The Research Plan

for the

First ISCCP Regional Experiment

FIRE

a U.S. program

in the context of the

International Satellite Cloud Climatology Project

ISCCP

Submitted to the

U.S. National Climate Program Office

November 1983

2

EXECUTIVE SUMMARY

Long recognized as a serious impediment to climate modeling, clouds remain one of the least understood yet highly influential components of the climate system. More reliable estimates of the long range effects of increasing atmospheric carbon dioxide, and increasing the potential of predicting large scale droughts and other interannual fluctuations, require better parameterizations of clouds in numerical models, and better data about cloudiness over the globe. The International Satellite Cloud Climatology ProJect (ISCCP) is obtaining global statistics of cloud parameters. The First ISCCP regional Experiment (FIRE) is a complementary U.S. effort that concentrates on improving the parameterizations, and obtaining the basic knowledge to better interpret satellite images of clouds on regional and smaller scales. An improved basis for interpretation is also important for applications such as compiling regional indices of precipitation or estimating local weather conditions in remote areas.

Given the importance of improving our ability to model clouds and radiation for climate, and given the relatively primitive state of our present understanding of the problem, a broad based research program is required. As planned, FIRE ls a broadly based research effort that concentrates on important issues for which significant progress seems achievable with the resources and tools likely to be available. FIRE is proposed as a 5 year multi-agency program with initial funding in late FY84. It involves a team of scientists drawn from a variety of universities and research laboratories. It facilitates the intensive processing of case study data sets constructed from existing satellite systems with complementary observations from the earth's surface and from aircraft. It concentrates on an area that includes the continental United States and surrounding oceans. Among its goals are the isolation and testing of distinct phases in the modeling and understanding of cloud/radiative processes. Also included are statistical characterizations of the structure and radiative properties of cloud systems as seen from space, and from a few locations at the earth's surface. Within FIRE two types of cloud have been singled out for intensive process experiments: marine stratocumulus, which is a mayor influence on the amount of solar radiation absorbed at the ocean surface, and cirrus, which is a mayor control on the flux of infrared radiation to space. To facilitate study by many independent investigators, FIRE includes a center for the aquisition, editing and assembly of all data relevant to the experiment. Thus FIRE includes five mayor elements:

Validatlon of cloudiness and radiation parameterizations

Studies of marine stratus and stratocumulus

Studies of processes that maintain cirrus

Statistical studies of cloud physical/radiative properties

Data set assembly

The resources required are 1) funding to support the research groups involved, 2) facilities such as aircraft and computers, and 3) data from operational satellites, including limited rescheduling of certain operations.

If successful, FIRE should provide a basis for evaluating and improving the performance of cloud/radiation parameterizations in climate models. It should enable better interpretations of cloud images, and it should validate the cloud retrieval algorithms used in ISCCP.

4

Table of Contents

Page

Executive Summary 2

Foreword 3

1. The Concept of FIRE 7

1.1 The Cloud-Climate Feedback Problem 7

1.2 Why FINE is needed 7

1.3 The Objectives of FIRE 8

1.4 The Proposal for FIRE 8

1.5 The Strategy for FIRE 9

Figure 1.1 12

1.6 Resources Required 11

1.7 The Experiment Team 11

1.8 Schedule 14

Figure 1.2 15

2. Validating Cloudiness and Radiation Parameterizations 16

2.1 Scientific Objectives and Methodology 16

2.2 Clear Sky Radiances and Synoptic Scale Climatology 17

2.3 Validating Models for the Radiative Modifications 17

Due to Clouds

Figure 2.1 18

Figure 2.2 19

2.4 Testing GCM Radiative Flux and Cloud Production Codes 23

2.5 Sampling Strategy 24

2.6 Data Requirements 25

2.7 Schedule and Resources 25

2.8 Expected Results 26

2.9 References 27

3. Studies of the Marine Stratus and Stratocumulus Systems 28

3.1 Introduction 28

3.2 Theoretical Background 30

3.2.1 What determines the factional cloudiness? 30

3.2.3 What is the distribution of radiative cooling? 31

3.2.4 What determines the entrainment rate? 31

3.2.5 What is the role of the mesoscale? 32

3.3 Satellite Observations 33

3.3.1 Statistical characterizations 33

3.3.3 Validation 35

3.4 Alrcraft measurements in the marine boundary layer 35

3.5 Island Studies of Marine Stratus and Stratocumulus 36

Systems

3.6 Radar observations 38

3.7 Summary and Candidate Experiment 39

3.8 Schedule 41

Table 3.1 43

3.9 References 44

Page

4.1 Introductlon 48

4.2 Present Knowledge 48

4.3 Objectives 49

4.4 Case Study Requirements 50

Table 4.1 51

4.5 Timing and Locatlons 54

4.6 Expected Results 54

4.7 Bibliography 56

5. Statistical Studies of Cloud Physical/Radiative Propertles 60

5.1 Purpose of the Statistical Program 60

5.2 Present Status 60

5.3 Strategy 63

5.4 Resource Requirements 65

5.5 Schedule 66

5.6 What will be achieved 66

5.7 References 67

6. Data set assembly

6.1 Archive of raw and case study data sets 69

6.2 Data Processing 69

h.3 The FIRE region 70

Figure 6.1 71

Figure 6.2 72

Table 6.1 74

6.4 Callbration 75

7. Relationship of FIRE to ISCCP 76

7

1. The Concept of FIRE

1.1 The Cloud-Climate Feedback Problem

Many aspects of life on earth are affected by climate, its year to year fluctuations and possible long term changes due to increasing atmospheric carbon dioxide and other manmade causes. With the hope of avoiding a potentially adverse climate resulting from man's activities and also of improving the management of dwindling resources, a hierarchy of climate models has been developed. The hierarchy ranges in complexity from simple, even analytic, formulations based on the radiative energy balance of the earth, to highly sophisticated and computationally intensive general circulation models. These models have been shown to simulate convincingly many elements, such as wind and temperature fields, of the observed climate. Though imperfect they are nevertheless invaluable tools for investigating the causes of interannual variability and the effects of long-term changes such as increasing concentrations of atmospheric carbon dioxide. However, mayor areas of uncertainty are the interactions between atmosphere and ocean, and the feedbacks between cloudiness, radiation and dynamics. This research plan is directed toward the second of these problem areas.

Owing to their strong influence on the earth's radiative energy budget, clouds undoubtedly play a mayor role in shaping the current climate and affecting the way it might change. Simple radiative energy balance calculations indicate for example that as small as a 4X increase in global cloud cover could be sufficient to offset the global warming predicted for the anticipated doubling of CO2. How clouds might be affected by changes due to a perturbation like increased CO2 and how in turn they would affect the response to the perturbation is still a mystery.

1.2 Why FIRE is needed

As already indicated, a mayor limitation on confidence in forecasts of climate changes such as that due to increasing atmospheric carbon dioxide is the current lack of realistic models for large scale cloud systems and their influence on radiation. Together with ISCCP, FIRE should help remove this limitation.

The inadequacy of current models for large scale cloud systems is directly linked to the lack of quantitative and objective observations of cloud systems and their properties. ISCCP and FIRE will provide, in some instances for the first time, observations that are the key to understanding the physical processes that govern cloud systems and that also affect their influence on the earth's radiation and moisture budgets. ISCCP is concentrating on obtaining the global data sets necessary to evaluate overall general circulation model (GCH) performance. FIRE is directed towards testing and improving cloud

8

parameterization schemes used in GCMs and increasing understanding of subgrid scale cloud radiative processes. They are thus complementary approaches to a broad ranging and deep seated problem.

Though primarily directed towards climate modeling, the knowledge gained through ISCCP and particularly through FIKE will also be useful for other purposes. For example, the statistics of cloudiness and its impact on radiation reaching the earth's surface are essential both to solar energy exploitation and to hydrology. Satellite images can also be useful for determining local weather conditions and atmospheric transmissivity in remote regions, but only if procedures for proper interpretation have been established. In addition, improved understand ing of the cloud properties as inferred from satellites should lead to better indices of precipitation. Such data could be used both for synoptic climate analysis and to establish patterns of probable crop yields.

1.3 The Objectives of FIRE

The central objectives of FIRE are to quantify the capabilities of current models for large scale cloud systems and for their effects on radiation, and to obtain the data and understanding necessary to improve these models. The primary source of information will be operational satellites supplemented by selected in situ and ground based observations. A secondary objective is to check and, where possible, improve the interpretation of global statistics on cloud parameters which will be collected by ISCCR.

1.4 The Proposal for FIRE

This plan proposes an integrated research program with 5 major elements:

9

1.5 The strategy for FIRE

Given the need for physically realistic models of clouds and their radiative effects, and given the relatively primitive state of current models and our understanding of the problem, a broad based research program is required. In order to maximize the likelihood of success, the program should concentrate on issues which are essential to our understanding of the problem and for which significant progress seems achievable with the resources and tools that are likely to be available.

The strategy is to isolate and test distinct phases in the modeling of the whole cloud/radiative process, in particular:

Cirrus and marine stratus/stratocumulus systems are selected for special examination because both systems strongly influence the earth's radiative energy budget. In the case of stratus and stratocumulus the effects are on the solar radiation reflected to space and reaching the surface. There is little compensation for these effects provided by thermal emission from these clouds. In the case of cirrus, the effect is on the radiation emitted by the earth, there being only a slight compensation by the solar radiation reflected from these clouds. For both systems significant improvements in our understanding appear to be within reach. In each case, models of key cloud processes exist but have not been adequately compared to observations.

Other cloud systems, like those associated with fronts and with

deep convection, are thought to be less widespread than are the large

scale cirrus and stratus systems. They are also more difficult to treat

physically. Efforts in FIRE on frontal and convective systems will be

confined initially to characterizing statistically their mesoscale

structure and radiative properties. Later FIRE may interact with field

programs now being planned in the U.S. for studying these systems.

The predictive capabilities of cloud and radiation parameterizations will be tested by breaking the process down to the extent possible into discrete parts, each with its own methodology and emphasis. The first part involves testing the directional radiances predicted in a variety of spectral bands, given a description of the

10

cloud field in a form consistent with that used by GCMs. The cloud field

will be inferred from a subset of the directional radiance observations

themselves. The second part concerns the ability of cloudiness

algorithms to reproduce the essentials of the field of cloud parameters,

given the synoptic scale distribution of temperature, humidity and

vertical velocity. The test uses as input, state of the art analyses of

conventional weather observations, and compares the algorithm output

with the cloud parameters inferred from the directional radiances in the

first part. Yet another question, not explicitly addressed in FINE, is

how well climate GCMs can reproduce the regional statistics of the

temperature, humidity and vertical velocity present in the actual

atmosphere.

The above elements would be seriously deficient without an effort aimed at obtaining representative statistics of real cloud fields in the region under consideration. Any practicable parameterization scheme will have to draw on observations to some extent. For example, what are typical three dimensional shapes of cloud elements and how do they influence the directional radiances, or how often are there mesocale horizontal structures present which bias the area averaged cloudiness? The compilation in FIRE of a regional climatology, classified both by location and by synoptic type, will be an important adjunct to the global view being provided by ISCCP. It will both encourage research on alternative cloud retrieval algorithms, and also provide intensive experience on thelr application. In particular it will allow intercomparisons with the operational procedures followed internationally. FIRE is to take advantage of more capable data sets than can be handled on a global basis. Only from such studies can it established whether it is possible to capture the basic behavior with smaller number of variables, and less intensive data processing.

Because the program *ill involve the collection of data from many different satellites and other instruments, and from investigators at many different institutions, a program element devoted to data set assembly is essential. The goal is to supply participating scientists with complete data sets tailored to selected situations, and to recapture for wider use significant analyses, derived parameters and other added information.

The experiment will be based primarily on intensive processing and case studies of data from existing satellite systems. This data will be combined with complementary observations from the earth's surface and aircraft. Special attention will be devoted to parts of the continental U.S. and the Eastern Pacific Ocean. Though sufficiently large and dispersed so that they provide a spectrum of cloud types and behavior, the special observing regions will be selected to provide a profusion of marine stratus/stratocumulus and continental cirrus samples. A broader sample of scenes for analysis will be drawn from the primary FIRE region extending from the equator to 50 N and from 70 W to 140 W. These data sets and research conclusions will also provide a basis for interpreting the algorithms used in the compilation of

11

global cloud climatologies in ISCCP.

The intensive effort will involve acquisition of some additional satellite data sets for limited periods (VAS soundings, AVHRR LAC data, and some stereo viewing), some ancillary observations (particularly aircraft soundings, additional radiosondes and special surface data related to clouds), and considerable additional data processing. This processing is to utilize fully the signatures of atmospheric profiles and three dimensional cloud structures which are present in the satellite data sets to provide as complete a picture as possible of the cloud populations, their impact on the radiation field, and the atmospheric state variables. Multiple views from different satellites are essential for this purpose. Routinely available correlative data on atmospheric state variables and the ancillary observations will be used in selected locations to confirm interpretations, and to obtain key addtional information which is not otherwise available. The observing system is shown schematically in Figure 1.1.

1.6 Resources Required

The resources required are of three kinds: funding to support the research groups involved, facilities such as aircraft and computers, and access to data from operational satellites, including limited rescheduling of certain operations. More specific details are given in sections 2 - 6.

1.7 Experiment Team

FIRE, as envisioned, promotes a number of orchestrated research efforts which require careful coordination. Although the FINE plan stresses the view that observations lead to model development, such development in turn often leads to improved data collection and processing algorithms. FIRE must provide the coordination necessary to maintain a strong bond between these activities. Likewise for the case studies of cirrus and marine stratus, success requires the coordination of surface and aircraft based observations with the available satellite observations. To achieve the desired coordination, FIRE requires a science team whose members are those participating in the experiment. The team is to plan the details of the coordinated phases of the experiment and to perform the experiment in a manner that takes advantage of the concurrent mixture of mutually supporting observations called for in this plan.

It is anticipated that, following publication of this plan and after an interval for preparation and review of proposals, an Experiment Team for FIRE will be formed from scientists involved in the FIRE data gathering and processing and/or in the use of the FIRE data for the improvement of cloud parameterizations in climate models. The experiment team will report to the NCPO or other designated agency and will be responsible for completion of the five elements of FIRE. The experiment team is expected to agree to a use of data and publication arrangement.

14

1.8 Schedule

Initial funding for FIRE is anticipated in FY 84. The initial funding is to be used mainly for detailed planning and pilot studies using existing data. The program will last 5 years, with the mayor data collection effort concentrated in a 3 year interval containing the field phases of the in situ observations. An approximate schedule is shown in Figure 1.2.

A number of trade-offs have been considered in formulating this schedule. First, owing to the expected lifetimes of current operational and special experiment satellites, the FY85 - FY89 period appears to be the best window for conducting FIRE. Second, because the satellite and field phases of the experiment are interdependent, delaying the start of a field phase would necessarily extend the satellite related work and thereby significantly increase the cost of the experiment. Third, although the intensive observations of cirrus and of marine stratus and stratocumulus systems take place in the same years, owing to the different climatologies of these cloud systems, the field experiments occur at different times within the year. As a result competition for equipment (aircraft, radars, etc.) is minimized. Finally, due attention must be paid to the lead times required by the intensive field programs.

16

2. Validating Cloudiness and Radiation Parameterizations

2.1 Scientific Objectives and Methodology

Any approach to improving parameterizations of clouds and radiation in GCM's must start with a realistic appraisal of present techniques. The first step is to establish objective measures of their performance and obtain evidence as to the source of their deficiencies. Their present status Ls admitedly primitive, but this situation will not improve unless objective measures are available. So far such appraisals have been limited to intercomparisons of the model cloud climatology, usually zonally averaged, with the estimates of London (1957) which were derived from visual observations of cloud type by surface observers. Only satellites can provide the sampling and global perspective necessary for more searching intercomparisons. Radiative outputs have been compared with maps of the monthly average fluxes at the top of the atmosphere, such as those compiled by Campbell and yonder Haar (1980) from broad band wide field of view instruments mounted on polar orbiters. Such comparisons can, however, provide only an overall check on model performance. They fail to provide information on whether discrepancies (or even agreement!) are due to inadequate simulations of synoptic structures, poor estimates of the clouds associated with these structures, improper radiation associated with those clouds, or a combination of all of these.

The strategy in FIRE for validating GCM radiative codes against observation is to separate, to the extent possible, the component parts of the total process, and devise independent tests for:

The core program for FIRE concentrates on the second and third of these, for which a possible approach is outlined below. Since some of the concepts are relatively new, the methodology is discussed in some detail. Alternative ways of achieving the same basic goals should also be considered in FIRE.

2.2 Clear sky radiances and synoptic scale climatology

Though present techniques for estimating radiances from a column free of cloud and haze are generally assumed to be adequate, they

17

have never been properly verified to the level of accuracy (a few

W m~2)

needed for absolute climate modeling. FIRE itself will not attempt

to

achieve such accuracy but will instead rely on efforts currently

underway to remedy this situation under the auspices of the

International Radiation Commission.

Likewise, though very much part of the overall problem of climate model validation, the ability of a GCh to simulate regional climatologies of synoptic processes and their associated fields of temperature, moisture and vertical velocity, will be considered as outside the scope of FIRE. FIKE will concentrate on the modifications introduced by the presence of clouds and our ability to produce those clouds.

2.3 Validating models for the radiative modifications due to

clouds

In validating models for the radiative modifications due to clouds, the strategy suggested is to use a state-of-the-art radiative transfer column model as a tool for assessing the merits of individual cloud/radiation models, and to serve as an intermediate standard between GCM codes on the one hand and observations, predominantly satellite narrow spectral band directional radiances, on the other.

A central issue in modeling clouds and radiation is the identification of a suitably limited set of cloudiness variables which is used to summarize the features of real cloud populations. This identification amounts to the selection of an idealized cloud model which is both suitable for computations of the radiative effects and which can be related in some way to large scale dynamical processes. Such a cloud model is implicit or explicit in every GCM parameterization. It is also inherent in any attempt to retrieve cloudiness variables from observations. A primary question, both for testing existing parameterizations, and for summarizing statistics of cloud populations, is whether the cloud model being used is adequate. For individual scenes drawn from a representative ensemble of synoptic situations, are there any values whatsoever for the chosen cloudiness variables that within a state-of-the-art radiative code can reproduce within expected uncertainties all of the observations? The column model procedure is intended to answer this question, and, incidentally, yield best fit estimates of what the values for the cloudiness variables actually are.

Thus Figure 2.1 shows the role this column model plays ln FIRE. The applications depicted by this figure are discussed in this and in later sections. Figure 2.2 shows in detail the radiation model validation procedure which is denoted by box A5 in Figure 2.1.

Figure 2.2 identifies two columns: cloud retrievals and GCM formulations. In one case the set of cloud models ((B5) in Figure

20

2.2) ranges freely over those used in state-of-the-art radiative transfer models. In the other case the models are constrained by limitations imposed by GCMs (B7) (e.g. vertical resolution, cloud properties constrained to those given by amount and liquid water content, etc. (B8)). For both types of cloud models, the directional spectral radiances from a high resolution radiative column model (B1) can be compared with several simultaneous satellite images (B3) of individual scenes approximately 250km square (64). Such comparisons would be made for scenes observed in a variety of spectral bands and from a number of viewing angles. Since the cloud variables are not known a priori, part of the information available from the observations must be used to estimate them (~4.1). The remaininy information should then be used to check whether the cloud fields so retrieved are internally consistent and physically reasonable (~4.2). Also, when available, surface and in situ observations (B2) provide necessary confirmation of the satellite retrievals. To the extent that consistent sets of cloud variables are obtained in a wide variety of cases, and sensitivity tests show that enough information is indeed available to provide a valid test, it will be inferred that both the procedures for computation of the radiance field and the retrievals themselves are satisfactory. Otherwise, alternative forms for specifying the cloud variables will have to be developed and tested (B9 - B10, B9 - B11). This approach has not been used in its entirety before, though elements can be found in studies by Rossow et al. (1983), Gordon et al. (1983a,b) and Bonnel et al. (1983).

The ultimate test of a given set of models is their ability to reproduce observations for a larye number of scenes. By exploring the ensemble of scenes for which a particular set of models provide successful retrievals, (B12), FIRE not only assesses the utility of the models, but also provides a set of validated cloud retrievals (B13) that may be applLed to other uses as is illustrated in Figure 2.1.

The specification of a complex cloud field by a limited number of parameters in order to calculate the radiation is essentially a parameterization or a model of the physical system. In any particular GCM the number of cloudiness variables at any one grid point is quite limited (e.g. the fractional cloud cover for low, middle and high clouds). Also additional properties essential for the radiative calculation are simply assumed (e.g. the emissivity of cirrus or the directional reflectance of low cloud). Thus the actual set of variables (B8) is a highly constrained subset of the potential set, and with an overdetermined system in which the number of genuinely independent relevant data substantially exceeds the number of unknowns it is possible both to estimate the members of the subset and to test how well the chosen constraints permit simulation of the radiation field. Such an approach would allow identification of systematic bias associated with the specification of the cloud variables, and thereby also allow a systematic search for improvements (B7, B8, ~4, B9, B11).

21

Since by far the most widely available observations are narrow spectral band directional radiances at satellite altitude, the standard methodology is designed to require them alone. Where possible the narrow band radiances will be augmented by broad band wide field of view satellite measurements. Also, where available, suitably sampled in situ measurements of surface or mid-atmospheric radiative fluxes, and ancillary observations such as cloud base height and radiosonde soundings may enable important additional intercomparisons.

A significant problem is our lack of understanding of how sub-pixel scale cloud structure affects the directional radiances, and hence also the overall fluxes. It will be necessary to experiment with different approaches to parameterizing the sub-pixel scale influence. The column model activity will therefore require a strong theoretical program on the radiative effects of real cloud populations. Such a research effort on fundamentals is also needed by the statistical program described in Section 5. In a limited sample of cases, very high resolution images such as those from LANDSAT will be particularly valuable for checking assumed cloud scale structures. Another check will be measurements from the yround of apparent cloud cover as a function of zenith angle.

The column model should include a high quality radiation code (B1) capable of calculating radiances appropriate to the spectral windows of the instruments under consideration, with due regard to filter characteristics. The model should have a vertical resolution sufficiently fine that it is not an issue. The number of potential cloudiness variables, e.g. height, fractional cover, optical depth, and alternative specifications of microphysics and cloud scale morphology, which are included in the computation should be substantially larger than will be varied independently in any one retrieval (B5). This degree of freedom permits particular cloudiness parameterization schemes to be introduced as constraints (B6). Changes in the cloudiness parameterization, and GCM modeling assumptions, like reduced vertical resolution of temperature and humidity, can then be made easily and without introducing numerical inconsistencies. The sensitivity of output radiances to variations in input parameters will need to be examined thoroughly. The extent to which retrievals for any given scheme and data set are overdetermined or underdetermined may then be examined.

Cloud retrievals are envisaged on the scale of individual pixels or small groups of pixels, within a broad sample of (250km)^2 scenes (B4.1). The basic synoptic scale temperature and humidity profile used for all the pixels in a scene will be obtained from satellite soundings in sub-areas estimated to be cloud free. All variations from the cloud free radiances will be supposed to be associated with clouds. Consistency of a retrieval will be Judged by the ability of the model to reproduce radiances in different spatial resolutions, directions or channels not used in the retrieval, and by the horizontal structure of the cloud fields (B4.2) (e.g. by whether clouds tend to appear in layers

22

of uniform cloud top temperature, or to form mesoscale structures

resembling penetrative convection).

It should be noted that model validation using this approach is necessarily somewhat indirect. Important aspects of the radiative calculations in the column model are not tested at all. For example, the basic approach does not insure the absolute calibrations of clear sky radiances to space nor does lt insure accurate estimates of the radiative flux divergence within the atmosphere. Other elements of FIRE will be undertaking relevant in situ observations and every opportunity will be taken to correct for these and similar limitations.

One use of the column model is to develop alternative formulations of the cloudiness variables. This use is essentially a search for improved cloudiness retrieval algorithms. It differs from many such attempts in that the number of retrieved parameters is deliberately kept as small as possible (B6), less than the number of relevant data available, in order to use the redundancy (B4.2) as a test of the procedure. It could thus be described as looking for the simplest scheme that seems to work. The test is: if on a representative sample of scenes, a parameterization consistently yields retrievals which fit the observations within probable error and does not violate other preconceptions about cloud fields (e.g. their tendency to occur in layers) it passes the test (B5, B6, B4, B9, B12). It must be remembered, however, that the information gained is limited to that subspace of the universe of cloudiness parameters for which the inversion matrix is formally overdetermined, i.e. to those aspects for which there is redundant but independent data from several channels or directions. If, on the other hand a parameterization yields inconsistent results, it is natural to vary the specification of the permitted cloudiness variables, and repeat the computations (B5, B6, B4, B9, B10). This constitutes an empirical search for a more satisfactory parameterization. If one is found which works well on the given sample of scenes it must then be tested on an independent sample. Given the primitive basis for presently existing parameterizations, this test -search - retest cycle may prove to be a frequent mode of operation.

The mesoscale cloud populations inferred from consistent retrievals using the column model can then be used in the applications illustrated by Figure 2.1. In particular they can be used in the statistical studies described in Section 5 (box A8 in Figure 2.1). They, of course, are not the only tool used in the statistical studies. For example, correlations of radiances in space and time that are fundamental to statistical studies, are likely to be ignored by models that provide the best pixel scale cloud retrievals. Nevertheless, the retrievals offer relevant cloud parameters whose statistics require examination.

The retrievals will also serve as a basis of comparison with the global ISCCP retrievals (A9). Likewise, conclusions on the directional

23

properties of clouds also may prove useful for ERBE. Where appropriate,

the output fron a number of neighbouring scenes will be integrated to

simulate what should be observed by the broad spectral band, wide field

of view ERBE instrument which will fly on polar orbiters and one

satellite with highly inclined orbit. The comparisons with ERBE

measurements will provide an important test of calibrations.

Immediate uses of the validated retrievals are, of course, tests of GCM radiative flux codes (A6) and GCM cloud production schemes (A7). These uses are described next.

2.4 Testing GCM radiative flux and cloud production codes

Since the radlatlon codes used ir1 many working GCMs provide only broad band radiative fluxes, the column model has to be used as an intermediate standard. Errors in the GCM computations can arise in a number of ways, but take for example the following: In the interests of efficiency undesirable shortcuts may have been taken in the radiative code. It is assumed that the radiative code of the column model is adequate in all material respects. If the cloudiness varLables and the vertical resolution for profiles of temperature and humidity are kept the same, then differences between the column model output, integrated over angle and spectral band, and the output of the GCM code may be ascribed to the shortcuts (A6).

To test the ability of a GCM to produce the clouds themselves (A7), consistent analyses of temperature, humidity and vertical velocity fields for a variety of synoptic situations (A4) may be obtained from yroups undertaking state-of-the-art 4-dimensional assimilation, as at the European Centre for Medium Range Weather Forecasting (ECMWF). These analyses will then be fed as input data into the GCM alyorithm used to predict cloudiness. The predicted cloudiness may then be compared with retrievals using the column model with cloudiness parameters constrained in the same manner as in the GCM. To the extent that the input analyses and cloud retrievals are correct, this procedure is a direct test of that algorithm. Since cloudiness algorithms are often tuned to the climatology of temperature and humidity in each particular model, separate checks (not part of the core FIRE program) will be needed of the extent to which the input analyses are consistent with-the model climatology. Likewise, as no analysis of the meteorological field is expected to be entirely accurate, comparisons of several state-of-the-art analysis schemes (e.g. the fGGf level 3 analyses from GFDL and ECMWF) would provide estimates of the sensitivity to realistic analysis errors.

In principle, the cloudiness alyorithm can be isolated for this purpose from the remainder of the GCM (Lonnel et al. 1983). Mowever, in many algorithms cloudiness is deemed to occur only when the gridpoint relative humidity exceeds some threshold close to 100%, and any excess moisture over saturation is intantaneously precipitated. The average

24

cloudiness inferred over a number of model timesteps thus depends as much on the model dependent vertical velocity and convective adjustment processes as on the precise initial distribution of moisture. It will therefore be necessary to run the model for a while to obtain representative values for comparison with observations. Care will therefore be needed in transferriny temperature and moisture fields based on an analysis-forecast cycle using a different model.

A complementary approach to testing cloud production schemes is to use the GCM to compute from the chosen analyses of temperature and humidity the broad band fluxes that should be observed by a wide field of view instrument in the location of an actual satellite equipped with such an instrument. Thus the cloudiness and radiation algorithms together will be tested. This approach yields less diaynostic information.

2.5 Sampling Strategy

For the column model retrievals, a representative sample of scenes is needed. These scenes should be taken from tropical, subtropical and midlatitudes, including both marine and continental environments. They should also include a variety of both local sun anyles and viewing angles. The highest importance is attached to simultaneous high resolution data from as many channels and distinct viewing angles as possible. These criteria should be met even at the expense, if necessary, of the number of cases considered.

Since all relevant polar orbiters (except ERBS) are in different sun synchronous orbits, the primary conjunction must be between images and multichannel soundings from a single polar orbiter and simultaneous views of the same location from two geostationary satellites (principally NOAA-7,8 and GOES East and West). At a given latitude, such simultaneity constrains the sun angle according to the local passage time of the polar orbiter. Fortunately, the sun angle also to some extent changes with season.

Two images are simultaneous If the clouds have changed little between them on a pixel by pixel basis. Simultaneity is thus governed by the resolution of the observations. As the typical resolution called for by FIRE is ~1km and as typical wind speeds for clouds "5Okm/hr, the time separation between simultaneous views should be small compared with 1/50 Hr, that is small compared with 1 minute. Such simultaneity has been achieved previously for the GOES satellites by coordinating the starting times of the scans for the two satellites. The conjunction of a polar orbiter and GOES images also dictates the starting time for the GOES scan. Of course, the requirement for simutaneity is less stringent when the resolution of the observation becomes large. Thus, for the 8km resolution of the GOES IR channel, the time difference between observations must be small compared with 10 min., that is -~1 min.

25

Note that it is also necessary to obtain for selected days simultaneous observations at the same location at intervals throughout the day from GOES East and GOES West, particularly including passage times of LANDSAT, FHBS, and DMSP. Such observations are necessary to sample the diurnal cycle.

It is not essential that the cloud production studies be undertaken on the same set of cases used for column model validation, though considerable overlap is highly desirable. The inital analysis of temperature and humidity based on routine weather observations will be nemispheric or global. The cloud production algorithm can then be run on a gridpoint by gridpoint basis, and compared with whatever satellite observations are available. A representative geographic spread of cases is needed. Of course, intercomparisons in the tropics and in regions where conventional data Ls sparse will naturally be less reliable.

2.6 Uata requirements

The column model studies require substantial amounts of satellite observations that are carefully coordinated between a number of different platforms. A preliminary observing plan is sketched in section 6. The assembly of such data sets, preliminary editing and determination of navigation and relevant calibration parameters will be the reponsibility of the Data Assembly Group (see Section 6).

For the cloud prediction studies, existing archived FGGE analyses provide a good sample of temperature and humidity fields. Appropriate wide field of view radiometer measurements are also available. For intercomparison with new satellite observations (particualarly the capabilities provided by stereo VAS), the analyses routinely compiled by ECMWF are probably the most suitable.

2.7 Schedule and Resources

The column model effort is central to several aspects of FIHE, and should be undertaken as soon as proposal review and funding permit (late FY 1984). Initlally the effort will use existing data. The effort may be expected to last 5 years. Continuity of scientific leadership and experienced support personnel is extremely important to lts success.

Support will be needed for the research group or groups involved. This support should include computer resources. It is assumed that the necessary data acquisition and editing will be undertaken within the data assembly program.

Support will also be needed for the basic theory and assessment of the radiative effects of real cloud geometries.

The testing of GCM cloud predictions is probably best encouraged within existing GCM groups. Its timing should be related to the availability of credible cloud retrievals and other group priorities.

26

2.8 Expected Results

GCM cloudiness and radiation parameterizations have not previously been effectively isolated for comparison with observations. If the approach described here can be made to work, it will be a very significant step forward. It is anticipated that it would immediately highlight deficiencies in existing parameterizations of cloudiness and in algorithms for estimating cloudiness. It is then expected to guide the search for better algorithms. A successful search would lead to better cloud retrieval algorithms as might be used by ISCCP, better statistical characterizations of cloud lifetimes and mesoscale structures, and many other applications.

27

2.9 References

28

3. Studies of Marine Stratus and Stratocumulus Systems

3.1. Introduction

Owing to their effect on the earth's radiative energy budget, clouds in the boundary layer over the world's ocean are likely to strongly influence climate and climate change. Their high albedos compared with that of the ocean background give rLse to large deficits in the absorbed solar radiative flux at the top of the atmosphere while their low altitude prevents significant compensation in thermal emission. Likewise at the surface, they considerably reduce the flux of solar radiation into the ocean while owing to the nearly saturated state of the boundary layer, they offer only a marginal increase in the downward thermal radiation. With respect to a well known climate problem, a change in global cloud cover of only 4%, if it were confined to an increase in low level cloud, would be sufficient to offset the 2-3 K predicted rise in global temperature associated with a doubling of CO2. Our present understanding of the processes controlling cloudiness is insufficient for confident predictions of the extent or absence of such changes.

In addition to the radiative properties of these clouds, one notes that when they exist, boundary layer clouds strongly influence the heat and moisture exchanged between the troposphere and the boundary layer. Clearly, these cloud systems are likely to play a mayor role in determining the climate, and models which reproduce their thermodynamical behavior are essential to advanced studies of climate dynamics and assessments of climate change.

A substantial fraction of boundary layer clouds is continuous stratus and broken stratocumulus, in which the cloud elements are intimately related to convective updrafts in the subcloud layer. Most of the remainder are trade wind cumulus, in which the motion is driven by condensation in the clouds themselves, and is only loosely coupled to the circulation below.

Stratus cloud layers, and to a lesser extent stratocumulus layers, have been modelled extensivley and with increasing complexity, by Lilly (1968), Moeng and Arakawa (1982), Deardorff (1980a) and others, and they have been observed in the field off the coast of California (Brost et al., 1982a, b) and as part of the Joint Air Sea Interaction Experiment (JASIN) off the British Isles (Pollard, 1978). Most of these studies, however, focused on horizontally homogeneous and isotropic stratus cases. On the other hand, satellite observations reveal that even extensive stratus and stratocumulus systems (~1000 km)2 exhibit considerable structure and fractional cover even at the smallest available resolution (1-4 km)2 (Coakley and 8retherton, 1982). Relatively few studies have yet focused on the mechanisms determining the fractional cloud cover (Sommeria and Oeardorff, 1977; Albrecht, 1981; Bouyeault, 1982) which are of central importance to modeling the

29

role of these clouds in climate. Given the existing basis of

experience, significant progress in this area seems possible. Therefore

the mayor goal of the marine cloud-capped boundary layer segment of FINE

is to develop physical models and parameterizations for fractional cloud

cover in stratocumulus layers.

The strategy for pursuing this goal should be to test candidates for the break-up of stratus systems that have evolved through theoretical studies and model simulations over the years. This will involve bringing to bear for the first time extensive satellite and radar and/or lidar observations of the cloud-capped marine boundary layer. The necessary observations of stratus and stratocumulus systems fall into two categories. The first is descriptive, involving satellite observations and surface-based observations from a lLmited number of stations. These observations should yield climatological properties of stratus and stratocumulus within the Pacific basin, following a sampling strategy designed to produce stable representative statistics of the properties. The second category of observations is process oriented. It combines the satellite and surface-based measurements with intensive radar, lidar and aircraft measurements. These intensive measurements will be taken during short periods and repeated at intervals throughout the course of FIRE. The emphasis should be on obtaining for a few cases oomplete sets of the relevant variables, with enough redundancy to per,nit cross checks and some instrument failures. The observations should provide insight into the physical processes that govern the existence and structure of stratus and stratocumulus systems. These insights should in turn lead to improved models of the physical processes.

For trade wind cumulus, it seems more appropriate at this time to re-examine the existing in situ and satellite observations, such as those taken during GATE, and to use new satellite data to characterize the nature and extent of variability, before undertaking a mayor new field program. Such studies are appropriate within the statistical programdescribed in section 5.

Following discussion of the scientific issues, this section concludes by summarizing the required observations in the form of a description of a candidate experiment, including a schedule of activities. The candidate marine stratus and stratocumulus experiment is confined to the Pacific basin, with the intensive observations on the western coastal region of the US. It runs for five years. The candidate experiment serves only to suggest a possible and likely pool of observations. It is not meant to exclude alternate approaches that might yield equivalent results. The variety of observations indicated clearly argues for an experiment team which includes experts from a number of normally exclusive disciplines.

30

3.2. Theoretical background

Numerical simulations of stratus and stratocumulus systems have been performed with simple mixed-layer models (Lilly, 1968), higher-order closure models (Moeng and Arakawa, 1980; Chen and Cotton, 1983), large-eddy models (Deardorff, 1980a; Sommeria, 1976), and general circulation models (Suarez et al., 1983). As a result of these studies, several key theoretical issues have emerged, and a carefully designed observational program is needed to resolve them. The issues are as follows.

3.2.1 What determines the fractional cloudiness?

Mesoscale (1-100 km) variations of cloudiness in both stratocumulus and trade cumulus regimes are well known (Agee et al., 1973), and although widely studied they have yet to be modeled with demonstrable precision. Potentially Just as important are microscale (0.1-1 km) variations in cloudiness, which can appear either as holes in a cloud deck or as patches of cloudy air in an otherwise clear marine layer. Even where the sky is completely obscured by cloud, fractional cloud cover can usually be found at some heights, notably near cloud top and cloud base.

Albrecht (1981) suggested that the trade cumulus cloud fraction can be parameterized as a function of the relative humidity, and he proposed a formula with reasonable behavior in the limits of very wet and very dry layers. A more advanced theory of fractional cloudiness must include a simple but explicit model for the small-scale variability of temperature and moisture within the marine layer. One possibility is to impose constraints on the variances of temperature and moisture, within the context of a "plume" model. A second approach is to use assumed Joint distributions of temperature and moisture (Som,-neria and Deardorff, 1977).

Because theories for fractional cloudiness are at best primitive, a broad-based data set is needed to explore the problem observationally. Of course, the most basic thing to document is the fractional cloudiness itself. In addition to satellite observations and photographic methods, downward-pointing airborn lidar (Uthe et al., 1980) can be used to map the topography of cloud top, as well as gaps in the cloud, with very high resolution. Where clear and cloudy columns exist side-by-side, the characteristic soundings of the two columns should be determined. An attempt should be made to document the extent to which individual clear or cloudy patches are associated with characteristic circulation systems such as convective cells. Conditional sampling techniques could be applied to aircraft data for this purpose.

31

3.2.2 What is the role of cloud-top entrainment instability in

determining cloud type?

Squires (1958) and Lilly (1968) pointed out that evaporative cooling can cause air that is entrained into a cloud to become negatively bouyant, unless the entrained air is very warm to begin with. Randall (1980a) derIved a stability criterion for this entrainment instability, and discussed its possible role in determining the distribution of cloudiness over the subtropical oceans. Deardorff (1980b) presented results from a high resolution three-dimensional numerical model, which indicate that the onset of the instability is accompanied by rapid entrainment, a significant reduction in cloudiness, and a decrease in the effectiveness of turbulent mixiny. Moeng and Arakawa (1980) and Chen and Cotton (1983) obtained similar results using hlgher-order-closure models. However, Fravalo et al. (1983) discuss observations during JASIN which may IndLcate that a solid or nearly solid cloud deck can persist in spite of the instability.

Observations suitable to investigate the instability and lts effect on the cloudiness would include measurements of the te;nperature and moisture "jumps" across cloud top, direct evidence of evaporatively cooled descending parcels, high-resolution maps of cloud structure, and soundings in regions where the instability is occuring. It would be particularly useful to document these properties of the marine layer across a transition zone in which the fractional cloudiness is decreasing sharply. The Saturation Point ideas of Betts (1983) may provide a useful framework for the analysis.

3.2.3 What is the distribution of radiative cooling?

The potential importance of the thickness of the layer of cloud-top radiative cooling for the dynamics of the marine layer, and particularly for determining the entrainment rate, has been discussed by Deardorff (1976), Kahn and Basinger (1979), Schubert et al. (1979), Randall (198Ob), Lilly and Schubert (1980), and Fravalo et al. (1980). The finite area-averaged thickness of the cooling layer arises from the finite opacity of the cloud, and also from the presence of small-scale fluctuations of the cloud-top height due to turreting and waves. Essential is documentation of the area-averaged thickness of the radiatively cooled layer, and the extent to which this thickness is controlled by fluctuations of the cloud-top height. Also useful are covariances of the radiative cooling rate with both buoyancy and vertical motion. These covariances influence the rate of production of buoyancy variance and buoyancy flux, respectively.

3.2.4 What determines the entrainment rate?

Entrainment closure assumptions for cloud-topped mixed layers have been proposed by Deardorff (1976), Kraus and Schaller (1979), Stage and Businger (1981a, b), and Suarez et al. (1983). A discussion of the en trainmen" problem is given by Randall (1983). All of the existing

32

theories are based on consideration of the turbulence kinetic energy balance, primarily emphasizing the role of buoyancy. Brost et al. (1982a, b) report observations suggesting that shear can also be important, and Suarez et al. (1983) have included shear in their closure.

Observations designed to test the entrainment theories should include the following:

a) Measurement of the entrainment rate. One approach is to examine the mass budget of the marine layer. Both the layer depth and the diveryence must be determined for a box of perhaps 10 to 30 km radius. The layer depth can be measured through a variety of methods, including acoustic sounding, lidar ranging, instrumented balloons, and aircraft stacks. Measurements of the divergence can be performed using doppler radar and/or doppler lidar.

A second approach is to measure the inversion-layer budget of a conserved variable, such as ozone mixing ratio (Lenschow et al., 1982).

b) Measurement of the buoyancy flux profile through the depth of the marine layer. For in situ measurement, usual problems with probe wetting must be overco,ne to achieve this. New techniques may make it possible to infer the buoyancy fluxes from doppler radar or doppler lidar measurements. Liquid water fluxes are important (Randall, 1980a), and might be obtainable from a combination of millineter-wavelength doppler radar and dual-channel microwave radiometer data (Hogg et al., 1983). A related issue is the role of entrainment in determining the cloud droplet distribution (Baker and Latham, 1982). For this aspect microphysical measurements arc needed.

c) Measurement of the shear stress profile, to allow the shear production term of the turbulence kinetic energy equation to be evaluated. Again, both in situ and renote-sensing techniques can be used here.

d) Observation of individual entrainment events. Tethered balloons (Roach et al. 1983) are ideal for this purpose.

e) Conditional sampling of buoyantly productive and consumptive parcels at several levels in the layer, as reported by Wilczak and Businger (1983) for the clear convective boundary layer. Such observations can only be done with aircraft.

3.2.5 What is the role of the mesoscale?

Agee et al. (1973) and others have described the observed frequent occurence of mesoscale cellular structures ln the marine boundary layer. A number of promisLng theoretical ideas have been developed concerning these structures (e.g., Fiedler, 1983). FIRE represents an

33

exciting opportunity to measure and document the structures of IndLvidual mesoscale cells, particularly the so-called "closed" cells which are associated with stratocumulus clouds. Useful observations will Lnclude doppler measurements oP the characteristic mesoscale wind fLeld associated with cells, and the variation of the marine layer depth, thermodynamic structure and surface fluxes within a cell. FIRE might also determine the vertical transports of sensible heat, moisture, and momentum associated with this type of convection.

3.3. Satellite Observations

Recent work demonstrates that satellite imagery can provide quantitative characterizations of low-level (presumably stratus and stratocumulus) cloud layers over oceans (Coakley and Bretherton, 1982). In addition, some have used the imagery to estimate the solar insolation reaching the ocean surface (Gautler and Masse, 1983). In pursuit of ,nodeling the fractional cover and concurrent radiative properties of stratus and stratocumulus systems, the imagery offers: 1) a determination of the time and space scales for the variability of cloud amount, cloud height and visible reflectivity, and 2) in conjunction with additional surface based and aircraft borne observations, the imagery furnishes the cloud cover and radiative properties that may be linked to the thermodynamic structure of the lower atmosphere. Of course, as techniques for analyzing the imagery are relatively new, their validation remains an essential component of mayor experiments such as FIRE.

3.3.1 Statistical characterizations

A first aim of satellite observations for the cloud-capped marine boundary layer is to determine the time (minutes-days) and space scales (1-3000 km) associated with varLability within stratus and stratocumulus systems. These scales would, for example, establish an observational foundation for modeling studies which now must assume that such systems and the physical processes that govern them are spatially homogeneous and isotropic (Sommeria and Deardorff, 1Y77). The questions to be answered by the observations are 1) over what time and space scales do the cloud properties remain uniform and what are their characteristic features; 2) on what scales are the assumptions of homogeneity and isotropy valid and 3) with what frequency do the characterLstic features recur? Such information, in addition to providing a foundation for modeling studies, will also provLde statistics useful for designing the field experiments described below.

As part of the time and space scales of characteristic structures, one important characterization of stratus systems is their diurnal evolution within given geographical regions. Geostationary satellites can of course provide statistics for this evolution. Typical diurnal variations, for example, of cloud top height have been predicted in a number of model stud Les (Albrecht, 1979; Schubert, 1979; Oliver et al, 1978; Hanson and Gruber, 1982).

34

In addition, satellite imagery provides the immediate opportunity of characterizing the radiative properties of stratus and stratocumulus systems on a variety of resolutions. Contrary to common belief, the systems do not appear to reflect solar radiation as plane-parallel cloud models would predict, but instead they exhibit reflectivities with a large degree of horizontal variability (Coakley and Bretherton, 1982). Presumably this variability is due to the small-scale structure of liquid water within what appear to be horizontally homoyeneous layers (Derr, 1982) or to small scale irregularities Ln the upper surface. Through multiwavelength observations one hopes to analyze the dependence of the reflectivity on the liquid water content (DeVault and Katsaros, 1983).

For broken cloud above a warmer background, the directional distribution of infra-red window radiances contains information on the three dimensional shapes of cloud elements. These shapes also affect the visible reflectivity, but in ways which depend on the direction of the sun and the droplet size distribution within the cloud. Since the scattering process depends in a highly non-linear manner on the cloud geometry, and every observational approach involves either averaying or sampling over unresolved small scale structure, a central need for radiative budget studies is to develop, usiny a combination of theory and observations, statistical models which relate visible and infra-red radiances to bulk measures of liquid water and cloud shape. Thus, to characterize the radiative properties of stratus and stratocumulus from the satellite data the first questions to be answered are 1) What is the spatial variablility of cloud reflectivity; 2) how is it linked to the spatial scales of the cloud cover and cloud height and therefore presumably to the total column content of liquid water; 3) how does the reflectivity vary with wavelength; 4) how does it depend on the sun-cloud-satellite geometry; 5) how do these dependencies change with spatial averaging and 6) how are such variations linked to the emitted infrared radiation?

3.3.2 Observations concurrent with in situ measurements.

The energy and moisture budgets of stratus systems argue for balances between entrainment, radiative cooling at cloud top, and surface fluxes in order for the systems to survive (Lilly, 1968 and Deardorff 1976). Imbalances will lead to instabilities which in turn lead to the break-up of the systems (Randall, 1980; Deardorff, 198Ob). One suspects, however, that during break-up these three fluxes may again establish a balance which maintains a certain fractional cloud cover or at least they govern the rate of break-up (Moeng and Arakawa, 1982; Randall and Hoffmann, 1983). Clearly, in conjunction with observations of the surface fluxes, the boundary layer, cloud and inversion layer flux profiles (Lenschow, 1973 and Brost et al. 1982a, b), satellite imagery offers the opportunity of linking cloud properties to the competing fluxes of energy and moisture. The question to be answered is how are

35

fractional cloud cover and cloud top variability related to the radlative and turbulence energy budgets. The surface and alrcraft borne observations required to measure the needed fluxes are described in later sections.

With regard to the satellite observed reflectivities by stratus, concurrent in situ observations of the water drop size spectrum and the liquid water content could provide a link between liquid water and reflectivity. In estabilshing this link the question to be answered ls how does spatial variability in liquid water content and microphysical structure of the clouds affect their reflectivity? 3.3.3 Validation.

Of course, inferences about cloud structure made from satellite imayes need valIdation. One approach is to compare results with very high resolution images such as LANDSAT. Such comparisons can provide confirmation of estimates of sub-pixel scale cloud yeometry and fractional cloud cover, and indicators of haze. Another approach is to obtain cloud top height at very high resolution along continuous tracks from airborne lidar or millimeter radar. Such measurements are needed not only to verify that the basic interpretation of the low resolution imayes in terms of cloud type and mean cloud height is approximately correct, but also to develop statistical characterizations of very small scale structures in ceal cloud populations and how these structures affect the directional properties of low resolution radiances. 3.4. Aircraft measurements in the marine boundary layer

Instrumented aircraft will play a mayor role in any field study of ,narine stratocumulus clouds. Aircraft can probe the structure of the entire boundary layer, as well as the capping inversion and the overlying free atmosphere. They provide both in situ and remote observations. The mobility of aircraft permits measurement of both horizontal and vertical variations in boundary layer structure in a time period short enough that significant time changes do not occur.

Observational capabilities required of aircraft included in a marine stratocumulus program are as follows:

a) fast response (-10 Hz bandwidth) measurements of air velocity components and scalar variables to be used for turbulence flux estimates and other turbulence statistics, b) mean in situ measurements of wind and scalar variables, and

c) remote sensing of atmospheric and surface structure. In addition, accurate aircraft position, both horizontal and vertical, is an important requirement.

36

Aircraft are presently available that can measure fast-response fluctuations in velocity components, temperature, and humidity with enough accuracy that vertical fluxes of these variables can be estimated throughout the marine boundary layer (Brost et al., 1982a,b). Other scalar fluxes can also be measured. For example, Lenschow et al. (1982) have measured fluxes of ozone in a marine boundary layer and show that measurements of the ozone flux and mean concentration profiles through the boundary layer and its capping inversion can be used to estimate the entrainment velocity at the top of the boundary layer. An advantage of ozone over temperature and humidity is that, on typical observing time scales, ozone is believed to be conserved in an unpolluted atmosphere. Fast-response instrumentation is also necessary to resolve the detailed structure of the entrainment region at the top of the boundary layer in order to study the physical mechanisms involved In the entrainment process (Mahrt and Paumier, 1982).

A mayor problem is measurement of both mean and fluctuating temperature and water vapor mixing ratio in clouds. At present, infrared and ultraviolet techniques seem to offer the most potential for these measurements; further development and testing is necessary, however to insure this capability.

Measurement of divergence in the marine boundary layer by integrating the mean horizontal wind field around a closed flight path may not be feasible because of time changes that can occur during the flight. Possibly this can be solved by using several aircraft simultaneously. In any case, very accurate mean winds need to be measured in order to have any hope of measuring divergence by this technique. This requires updating of standard INS-based wind systems (wind measurement using an inertial navigation system to measure the aircraft velocity) with positional information from radio or satellite navigation systems.

Remote sensing offers probably the most potential for future improvements in aircraft instrumentation. One important variable is remote measurement of inversion height while the aircraft is flying either above or below the top of the boundary layer. This can be done by lidar (e.g., Browell et al., 1983) or possibly by radar. Remote measurements of cloud liquid water, ocean wave heights, and aerosol distribution are other useful remote measurements that are feasible with present-day technology.

3.5. Island studies of Marine Stratus and Stratocumulus Systems

While instrumented aircraft are an excellent method of obtaining turbulence, microphysical, and radiation profile data for MBL cloud dynamics research, such data essentially represent hlghly detailed case studies (e.g., Brost et al., 1982a,b). Compiling a new climatologically representative aircraft data set for the development of relevant cloud/ radiation parameterizations is fiscally impractical (at least in the context of FIRE). An alternate approach is a modest program of surface

37

based measurements (moored ship or island) in a logistically attractive region with a reasonable incidence of marine dominant flow. A good candidate foc such a study of MBL stratus and stratocumulus is San Nicholas Island (SNI) which is located approxLmately 100 km SW of Los Angeles. SN[ is owned by the U.S. Navy where a small hase (airport, harbor, housing, food services, etc,) is maintained pri,narily for radar tracking. The NW tip of the island has a fully instrumented micrometeorological tower (at the high tide line) that has been used for marine surface layer research (Blanc, 1980). Due to the prevailing NW flow, the island is situated in a classic marine boundary layer about 50X of the time and has been used in conjunction with several MBL studies (e.g., Davidson et al., 1983).

Of course, island measurements may not be representative of conditions over the surrounding ocean. Careful examination will be needed of satellite images and aircraft data to characterize any island effects. The proximity of the continent will also require cautious interpretation when the wind Ls from the North-East quadrant.

The Naval Postgraduate School and the Naval Research Laboratory are developing plans to cooperate with several universities in a multi-year field program of nearly continuous MBL studies at SNI which would Lnvolve the installation of a three wind component doppler SODAR and augmentation of the tower with a full suite of longwave and shortwave surface radiation measurements. An existing network of radiosonde stations (two island, four coastal) can be used to define horizontal variability and lower tropospheric thermodynamics. Since the program is at present only a concept, a full list of instrumentation and participants is not available. For the purposes of FIRE, some method of obtaining cloud liquid water profiles is an absolute minimum requirement. Although a vertically pointing millimeter wavelength radiometers are one possibility (Hogg et al., 1983), a properly instrumented tethered balloon could provide in situ sampling of liquid water, temperature, water vapor, and radiation (note the excellent balloon studies of stratus by Caughey et al., 1982 and Slingo et al., 1982). The information obtainable from such a collection of instruments is summarized below (T = Tower; S = SODAR; B -- Balloon; R = Radiosonde at SNI).

40

B. Thermodynamic Structure

The deployment of these observations should be designed to provide a stable statistical ensemble of characteristics for stratus and stratocumulus systems. Thus satellite observations might involve the analysis of 100 cases dispersed throughout each of several years for (1000km)^2 to (2500km)^2 regions. The surface observations would also be Judiciously sprinkled over a several-year period, in order to develop a similar ensemble of cases for the particular surface station.

The investment required for these studies is relatively modest. The results should lead to realistic climate model parameterizations of existing stratus and stratocumulus systems. In order to model changes ln these systems, as might for example evolve in a climate change, we require models for the dynamics of the system. The process oriented observations described next are designed to gain these models.

The process oriented observations belong to Intensive field phases of the experiment. These field phases are for short duration (several weeks) but in order to develop confidence in the results, should be repeated (at least once) during the course of F[RE.

41

II. Aircraft:

A. Direct Observatlons

1. In-situ entrainment rates

B. Validation platform for remotely sensed parameters

3.8. Schedule

Table 3.1 shows a time line (years after grants awarded and experiment team formed) of the activities wLthLn the cloud-capped marine boundary layer segment of FIRE. There are three parts to this segment: 1) theoretical and nodeling stu.1ies; 2) ,atellite observations, and 3) field experiments.

Owing to the many degrees of freedom available to stratus and stratocumulus systems, model studies are far from exhausting the range of significant; sensitivity and simulation tests that might be performed. Such work should therefore continue throughout the experiment. At first the modeling is apt to be confined to rudimentary models of fractional cover and the development of three-dimensional eddy simulation models. As the experiment evolves one anticipates detailed comparisons of fiel1 observations with the eddy simulation models. With regard to the radiative properties of the stratus and stratocumulus systems initial studies should aim at supporting the later field measurements, while later studies

42

should seek to explain satellite and field observations.

The utilization of satellite observations to characterize the marine cloud-capped boundary layer as well as satellite validation work can begin immediately. Indeed, suitable studies would use archived data to develop initial statistics while awaiting the field phase of the program. Also possible within the FIRE concept is the processing of archived data that is concurrent with previous field experiments such as JASIN and EPOCS. As many of the techniques applied to imagery are new, their further development and even the development of yet-to-be-thought-of techniques are anticipated throughout the experiment.

The field experiments are preceeded by almost two years of preparation. Radar applications are new. As much of the experiment plan relies on the successful use of millimeter doppler radars, a preliminary pilot observing period deslgned to assess their potential usefulness seems warranted. Such an observing period should be included in the two year planning phase. Aircraft and surface based observations, though the traditional mainstay, would undoubtedly enjoy better success with a two year preparation period.

At least two years for the intensive field experinents with 3-6 weeks each year are anticipated. This multi-year plan anticipates less than optimum operation for the first year, and based on the first year's experience, closer to optimum operation during the second year. A one year evaluation period is placed between the two field experiment periods to further insure the success of the second year. More than one year of field work is required to establish a statistical ensemble (9-18 cases) suitable for intensive studies. A final year follows the field work for researchers to complete initial analyses of the field data.

43

Table 3.1. Schedule of activities for the marine stratus and stratocumulus component of FIRE. It appears that "year 1" will be FY85.

Year after grants awarded

1 2 3 4 5

Theoretical and modeling studies

a) Radiative models x x x x x

b) Cloud models x x x x x

Satellite observations

a) Statistical characterizations x x x x x

b) Validation x x x x x

c) Concurrent with field experiments x x

Field experiments

(instrumented island, radar, aircraft)

Preparation x x x

Field x

Analysis of results x x x x

44

3.9 References

46

__________, and W.H. Schubert, 1980: The effect of radiative cooling in a cloud-topped mixed layer. J. Atmos. Sci., 37, 482-487.

918-930.

____________, 1983: Buoyant production and consumption of turbulence kinetic

energy in cloud-topped mixed layers. Submitted to J. Atmos. Sci.

47

_____________, and J.W. Deardorff, 1977: Subgrid scale condensation in models of non-precipitating clouds. J. Atmos. Sci., 34, 345-355.

_________, and ____________, 1981: A model for entraininent Into a

cloud-topped marine boundary layer. Part II: Discusslon of model

behavior and comparison with other models. J. Atmos. Sci., 38,

2230-2242.

48

4. Processes Responsible for the Extent, Radiative Characteristics and

Maintenance of Cirrus

4.1 Introduction

In spite of thelr often tenuous nature and innocuous appearance, cirrus clouds can have a pronounced influence on our climate. This fact has been increasingly recoynized in recent years as yeneral circulation and climate modelers have attempted to include the effects of cloudiness in their models. It has not come as a surprLse to radiation experimentalists and theorists, who have long been aware of the remarkable impact that cirrus clouds have on the radiation budgets at the top of, within, and at the bottom of the earth's atmosphere.

The importance of cirrus clouds is emphasized in climate studies by three factors; first, cirrus cloud systems often have a large areal extent and may last for significant periods of time; second, the impact that cirrus clouds actually have on the climate can be large or small, and of a positive or negative sign, depending on the highly variable radiative properties of the cloud and on the height or radiating temperature of the cloud itself; and third, whether a cirrus cloud exists or not is determined by a very fine balance of the water budget of the upper troposphere, a balance which is largely undetectable with the coarse descriptions of the water budget in large scale models.

4.2 Present knowledge

Although the importance of cirrus clouds is widely recognized, relatively little is actually known about them. This paucity of information is a result of several factors. Only recently has the potential importance of cirrus become apparent, thus stimulating scientific investigation into the subject. Cirrus clouds are located high in the troposphere and, as a result are difficult to observe in situ. Indeed, prior to the mid-1970's, most studies of cirrus were limited to photographic image studies made from either the yround or satellites.

FIRE represents a unique opportunity to study an important phenomenon about which relatively little is known. This opportunity presents itself because technology is now becoming available which can be used to make meaningful in situ and remote observations of cirrus clouds. Also the rudiments of an understanding of the basic processes involved in the formation, maintenance and dissipation of cirrus are emerging from the current research.

The microphysically oriented studies of Heymsfleld marked some of the first quantitative studies of cirrus clouds. Although Heymsfleld's emphasis was on the microphysics, he did begin to formulate a consistent model of how a certain class of cirrus cloud (cirrus uncinus) propagates itself. More recently, Starr (1980) reported a modeling study which

49

provides a physical numerical model of the dynamic, thermodynamic, microphysical, and radiative processes acting upon the cirrus cloud layer. This model, although rather crude in some respects, suggests a conceptual 1eslyn for an observational program aimed at the processes.

Initial experiments with the model show that a critical parameter is the number of large ice crystals in the cloud layer. If there are too many, they will fall out and the cloud will dissipate faster than the ice population can be regenerated from supersaturation due to large scale upward motion or radiative cooling. However, some moisture ls carried back from the subcloud layer into the cloud by mesoscale convective circuIations with a horizontal scale of a few hundred meters to a few kilometers, driven by radiative cooling and evaporation into entrained air at cloud top. Such motlons can also cause enhanced crystal yrowth in updrafts.

In order to test the model it is necessary to supply as initial conditions vertical profiles of larye scale temperature, horizontal wind, ice distribution within the cloud and relative humidity outside the cloud boundaries, as well as the large scale vertical velocity, and then to follow its evolution for several characteristic recirculation times (i.e., several times 20 ;nins), measuring as many model variables as possible. Here larye scale refers to averages over the model integration area, i.e., over 10 km. Because much of the instrumentation, and hence the observation area, will be fixed relative to the earth's surface, whereas the mesoscale structures will drift with the wind, it will be necessary to approach the intercomparison statistically. This involves selecting from a lonyer observing period intervals of perhaps 1 hr when the large scale variables in the observation region are reasonably constant in time, and appear from satellite images and in situ samples outside the region to be homogeneous in space. Particularly critical to assessing model performance are the statistLcs of the mesoscale motion and ice distribution (maynitude and scale of fluctuations, horizontal plan form) and the spectra of crystal sizes as an indicator of the microphysical regeneration processes. From another perspective, it is Lmportant also to verify that the mean cooling of the top of the layer due to vertical radiative flux divergence is being correctly estimated in terms of the crystal habit and size distribution, and that in situ radiation measurements agree with those from remote sensing.

4.3 Objectives

The primary purpose of the overall cirrus program is to understand how these clouds form, are maintained and dissipate. The case study program will concentrate on the processes mnaintaining cirrus against dissipation by ice crystal fallout. This focus has been chosen as probably the most important determinant of cloud system lifetime and hence overall extent, and as a simpler, more tractable, problem than modeling the moisture injections associated with initiation. The case study program

50

must be considered in conjunction with the statistical analysis of

cirrus systems from satellite data and ground based remote sensing

described in Section 5.

The case study program has four interrelated yet distinct objectives. It will provide ground truth to support the validation of satellite radiation data. With the aircraft microphysical and radiation measurements it will allow intercomparison of radiation calculations and measurements ln the presence of clouds. These observations will be essential bench marks for testing existing radiative transfer codes as well as in the development of new ones eventually to be applied to the climate problem. Third, the total case study data will support phenomenological studies of the upper tropospheric cloud systems. The final objective is to provide an initialization verification data set for use in testing existing models of cirrus.

Though the primary emphasis is on cirrus clouds, the observation system requirements and basic issues are similar for extended middle tropospheric layer clouds composed of liquid water. These systems will form a backup objective for those occasions when the atmosphere is consistently uncooperative.

4.4 Case Study Requirements

Table 4.1 is reproduced from the First ISCCP Regional Experiment workshop report. This table summarizes the variety of measurements and platforms required for the case study experiments referred to in both the middle and upper tropospheric cloud regimes as well as for the marine stratus and stratocumulus studies. Additional details about these measurement requirements are available in the workshop report.

The terms FIRE Macro-Scale and FIRE Meso-Micro Scale used in Table 4.1 refer to the spatial scale of required observations. The use of the term macroscale is a slight departure from convention in that we are referring to spatial scales on the order of one hundred to five hundred kilometers; the terms meso and micro refer to spatial scales of kilometers or a few hundred meters and to the microphysical scales of ice crystals (10 to 1000 micrometers), respectively.

While most of the observations listed in Table 4.1 are self explanatory, some deserve additional comment. Accurate measurements of water vapor in the upper troposphere have traditionally been difficult to obtain. The primary efforts in obtaining this data will probably rely upon aircraft observations using both frost-point and Lyman-alpha devices. The number density of both large and small ice crystals is important to model validation, requiring a variety of instruments to obtain the complete size spectrum.

Wind component informatLon is needed on several spatial scales. On the scale of several tens to hundreds of meters vertical motion

51

Table 4.1: Intensive Case Study Observations

Observation Platform

Parameter FIRE Macroscale FIRE meso-micro scale

Cloud Amount GOES GOES

Polar Orbiter Polar Orbiter

Landsat Landsat

Sfc. Obs. Aircraft

Photography

Sfc. Obs.

Cloud Top Height GOES GOES

(JR, stereo)

Polar Orbiter Polar Orbiter

Sfc ranging (lidar)

Lidar (Aircraft)

Aircraft direct

Surface-generic

Temperature Profile GOES VAS GOES VAS

Polar Orbiter Polar Orbiter

NWS-rawinsonde NWS-rawinsonde

conventional enhanced

Aircraft remote &

direct

Sfc: microwave

profiler

Water vapor profile NWS-rawinsonde NWS-rawinsonde

conventional enhanced

Aircraft remote

direct

Sfc. microwave

profiler

Wind components GOES-cloud tracking GOES

-gradient wind -gradient wind

Polar Orbiter Polar Orbiter

-gradient wind -gradient wind

NWS enhanced NWS enhanced

Doppler radar Doppler radar

Radiative Divergence Satellite Satellite

-multi-parameter -multi-parameter

Aircraft

Radiometersonde

52

Cloud thickness/aspect ratio GOES-stereo GOES-stereo

-optical thickness -optical thickness

Polar orbiter Polar Orbiter

-optical thickness -optical thickness

NWS-rawinsonde NWS-rawinsonde

conventional enhanced

Sfc-generic Sfc-generic

Sfc -lidar

Aircraft -lidar

-direct

-photo

Cloud Organization GOES GOES

LANDSAT Aircraft -photo

-microphysical

Sfc -lidar

Sfc -photo

Cloud liquid or lce water Polar orbiter Polar Orbiter

-microwave Sfc -microwave

(maritime, profiler

liquid only) Aircraft -hot wire

Cloud particle size Aircraft -size

spectrometers

Aerosol loading GOES Aircraft -in situ

-remote

Broadband infrared GOES Aircraft

cloud emittance Polar Orbiter Radiometersonde

Sfc -downward spectral

irradiance

Broadband solar cloud Sfc Sfc -lidar

transmittance -radiometer

Aircraft

reflectance GOES,ERBE Aircraft

absorptance GOES,ERBE,Sfc Alrcraft

Infrared & solar GUES GOES

spectral intensity ERBE ERBE

[10 - 12 micron] Polar Orbiter Polar Orbiter

[visible, near IR] Aircraft

53

estimates within the cloud itself are required. These estimates will be

qualitatively conpared with model va1ue. as well as to delIneate the

horizontal scale of the vertical circulations within the cloud layer.

For tnese observations reliance will probably have to be upon remotely

sensed particle motions using surface-based doppler lidar and a

microwave profiler designed by the Wave Propagation Laboratories of ERL,

NOAA. It is unlikely that aircraft will be able to resolve the more

subtle vertical circulations ln the cloud directly.

Although very few models of upper tropospheric clouds presently exist, they universally require as a boundary condition the large scale lifting field, where large scale refers to scales larger than the cloud model domain. This would indicate that vertical motion, or horizontal divergence fields, be defined on the scale of about 10 kilometers. A signifIcant magnitude, the dynamical equivalent of radiative cooling of 10ïC/day, is 1 cm/s. This is a very difficult task; it ls a scale smaller than that resolved by the National Weather Service analyses which use radiosonde data; aircraft are vLrtually useless for this type of measurement and lt is a significantly larger domain than the surface remote sensing devices can normally sample.

There are several possible solutions to this dilemma. Probably the mvst suitable is to use vertically pointing doppler lidar or microwave, averaging over time as the cloud layer drifts past the instrument and subtracting a mean particle fall speed estimated from the size distribution spectrum. It will clearly be difficult to achieve the accuracies necessary. Direct measurement from aircraft places very strinyent requirements on relative motion sensiny and attitude determination. If the cloud layer is demonstrably homogeneous it may be possible to aggregate together laryer areas and obtain estimates from dynamical models based on synoptical scale analyses. Flnally, it may be necessary to accept that the larye scale velocity cannot be measured directly and treat it as a parameter to be estimated during the intercomparison between the cirrus model and observations.

Radiative divergence determinations are also required on several scales. These scales ranye from a system of clouds of up to several hundred kLIometers on a sLde to detailed vertical radiative divergence ln the vicinity of cloud tops and bases. The larger scale system radiative divergence estimates will be recreated from observations of independent variables such as moisture and temperature profiles and cloud structure determined from a combination of satellite, surface and aircraft data. Some of these data may be of a statistical nature; the cloud microphysical characteristics, cloud base height or cloud thickness are examples of variables which may require a statistical representation on the larger scale.

On a smaller scale, aircraft observations become very useful for the direct observation of radiative divergence. The importance of simultaneous sampling of microphysical and radiative properties cannot

54

be overemphasized. These simultaneous observations will enable the testing of sinyle colunn type algorithms against real data; this data set will also allow some tuning to better represent reality of the simpler algorithms commonly used in models.

4.5 Timing and Locations

The middle and upper-tropospheric case study investigations would be staged in two separate tlme periods. The strategy for selecting these time periods would be based upon three primary criteria: frequency of occurrence of phenomenon to be observed; availability of required observing platforms; opportunity to exploit knowledge gained from previous observing period.

Data supportiny the cirrus cloud investigations of FIRE would be collected intermittently during the period beginning FY86 and ending FYfl9 Statistical data will be gleaned fro,n satellLte and ground based observations throughout the period as well as from some historical data collected prior to 1986. The cirrus field experiments will be held in either early spring or fall of 1986 and 1988. The field experiments have been purposely scheduled 1 year apart to allow some initial analyses of the Pirst field experiment; this delay will enable us to assess our initial succcesses and (allures and thereby modify the second experiment, it necessary; the one year down period will also allow for some modificatlon of hardware, should it appear advisable. Each field experiment time window would be approximately three weeks in length with the objective of sampling at least five occurrences within that window. In addition, within the experiment window, we anticipate the opportunity to sample other cloud phenomena than those of primary interest. These data well be particularly valuable for the ISCCP and FIRE verification and the radiative transfer algorithm verification objectives.

The locations of the case study experiments are primarily dictated by where the phenomena of interest are likely to occur, with secondary constraints imposed by observing platform capability and minimum interference with civilian air traffic. While considerably more study is needed before specific experiment sites are chosen, initial indications are that mid-latitude middle and upper-tropospheric cloud experiment sites would include the upper-midwestern section of the U.S. during the early spring season, whLle the south-central coastal states or the southwest would be appropriate for the study of subtropical upper-tropospheric cloud systems. Wc have confined our initial attention to areas which already have a substantial rawinsonde network.

4.6 Expected Results

The case study program should establish whether present models of the processes maintaining existing cirrus cloud are understood to first order and can be modeled successfully. This is a prerequisite to developing soundly based algorithms suitable for use in GCM's. It

55

should also validate estimates of cirrus characteristics such as emissivity fron sateIIltes. The statistical program related to cirrus should determine the mayor source regions and lifetimes of cirrus systems within the region under consideration. These results should lead to improved understanding of the role of cirrus in the climate system and improved modelling of this in GCMs. A demonstrated ability to characterize cirrus cloud remotely, and to model its microphyslcal and radiative properties, should also lead to applications such as the prediction of nighttime surface cooling and frost.

56

4.7 Bibliography

Berson, F., 1967: Layer echoes from under a canopy of high cloud.

Aust. Meteor. Mag., 15, 205-224

____________, and A. A. Barnes, 1980: Clrrus particle distribution study,

____________, 1981: Cirrus particle distribution study, Part 8.

__________, 1971: Cirrus clouds and the climate. 3. Atmos. Sci., 28,

__________, 1973: Infrared heatiny calculations with a water vapor pressure

57

_________________, and R. G. Knollenberg, 1972: Properties of cirrus

generating cells. J. Atmos. Sci., 29, 1358-1366.

_________________, 1974: Ice crystal growth in deep cirrus systems. Preprints Vol., Conf. on Cloud Physics, Tucson, Amer. Meteor. Soc., 311-316.

________________, 1975b: Cirrus uncinus yenerating cells and the evolution of

_______________, 1977: Precipitation development in stratiform ice clouds: ~

microphysical and dynamical study. J. Atmos. Sci., 34, 367-381.

_____________, 1956,: The forms of ice clouds, II. Q art. J. Roy. Meteo Soc.,

82, 257-265.

Oddie, B. C. B., 1959: Some cirrus cloud observations made by the Westminster Shiant Isles Expedition, 1958. Weather, 4, 204-208.

58

______________, and G. l. Stephens, 1980: The interpretation of remotely

____________, and S. K. Cox, 1980: CharacterlstLcs of middle and upper

troposphere clouds as deduced from rawinsonde data. Atmos. Sci. Paper 327, Colo. State Univ., Fort Collins, 72 pp.

____________, 1982: Numerical Experlments on the Formation and Maintenance of

Cirriiform Clouds; Ph.D. dissertation, Colo. State Univ., Fort Collins, 352

pp.

____________, and S. K. Cox, 1984b: Clrrus clouds, Part 2: Numerical

experiments on the formation and maintenance of cirrus. J. Atmos.

Sci., 41.

Stephens, G. L. and P. J. Webster, 1979: Sensitivity of radiative forcing to

variable cloud and moisture. J. Atmos. Sci., 36, 1542-1556.

59

____________, 1980: Radiative transfer on a linear lattice: Applicatlons to

anisotropic ice crystal clouds. J. Atmos. Sci., 39, 2095-2104.

_____________, 1978b: CLrrus particle dLstrLbutlon study, Part III. AFGL-TR-

78-0305, 67 pp.

_____________, and D. M. Brooks, 1Y78: CLrrus particle distribution study,

Part II. AFGL-TR-78-0248, 108 pp.

_____________, and A. A. Barnes, 1979: Cirrus particle distrLbution study, Part IV. AFGL-TR-79-0134, 91 pp.

Welch, R. M., J. G. Geleyn, W G. Zdunkowski and G. Korb, 1976:

Radiative transfer of solar radiation in model clouds. Beltr.

Phys. Atmos., 49, 128-146.

__________, S. K. Cox and J. M. Davis, 1980: Solar Radiation and Clouds.

Meteorological Monographs, Vol. 17, No. 39, Amer. Meteor. Soc., 93 pp.

60

5. STATISTICAL STUDIES OF CLOUD PHYSICAL/RADIATIVE PROPERTIES

5.1 Purpose of Statistical Program

Cloudiness and lts radiative impact are influenced by many processes not expilcitly described in atmospheric general circulation models. A proper parameterization will require knowledge of characteristics for scales ranging from a few kilometers to hundreds of kilometers. Even with such knowledge, the most comprehensive algorithm will inevitably have some semi-empirical elements. Establishing a properly sampled statistical data base for inferring empirical connections and testing new algorithms must be a high priority objective for FIRE. Such a data base is also needed to compare with the global statistics being compiled by ISCCP. Indeed, to the extent that empirical relations between typical cloudiness/radiative variables and large scale synoptic variables can be established, they may serve to suggest theories of the processes involved and guide the development of process studies. When accompanied by appropriate understanding, such statistical relations may even become components of a paramaterization scheme.

The gathering of a statistical data base differs from the intensive case studies by concentrating less on developing state-of-the-art techniques for understanding cloud and radiative processes and more on obtaining an adequate representative data set from which reliable statistics may be inferred. It must also be recognized, that the meaningful characterization of cloud populations from satellite data is still in its infancy, and therefore a variety of statistical characterizations would be appropriate. Because certain properties such as the height of cloud base which are important for the radiative budget at the earth's surface can only be measured from satellites with extreme difficulty, if at all, some in situ observations will be required.

Present GCMs commonly assign clouds a lifetime equal to model timestep, after which all liquid water is precipitated. They thus assume a very different thermodynamic and moisture budget from that applicable to non-precipitating clouds. Thus from a modeling standpoint it is also important in the statistical program to distinguish precipitating from non-precipitatiny clouds.

5.2 Present status

Global statistics of cloud distributions using routine weather reports from ground based observers have been compiled by London (1957), Hahn et al (1983), and others, but these are difficult to interpret in terms of the quantitative modeling of cloud radiative effects (Gordon et al 1983). Furthermore, because of the scatter introduced by point sampling and individual observer biases, only hlghly aggregated quantities such as seasonal mean high level cloudiness are probably reliable. Sadler (1981) has pioneered attempts to use indices of satellite brightness for measures of interannual changes in cloudiness over the Pacific Ocean. However, there ls currently little published information available

61

about characteristic distributions of cloud parameters on scales

intermediate between a dally climatology for one station and a seasonal

climatology for a region.

Operational weather satellites are providing images in great profusion, yet the information in these images is not being fully exploited to describe cloud fields. Particular features are being looked at for specific objectives, (e.g., the size of cumulonimbus anvils has been correlated with precipitation, and the solar flux at the surface has been inferred from the apparent albedos) but systematic efforts to characterize the morphologles and radiative properties of real cloud populations are few (e.g. Cahalan et al. (1982), Chahine (1982), Schiffer and Rossow (1983), Taylor and Stowe (1983), Susskind et al. (1983), Coakley and Baldwin (1983)). There are a number of reasons for this situation.

Satellite images and multichannel sounding instruments provide a significant but essentially incomplete picture of the cloud population within the field of view, and of the radlatlon at the earth's surface. They are essentially restricted to a top down view, with a spatial resolution which ls usually inadequate to disentangle the details of lndividual cloud elements or to establish unequivocally the radiative properties of the clear sky in small gaps between the clouds. For optically thick clouds, only properties near the upper surface are accessible. For partially transparent clouds such as cirrus, an even more serious problem ls assessing the degree of tranparency.

Yet much information about variables such as the fractional cloud cover, cloud type and cloud top temperature is indeed present ln the operational satellite data stream. It ls contained ln the relationships between radiances from the same pixel at different wavelengths, in the spatial structure and temporal development of the radiance fields, and in systematic differences with different viewing angles. Experienced analysts inspecting individual images and image sequences are able to interpret in them a wide variety of meteorological processes by lnferring and using the apparent cloud structures. Though such analyses are inevitably somewhat personal, a useful degree of consistency ls apparent (Stowe, 1983).

Yet our ability to automate such analyses, to apply them consistently to a large sample in a manner appropriate for climate studies, has lagged. Insufficient effort has been devoted to defining what patterns are typical, and to isolating operational procedures for assigning numbers (even simply occurrence or non-occurence of a property) to any given situation. Only ln terms of such numbers, and ln terms of properly controlled sampling, can meaningful statistics be compiled. For example, in a given synoptic situation, such as the subtropical Jet stream, what are the appropriate parameters on which to construct a statistical summary of the occurrence of a cirrus system? One could use areal extent, mean apparent infrared emissivity, region of origin and lagrangian lifetime, or one could propose a whole variety of

62

alternatives. Precise operational definitions need to be established, which capture within known reproducibility the essence of what an analyst would mean by these terms. Though these issues are normal and manageable, there has been relatively little experience with them using satellite derived data.

Another reason why satellite analyses have lagged is that satellite data comes in large quantities and requires sophisticated computer equipment to handle it. The effort required to follow through an adequate sample on even one precisely defined classification scheme is substantial, and few individuals or groups have so far undertaken it. However, the necessary facilities and data handling techniques are becoming less expensive and more widespread. As a result, computing resources should no longer be a limiting factor on progress in this area.

The development of algorithms for retrieving cloud variables has tended to concentrate on pixel by pixel estimates using thresholds or multichannel inversions rather than on the spatial or temporal structures. There are good practical reasons for this choice. Threshold methods work well lf each pixel either is totally filled with optically thick homogeneous cloud or is clear sky above a known background. However, a large fraction of pixels are only partially filled by clouds. More recently, approaches have been developed (Coakley and Bretherton (1982)) which estimate the impact of partially filled pixels, but these techniques require other assumptions about the cloud field (e.g., spatlally uniform cloud top temperature). Using simultaneous stereo views from two geostationary satellites, cloud heights can be determined geometrically, but this approach places extreme demands on the simultaneity and navigation of the views from both satellites and has not yet been widely applied. Other approaches using infrared sounder data provide clear sky and cloud top temperatures for an ensemble of sounder points (Smith and Wolf, (1976), Chahine (1977), Smith and Platt (1979)) but with poor accuracy in some cases (Wieleckl and Coakley (1981)). Particularly troublesome for all approaches to fields of convective clouds is the angular variation of the reflected and emitted radiances caused by the three dimensional geometry of the cloud elements on a sub-pixel scale (Naber and Weinman, 1983). Thus, though some items (such as the product of cloud top infrared brightness and fractional cloudiness within a pixel) are clearly well determined under most circumstances and others (such as cloud base) are only occasionally, if ever, determinable, there ls no consensus about optimal retrieval algorithms for even the simplest cloud parameters. A fundamental assumption of FIRE is that by using: 1) multiple views of the same scene from different directions, 2) all the spectral channels and spatial resolutions available, 3) a priori knowledge about cloud fields (such as their tendency to occur ln layers), and 4) experience about typical patterns in the images, we can achieve a mayor advance in the determination of pixel scale cloud parameters and the objective interpretation of processes at work in a scene.

63

It is now possible to detect precipitation ln excess of about 5mm/hr from polar orbit satellites using the SMMR on Nimbus 7 on a spatial resolution of about 25km. Though this satellite is beyond lts expected lifetime, and cannot be counted on during FIRE, the UMSP scheduled for launch in 1985 will carry a rather similar instrument, plus a 95 CHz channel which will distinguish ice clouds from water clouds. At the passage times of these satellites, it will thus be possible to obtain important additional information on cloud properties, which must be related to the analyses based on the visible and infrared.

For yround based observations new technology provides opportunities to remotely sense important variables which cannot be obtained systematically in any other way. For example, vertically pointing millimeter radars, lidars, and microwave sounders can detect cloud base, cloud thickness and integral measures of the droplet or ice crystal size distribution. Of course, some in situ measurements are needed to validate the interpretation of such remotely sensed observations. Continuous vertical profiles of wind, temperature and humidity can also be remotely sensed. Scanning video cameras may be used to measure the apparent fractional cloud cover as a function of zenith angle. Such information may prove useful in modeling the angular distribution of radiances at the top of the atmosphere. Forward scattering of solar radiation also contains information about cloud droplet size distributions for thin clouds.

5.3 Strategy

Given the present state of knowledge and the importance of improving it, the approriate strategy would appear to be first to encourage simultaneously a broadly based research program into technlques for the analysis of satellite data in terms of pixel scale cloudiness parameters and mesoscale scale cloud morphology. Second would be to encourage a more structured effort to compile useful and reliable statistics on the properties of real cloud populations, as viewed both from satellites, and, to a much more limited extent, from the earth's surface. [SCCP will provide aggregate statistics on the large scale, but FIRE must address mesoscale structures and cloud system lifetimes. It must also strengthen the basis for interpretation of the cloud retrieval algorithms used in the international program.

The basic research program will overlap significantly with the testing of cloud parameter retrieval schemes and radiance computations using a column model, as described in section 2.3, and with the process studies described ln sections 3 and 4 (see in particular section 3.3). It will not, however, be constrained by the specific obJectives of those elements. The emphasis should be on methods for interpreting satellite images meaningfully in terms of real cloud processes and populations. Some of the significant questions which may help guide this research are: How to determine cloud type, including fractional cloud cover in layered systems and vertical development when present and whether precipitating; can cloud radiative properties be related statistically to models of cloud microphysics and to mechanistic thermodynamic models of cloud structure; what gives rise to mesoscale variability.

64

For the structured part of the program, the lnformatlon required is of two kinds:

The value of a statistical summary depends not only on the validity of the data and the concepts around which they were organized, but also on the sampling with which the data were selected. Though a detailed plan must await a review by the analysis groups concerned, it is worth considering the principles that should govern the sampling. For satellite data, the timing and ground coverage is essentially predetermined by the orbit. Nevertheless, a critical step is scheduling the operational deployment of high information rate modes such as VAS dwell soundings or AVHRR LAC data. In addition, after this data has been captured in an archive, another key decision is to select the subset to be subjected to intensive processing. The simplest procedure is to preselect geographic scenes some 1000km square for analysis, and to sample them regularly through the seasons for several years. A subset of these samples should be at 1 hr intervals (GOES data only) to establish the diurnal cycle. Such a geographic/calendar description provides an unbiassed picture of what was present in those areas but is of relatively little use in generalizing to other situations. A different, complementary, approach ls to classify scenes by synoptic category (e.g. Jet stream cirrus, deep convective systems, suppressed boundary layers) and to ensure a representative sampling of each of ten or so categories. Geographic sampling will be used to determine the frequency by season for each category at that location, and to detect any biases introduced through conditional sampling. Conditional sampling will then be used to determine typical cloud parameter statistics in each category, and to characterize mesoscale structures appropriate to the situation.

For the ground based observations, emphasis should be on as complete a set of variables from the list above as can be obtained at reasontable cost,

65

from a small number of stations, some of which would be colocated with the process case studies. Such stations should be malntalned for 3-5 years with frequent regular sampling coordinated with that from satellites. The ability to relate systematically, at even a few locations, cloud height, thickness and microphysics to wind, turbulence profiles, radiation at the surface and radlatlon to space has great potential in verifying satellite Interpretatlons, and in establishing relationships important to GCM modelling. Emphasis should be on simple measurements maintained essentially continuously rather than on sophisticated observations that are poorly sampled. These measurements should of course be augmented as appropriate during the intensive Field phases of the process studies.

5.4 Resource requirements

Satellite Data

This program is probably the most demanding within FIRE on the acquisition and processing of satellite data sets. Collectlon will have to be scheduled in advance. Decisions on assembly and processing of specific scenes can, however, await classification by synoptic category.

The scale of effort required will depend strongly on decisions yet to be made about the algorithms to be used in retrieving cloud parameters for the structured statistical program. At this time It appears likely that the directional lnformatlon obtained from simultaneous views from two GOES/VAS dwell soundings and the NO M/AVHRR LAC and TUVS data will be essential to give sufficiently unambiguous inferences of cloud parameters. It may prove possible to relax this requirement, but such a possibility has not yet been established. LANDSAT and ERBS data will be used on a pilot basis in research programs. NIMLUS 7/SMMR data and DMSP imagery and microwave data will be used where available.

For the purpose of scaling the data collection and processing effort assume:

Calendar Sampling

Synoptic Category

10 categories

100 cases/category spread over 4 years

The above strategy requires that the simultaneous data (see section 2.5 for a definition) be collected over the region of overlap between the geostatlonary views, from ON to 50N and from 70W to 140W at

66

the equator, for about 80 days per year for 4 years, or the equivalent

with a concentration in a 2 year interval. In addition, on about

20days/year sampling the diurnal cycle at 1hr intervals using simultan

eous GOES. To the extent that this requirement is relaxed, the primary

FIRE region can be expanded if necessary to include the complete field

of view of the satellites concerned.

Surface Observations

To obtain adequate coverage 1 island and 2 continental stations are required for 3-5 years.

5.5 Schedule

After appropriate proposal review, research into procedures for retrieving cloud parameters and a pilot study for the structured statistical program from satellite data should be undertaken as soon as possible (FY 84). These should start with existing data and not await availability of simultaneous VAS. At least 1 surface station should be committed ln FY 84 and established early ln FY 85.

The full research program should be funded from FY 85 and continue at a roughly constant level for the remainder of FIRE. Systematic data collection for the structured program should start late FY 85 and continue for 4 years.

5.6 What will be achieved

Because the statistical characterization of clouds ls still In its infancy, there are many uncertain factors ln a program of this type. Nevertleless, lf lt ls undertaken at the level described, lt is reasonable to expect that lt will:

67

5.7 References

Schiffer, R. A. and W. B. Rossow, 1983: The International Satellite

Cloud Climatology Project (ISCCR): The first project of the world

climate research programme. Bull. Amer Meteor. Soc., 64, 779-784.

68

Taylor, V.R. and L.L. Stowe, 1983: Heflectance characteristics of uniform

Wielicki, B. A. and J. A. Coakley, 1981: Cloud retrieval using infrared sounder data: Error Analysis. J. Appl. Meteor., 20, 157-169.

69

6. Data Set Assembly

The data set assembly activity consists of two separate but interrelated tasks -- archival and processing.

6.1 Archive of raw and case study data sets

The need to collect and assemble an archive of candidate data sets arises from the fact that data for FIRE will come from a variety of sources. For example, FIRE calls for simultaneous views of a given scene from GOES and polar orbiter platforms. To provide research scientist with a comprehensive data set for that scene the GOES and polar orbiter data must be edited and the appropriate parts combined to form a readily usable data set. It in situ and surface based observations are available, they too must be edited and combined with the satellite data. The goal is to provide research scientists with comprehensive sets of data for the cases of interest and at the same time to tailor these data sets so that the crucial information needed for research is readily accessible. In addition to archiving the raw data streams from which cases are selected, FIRE will also archive the case study data sets. As these data sets, both the raw and the case study data sets, are likely to prove useful for post FIRE research, they must also be transferred from the FIItE archive to permanent archives at the completion of FIHF.

In order for FIRE to meet its obJectives and to collect the needed data, It must be remembered that an essential requirement is the appropriate scheduling of satellite operations to cover the desired areas at the approriate times. In addition, it will be necessary to schedule the appropriate satellite operatLonal modes in order to obtain the most desirable data. The latter is especially true for the VAS on the GOES satellites. The mode ls controlled by the satellite control operations at the Wallops data acquisition station.

6.2 Data Processing

A data processing activity is also needed in order to extract the much more limited data set for individual case studies mentioned above. Here, data from several satellites and possibly from in situ and surface observations will be assembled into one or more computer compatible tapes for use by individual scientists. These data sets will require processing to establish resolution, locations, sun angle, viewing angle, calibration, etc. Such lnformatlon must be included, for example, when several satellites view the same scene in Order to establish angle dependence. In many such instances, It may prove desirable to map the individual images to a common proJectlon.

The basic philosophy to be used in assembling the case study data sets includes the following:

wing:

70

6.3 Region of the Experiment

A fundamental tenet of FIRE ls that satellites will be the primary source of data and that a limited amount of ancillary data will be added as needed. The strategy for the column model studies and the statistical program calls for intensive processing of scenes some 1000 km across with simultaneous views from as many satellites as possible. In practice this means GUES Last, GOES West, NOAA-7, NUM-8, occasionally ERBS, LANDSAT and possibly DMSP.

In order to better understand the constraints imposed by the need for nearly simultaneous observations from different satellites, it is helpful to examine Figure 6.1 which represents the coverage from two ascending node orbits of the NOAA satellites and Flgure 6.2 which shows the area of overlap of all geostationary satellites. The region of overlap for GOES Last and GUES West Is cross hatched.

Figure 6.1 shows the footprint of the TOVS microwave (circles) and Ilt (dots, about 30km across at the sub satellite point) sounders. The AVHRR imager has 1 km resolution ln the visible and Infrared windows. The swath width ls about the same for the three instruments. The amount of overlap between adjacent orbits Is trivially small at latitudes equatorward of 50 N. Moreover, they differ in time by 102 minutes, hardly a simultaneous observation.

The situation with GOES East and West Is more satisfactory with regard to sImultanelty. The overlap region not only covers most of the

73

U.S. but a considerable portion of the eastern Pacific as well. (San

Nicholas Island, where the marine stratus and stratocumulus field phase

is planned, is ln an excellent position to be covered by both GOES

satellites, as well as the polar orbiter.)

The most useful region in the overlap areas is not at the center, for here the viewing angles are the same. Regions near the edges of the overlap area are best since the viewing angles are quite different from each satellite. The Mldwest and the Pacific Coast are more satisfactory than areas near the 100th meridian. Notice that a typical swath from the polar orbiter covers only about half of the overlap region; however, the range of viewing angle is much greater for the polar orbiter.

In summary, the coverage from both geostationary and polar orbiter satellites is near ideal for the two areas selected for intensive process studies and ancillary observations; l.e., off the coast of Californla and the upper Mld-West.

Finally, it should be mentioned that Figure 6.1 shows two ascending node orbits. There are two descending nodes twelve hours later at night where the swaths are oriented from the northeast. Obviously in order to maximize the overlapping coverage from the two COES and the polar orbiter satellites, a different schedule than the ordinary operational ones for the GOES satellites will be needed to match the simulataneous overflight of the polar orbiter. These new schedules can easily be planned weeks in advance.

The number of satellites and auxiliary information possible from conventional sources is considerable. Table 6.1, "Present Archive Status", gives many of them. Not all sources will be required at any one time, but when a match is possible, the value of that data set will be very much greater than the sum of individual observations. Clearly, when FIRE is underway, it is imperative that specifics be given in every case. The list is presented here for illustrative purposes only, and is not intended to be complete. ERBS and future DMSP satellites should be added to the list since the timing of FIRE will extend into those satellite flight programs.

74

TABLE 6.1 PRESENT ARCHIVES STATUS

TYPE LOCATlON ARRANGEMENTS SPECIAL PROBLEMS

GOES VISSR SSEC Continuous Scheduling of Stereo

Scans

GOES VAS SSEC & NOAA Scheduled Scheduling of PDL

Intervals

NOAA7,8 NESDIS Continuous AVHRR(GAC),HIRS,MSU

NOAA7,8 Wallops Special arranyements for

AVHRR(LAC) Redwood City On overpass recording LAC data

Nimbus 7 GSFC Continuous Available 3 Years behind

real tLme

Landsat EROS So far mainly Expensive

Sioux Falls over land

Radar None Dial-Up Requires intensive

(digital) Kavouras collection effort

Radar Ashville Continuous Poor quality

(15 mm film)

DMSP None AFGL? ?

(digital)

DMSP WDC-A (film) Continuous ?

Boulder

Surface, SSEC Continuous Holes when computer down

radiosonde (604)

Surface, Ashville Continuous Slow response for

radiosonde new data

Surface DOE? ? ?

radiation

Research NCAR & others For experiment Requires special

aircraft only processing

6.4 Calibratlon

FIRE is concerned primarily with the relative Impact of clouds on the radiances observable from satellites, rather than with the absolute accuracy of radiative fluxes from clear sky. As such, it is somewhat less demanding on the callbratlon of the satellite instruments than is, for example, the Earth Radiation Budget Experiment. Nevertheless, obtaining the angular distribution of radiances observed from different satellites, and making Lnferences from the relative radiances in different spectral bands from the same instrument requires considerable attention to the stabilLty of the instrument responses and to systematic intercomparisons between instruments and channels on the same instrument. Though much can be achieved in this regard by careful accumulation of statistics during the data analysis, matters would be greatly simplified and significantly Improved by a properly planned calibration program. Such a program would also benefit other uses of the same data streams. Of hlghest concern are pre-flight documentation of instrument response characteristics, accurate records of the timing and magnitude of operational changes in electronic gain and offset controls and systematic intercomparisons between satellites and checks against ground based control points to establish slow degradation or sudden changes in response due to unknom causes. FIRE will rely heavily on NOAA and other satellite operators for assistance Ln this reyard.

Calibration and quality control of the ln situ and ground based remote sensing will be the responsibilLty of individual principal investigators. Observing system intercomparlsons will be required as part of the intensive process and statistical studies.

76

7. Relationship of FIRE to ISCCP

FIRE is complementary to ISCCP in several ways. Both are working towards similar objectives from different directions and neither is complete without the other.

FIRE is directed towards the Lmprovement of parameterizations of cloudiness and radiation algorithms in GCMs. ISCCP is directed more towards the evaluation of overall GCM performance in predicting cloudiness parameters.

Within one region, FIRE will describe the mesoscale features and lifetimes associated with cloud populations and contribute to the understanding of the processes giving rise to them. ISCCP will contribute low resolution global statistics of cloudiness varLables derived from satellite data.

Both FIRE and ISCCP are concerned with evaluatiny and improving algorithms for the retrieval of cloudiness from satellite observations. The column model sub-element in FIRE wLII provide a systematic approach to testing the consistency of a variety of algorithms (including those befog used for ISCCP), and to developing new algorithms within a GCM modeling framework. The statistical studies wLII provide a more free ranging approach to this problem.

The ground based ancillary observations in FIRE will, for a very few statLons, provide an independent statistical base of observations of cloudiness parameters such as cloud top height and thickness, which can be compared both to the ISCCP product and to the FIRE interpretations of multi-satellite views for those locations and times. They will also contribute Lnformation on parameters such as cloud base which are Lmportant for GCM model parameterizations but can be obtained from satellite observations only with great difficulty, if at all. Aircraft observations during the FIRE process studies will also provide direct confirmatlons on a very small sample of cases.

The low resolution data sets being prepared within ISCCP will be useful for planning the more intensive processing within FIRE.

The column model and the statistical characterization efforts in FINE will be accelerated to provLde as early feedback as possible to ISCCP.