Richard C. J. Somerville
Catherine Gautier
Co Chairs
Cloud Dynamics and Microphysics
For many years, virtually all general circulation models (GCMs) treatments of clouds were based on simple algorithms relating cloud amount to relative humidity. Such parameterizations usually produced positive global-average cloud-radiation feedbacks in numerical experiments simulating greenhouse-induced climate change. For example, in a typical integration performed with a GCM developed a decade or two ago, a climate warming due to increased atmospheric carbon dioxide concentrations would lead to increased average cloud heights and/or decreased average cloud amounts. It is easy to understand qualitatively why such feedbacks were positive. First, higher clouds are colder and so less effective infrared emitters, and they generally have lower albedos than lower clouds, so the cloud-height feedback was positive (i. e., the change in clouds produced by the warming tended to amplify the warming). Second, average model clouds, like average real clouds, contribute more strongly to the planetary albedo than to the planetary greenhouse effect (the shortwave cloud forcing is larger than the longwave cloud forcing by about 20 Watts per square meter (Wm-2). Hence, a reduction in cloud amount reduces the shortwave effect more than the longwave effect of clouds. Thus, the cloud amount feedback is also positive.
Climate models are now more numerous and more complicated, however, and model responses to increased greenhouse gas concentrations are more varied. GCMs today attempt to take into account a broader range of physical processes involved in cloud- radiation feedbacks. The climate modeling community now realizes clearly that cloud feedback processes are not limited to macrophysical cloud properties, such as cloud amount and cloud altitude. In recent years, many GCMs have begun to include cloud parameterizations which include explicit treatments of cloud physics. Several talks were concerned with the connections between climate and the microphysical aspects of clouds.
K.-N. Liou used a radiation model, together with theoretical and observational estimates of the temperature dependence of cirrus ice crystal size distribution and ice water content, to study feedback processes associated with ice microphysics. In a one-dimensional radiative-convective model, he found that the positive longwave feedback dominated the negative shortwave one. P. Norris has analyzed ASTEX data, including liquid water and radiative flux measurements. He estimated optical depths from aircraft liquid water and effective radius measurements and found that a theoretical albedo calculation based on these observations agreed well with direct albedo measurements. Norris is also developing a three-dimensional non-hydrostatic numerical model of a cloud- topped marine stratocumulus boundary layer, based on a code developed to simulate laboratory cellular convection.
K. Stamnes also discussed using a combination of observational estimates of cloud microphysical properties and theoretical calculations of their radiative effects in order to infer the climate sensitivity to variables such as drop size. He points out, among other results, that the infrared properties of clouds are sensitive to cloud scattering, and so clouds ought not to be treated as black bodies. S. Twomey also provided estimates of the partial derivatives which characterize the sensitivity of climate change to factors such as extinction and absorption. These cloud radiative properties are themselves sensitive to cloud microphysical aspects such as droplet concentrations. The sensitivity is strong enough to raise serious questions as to how the predictability of climate might be affected by relatively small changes in microphysical quantities.
S. Warren summarized recent improvements in the well-known surface-based cloud climatology which he and others have developed over a period of years. Among many other refinements, the observational estimates of the diurnal cycle of cloud cover have benefited from the use of a moonlight criterion to distinguish nights on which adequate illumination was available. The current best value for global average cloud cover is 64%.
Atmosphere-Surface Interactions
Recent years have seen renewed interest in the simple question of which physical processes are responsible for the observed large-scale upper bound of about 304K on sea surface temperature (SST). In a sense, this question itself is an indicator of our lack of understanding of fundamental properties of the climate system, especially of the role of clouds. Many other such questions, seemingly simple in form but impossible to answer conclusively, could be posed. For example, why is the global cloud cover about 60%, and why is the planetary albedo about 30%? Because we cannot account theoretically for these observed properties of the present climate, we are at a loss to explain convincingly how they might change in some future climate, such as one modified by increased greenhouse gas concentrations.
R. Grossman used Coupled Ocean-Atmosphere Response Experiment (COARE) and Central Equatorial Pacific Experiment (CEPEX) data, together with simple models, to explore the role of mesoscale convective systems in regulating sea surface temperatures in the equatorial Pacific. He finds that no single simple mechanism can account for the observed limits on SST variability. Instead, a suite of processes appear to be involved, including not only cirrus shading and the super-greenhouse effect, but also evaporation, ocean mixing and sensible heat transfer. S. Sherwood has developed a simple box model with which to explore atmosphere-SST feedbacks. His preliminary results suggest that in regions of deep convection, the dominant physical processes affecting SST variations are shortwave cloud forcing and surface fluxes, but that cloud optical properties or cloud lifetimes must also be involved if either of these processes is to be effective in stabilizing tropical mean SST changes.
E. Smith reported on results from FIFE, an experiment in central Kansas aimed at assessing the ability of GCMs to simulate surface fluxes. He was able to evaluate both a biosphere model and a suite of conventional GCM turbulence closure schemes. The biosphere model showed promise as both a route to improved flux accuracy and a theoretical tool for improving understanding of GCM results. Nevertheless, at their present stage of development, the biosphere models are still much too complex by GCM standards, and there is also a need to solve the problem of a scale mismatch with GCMs. The turbulence closure schemes, by contrast, in general seemed to be too simplistic to be applicable to the diversity of situations which occur in actual atmospheric boundary layers. For example, under unstable conditions, substantial overestimates of sensible heat fluxes occurred with all the tested schemes, in part because the schemes had not been adequately calibrated against data from sources such as FIFE.
Parameterizations
One way to define the awkward term parameterization is simply as an algorithm uniquely relating the statistical effect of small-scale processes on large-scale fields, with the critical restriction that the algorithm must be an explicit function of the large-scale fields themselves. The common GCM expedient of making cloud amount dependent on relative humidity illustrates the nature of the parameterization problem. Relative humidity is calculable as an explicitly predicted model variable on the GCM grid scale. Cloud amount has substantial subgrid variability, however, and there is no obvious way to relate cloud amount to relative humidity based on first principles. In general, a sufficiently moist but subsaturated GCM grid volume will contain some clouds, and a saturated one will presumably be overcast, but there is no evident route to specifying a universal and deterministic relationship between cloud cover and relative humidity.
Not only are we uncertain how much of the observed variability of clouds can be related to relative humidity, we are also unable to say with any confidence whether other large-scale variables, such as vertical velocity, need to be invoked. Indeed, the fundamental question of parameterizability, the determination of the extent to which parameterization is possible in principle, is unanswered. Thus, a great variety of ad-hoc formulas have been devised, justifiable only empirically to the extent that they are justifiable at all.
A conspicuous feature of existing parameterizations is that there is a noticeable similarity between the approaches taken by the different GCM groups. In particular, the cloud parameterizations developed by the various groups tend to have many properties in common. For example, virtually all current parameterizations of cloud amount attempt to relate gridpoint quantities, such as cloud cover, to the GCM variables at that gridpoint alone, neglecting large-scale structure. This practice of treating each gridpoint independently sometimes leads to undesirable results in the form of substantial horizontal gradients on the smallest resolvable GCM spatial scales. In addition, modern developments, such as the advances made in fractal and multifractal representations of variability, have not yet found their way into common practice in GCMs.
Many of the participants in this session are actively involved in one form or another of parameterization research, and there was extensive discussion of the development and validation of a wide diversity of parameterizations. The participants themselves represented a broad variety of backgrounds and perspectives. Thus, the presentations ranged from observational tests of existing algorithms to theoretical treatments of novel proposed approaches.
H. Hanson described an attempt to parameterize shortwave transmittance through clouds, using nondeterministic characterizations of cloud populations, together with an attempt to take the distribution of all three phases of water into account in a unified manner. Harshvardhan discussed the use of satellite remote sensing data. Using Landsat and ISCCP measurements, he finds that cloud liquid water path (LWP) is almost invariant with cloud fraction over a wide range of cloud fractions and pixel sizes, so that the mean LWP can be regarded as the average over a population of clouds, each of which has essentially constant LWP. An important implication of this result is that cloud fraction can in principle be inferred from knowledge of the gridpoint average LWP.
J. Kiehl summarized the present state of one especially active research area, that concerned with so-called anomalous absorption of solar radiation in the atmosphere. He reported on GCM experiments in which enhanced absorption had been incorporated by modifying the cloud single scattering albedo. By tuning the top-of-atmosphere solar radiation budget back to the observed values, using LWP and cloud amount as free parameters, he finds that the partitioning of the absorbed solar radiation between surface and atmosphere is changed by an amount equivalent to half the global average latent heat flux. Kiehl emphasized that the physical mechanisms responsible for this anomalous absorption are still unknown.
E. Roeckner described the current state of the radiation budget simulated by the ECHAM atmospheric GCM. This model is based on the ECMWF global numerical weather prediction model, to which modified physical parameterizations have been added. The ECHAM model has been extensively tuned to match ERBE observations of top-of-atmosphere radiation budget quantities. The global longwave budget is well simulated, but there still are noteworthy areas of unrealistic shortwave forcing, associated with failure to realistically simulate certain types of cloud, such as marine stratus off the west coasts of North and South America.
B. Soden discussed a cirrus parameterization scheme in which cirrus ice water path (IWP) is diagnosed as a function of temperature, pressure, vertical velocity and lapse rate. These parameters are in turn obtained from ECMWF analyses of conventional meteorological data, while ISCCP retrievals provide information on the occurrence of cirrus and on cirrus optical depth. Among other results, he finds that monthly-average cirrus occurrence can be predicted from relative humidity, provided that corrected relative humidity fields, in which satellite data supplement the ECMWF analyses, are used. On a daily basis, however, cloud cover and relative humidity are not well correlated.
R. Somerville described the use of a single-column diagnostic model and a GCM in validating cloud-radiation parameterizations against ARM observational data. The single-column model accurately mimics one column of a GCM in terms of physical parameterizations, but it is forced and constrained with horizontal flux convergences from observational data. Products of the model, such as net surface solar irradiance, which are sensitive to cloud occurrence and cloud radiative properties, can be compared with observational data to test the parameterizations. He noted that the corresponding GCM experiments with the NCAR CCM2 model, in which a liquid water budget parameterization was tested in an inverse climate change experiment driven by SST perturbations, led to a strong temperature increase in the upper troposphere. This phenomenon could be traced to the large vertical heat transport produced by the CCM2 mass flux convection scheme, illustrating that other model-dependent properties could strongly affect the behavior of a given cloud parameterization in any specific GCM.
C. Walcek reported on an extensive series of observational tests of the relative humidity dependence of cloud cover. He finds that relative humidity is the best single indicator of the occurrence of cloud. However, it appears that cloud coverage decreases exponentially as relative humidity drops below 100%. Additionally, cloud cover is not zero below a fixed relative humidity threshold, as is often assumed in GCM algorithms. Among many other interesting results, Walcek has determined that the lower planetary boundary layer is the atmospheric region in which cloud cover is most sensitive to relative humidity.
Radiative Transfer Developments and Investigations
One of the most important recent developments in radiative transfer involves three-dimensional radiative transfer computations through complex distributions of liquid water (or ice). New formalisms are slowly emerging to facilitate radiative transfer computations in three dimensions, but a main thrust of research activities is in the use of Monte-Carlo models (a direct simulation of the physical processes involved in radiative transfer in which the path of a photon is described by probability functions) and of approximate radiative transfer methods. W. Wiscombe and W. O'Hirok discussed Monte- Carlo techniques and their application to the calculation of radiation interaction with complex multifractal cloud liquid water distributions whose properties have been derived from observations. F. Evans presented a backward Monte-Carlo approach to estimate photon path length probability distribution. This approach is based on the order of scattering solution of a deterministic system and expresses the radiative response explicitly in terms of the medium optical properties. While a general solution would include all of the paths, the approximation presented by Evans assumes just two successive paths and seems to be accurate when compared to aircraft observations. P. Gabriel discussed two approximation methods to calculate the domain-averaged bulk radiative properties of clouds such as albedo, flux divergence and mean radiance without using cloud fraction as a specifier of cloud inhomogeneity. N. Byrne discussed his stochastic radiative transfer through a mixture of binary material and investigations of classes of solutions with different closure approximations.
Many of the proposed stochastic approaches presented have substantial appeal for the development of new GCM parameterizations of radiation transfer through inhomogeneous clouds. The most important aspects of the spatial variability of clouds for radiative transfer (e.g., photon path probability distributions, multifractal parameters) could be characterized from observations (e.g., cloud probes, millimeter-wave radars, microwave radiometers) and then expressed with a few key parameters that GCMs predict (e.g., cloud fraction). It seems likely that fast radiative transfer methods based on these parameters could be developed to include the effects of cloud inhomogeneity in cloud radiation GCM parameterizations.
Results from a number of studies using one- to three-dimensional radiative transfer models applied to a variety of problems were presented. One topic investigated by several participants was the so- called anomalous absorption of solar radiation by clouds or possibly by other constituents. J. Kiehl investigated it from the GCM point of view, addressing questions of absorption sensitivity to microphysical parameters, particularly the single scattering albedo (in terms of co- albedo), hypothesizing that it could differ from those computed by Mie scattering theory if cloud droplets were composed of mixtures of water and absorbing medium (aerosols). Other possible mechanisms suggested for cloud anomalous absorption include vapor-droplet overlap, finite cloud effects and continuum absorption. Kiehl also presented implications for GCM simulations of increased atmospheric absorption. C. Gautier performed a series of studies of total column atmospheric absorption of solar radiation due to the presence of cloud, attempting to assess the sensitivity of this absorption to micro- and macrophysical cloud parameters, as well as to atmospheric and surface parameters. She found that the absorption sensitivity was largest to cloud effective radius (and cloud phase), and also to cloud altitude. This suggests that cirrus and stratus, for instance, have very different effects on the absorption of solar radiation in the atmosphere.
K. Stamnes presented results from observational and radiative transfer modeling studies on atmospheric absorption of solar radiation in high latitudes, particularly over highly reflective snow and ice surfaces. He found that the absorption was harder to characterize with conventional approaches under these conditions. W. O'Hirok studied the role of cloud inhomogeneity on total column atmospheric bulk absorption of solar radiation and found that larger absorption could be expected with similar cloud microphysical properties for inhomogeneous clouds than for homogeneous (one- dimensional) clouds.
Two studies addressed climate sensitivity and feedback issues with highly detailed radiative transfer models. K.-N. Liou used a radiation model that includes the delta-four-stream approximation for radiative transfer in nonhomogeneous atmospheres and ice clouds to investigate the impact of cloud microphysics on climate. His preliminary results indicate that a net positive temperature- emissivity feedback dominates the net negative temperature-albedo feedback. K. Stamnes used a radiative convective model with an accurate treatment of radiative transfer including clouds to study the climate sensitivity to changes in mean drop size and optical thickness. He also used a radiative transfer model for the coupled atmosphere/sea ice/ocean system to study the partitioning of radiative energy between the three strata, and discussed the potential for testing such a model in terms of planned experiments in the Arctic.
Finally, S. Lovejoy and D. Schertzer addressed the issue of radiative transfer through mono- and multifractal clouds. When monofractal clouds occupy only a fractal subset of the space, two fundamental limits exist: the optically thick and optically thin cases. For sufficiently thick clouds, they found that plane parallel predictions could be seriously inaccurate. In the case of multifractal clouds there is a fundamental qualitative difference between clouds with many and few low-density regions. For thick clouds, the near linearity of the photon path moment scaling function allows the renormalization of the optical density to an "equivalent" plane parallel density. These stochastic radiative transfer results can explain the success of first- order Markov approximations which ignore high-order correlations in scatterings.
Multifractal and Stochastic Cloud Analysis and Modeling
Clouds have long been described as plane parallel infinite layers of liquid water or ice despite the fact that such clouds can never be found in nature. However, almost all measurements of cloud liquid water content (LWC) from aircraft show intermittent dry patches embedded within clouds. While different LWC records appear quite distinct in terms of the amount of variability, their power spectra are generally quite similar over a wide range of scales. The analysis of satellite observations of clouds in the visible or infrared spectral bands show similar scaling characteristics.
These observations taken together suggest that, although clouds have a complex structure, that structure can generally be described by probability distributions fully characterized by a small number of parameters (i. e., three) derived from multifractal theory. The observed differences among clouds can be attributed to the anisotropy of the fields in which they are embedded; the anisotropy resulting from the differential vertical stratification of the atmosphere and the earth's rotation, according to D. Schertzer and S. Lovejoy. They have developed a general framework for anisotropic scaling expressed in terms of isotropic self-similar scaling. Fractal and multifractal concepts have been introduced which explain why only three parameters are sufficient to fully describe the statistics of highly intermittent cloud liquid water fields. These parameters represent: (1) how non-conservative the mean field is; (2) how fractal the mean field is; and (3) how multifractal the field is. These three characteristics describe the statistical nature of the field fully. Obviously, other statistical characterizations of cloud distributions can be and have been used which simplify the description even more. Such measures still contain information about the statistics of the distribution that characterize the average distribution of particular realizations.
N. Byrne, for instance, presented models of stochastic radiative transport in which the atmosphere is represented as a mixture of two materials (cloud and clear air), each having unique and definite radiative properties. Statistical approaches to studying clouds are appropriate because cloud observations are inherently statistical, and it is the mean radiative effect of complex 3-dimensional cloud structure that is usually desired.
The multifractal framework, together with simpler examples, has been used to simulate cloud properties (e. g., liquid water distributions) which have characteristics that are similar to those observed. Several methods exist to generate fractal and multifractal LWC fields with both intermittency and non-stationarity. Among these are the bounded cascade model used by W. Wiscombe and colleagues and the fractionally-integrated cascade used by Schertzer and Lovejoy's group. These LWC fields (or other cloud descriptors) can then be used to determine the dependence of cloud radiative properties on the fractal characteristics of the clouds.
Observations
Most of the participants presented data in one form or another and most were users of data, not instrument developers. The one exception was E. Eloranta, who built a High Spectral Resolution Lidar (HSRL) that resolves the high spatial variability of optical depth in clouds. The inversion of the backscatter signal necessary for estimating optical depth (a process which can be complicated by the fact that both molecular and aerosol backscatter signals are present) is facilitated by the HSRL. Its large dynamic range permits the study of aerosols and clouds with optical depths varying from 0.01 to 3. Depolarization measurements which are used to determine the nature of hydrometeors present (i. e., water vs. ice) show that water clouds must almost always be taken into account during cirrus observations. One of the most promising new developments with this instrument is the possibility of measuring effective radius via diffraction peak width and variable field-of-view measurements.
Cloud liquid water content (LWC) measurements from a number of sources were used in many of the results presented. W. Wiscombe discussed the many ways to measure cloud LWC, ranging from traditional aircraft hot wire probes, such as the Johnson-Williams or King probes, to the FSSP which provides cloud droplet spectra to estimate LWC. Measurements from a new optical probe, the Gerber probe, were emphasized by several investigators. The Gerber probe is a promising instrument which can provide high quality LWC and effective radius measurements at much improved rates. Both Wiscombe and F. Evans analyzed data from that instrument taken during the Atlantic Stratocumulus Transition Experiment (ASTEX). Evans used Gerber cloud probe data from ASTEX to compute path probability distributions in clouds. He found good agreement between ASTEX cloud data and his Monte Carlo stochastic radiative transfer results. P. Norris used profiles of cloud liquid water and particle effective droplet size (and derived cloud optical depth) and radiative flux data from aircraft during ASTEX. K.-N. Liou used aircraft measurements of crystal size distribution and ice water content (IWC) in mid-latitude cirrus clouds to show that the ice crystal size distribution and ice water content (IWC) are systematically dependent on temperature. Other cloud liquid water observations from the surface were discussed, such as those from microwave radiometers which provide a good estimate of cloud liquid water path.
Measurements from the Multi-Filter Shadowband Radiometer (MFRSR), which measures direct and diffuse solar radiation deployed in the ARM program, were exploited by both N. Byrne and K. Stamnes. Byrne used them in conjunction with cloud cover estimates from GOES to test the theory which he developed together with F. Malvagi, G. Pomraning and R. Somerville. The theory describes spatial averages at one time, but observationally only one suitable radiometer is available at the Oklahoma ARM site. Therefore, he analyzed a time series from about a dozen half-days, for which MFRSR and GOES data were both available. He showed that the stochas-tic descrip-tion provided a some-what better fit to the data than a fractional cloud cover model, but far more data will be re- quired to settle the issue. Stamnes explored the potential for deriving optical depth from narrowband measurements and mean drop size from bispectral transmittance measurements in terms of the channels available in the MFRSR. The optical depth can be reliably inferred from the 862 nm channel (which is less influenced by atmospheric aerosols than channels at shorter wavelengths), while the mean drop size could be determined from a combination of measurements in the 862 nm channel and a channel centered at 2.2 microns. While the latter channel is currently not available, it would be a valuable addition to narrowband instruments such as the MFRSR.
Conventional surface observations such as sea surface temperature (SST) were reported by themselves but were most often utilized in conjunction with large-scale satellite observations from ERBE or ISCCP. R. Grossman reported on his use of TOGA-COARE measurements of atmospheric and oceanic variables in the western tropical Pacific and found that SST shows a cycle of 3-4 months. However, clearly, a longer observation period would be necessary to confirm this result. He also found indications that, at least some of the time, high SSTs were associated with low wind speeds, and low SSTs followed periods of high wind speed.
High-resolution satellite observations (Landsat data) were used by Harshvardhan but in a degraded form to simulate the various resolutions used by ISCCP. They were then thresholded to classify pixels into clear and cloudy. Liquid water path (LWP) in cloudy pixels was retrieved from these data and found to be essentially invariant to the cloud fraction, at least in the range 0.2 - 0.8 for any pixel resolution. At high cloud fractions (greater than ~0.8) the LWP can be considerably higher than the relatively constant value for lower fractions.
S. Sherwood used monthly averaged ERBE observations of longwave cloud radiative forcing (CRF) during the 1985-89 period, together with Reynolds' analyses of SST during the same period, to investigate the influence of SST on the tropical atmosphere. The SST components which average to zero over a large area were found to be associated with large shifts of CRF (20-25 W/m2 /K) toward higher SSTs. These include the annual component of the seasonal cycle and the time- average distribution within the Pacific warm pool region. Conversely, components involving mean SST changes over large areas (areas that include most deep convective activity) are not associated with significant changes in CRF. These components include El Niñ o, La Niñ a, and the biannual component of the seasonal cycle.
E. Smith used the data from a network of surface flux stations operated during the First ISLSCP Field Experiment (FIFE) for 143 days in 1987 and 21 days in 1989. Annual, intraseasonal, synoptic, and diurnal time scales were found to be the four predominant temporal scales of variability for the fluxes. Cloudiness was found to be the dominant control on flux magnitudes. Precipitation and its resultant effects on soil moisture distribution were found to be the dominant control on evaporative fraction or Bowen ratio. The processes of burn treatment, grazing conditions, topography, and cloudiness on radiative, sensible heat, and moisture fluxes had a much smaller effect than cloudiness, which was found to be the dominant control on the modulation of sensible and latent heat fluxes. For sensible heat, the amplitude of the effect of cloudiness was largest during the senescent period, while for latent heat, it was largest during the growing season. The RMS uncertainties in the measured fluxes were estimated to be approximately 30 Wm-2. When a persistent gradient of soil moisture was observed across the site, a gradient in evaporative fraction and thus a cross-site difference in sensible heating of the boundary layer were found. A resulting secondary boundary layer circulation was established with significant daytime vertical velocities.
D. Sowle discussed measurements from the Unpiloted Aerospace Vehicle (UAV) program. Four new instruments under development were briefly presented: HONER, a novel net flux radiometer; MPIR, a multi-spectral cloud imaging radiometer; CDL , a cloud detection lidar; and UAV-AERI, an IR interferometer. All UAV flights include a standard meteorological package to measure temperature, pressure, and relative humidity. K. Stamnes used broadband surface albedo and solar irradiance measurements from the NOAA/CMDL station in Barrow, Alaska. The seasonal variation in cloud optical thickness at Barrow, Alaska was derived using these data for the period April 1988 through August 1988.
Large scale surface and satellite observations of clouds were used by a number of investigators to study cloud climatology or relate cloud variability to that of other parameters. In particular, C. Walcek employed the U. S. Air Force database (so-called 3DNEPH) of cloud data to investigate its correlation with relative humidity fields produced by assimilating radiosonde observations using a mesoscale meteorology model. He found that, in contrast to current GCM methodologies, clouds exist over a wide range of relative humidities, rather than disappearing below some arbitrarily defined threshold, typically 60-80%, depending on height in the atmosphere. S. Warren used surface weather observations from stations on land and ships in the ocean to obtain the global distribution, at 5° x5° latitude-longitude resolution, of total cloud cover and the average amounts of the different cloud types: cumulus, cumulonimbus, stratus, stratocumulus, nimbostratus, altostratus, altocumulus, cirrus, cirrostratus, cirrocumulus, and fog. Diurnal and seasonal variations were derived, as well as interannual variations and multi-year trends were then estimated. Great emphasis was put on the difficulty of detecting clouds at night due to inadequate illumination of the clouds, and on how to remove this bias by selection of observations made under sufficient moonlight.