The fourth intercomparison is actually one of the earliest conducted by DOE. ICRCCM is a program co-sponsored by DOE, the World Meteorological Organization (WMO), and the International Radiation Commission (IRC). The late Fred Luther gave the best description of the rationale for the program:
"Since the transfer of solar and longwave radiation is the prime physical process that drives the circulation of the atmosphere and its temperature structure, it is natural that an evaluation of the modeling of physical processes important to climate begin with radiation."
(Luther 1984)
The purpose of this international effort was to evaluate and improve solar and longwave calculations used in climate models. The name is partly a misnomer because the comparison has involved many radiation models too detailed for use in a practical climate model.
The first ICRCCM workshop examined a total of 42 separate sets of model calculations. These were intercompared for 37 specified clear-sky control cases. Relatively large discrepancies (10% to 20%) among different models surprised the participants because the prevailing opinion had been that clear-sky longwave radiation was a solved problem. Suspecting that code or input errors might be responsible for the discrepancies, participants reviewed and modified their calculations. As a result, in some cases the range of discrepancies in model results became smaller. However, the main conclusions of the workshop report remained intact. They included a large spread among less detailed models (the kind actually used in climate and weather prediction models). ICRCCM disproved the assumption that the physics of molecular absorption and absorption line data were well understood
is doubled, the median band model sensitivities differ by up to 18% from line-by-line values, while
their spread is an order of magnitude larger.
only and 
only, the spread in results among band models increases considerably compared to the case when all absorbing gases are included;
this indicates that the success in the latter case is partly fortuitous because of the way absorbing bands overlap in the Earth's atmosphere.
A great impediment to improving line-by-line models is that there is no accepted theory for continuum absorption. Varanasi (1988) is pessimistic about the prospects for an accepted theory any time soon. Thus, modelers must rely on empirical formulations based on laboratory measurements. Almost all radiation modelers use the empirical continuum formulation of Roberts et al. (1976). However, new laboratory (Burch and Alt 1984) and field (Cutten 1985; Kneizys et al. 1984) measurements of the continuum show significant disagreements with the Roberts formulation. (It is common for the continuum numbers to change every few years.) Clough changed Roberts' formulation in the latest release of FASCOD, a commonly used high-resolution transmittance model, and developed a treatment of the continuum which is consistent with tabulated line intensities.
However, FASCOD had to be adjusted by using 10% more water vapor than measured to agree with high-resolution interferometer sounder (HIS) spectrometer measurements in the continuum. The HIS also revealed 60% errors in the foreign (air) broadened portion of the FASCOD continuum (Figure 6). The adjustments to the self-broadened water vapor coefficients are within the uncertainty of the radiosonde observations. Nevertheless, the results displayed in Figure 6 illustrate the continuing uncertainty about the accuracy of present-day empirical formulations of the continuum and about how to interpret the measurements on which they are based.
Two major components of the World Climate Research Program, TOGA (WCP-92 1984) and ERBE, have called for radiative flux accuracies of 10 
or better. Existing
radiation models cannot provide that accuracy. Except in unusual situations, typical sensible and latent heat fluxes are no larger than about 200 . The
intermodel radiative flux disagreements are thus a significant fraction of normally observed energy fluxes. In the final analysis, it is these energy fluxes
which control the climate.
These large disagreements among radiation computations do not manifest themselves as different computed climates because most climate models are tuned to give
the "right" answer. This tuning is evident in the climate models' omission of water vapor continuum absorption through the 1970s. While the exact magnitude
and temperature-dependence of water vapor absorption are still outstanding problems, the existence of the absorption is indisputable. Omitting the continuum
causes differences of roughly 80 in net surface longwave flux in the equatorial region, tapering off to zero at 60N and 60S.
Having identified these discrepancies, the ICRCCM participants considered the usefulness of existing data sets for differentiating among the more accurate models. They concluded that the broad-band flux data gathered during typical field programs were not useful for this purpose for the following reasons:
After some 50 years of development, it is still impossible to push broad-band flux radiometer errors below 5%, even in the most capable hands. Broad-band flux measurements are averages of spectral radiance over all angles and wavelengths. When the same averaging is performed on the model results, it becomes almost impossible to trace the reason for the inevitable model-measurement disagreements.
Theoreticians believe their models have outdistanced current field data. They recognize that such data have bearing on other disciplines but consider them useless for doing research on radiation. The estrangement between the needs of modelers and the design of typical field programs is viewed with concern by the DOE.
Satellite observations may be considered as a means of constraining the various models. However, operational satellite radiometers tend to be poorly calibrated or uncalibrated and have been shown by underflights to exhibit drifts of 20 to 30% over just a few years. Atmospheric profiles are rarely taken at the locations of the satellite measurements. ERBE measurements are probably the most accurate and precise to date but, lacking spectral resolution, they are of little assistance in correcting the factors leading to intermodel disagreements. Also, satellite observations do not address the problem of surface energy budget and atmospheric heating and cooling rates.
Many spectrally resolved observations have been made from aircraft and stratospheric balloons, mostly to help better understand atmospheric chemistry. These observations are sometimes of high quality, especially the spectra taken from stratospheric balloons to look for exotic molecules. But they cover only a portion of the spectrum, and profile information is lacking. Downward-looking spectra experience `surface clutter' problems, making it impossible to distinguish effects of varying surface temperature and emissivity from defects in radiative model formulation.
Existing surface observations with reasonable spectral detail and covering most of the longwave spectrum were reported throughout the 1950s, coincident with the development of Fourier transform spectrometers. These spectra were exhibited rather than compared with calculation, because atmospheric profiles were rarely measured concurrently. Highly detailed surface IR spectra are still occasionally taken by astronomers for their own purposes. Some observations are used for the study of the atmosphere (Stokes et al. 1983). There are, however, very few emission spectra of the atmosphere, notable exceptions being the recent observations reported by LaPorte et al. (1988) and Smith et al. (1990).
Laboratory spectroscopic measurements are another potential source of radiative transfer information. There are many practical difficulties associated with this approach to the problem. Primarily, there is a lack of interest in the spectroscopic community to study gases whose spectra are thought to be fairly well known. As a result, atmospheric radiation projects have suffered the loss of valuable research potential. However, laboratory spectroscopy alone cannot resolve all the relevant difficulties associated with the radiation codes.
Atmospheric conditions, especially cold temperatures and/or high humidities, are difficult if not impossible to reproduce in the laboratory. This is particularly true in the vital area of continuum absorption. Studies at relative humidities over 70% are a persistent problem. This is the threshold for condensation on hygroscopic dust particles and therefore for fogging of optical elements. Furthermore, spectroscopists have reached an impasse in the area of line wings and the continuum that prevents progress in line-by-line modeling (Varanasi 1988).
Thus, conventional data sets, while interesting for some questions, cannot decisively resolve the widespread model disagreements. Climate modelers have reached a critical impasse. Their models disagree at a level significant for climate, yet no absolute standard exists to arbitrate the disagreements. Only empirical information can resolve the difficulties, yet existing data are inadequate for the task. Recognizing these facts, the radiation transfer community repeatedly recommended a sophisticated observational strategy:
"A dedicated field measurement program is recommended for the purpose of obtaining accurate spectral radiances rather than integrated fluxes as a basis for evaluating model performance."
(Luther 1984)
Following the 1988 ICRCCM workshop, the ICRCCM recommended, and the IRC and the WCRP endorsed, a second phase of the project. The primary purpose of the second phase would be the validation of radiation models through comparison with observations. The objective was specifically to:
"Determine the requirements for real in situ data for validation of high spectral resolution models and other radiative transfer computations and explore ways of obtaining these data by either a specific dedicated measurement programme or by appropriate enhancement of other experimental activities, such as may be part of ISLSCP and ISCCP regional experiments."
The radiation instruments and atmospheric profiling technology necessary to address the problems raised by ICRCCM are now available. Furthermore, the experimental framework necessary to address those problems is compatible with that necessary to address the GCM cloud-radiation, prediction and parameterization problems. Thus, ARM has evolved to a program directed at the improvement of GCM radiation and cloud models. The details concerning the requirements to address these problems are discussed in the Experimental Design Section, and information concerning the necessary instrumentation is discussed in the Program Requirements Section of this report.
