British Scientists Find Increase in Outgoing "greenhouse" Radiation Waves

Increases in greenhouse forcing inferred from the outgoing longwave radiation spectra of the Earth in 1970 and 1997

JOHN E. HARRIES, HELEN E. BRINDLEY, PRETTY J. SAGOO & RICHARD J. BANTGES

Space and Atmospheric Physics Group, Blackett Laboratory, Imperial College, London SW7 2BW, UK
Correspondence and requests for materials should be addressed to J.E.H. (e-mail: j.harries@ic.ac.uk).

The evolution of the Earth's climate has been extensively studied^{1, 2}, and a strong link between increases in surface temperatures and greenhouse gases has been established^{3, 4}. But this relationship is complicated by several feedback processes—most importantly the hydrological cycle—that are not well understood^5-7. Changes in the Earth's greenhouse effect can be detected from variations in the spectrum of outgoing longwave radiation^8-10, which is a measure of how the Earth cools to space and carries the imprint of the gases that are responsible for the greenhouse effect^11-13. Here we analyse the difference between the spectra of the outgoing longwave radiation of the Earth as measured by orbiting spacecraft in 1970 and 1997. We find differences in the spectra that point to long-term changes in atmospheric CH₄, CO₂ and O₃ as well as CFC-11 and CFC-12. Our results provide direct experimental evidence for a significant increase in the Earth's greenhouse effect that is consistent with concerns over radiative forcing of climate.

Starting in October 1996, the Interferometric Monitor of Greenhouse Gases (IMG) instrument¹⁴, on board the Japanese ADEOS satellite, produced about nine months of global observations of the spectrum of outgoing longwave radiation. Some 27 years earlier, NASA had flown a similar instrument (IRIS—Infrared Interferometric Spectrometer¹⁵) on the Nimbus 4 spacecraft, between April 1970 and January 1971. Details of the performance of the two instruments are given in Table 1. Although these two experiments were on board separate spacecraft almost 30 years apart, the existence of the two data sets allows the examination of the outgoing longwave radiation spectra to detect whether any significant change had occurred between the two measurements. Comparing averaged spectra for the same region, allowing for the different spatial and spectral resolutions of the two instruments, reveals that they are extremely similar and highly reproducible. Nevertheless, on closer inspection, informative differences are observed.

To illustrate this, Fig. 1a shows an averaged IRIS cloud-cleared brightness temperature spectrum superposed on an averaged IMG cloud-cleared spectrum, for the same three-month period (April–June) and a central Pacific area (latitude 10° N–10° S, longitude 130° W–180° W) for the range 710–1,400 cm^-1. The spectral resolution of the IMG data has been degraded to match that of IRIS, instrument field-of-view effects have been taken into account, and land/island areas have been masked out. For cloud clearing, a two-step approach was used¹⁶. Figure 1b shows three curves: upper, the difference IMG-IRIS taken directly from Fig. 1a; middle, a theoretical difference spectrum, corresponding to conditions for the central Pacific; and lower, the observed difference spectrum for a 'near-global' case (60° N–60° S) for the same period, for comparison. The curves in Fig. 1b are displaced by 5 K from one another. Figure 1c shows the component of the simulated spectrum that includes only the effect of trace-gas changes between 1970 and 1997 (omitting temperature and humidity changes), to aid interpretation.

The agreement between the upper and middle curves of Fig. 1b is very good. This result, given that the IMG and IRIS data were recorded using two separate spectrometers, 27 years apart, and that the simulation is completely independent of IRIS/IMG observations, gives confidence in the quality of the data. The internal consistency of the data is further illustrated by the inclusion of the bottom, quasi-global result. Despite the fact that the averages in the top and bottom curves contain very different numbers of spectra, the consistency between these results is notable: the major features of the observed difference spectra appear consistently in all the difference spectra we have studied.

Our interpretation of Fig. 1 is as follows. We consider first the sharp spectral features. A negative-going brightness

temperature difference is observed on the edge of the CO_{2 2} band, between 710 and 740 cm^-1, in accord with the known increase in atmospheric CO₂ concentrations between 1970 and 1997¹. The O₃ band centred at about 1,060 cm^-1 also shows a negative-going difference from the background window signal, which can be attributed to the known changes in ozone¹⁷ and in temperature¹⁸. A strong, negative Q-branch is observed at 1,304 cm^-1 in the CH₄ band, due mainly to increases in tropospheric CH₄ concentrations in the period between the observations, which causes emission from higher, colder layers of the troposphere. Negative-going lines due to ₂-band H₂O absorption are seen between 1,200 and 1,400 cm^-1. There is also evidence of weak features due to CO₂, CFC-11 and CFC-12 in the 700–1,000 cm^-1 range.

The simulations shown in Fig. 1b and c were calculated as follows. Profiles of atmospheric temperature and water vapour were extracted covering the same region and three-month time periods from the NCEP (National Centers for Environmental Prediction, Washington) reanalysis project¹⁹. Stratospheric ozone changes were estimated using measured trends²⁰ extrapolated back to 1970, whereas tropospheric ozone changes were calculated using a three-dimensional chemical transport model, forced by realistic emission scenarios²¹. Remaining gaseous concentrations were taken from the relevant IPCC values for CO₂, CH₄, N₂O, CFC-11 and CFC-12, in 1970 and 1997. The MODTRAN3 code²² was used to calculate the expected radiance spectra in 1970 and 1997.

First, we have examined the central issue of the accuracy and precision of the two data sets, and whether our analysis is justified on this count. We believe it is, for the following reasons. (1) We have reported above the agreement between the observed difference spectra and simulated spectra, which are calculated quite independently from basic knowledge of the atmospheric state. (2) We have derived difference spectra over a wide variety of regions and times (including the east and west Pacific, Atlantic, and Indian oceans) and see consistency in the principal absorption features. (3) Random error is reduced by averaging: in the extreme case in which all the error in Table 1 is assumed to be random, we obtain a value of 0.058 K for the central Pacific difference spectra. (4) Absolute accuracy is potentially a more serious issue: if all the error quoted in Table 1 were systematic, this would indicate a maximum absolute peak–peak error for the combined data of about 0.45 K at the centre of the IRIS/IMG pass bands, increasing to about 0.75 K at the edges. Such errors of absolute calibration vary slowly with wavenumber, and could produce small 'ghosts' at the positions of the observed spectral features, though these are very unlikely to be significant. Slowly varying systematic errors²³ may also contribute to the differential window signal, but only at this same level of <0.45 K. (5) Normalization of each averaged spectrum to the intensity at one selected wavenumber indicates no significant multiplicative error between IMG and IRIS above the 0.5 K level. (6) Examination of the data shows that neither spatial nor temporal sampling is seriously biased. We conclude that the main features of the observed spectra cannot be accounted for by instrument errors, but that the absolute calibration of the difference spectra might be in error by up to about 0.5 to 0.75 K peak–peak.

Using available aerosol data²⁴, we have shown for typical conditions that aerosol absorption is unlikely to be a significant source of error (0.1 K). However, recent work²⁵ has shown that ice cloud, particularly if composed of small crystals, does exhibit stronger absorption in the 800–1,000 cm^-1 than the 1,100–1,200 cm^-1 window. It is quite possible that small residual amounts of ice cloud absorption remain in both sets of data. Owing to the larger field of view, the IRIS spectra have a much higher probability of being contaminated than their IMG counterparts. The observed 1 K or so enhancement of the 800–1,000 cm^-1 difference signal would be consistent with this, and could also arise from a change in the mean cirrus microphysical properties. We cannot separate these two effects, but we do conclude that the observed window difference spectra strongly indicate an effect involving residual small ice crystal effects, incompletely cleared from the data. R.J.B. has performed further calculations, following on earlier work²⁶, which confirm that window difference spectra of the magnitude observed can easily arise from small changes in amount, size or shape of small ice crystals: these studies also indicate that the difference spectrum should be larger below about 920 cm^-1, which is consistent with the observed data, especially the global case (Fig. 1b). Further work on these and other cloud effects in the data will be performed separately: for the present, we believe we have demonstrated a sufficient understanding of the observations to give confidence to the principal findings of this work regarding changes in radiative forcing due to CH₄, CO₂, O₃ and the chlorofluorocarbons.

Before any further analysis, the IMG data were smoothed to match the IRIS resolution. In addition, the wavenumber scale in the two instruments is slightly different, owing to differing solid angles within the instruments²⁷. An empirically derived multiplicative factor of 0.9995 was applied to the IRIS wavenumber scale. Our work indicates that this process needs to be performed extremely accurately in order to avoid artefacts appearing in the difference spectrum.

Cloud clearing was achieved in two steps¹⁶. First, for each spectrum, the brightness temperature at 1,126.6 cm^-1 surface to top of atmosphere transmittance 0.94) was compared to the underlying sea surface temperature taken from NCEP for the same location and time. If the difference exceeded a set threshold, the spectrum was rejected as being cloud-contaminated. The threshold for both IMG and IRIS was set at 6 K. This value was found by testing the variation of the standard deviation of the spectra with the threshold value.

The remaining spectra were subjected to a second filtering, to try to remove any residual spectral effects of cirrus clouds. The brightness temperature difference between two wavenumbers (909.8 and 1,250.4 cm^-1) was compared with a threshold value¹⁶. The threshold value was chosen as before, by noting the temperature difference at which the standard deviation of the set of IRIS, or IMG, spectra fell to a near-constant value. A value of 8 K was used for IRIS spectra and 7 K for IMG. More severe cloud-clearing thresholds led to either all IRIS spectra being rejected as contaminated, or insufficient numbers of IMG spectra being retained to give the requisite spatial and temporal coverage.

Numbers of retained spectra in Fig. 1 The numbers of spectra retained and used in averaging to produce Fig. 1, after cloud clearance, were as follows. Global: IMG, 4,061; IRIS, 529. Central Pacific: IMG, 213; IRIS, 28.

Random and absolute errors The total uncertainty of IRIS and IMG are quoted in Table 1. To examine the random and absolute components of this, we have taken the quoted errors to represent either the maximum random error (for example, caused by noise in the detector or electronics system) or the maximum systematic error (for example, multiplicative or additive error, such as absolute errors in the transmission or reflectivity of optical components such as filters and mirrors, or the temperature or emissivity of the blackbody targets used to calibrate the instruments, or an offset due to vignetting in the field of view). Such errors would typically be slowly varying with wavenumber.

For the random error, we have divided the errors quoted in Table 1 by the square root of the number of spectra left in each cloud-cleared average, and combined IMG and IRIS errors in a root-mean-square sense. The error in the difference spectrum of Fig. 1 amounts to 0.058 K. For systematic error, from Table 1 the maximum peak–peak error that could arise, and which would not be reduced by averaging, would be 0.45 K in the band centre and 0.75 K towards the edge of the band.

References

Acknowledgements. We thank L. Chen and R. Goody for providing access to IRIS data, and S. Taguchi and H. Kobayashi for providing access to IMG data.