399Bulletin of the American Meteorological Society1. IntroductionThe atmospheric science community includes alarge and energetic group of researchers who deviseand carry out measurements in the atmosphere. Thiswork involves instrument development, algorithmdevelopment, data collection, data reduction, and dataanalysis. The data by themselves are just numbers. Tomake physical sense of the data, some sort of modelis needed. This might be a qualitative conceptualmodel, or it might be an analytical theory, or it mighttake the form of a computer program.Accordingly, a community of modelers is hard atwork developing models, performing calculations, andanalyzing the results by comparison with data. Themodels by themselves are just “stories” about the at-mosphere. In making up these stories, however, mod-elers must strive to satisfy a very special and ratherdaunting requirement: the stories must be true, as faras we can tell; in other words, the models must beconsistent with all of the relevant measurements.This essay deals with the relationships betweenmodels and measurements in the atmospheric sci-ences. In view of our own backgrounds and interests,we emphasize cloud and dynamical processes. Thetone of the essay is playful, and we make no attemptto be rigorous or comprehensive, but the subject isimportant and is not often explicitly discussed.2. MeasurementsIt is useful to distinguish two kinds of measurements.a. Measurements of universal functionsSometimes we measure a number or function thatis believed, on theoretical grounds, to be universal, sothat in principle it should only have to be measuredonce. The surface-layer similarity functions of Moninand Obukhov (e.g., Monin and Yaglom 1971) areamong the few good examples of this type of mea-surement in atmospheric science. Note that these func-tions describe certain statistics of the turbulent surfacelayer rather than the outcomes of measurements madeat single points in space or single instants in time.Additional examples of universal functions and/orconstants, which are of interest to a community thatencompasses but goes far beyond atmospheric sci-ence, include the physical properties of water and airMeasurements, Models,and Hypotheses in theAtmospheric SciencesDavid A. Randall* and Bruce A. Wielicki+ABSTRACTMeasurements in atmospheric science sometimes determine universal functions, but more commonly data are col-lected in the form of case studies. Models are conceptual constructs that can be used to make predictions about theoutcomes of measurements. Hypotheses can be expressed in terms of model results, and the best use of measure-ments is to falsify such hypotheses. Tuning of models should be avoided because it interferes with falsification.Comparison of models with data would be easier if the minimum data requirements for testing some types of modelscould be standardized.*Department of Atmospheric Science, Colorado State University,Fort Collins, Colorado.+Radiation Sciences Branch, Atmospheric Sciences Division,NASA Langley Research Center, Hampton, Virginia.Corresponding author address: David A. Randall, Departmentof Atmospheric Science, Colorado State University, Fort Collins,CO 80523.E-mail: [email protected] final form 30 August 1996.©1997 American Meteorological Society400Vol. 78, No. 3, March 1997(such as the saturation vapor pressure of water vaporas a function of temperature) and the optical proper-ties of water vapor, carbon dioxide, and ozone.b. Case studiesMost of the time, we make measurements in a case-study mode, accumulating data on particular se-quences of atmospheric events, which can becompared in detail with simulations of the samecases. Operational weather forecasting makes use ofthis type of measurement, and most atmospheric mea-surement programs such as the First ISCCP (Interna-tional Satellite Cloud Climatology Project) RegionalExperiment (FIRE), the Atmospheric Radiation Mea-surements Program (ARM), and the Tropical Ocean–Global Atmosphere Coupled Ocean–AtmosphereExperiment (TOGA COARE) also fit into this cat-egory. The simplest application of case-study mea-surements is to quantitatively document and/ordescribe what nature is doing. This is particularly in-teresting when the measurements deal with physicalsituations or physical variables that lie outside therange of previous measurements. For example, overthe past several decades satellites have provided dataon the global distribution of cloudiness (e.g., Schifferand Rossow 1983) and the effects of clouds on theearth’s radiation budget (e.g., Ramanathan et al.1989). Of course, efforts are being made to under-stand, model, and interpret these data, but the first or-der of business has been simply to piece together aquantitative description of the geographical, seasonal,and interannual variations of cloudiness and the ef-fects of clouds on the longwave and shortwave radia-tion at the top of the atmosphere.Occasionally, data that have been collected in acase-study mode can definitively show whether an im-portant idea is right or wrong. This rarely happens inthe atmospheric sciences, but once in a while it does oc-cur. For example, Matsuno (1966) theoretically predictedthe existence in the atmosphere of mixed Rossby grav-ity waves and Kelvin waves. A short time later, bothtypes of waves were discovered in the data (Yanai andMaruyama 1966; Wallace and Kousky 1968).3. ModelsA model essentially embodies a theory; this is trueeven for numerical models. A model (or a theory) pro-vides a basis for making predictions about the out-comes of measurements. Atmospheric models can beconceptually grouped in various ways; one such clas-sification follows.a. Elementary modelsThe disciplines of fluid dynamics, radiative trans-fer, atmospheric chemistry, and cloud microphysicsall make use of models that are essentially direct ap-plications of basic physical principles to phenomenathat occur in the atmosphere. Many of these “elemen-tary” models were developed under the banners ofphysics and chemistry, but some are products of theatmospheric science community. Elementary modelstend to deal with microscale phenomena (e.g., theevolution of individual cloud droplets suspended inor falling through the air, or the optical properties ofice crystals), so that their direct application to practi-cal atmospheric problems is usually thwarted by thesheer size and complexity of the atmosphere. Becauseof their generality, elementary models often