated using least squares linear regression. Using a Student’s t-test (Spiegel 1988) and assuming the annual means to be independent of one another, this negative trend was found to be significantly different from zero at the 99% level. Figure 7b indicates the effect of employing the NOAA-9 correction to the monthly data before performing the yearly averaging. The effect on the magnitude of the linear trend is negligible, while the reduced variance in the data increases the trend significance. The drop in global mean SSU1 TB of 0.7 K from 1979 to 1988 is in agreement with the value of 0.6 ± 0.2 K reported by Nash and Edge (1989). Finally, Fig. 7c illustrates an attempt to incorporate both the NOAA-9 correction and the effects of changes in mean cell pressure (p0) over time upon the SSU1 measurements. From our earlier arguments (section 2), the general decrease in p0 toward the end of the data (Fig. 2a) would be expected to artificially increase the magnitude of any downward trend in the observations if not accounted for. In order to address this, the tropical worst-case multiplying factors for each year (Fig. 2b) were applied to each global, annual mean value. Obviously, the use of a single theoretical global mean value for each year to account for the cell pressure effects is not ideal; in reality the impact would vary on smaller temporal and spatial scales. However, given that the variation in stratospheric lapse rate generally reduces with latitude, the factors applied here should represent an upper bound in the degree to which they affect the data trend. From Fig. 7c the reduction in the magnitude of the linear trend is clearly apparent. Despite this, the trend is still significant at the 97.5% level.

Having considered the global mean and climatological picture, annual mean globally resolved plots of the NOAA-9 corrected SSU1 observations for selected years are presented in Fig. 8. As illustrated by Fig. 5a, throughout the time series the TB field is dominated by distinct stationary features associated with planetary-scale wave activity. These exhibit a degree of variability in their magnitude and spatial extent (compare Fig. 8b with Fig. 8c) but remain centered at specific locations poleward of 30°N and 30°S. In the Northern Hemisphere, a TB maximum is found over northeastern Asia, stretching across the Bering Strait into Alaska. The corresponding minimum extends over northern Europe and the North Atlantic Ocean. The pattern is similar in the Southern Hemisphere, the TB maximum running across the southern Indian Ocean, while a minimum is located over the South Atlantic and South Pacific Oceans.

Within the Tropics the dominant influence upon the stratosphere is provided by the quasi-biennial oscillation (QBO), a roughly 28-month cycle in the direction of the zonal wind (Holton 1992). However, a Fourier transform performed on the SSU1 observations between 15°N and 15°S showed little evidence of the periodicity that might be expected due to the QBO, possibly because of the limited vertical resolution of the instrument (∼17 km; Fig. 1b). Interannual variability is also at a minimum throughout the majority of the time series, brightness temperatures remaining between 227 and 228 K. The exception to this pattern occurs at the very end of the observation period during 1993 (Fig. 8d) and 1994, when a noticeably lower than average tropical TB field precipitates the drop seen in global mean values (Figs. 7a–c). The question as to whether this might be a spurious result due to the observed reduction in p0 during the lifetime of NOAA-11 can be addressed by comparing the progression of the TB fields for the years 1992 and 1993 (Figs. 8c,d) to that seen for 1986/87 (Fig. 8a,b) (when the rate of decrease of p0 aboard NOAA-9 was even greater). The increase in TB seen within the Tropics over the latter period suggests that the marked decrease from 1992 onward is unlikely to be an artifact of instrument performance. Indeed, the observed drop in SSU1 TB is in agreement with the findings of McCormick et al. (1995), who note that the 1993 stratospheric temperatures measured by the microwave sounding unit are the lowest seen over the instrument’s lifetime. They tentatively suggest that this could be a consequence of reduced solar heating arising as a consequence of enhanced ozone depletion due to heterogeneous chemical reactions taking place on the surface of stratospheric aerosols injected by the Pinatubo eruption.

The results discussed in the previous two paragraphs are reinforced by Fig. 9a, which illustrates the standard deviation in the annual mean fields over the 16-yr observation period. The largest variability in the SSU1 measurements is located at the boundaries between the extratropical maxima/minima. Conversely, the limited nature of the TB fluctuations noted within tropical regions is reflected in the small standard deviations at latitudes less than 30°N and S. Figure 9b presents the linear trends seen within the data at each grid point. These range from +0.025 K yr−1 within the Southern Hemisphere TB maximum to −0.1 K yr−1 over much of the United States and China, the overall tendency being toward TB decreases with time. Shaded areas denote where the gridpoint trends are significantly different from zero at the 90% level. Unsurprisingly, the regions of statistical significance are generally coincident with those areas showing the lowest interannual variability, incorporating the vast majority of the Tropics and a substantial portion of North America, Southeast Asia, and Australia. In these regions, SSU1 TB trends have a magnitude that is generally greater than 0.075 K yr−1.

4. Model predictions

a. Derivation of simulated TB fields

Global temperature and specific humidity fields representative of 10-yr mean atmospheres centered on 1865 and 2045 were obtained from a transient climate change integration of the Hadley Centre Climate Model (HADCM2). The model run included both the effects of increases in greenhouse gases, parameterized through the use of a single effective carbon dioxide concentration, and changes in tropospheric sulfate aerosols (Mitchell et al. 1995). HADCM2 is a coupled ocean–atmosphere general circulation model, with 20 layers in the ocean and 19 in the atmosphere. The horizontal resolution is 2.5° latitude × 3.75° longitude. Further details concerning model spinup, simulation of present-day climate, and variability are given by Johns et al. (1996). In order to ensure compatibility between model output and satellite coverage, all model values were bilinearly interpolated onto the same 5° × 5° grid. For ease of notation, the TB fields derived using the decadal mean model atmospheres will be referred to here as M1865 and M2045.

SSU1 brightness temperatures were calculated at each grid point for both HADCM2 atmospheres using the GENLN2 model as outlined in section 2. Given the results of the earlier sensitivity studies only the temperature profiles from HADCM2 were required; using representative Intergovernmental Panel on Climate Change hindcasts and predictions (Intergovernmental Panel on Climate Change 1990) carbon dioxide concentrations were set at 280 and 500 ppmv for the 1865 and 2045 cases, respectively, and assumed to be spatially invariant.

The impact of using temporally meaned model fields to generate SSU1 TB predictions must be addressed. Assuming carbon dioxide amounts are invariant, sensitivity analyses reveal that the difference between the SSU1 TB calculated using a mean atmosphere, and that calculated from the average of the individual atmospheres used to obtain the mean, are negligible. Hence, for M1865, where CO2 amounts can be considered constant, use of the decadal mean should have no impact on TB. The situation appears to be more complicated for the M2045 case due to the rapid increase in CO2 predicted to occur over this decade and due to the dependence of the weighting function on CO2 amount [Eq. (6)]. However, sensitivity studies again indicate that the use of mean quantities has a negligible effect on the calculated TB, with the shift in the weighting function peak being less than 0.7 mb over the 2040–50 decade for typical CO2 predictions.

b. Patterns in simulated SSU1 TB fields

Simulated SSU1 TB fields using the HCM2 M1865 and M2045 model predictions are presented in Fig. 10a and Fig. 10b, respectively. The most obvious feature to note when comparing the two fields is the predicted fall in brightness temperatures at all locations between 1865 and 2045. Looking in more detail, the stability in the positioning of the major features is evident. In both M1865 and M2045, peak temperatures are located over the Indian Ocean with a banded zonal structure extending across the equatorial regions from 30°N to 30°S, temperatures generally decreasing with increasing latitude. A more disturbed picture is present in the extratropics, with smaller peaks over the Asian plateau and eastern Southern Ocean, this latter feature being much reduced in its spatial extent in the M2045 relative to the M1865 case. Minima are apparent to the west of these secondary maxima in both atmospheres. The pattern stability will be due in part to the temporal averaging of the atmospheric data used to generate the fields, since a degree of smoothing will have been introduced through the meaning process. However, the regular occurrence of quasi-stationary features was also noted in the observed annual mean TB fields. Indeed, the positioning of the model-predicted extratropical maxima and minima is consistent with the measured fields, the only slight difference being the reduced extent and shift eastward of the Southern Hemisphere peak in the simulated fields. The major inconsistency between the model simulations and observed SSU1 TB fields occurs within the equatorial regions. Here, the Indian Ocean maximum and generally larger than average values of TB across the Tropics evident in the modeled fields are simply not seen in the observations.

The difference between the two model-generated TB fields is shown in Fig. 10c. This confirms the earlier assertion that the predicted TB change from 1865 to 2045 is uniformly negative, with an average background decrease in simulated TB of approximately 4 K. Of particular interest for spatial pattern correlation studies (section 5) are those areas that diverge relative to this average response. Such an area is found over much of the eastern Southern Hemisphere, with an enhanced reduction that generally increases with latitude. A second area of larger than average TB reductions is also evident over the North Atlantic. Given the dichotomy between model predictions and observations within the Tropics, the absence of a strong deviation from the model-predicted average TB decrease across this region is notable.

From section 2 it was seen that if instrument parameters remain constant, the principal influences upon SSU1 brightness temperatures are the stratospheric temperature profile and the atmospheric concentration of carbon dioxide. Assuming a stratospheric temperature profile that increases monotonically with height, an increase in carbon dioxide alone would have the effect of increasing SSU1 TB via a rise in the effective emitting level. However, since the stratospheric temperature is predicted to cool with increasing carbon dioxide concentrations, the actual response is a combination of two opposing effects. To illustrate this point, Fig. 11 shows the global mean temperature profile for M2045 and M1865 constructed using area weighting of the individual gridpoint profiles. Also indicated is the level of the peak of the analytical SSU1 weighting function associated with the global mean CO2 concentrations representative of each atmosphere. The increase in altitude of the weighting function peak from 1865 to 2045 is apparent; however, the stratospheric temperature decrease over the same period is such that the temperature corresponding to this level is approximately 3 K cooler than that at the peak of the 1865 function. Hence the substantial decrease seen in the simulated SSU1 brightness temperatures from M1865 to M2045 is expected.

The strong dependence of the spatial pattern of the simulated TB changes upon carbon dioxide concentration and stratospheric temperature is indicated even more clearly by Figs. 12a–c. These plots show, respectively, the global distribution of the temperature at the weighting function peak, TPEAK, for M1865 and M2045, and the change in this quantity between the two simulations. Fields of TPEAK were estimated by using Eq. (6) to find pPEAK at each location, then exponentially interpolating to the temperature corresponding to this pressure. Note the coherence between regions of TB (Figs. 10a,b) and TPEAK (Figs. 12a,b) maxima and minima for both M1865 and M2045. This leads directly to the match seen between the difference fields (Figs. 10c and 12c), with regions of larger than average decreases in TPEAK generally corresponding to the areas of largest TB deviations. In essence then, the modeled SSU1 fields incorporate information concerning both the initial climate forcing mechanism (the CO2 increases) and the climate response (the reductions in stratospheric temperature).

5. Pattern correlation statistics

Following the example of Geer et al. (1999) we employ two measures, the “centered,” R(t), and “uncentered,” C(t), pattern correlation statistics, to assess the time-evolving level of agreement between the SSU1 annual mean observations and the simulated TB fields. The former measure was first defined by Santer et al. (1993), the latter by Barnett (1986), and each is expressed in the forms used here by Eqs. (7) and (8). The underlying rationale behind the use of the two statistics is that while C(t) can contain information pertaining to both changes in the spatial mean and changes in the anomalies relative to this mean, R(t) is only influenced by changes in the anomaly pattern:
i1520-0442-12-11-3197-e7
Here, x is an index over each spatial point, with the model-predicted spatial pattern of change, ΔM(x), defined as
Mxxx
and ΔD(x, t), the evolving observational anomaly field, given by
Dx, tDx, tD0x
where D(x, t) are the annual mean SSU1 observations, and D0(x) is the mean of this array over the measurement period. This definition was chosen to be consistent with Geer et al. (1999). Essentially we are constrained by the length of the observation period. Sensitivity studies indicate that employing anomalies calculated using 2-yr overlapping averages relative to a 2-yr mean reference state (1979 and 1980) has a negligible effect, reducing the R(t) and C(t) trends by less than 3%. The area weighting field, W(x), is identical to that given in Geer et al. (1999), the observed spatial mean anomaly field for each year, Δ(t), being defined as
i1520-0442-12-11-3197-e11
with a similar expression for the spatial mean of the predicted change, ΔM̂. Finally, sD(t) and sM are, respectively, the spatial standard deviations of the observed and modeled fields, the former being given by
i1520-0442-12-11-3197-e12
with the latter being defined similarly.
Figure 13 gives examples of the observed SSU1 TB anomaly fields, ΔD(x, t). Apparent in the global mean sense is the increasing cooling with time, a finding that corresponds with the modeled response. This initial perusal is borne out by the positive trend seen in the uncentered C(t) statistic (Fig. 14a), significantly different from zero at the 99% level. However, the same plot also illustrates a negative trend in the centered R(t) statistic, also significant at the 99% level. Hence it appears that while the C(t) correlation is dominated by the observed global mean spatial change, the pattern about this mean change becomes less like that shown in the model results with time. This theory can be investigated in more detail by recourse to Eq. (13) (Santer et al. 1993), which decomposes C(t) into spatial mean and pattern anomaly components,
CtaRtbt
where
i1520-0442-12-11-3197-e14a
Plots of the various components of Eq. (13) against time (Fig. 14b) illustrate the dominance of the spatial mean change, Δ(t), in determining C(t) with the change in the pattern about the mean exerting little or no influence. Note in particular the influence of the warming due to El Chichón and Pinatubo being manifested in the distinct local minima within the upward trend of C(t).

To explain the negative trend in R(t), recall that from the previous section the modeled pattern response exhibited a generally featureless picture in the Tropics, with larger than average TB decreases located in the eastern Southern Ocean and in the North Atlantic (Fig. 10c). A similar picture is seen in several of the anomaly fields through 1979–94, particularly in the years 1987 and 1989 [corresponding to the peaks on the R(t) plot]. However, an increasingly evident disagreement between the spatial anomaly patterns seen in the measured and simulated fields is apparent from 1990 onward. This is generally engineered through the persistent occurrence of positive TB anomalies in the observed fields over the eastern Southern Ocean. Indeed, 1994 gives a particularly marked negative correlation, exhibiting an anomaly pattern that is almost the complete inverse of the modeled response (Fig. 13d).

From the combination of the C(t) and R(t) trends it would appear that over the period studied, while the model predicted changes in SSU1 TB are in agreement with observations in terms of the global mean sign of the change seen, the time-evolving spatial anomaly pattern in the data is steadily diverging from the simulated picture. Three possible mechanisms for this divergence immediately present themselves. The first concerns the use of equivalent CO2 as a proxy for the effects of other radiatively active trace gases. The second arises as a result of natural climatic perturbations (e.g., volcanic eruptions), whose “signals” are present in the data but not in the model runs. Similarly, the third involves the absence of predicted changes in stratospheric ozone from the model simulations.

Addressing the first mechanism, Wang et al. (1991) use GCM simulations to show that such an approximation will tend to produce a large overestimate in stratospheric cooling, while underestimating the degree of surface and upper-tropospheric heating. Of particular relevance for this study is the finding that the use of equivalent CO2 tends to produce a response that is only weakly dependent on season and latitude. In contrast, the explicit inclusion of trace gases leads to an atmospheric response with a markedly stronger seasonal and latitudinal dependence.

The direct impact of the second mechanism has already been indicated in terms of the SSU1 TB increases associated with the El Chichón and Pinatubo eruptions reducing the C(t) correlation during 1982 and 1991. However, because of their global extent these annual mean TB increases do not cause significant deviations in the R(t) time series. Nevertheless, as noted earlier, the temperature changes engendered by, in particular, the Pinatubo eruption have the potential to influence both stratospheric chemistry and dynamics. It is entirely feasible that these “feedback” effects may in turn perturb stratospheric temperatures and impose a further indirect effect on SSU1 TB.

The third mechanism relates to the observed changes in stratospheric ozone over the observation period. Measurements from the Total Ozone Mapping Spectrometer indicate a significant decrease in ozone totals poleward of 20°N/S, the negative trends over the period 1979–90 increasing with latitude, particularly toward Antarctica (Stolarski et al. 1991). This pattern is confirmed by Stratospheric Aerosol and Gas Experiment II data (McCormick et al. 1992), which suggests the greatest ozone depletions occur within the lower stratosphere, at altitudes below 23 km. Ramaswamy et al. (1996) investigated the effects of these observed ozone losses in a GCM experiment and concluded that the results obtained corresponded well with the spatially and seasonally varying temperature observed in the lower stratosphere. One notable result was the tendency for dynamical changes induced by the ozone loss to produce, at altitudes above the ozone depletion, a cooling in the Tropics and a heating at high latitudes, especially in the Southern Hemisphere. This could have particular relevance for the SSU1 observations given the anomaly patterns seen post-Pinatubo (Fig. 13d).

Although the absence of a direct effect of ozone absorption upon SSU1 TB has already been noted (section 2), and the peak region of ozone loss occurs below the nominal SSU1 sounding level, simple sensitivity studies utilizing a 1D radiative–convective model suggest that the incorporation of representative decadal stratospheric ozone changes can significantly enhance the SSU1 TB decrease via the induced stratospheric temperature changes. Thus the incorporation of ozone changes in model simulations becomes a key requirement in future attempts to pattern match observed and simulated radiance fields.

6. Summary

As a companion to the papers of Geer et al. (1999) and Harries et al. (1998), this study applies the technique of spatial pattern matching to an alternative radiance dataset, nominally sounding the middle stratosphere. The benefits of utilizing radiance measurements include the absence of a dependence on any retrieval algorithm, and the chance to look directly at perturbations to the earth’s radiation field, believed to be the mechanism driving climate change.

The trends and spatial patterns seen within brightness temperature, TB, data from channel 1 of the Stratospheric Sounding Unit (SSU1) have been investigated for the period 1979–94. We have indicated a method for simulating SSU1 TB using the GENLN2 line-by-line code and a two-cell approximation to model the pressure modulation cycle of the instrument. Sensitivity analyses show that, in terms of gaseous absorption, CO2 concentrations play the dominant role in determining SSU1 TB and fluctuations in cell length, temperature, and pressure modulation amplitude of the magnitude reported within the literature should have a minimal effect on the observations. Variations in the mean cell pressure, particularly noticeable when the observing satellite changes, while not producing large effects in themselves, are shown to have increased importance when considering the long-term trend in the measurements.

The global mean SSU1 TB progression from 1979 to 1994 shows a significant trend of −0.07 K yr−1. Accounting for the possible effects of alterations in mean cell pressure tends to reduce this trend, but it remains significantly negative. Apparent within the data are two periods of stratospheric warming associated with the volcanic eruptions of El Chichón and Mount Pinatubo. Resolving the fields spatially reveals that the variability within the dataset is dominated by the strength and extent of midlatitude features associated with planetary wave activity. Within the Tropics, little interannual variability is evident, although it is the fall in brightness temperature over this region that precipitates the coldest global mean values seen over the measurement period in 1993 and 1994. There is a strong possibility that this brightness temperature decrease is a response to the radiative and dynamical influence of post-Pinatubo ozone depletion.

Simulated SSU1 brightness temperature fields using results from the HADCM2 climate change run show a tendency for large decreases with time, a consequence of the predicted increases in carbon dioxide and associated stratospheric cooling. These last two factors are shown to play an opposing role in determining the derived brightness temperature. The simulated fields show the influence of planetary wave activity but show unexpectedly high brightness temperatures in the Tropics, a feature at odds with the observed data. The spatial pattern of change is uniformly negative, with maximum decreases over the eastern Southern Ocean and over the North Atlantic superimposed on an average background drop of 4 K.

Pattern matching between data and simulations indicated two opposing trends depending on which statistical measure was being considered. For the uncentered statistic, an increasing trend was noted and shown to be primarily a result of the decrease in both model and observed SSU1 brightness temperatures with time. When the influence of the spatial mean was removed from both datasets the level of agreement between them was seen to decrease with time. It is suggested that there are two main mechanisms for the lack of temporally evolving agreement in the spatial patterns: first, the use of equivalent carbon dioxide amounts in the model simulations, and second, the exclusion of ozone changes from the simulations. From the results obtained here each mechanism would be expected to exert a significant influence on SSU1 brightness temperature fields.

In essence this study has served as an introduction to an alternative method of exploiting radiance measurements. We have sought to identify the chief influences acting upon the channel under investigation and to highlight and explain regions of model/observational discrepancies as an aid for those involved in future simulations. In terms of the correlation statistics, both the shortness of the observational period, and the absence of multiple unforced GCM simulations with which to test trend significance are a barrier to unambiguous climate change detection. However, the development of additional sources against which to test model predictions can only lead to a tightening of the attribution question.

Acknowledgments

We thank J. Mitchell, A. Slingo, and M. J. Webb of the Hadley Centre for supplying the HADCM2 data; the BADC for supplying the SSU1 data;and P. Evans at RSI for providing access to SSU1 instrument details. We also thank two anonymous reviewers for their helpful comments.

REFERENCES

  • Anderson, G. P., S. A. Clough, F. X. Kneizys, J. H. Chetwynd, and E. P. Shettle, 1986: AFGL atmospheric constituent profiles (0–120 km). Rep. AFGL-TR-86-0110, 47 pp. [Available from AFGL, Hanscom AFB, MA 01731.].

  • Bailey, M. J., A. O’Neill, and V. D. Pope, 1992: Stratospheric analyses produced by the U.K. Meteorological Office. Climate Research Tech. Note 24, 50 pp. [Available from UKMO, London Rd., Bracknell, Berkshire RG12 2SY, United Kingdom.].

  • Barnett, T. P., 1986: Detection of changes in the global tropospheric temperature field induced by greenhouse gases. J. Geophys. Res.,91, 6659–6667.

  • Bates, J. J., X. Wu, and D. L. Jackson, 1996: Interannual variability of upper-troposphere water vapor band brightness temperature. J. Climate,9, 427–438.

  • Brownscombe, J. L., J. Nash, G. Vaughan, and C. F. Rogers, 1985: Solar tides in the middle atmosphere. I: Description of satellite observations and comparison with theoretical calculations at equinox. Quart. J. Roy. Meteor. Soc.,111, 677–689.

  • Charlock, T. P., 1984: CO2 induced climatic change and spectral variations in the outgoing terrestrial infrared radiation. Tellus,36B, 139–148.

  • Edwards, D. P., 1992: GENLN2 A general line-by-line atmospheric transmittance and radiance model. Version 3.0 description and users guide. NCAR Tech. Note NCAR/TN-367+STR, 147 pp. [Available from NCAR, P.O. Box 3000, Boulder, CO 80307-3000.].

  • ——, and L. L. Strow, 1991: Spectral line shape for limb temperature sounders. J. Geophys. Res.,96, 20 859–20 868.

  • Gaffen, D. J., 1994: Temporal inhomogeneities in radiosonde temperature records. J. Geophys. Res.,99, 3667–3676.

  • Geer, A. J., J. E. Harries, and H. E. Brindley, 1999: Spatial patterns of climate variability in upper-tropospheric water vapor radiances from satellite data and climate model simulations. J. Climate,12, 1940–1955.

  • Harries, J. E., H. E. Brindley, and A. J. Geer, 1998: Climate variability and trends from operational satellite spectral data. Geophys. Res. Lett.,25, 3975–3978.

  • Holton, J. R., 1992: An Introduction to Dynamic Meteorology. Academic Press, 511 pp.

  • Intergovernmental Panel on Climate Change, 1990: Climate Change:The IPCC Scientific Assessment. J. T. Houghton, G. J. Henkins, and J. J. Ephraums, Eds., Cambridge University Press, 365 pp.

  • ——, 1995: Climate Change 1995: The Science of Climate Change. J. T. Houghton et al., Eds., Cambridge University Press, 572 pp.

  • International TOVS Working Group, 1997: TOVS data in climate studies. A Report on the Ninth Int. TOVS Study Conf., Igls, Austria, ITWG, 9–13.

  • Johns, T. C., R. E. Carnell, J. F. Crossley, J. M. Gregory, J. F. B. Mitchell, C. A. Senior, S. F. B. Tett, and R. A. Wood, 1996: The second Hadley Centre coupled ocean–atmosphere GCM: Model description, spinup and validation. Climate Dyn.,13, 103–134.

  • Kidwell, K. B., 1995: NOAA Polar Orbiter data users’ guide. NOAA/National Environmental Satellite, Data, and Information Service, 250 pp. [Available from NOAA/NESDIS, Satelllite Data Services Division, Washington, DC 20233.].

  • Kiehl, J. T., 1983: Satellite detection of effects due to increased atmospheric carbon dioxide. Science,222, 504–506.

  • ——, 1986: Changes in the radiative balance of the atmosphere due to increases in CO2 and trace gases. Adv. Space Res.,6, 55–60.

  • Labitzke, K., 1994: Stratospheric temperature changes after the Pinatubo eruption. J. Atmos. Terr. Phys.,56, 1027–1034.

  • ——, B. Naujokat, and M. P. McCormick, 1983: Temperature effects on the stratosphere of the April 4, 1982 eruption of El Chichon, Mexico. Geophys. Res. Lett.,10, 24–26.

  • McCormick, M. P., R. E. Veiga, and W. P. Chu, 1992: Stratospheric ozone profile and total ozone trends derived from SAGE I and SAGE II data. Geophys. Res. Lett.,19, 269–272.

  • ——, L. W. Thomason, and C. R. Trepte, 1995: Atmospheric effects of the Mt Pinatubo eruption. Nature,373, 399–404.

  • Miller, D. E., J. L. Brownscombe, G. P. Carruthers, D. R. Pick, and K. H. Stewart, 1980: Operational temperature sounding of the stratosphere. Philos. Trans. Roy. Soc. London,296, 65–71.

  • Mitchell, J., C. Johns, J. M. Gregory, and S. Tett, 1995: Climate response to increasing amounts of greenhouse gases and sulfate aerosols. Nature,376, 501–504.

  • Nash, J., and J. L. Brownscombe, 1983: Validation of the stratospheric sounding unit. Adv. Space Res.,2, 59–62.

  • ——, and G. F. Forrester, 1986: Long-term monitoring of stratospheric temperature trends using radiance measurements obtained by the TIROS-N series of NOAA spacecraft. Adv. Space Res.,6, 37–44.

  • ——, and P. R. Edge, 1989: Temperature changes in the stratosphere and lower mesosphere 1979–1988 inferred from TOVS radiance observations. Adv. Space Res.,9, 333–341.

  • Oort, A. H., and H. Liu, 1993: Upper-air temperature trends over the globe, 1958–1989. J. Climate,6, 292–307.

  • Parker, D. E., 1986: Strategy for early detection of climate change. Adv. Space. Res.,6, 27–36.

  • ——, and J. L. Brownscombe, 1983: Stratospheric warming following the El Chichón volcanic eruption. Nature,301, 406–408.

  • Pick, D. R., and J. L. Brownscombe, 1981: Early results based on the stratospheric channels of TOVS on the TIROS-N series of operational satellites. Adv. Space Res.,1, 247–260.

  • Ramaswamy, V., M. D. Schwarzkopf, and W. J. Randel, 1996: Fingerprint of ozone depletion in the spatial and temporal pattern of recent lower stratospheric cooling. Nature,382, 616–618.

  • Reynolds, R. W., 1983: A comparison of sea surface temperature climatologies. J. Climate Appl. Meteor.,22, 447–459.

  • Salby, M. L., 1996: Fundamentals of Atmospheric Physics. Academic Press, 627 pp.

  • Santer, B. D., T. M. L. Wigley, and P. D. Jones, 1993: Correlation methods in fingerprint detection studies. Climate Dyn.,8, 265–276.

  • Smith, W. L., H. M. Woolf, C. M. Hayden, D. Q. Wark, and L. M. McMillin, 1979: The TIROS–N Operational Vertical Sounder. Bull. Amer. Meteor. Soc.,60, 1177–1187.

  • Spencer, R. W., and J. R. Christy, 1990: Precise monitoring of global temperature trends from satellites. Science,247, 1558–1562.

  • ——, and ——, 1993: Precision lower stratospheric temperature monitoring with the MSU: Technique, validation, and results 1979–1991. J. Climate,6, 1194–1204.

  • Spiegel, M. R., 1988: Theory and Problems of Statistics. 2d ed. Schaum’s Outline Series, McGraw-Hill, 504 pp.

  • Stolarski, R. S., P. Bloomfield, and R. D. McPeters, 1991: Total ozone trends deduced from NIMBUS 7 TOMS data. Geophys. Res. Lett.,18, 1015–1018.

  • Taylor, F. W., J. T. Houghton, G. D. Peskett, C. D. Rogers, and E. J. Williamson, 1972: Radiometer for remote sounding of the upper atmosphere. Appl. Opt.,11, 135–141.

  • 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 observation. Science,274, 1170–1173.

  • Wang, W.-C., Z. Zeng, and T. R. Karl, 1990: Urban heat islands in China. Geophys. Res. Lett.,17, 2377–2380.

  • ——, M. P. Dudek, X.-Z. Liang, and J. T. Kiehl, 1991: Inadequacy of effective CO2 as a proxy in simulating the greenhouse effect of other radiatively active gases. Nature,350, 573–577.

Fig. 1.
Fig. 1.

Aspects of the SSU1 response function: (a) Wide-band (dashed line) and pressure-modulated (solid line) transmission filters; (b) weighting function showing effects of cell pressure, p0, changes. Mean cell pressure is given in Table 1, maximum/minimum values are derived by adding/subtracting the associated error.

Citation: Journal of Climate 12, 11; 10.1175/1520-0442(1999)012<3197:CVATIS>2.0.CO;2

Fig. 2.
Fig. 2.

(a) Monthly mean measured SSU1 cell pressure. Vertical dotted lines indicate where a switch to a new satellite has been made. (b) Simulated effects of cell pressure, p0, history on SSU1 TB for two atmospheric cases. The effect is given in terms of a ratio between brightness temperatures calculated with p0 varying according to Fig. 1a, and those obtained when p0 was held constant over the measurement period.

Citation: Journal of Climate 12, 11; 10.1175/1520-0442(1999)012<3197:CVATIS>2.0.CO;2

Fig. 3.
Fig. 3.

(a) Time series of SSU1 global monthly mean TB from Jan 1979 to Dec 1994. (b) Power spectrum of (a). Horizontal axis gives the frequency component, f, equal to t/N, where t is the time period, and N is the total number of months in the time series. Hence, for a period of 12 months, f = 0.083.

Citation: Journal of Climate 12, 11; 10.1175/1520-0442(1999)012<3197:CVATIS>2.0.CO;2

Fig. 4.
Fig. 4.

Climatological monthly mean SSU1 TB fields constructed using the complete 1979–94 time series: (a) Jan; (b) Jul.

Citation: Journal of Climate 12, 11; 10.1175/1520-0442(1999)012<3197:CVATIS>2.0.CO;2

Fig. 5.
Fig. 5.

(a) Temporal progression of monthly mean zonally averaged SSU1 brightness temperatures. (b) Deseasonalized monthly mean globally averaged SSU1 TB anomalies. (c) As in (b) but using data adjusted for the NOAA-9 radiometric error. Deseasonalized anomalies are constructed at each grid point by subtracting the relevant monthly climatological field from each corresponding monthly mean field over the time series. These gridpoint values are then averaged over the requisite spatial scale.

Citation: Journal of Climate 12, 11; 10.1175/1520-0442(1999)012<3197:CVATIS>2.0.CO;2

Fig. 6.
Fig. 6.

(a) Temporal progression of monthly mean zonally averaged SSU1 TB anomalies (original data). Vertical lines indicate operative period of NOAA-9. (b) As in (a) using data adjusted for the NOAA-9 radiometric error. Vertical lines indicate times corresponding to El Chichón (EC) and Mount Pinatubo (MP) eruptions. (c) As in (a) for 20°N–20°S. (d) As in (b) for 20°N–20°S.

Citation: Journal of Climate 12, 11; 10.1175/1520-0442(1999)012<3197:CVATIS>2.0.CO;2

Fig. 7.
Fig. 7.

Time series of SSU1 annual mean globally averaged TB. (a) Original data. (b) Adjusted for NOAA-9 radiometric error. (c) Adjusted for NOAA-9 radiometric error and cell pressure effects.

Citation: Journal of Climate 12, 11; 10.1175/1520-0442(1999)012<3197:CVATIS>2.0.CO;2

Fig. 8.
Fig. 8.

Annual mean spatially resolved SSU1 TB fields for selected years: (a) 1986, (b) 1987, (c) 1992, (d) 1993.

Citation: Journal of Climate 12, 11; 10.1175/1520-0442(1999)012<3197:CVATIS>2.0.CO;2

Fig. 9.
Fig. 9.

(a) Temporal standard deviations of annual mean SSU1 TB data in K. (b) Linear trends in SSU1 TB fields from 1979 to 1994 obtained using a least squares fit. Shaded areas indicate trends of greater than 90% significance.

Citation: Journal of Climate 12, 11; 10.1175/1520-0442(1999)012<3197:CVATIS>2.0.CO;2

Fig. 10.
Fig. 10.

SSU1 TB fields simulated by GENLN2 using HADCM2 decadal mean atmospheres as input: (a) M1865, (b) M2045, (c) M2045 − M1865.

Citation: Journal of Climate 12, 11; 10.1175/1520-0442(1999)012<3197:CVATIS>2.0.CO;2

Fig. 11.
Fig. 11.

Global mean temperature profiles for 1865 and 2045 from HADCM2. Dashed lines indicate the level corresponding to the peak of the SSU1 weighting function in each case.

Citation: Journal of Climate 12, 11; 10.1175/1520-0442(1999)012<3197:CVATIS>2.0.CO;2

Fig. 12.
Fig. 12.

Temperature at the peak of the SSU1 weighting function: (a) M1865, (b) M2045, (c) M2045 − M1865.

Citation: Journal of Climate 12, 11; 10.1175/1520-0442(1999)012<3197:CVATIS>2.0.CO;2

Fig. 13.
Fig. 13.

Selected annual mean spatially resolved SSU1 TB anomaly fields, ΔD(x, t) in K: (a) 1979, (b) 1984, (c) 1989, (d) 1994.

Citation: Journal of Climate 12, 11; 10.1175/1520-0442(1999)012<3197:CVATIS>2.0.CO;2

Fig. 14.
Fig. 14.

(a) Pattern similarity statistics R(t) and C(t) with corresponding least squares linear trends. (b) Spatial mean [b〈ΔD(t)〉] and pattern anomaly [aR(t)] components of C(t).

Citation: Journal of Climate 12, 11; 10.1175/1520-0442(1999)012<3197:CVATIS>2.0.CO;2

Table 1.

Details of SSU channel 1 parameters governing gas cell response function.

Table 1.
Table 2.

Sensitivity of SSU channel 1 brightness temperature, TB, to gaseous absorbers, spectral resolution, and CO2 line mixing.

Table 2.
Save