1. Introduction

The detection and attribution of climate change is an important goal. A number of papers have been published that seek to use the long record of surface or near-surface temperature measurements for this purpose (e.g., Santer et al. 1995; Tett et al. 1999). Alternative parameters, such as the vertical temperature structure, sea level pressure, rainfall, and so on have also been investigated as suitable variables of climate change (e.g., Barnett and Schlessinger 1987; Santer et al. 1996). Other workers (Kiehl 1983; Charlock 1984; Goody et al. 1996; Slingo and Webb 1997; Harries et al. 1998) have presented a variety of studies of the information content of the earth's outgoing longwave emission spectrum. This is the flux of energy between about 3- and 100-μm wavelength (3300 and 100 cm⁻¹), which cools the earth to maintain energy balance with the incoming shortwave radiation from the sun, absorbed by the planet. The characteristics of this spectrum are described in a number of publications (e.g., Goody and Yung 1989).

In a recent paper, (Harries et al. 2001, hereafter H01), we presented the results of a study in which we carefully intercalibrated and compared IR spectra obtained from the Nimbus-4 spacecraft in 1970, by the Infrared Interferometer Spectrometer (IRIS), with spectra from the Interferometric Monitor of Greenhouse Gases (IMG), flown in 1997 on the Advanced Earth Observing Satellite 1 (ADEOS-1) spacecraft. This work showed that, over large regions of the earth, the clear-sky emission spectrum showed detailed changes, which agreed well with theoretical expectations based on the known changes of greenhouse gases such as CO₂, CH₄, O₃, and chlorofluorocarbons 11 and 12. In this way it has been experimentally confirmed for the first time that the greenhouse forcing of the earth has, indeed, been changed through the growth of greenhouse gases.

Climate change is much more than just a change in this greenhouse forcing. It also involves the response of the whole system, as represented in the radiation field by the radiation feedbacks due to a variety of causes. The most important atmospheric radiative feedback processes are due to water vapor and clouds (Houghton et al. 2001). Since completing our work on the clear-sky spectrum, we have studied the possibility that changes in the earth's outgoing spectrum may also contain information about changes in these feedback processes. It would be a major step forward in confirming our understanding of climate change if such changes in the dominant feedback processes could be observed.

However, a major difficulty in using the IR spectra measured from space for this purpose is sampling in both time and space. When sampling the cloud-free atmosphere, evidence indicates that the spatial variability beneath the satellite track may usually be regarded as small enough that spatial sampling differences between different instruments are not a significant source of uncertainty. Likewise, temporal variability, as long as major effects like the diurnal cycle are taken into account, need not introduce large uncertainties. However, when observing cloudy skies, we can expect larger uncertainties from sampling differences. This arises because of the high-contrast spatial structure, at all scales, which can exist in the cloud field and because of their transient nature. Under broken clouds the surface variability will also contribute to the problem with the precise location of the clouds, in addition to their properties, determining the observed brightness temperature.

In this and a companion paper (Brindley and Harries 2003, hereafter BH), we therefore present the results of studies that seek to make quantitative estimates of the uncertainty associated with the change in the all-sky spectral outgoing longwave radiation (OLR) derived from the IRIS and IMG measurements. In BH, we investigated the uncertainty arising from the differing instrumental spatial fields of view. In the present paper, we present the results of a study in which we examine the uncertainty caused by both spatial and temporal sampling differences between the two sensors. We find that, indeed, this uncertainty is significant compared with the change signal we might expect to observe today; but, given a steady development of global climate change, and also given the new generation of spectral sensors being launched into space, we believe that within about a decade, some signals of feedback processes may be detectable.

2. Observations and methodology

Both Nimbus-4 and ADEOS-1 were sun-synchronous satellites, and hence would be expected to sample each point on the earth once every 12 h at the same local time. Work by Salby and Callaghan (1997) and Engelen et al. (2000) has shown that such a sampling pattern can result in systematic biases in atmospheric parameters, even when these are averaged over the timescale of a month. In fact, for the IMG instrument in particular the observing pattern is even more irregular, since the operational schedule of the instrument resulted in measurements that are clustered over a period of a few days per month (Kobayashi 1999).

The approach we adopt to investigate quantitatively the uncertainty that derives from differences in sampling by the two different satellite sensors is as follows. Three-hourly, 0.5° latitude–longitude fields of International Satellite Cloud Climatology Project (ISCCP) 11-μm brightness temperatures (denoted by T_B11), and corresponding clear-sky brightness temperatures (T_B11clr) were obtained (Rossow et al. 1996). These data were taken during two regional campaigns, namely the Tropical Ocean Global Atmosphere Coupled Ocean–Atmosphere Response Experiment (TOGA COARE) and the Fronts and Atlantic Storm Track Experiment (FASTEX). The former took place from November 1992 to February 1993, and was situated over the warm pool region with the aim of better understanding the ocean–atmospheric coupling within the study area and describing the interactions, which might extend its influence to other regions (Webster and Lukas 1992). The latter, primarily designed to improve forecasting of end-of-storm-track cyclogenesis (Joly et al. 1997), occurred during January and February 1997 over the North Atlantic Ocean. While being too coarse to resolve fast subpixel-scale processes (e.g., the development and passage of individual cloud cells), the temporal and spatial resolution of the data should be sufficient to track longer-lived systems and to provide a good representation of the diurnal cycle (Salby and Callaghan 1997). In the TOGA COARE case, 2 months of ISCCP data were extracted (December 1992 and January 1993) in order to obtain a feel for the month-to-month variability in the T_B11 fields. Over the FASTEX region only data for February 1997 were available.

For each available month the true all-sky monthly mean value of ISCCP 11-μm brightness temperature was calculated, assuming perfect sampling, and denoted as T^*_Btrue. In order to be consistent with the IMG and IRIS comparisons reported in H01, each available month of ISCCP data was sampled according to the IMG (IRIS) observing times and locations for April–June 1997 (1970), including only those observations taken over ocean. These subsets of the data were then averaged to produce T^*_BIMG and T^*_BIRIS—the monthly mean all-sky values of brightness temperature that would have been obtained by each instrument. Thus, each of the 3 months of ISCCP data has one T^*_Btrue, but three T^*_BIMG, and three T^*_BIRIS corresponding to the three different sampling patterns (April, May, June). The same procedure was then applied to the T_B,11clr data to generate corresponding clear-sky quantities. In this case the true clear-sky monthly regional mean was denoted T^*_Bclr,true, with the values obtained using the IMG and IRIS sampling characteristics labeled T^*_Bclr,IMG and T^*_Bclr,IRIS, respectively.

For clarity, the results for each campaign are discussed separately, with the findings for the TOGA COARE region presented in section 3, and for the FASTEX area in section 4. The actual observed IMG and IRIS difference spectra for both regions are introduced in section 5 in order to assess whether the signals present in the spectra are sufficient to overcome the sampling bias defined by our simulations. In section 6 we discuss the abilities of the Advanced Infrared Sounder (AIRS) instrument to accurately capture the true atmospheric state given its sampling characteristics. Conclusions to be drawn from this study are then provided in section 7.

3. TOGA COARE results

4. FASTEX results

5. Observed IMG–IRIS spectra

If the change in sampling between IRIS and IMG was not a major issue the values of ΔT_B calculated in the previous sections should always approach zero. Clearly, for several of the cases considered earlier this is not the case, and so it is of interest to investigate whether the observed IMG–IRIS brightness temperature differences exceed the levels of uncertainty quoted. Starting with the TOGA COARE region, the area average all-sky monthly mean brightness temperature (T_B) spectrum for each instrument for April is shown in Fig. 7a. IMG observations have been smoothed to match the spectral resolution of IRIS using the appropriate instrument function. The spectral range is limited to 710–1400 cm⁻¹ because of the high noise levels of IMG at lower wavenumbers, and IRIS at upper values. Since the results quoted in the previous sections are based on 11-μm brightness temperatures, the normalized channel spectral response function is overplotted to provide an indication of its extent. Figure 7b indicates the spectral standard deviations associated with the mean spectra. There are two main reasons that could explain why the IMG standard deviations are higher than the corresponding IRIS values. First, even if both instruments had identical sampling patterns, the smaller field of view of IMG would mean that it would tend to “see” more extreme brightness temperatures than IRIS (see BH). Second, the different instrument sampling patterns may have resulted in IMG viewing a wider range of scenes relative to IRIS purely due to the timing and position of the formers' observations. Figure 7b also provides justification for using T_B11 as an indicator of the error due to sampling considerations since for both instruments close to the maximum variability is seen in this region of the spectrum under all-sky conditions. Results for May and June show the same pattern of behavior.

IMG–IRIS monthly mean brightness temperature differences for April–June averaged over the TOGA COARE region are shown in Fig. 8. The horizontal dashed lines on each plot show the likely error range due to the sampling of the two instruments based on the calculations in section 3a. Looking first at the individual spectra, the signatures of gaseous absorption due to CH₄, O₃, and CO₂ noted by H01 in the clear-sky results are still apparent in all the traces. Note that under all-sky conditions one would expect to see smaller sampling errors associated with these regions than that provided by the T_B11 estimate. All also exhibit a slight slope across the window region (800–1250 cm⁻¹), with differences tending to be more positive at the smaller wavenumbers. Across the window, the magnitude and sign of the differences seen vary greatly over the 3 months, but it is only in April that they can be said to emerge from the envelope of uncertainty associated with sampling. Even in this case, the differences are only of a larger magnitude than the error bounds at wavenumbers greater than 1080 cm⁻¹.

Figure 9 presents data analogous to Fig. 7 but for clear-sky conditions. The removal of cloudy spectra is achieved using the double-threshold method described in H01. Under clear conditions the pattern of variability is such that a sampling error based on the 11-μm channel can only really be considered representative of wavenumbers within the window region from 800–1250 cm⁻¹, excluding the 1040 cm⁻¹ ozone band. At higher (lower) wavenumbers, variations in the distribution of water vapor (mid-upper-tropospheric temperature) play the dominant role and a higher error would be appropriate (Fig. 9b). Monthly mean clear-sky difference spectra are shown in Fig. 10, along with the 11-μm error bound limited to the spectral region to which it applies. In all cases, at wavenumbers between 800 and 1000 cm⁻¹ the window difference signal easily exceeds the error magnitude. In contrast, on the high-wavenumber side of the ozone band (1080–1250 cm⁻¹) the difference is systematically smaller, with values at, or just exceeding the error bounds. As noted in H01 some of this discrepancy could be due to the presence of residual cirrus cloud contamination in the nominally clear IRIS spectra.

Corresponding IMG–IRIS difference spectra for the FASTEX region under all- and clear-sky conditions are shown in Figs. 11 and 12, respectively. Again the appropriate error bounds are indicated on each plot. Similar to the TOGA COARE results, only the April all-sky differences exceed the sampling error (Fig. 11a), although in this case the magnitude of the observed change is substantially larger than the error estimate. This pattern is reversed under clear conditions, where April differences are within the calculated error bounds over the window region (Fig. 12a). For May and June the clear-sky changes over the same spectral range are larger in magnitude and are always above or just equal to the appropriate error estimates (Figs. 12b,c).

6. Sampling ability from AIRS

Clearly, one of the major barriers to extracting unambiguous information from an all-sky comparison of IRIS and IMG data is the limited sampling associated with, in particular, the latter instrument. Given the potential of spectrally resolved data to determine changes in, for example, vertical temperature structure, constituent concentrations, and cloud amount and properties, in this section the ability of the AIRS earth observation instrument to accurately capture the atmospheric state is investigated in the context of its sampling characteristics.

The AIRS instrument is a high-spectral-resolution grating spectrometer, measuring the earth's outgoing radiation in 2378 infrared channels (3.7–15.4 μm) at a spectral resolution (λ/Δλ) of 1200. A visible/near-infrared photometer also provides imaging capability in four channels from 0.4–1.0 μm. Mounted on the Aqua platform, AIRS views the ground through a cross-track rotary scan mirror that provides ±49.5° ground coverage along with views to onboard spectral and radiometric calibration sources every 2.67-s scan cycle. The AIRS infrared spatial resolution is 13.5 km from the nominal orbit height of 705.3 km. Further instrument details can be found in Aumann and Pagano (1994). The Aqua orbit is polar sun synchronous, with an equator crossing time of 1330 in the ascending node and a 16-day repeat cycle.

Appropriate satellite tracks were calculated for April–June 2003 based on the Aqua orbital characteristics, and subsampled every 3 s to account for the AIRS scan cycle. These tracks were then used to sample the ISCCP data for both the TOGA COARE and FASTEX regions under all-sky (clear) conditions to obtain an estimate of the regional monthly mean 11-μm brightness temperature, T^*_BAIRS(T^*_Bclr,AIRS). Table 7 provides details of the level of agreement between the true and AIRS-sampled values in each case. The agreement is of the order ±0.5 K or better for all the scenarios considered, which is a notable improvement on the IRIS and IMG capabilities. Indeed, under all but 3 of the 18 cases investigated, the differences between the true and AIRS-sampled values have a magnitude of less than 0.25 K.

7. Conclusions

The aim of this paper has been to provide some insight into the likely usefulness of IRIS and IMG observations in determining accurately any change in spectrally resolved brightness temperatures seen under clear- and all-sky conditions. The findings suggest that the sampling of IMG in particular is too coarse to provide an accurate representation of the atmospheric state. The exceptions to this general rule occur during April 1997, where the number of samples was increased threefold relative to the other months when observations were made, and under clear conditions in regions exhibiting low spatial and temporal variability in surface temperature.

For IRIS, temporal and spatial sampling is much more systematic and consistent. Over the regions considered, errors in the all-sky fields are a maximum of 1.5 K compared to values obtained given perfect sampling, the majority of these errors being less than 0.75 K. Assuming the signal magnitude remained unaltered, if similar accuracy were available from IMG the observed all-sky differences would exceed the uncertainty associated with sampling in most of the cases considered. Under clear conditions the true value is estimated to within tenths of a degree. The observed clear-sky IMG–IRIS window signal is such that it is only in the Tropics that it consistently exceeds the calculated sampling error. Whether this behavior is likely to be a result of ENSO-induced variability within the climate system will be investigated in a forthcoming publication.

While the IRIS and IMG observations provide a useful resource for analyzing the spectral forcing of the climate system due to enhanced levels of well-mixed greenhouse gases, a quantitative interpretation of the climatic response must be undertaken with some care. Besides issues concerning the limited length of the two datasets, it has been shown here that the sampling of both is only really sufficient to analyze clear-sky differences in terms of their being a true reflection of changes in the atmospheric state. A more all-encompassing comparison against IRIS may become possible with the availability of data from new spectrally resolved instruments such as AIRS. For the majority of regions and months considered here, our results suggest that if a difference between all-sky 11-μm monthly mean brightness temperatures measured by IRIS and AIRS was in excess of ∼1 K it could be considered to be a true signature of change and not a sampling artifact. The question as to whether such a change could be considered to be representative of a climatic response to a given forcing, or was itself a natural fluctuation, would require further detailed analysis.