GEWEX Cloud System Study (GCSS) Working Group 4: Precipitating Convective Cloud Systems

Transcription

1 GEWEX Cloud System Study (GCSS) Working Group 4: Precipitating Convective Cloud Systems Mitchell W. Moncrieff,* Steven K. Krueger, + David Gregory, # Jean-Luc and Wei-Kuo Tao & ABSTRACT The authors present the objectives of the working group on precipitating convective cloud systems. These center on developing physically based parameterizations for global models in which basic research into the large-scale role of cloud systems is an important part. The approach calls on a range of expertise: cloud-resolving modeling and contributing research, observational evaluation of the model results, and tests of parameterizations in single-column models. Ongoing studies focus on oceanic cloud systems in Tropical Oceans Global Atmosphere Coupled Ocean Atmosphere Research Experiment (TOGA COARE). First, cloud-resolving modeling of organized convection on a timescale of a few hours concentrates on momentum transport and surface fluxes. Results are evaluated against data obtained during the 22 February 1993 Intensive Observation Period, which include airborne Doppler radar measurements of a squall line. Second, multiday simulations focus on the environmental effects of cloud systems as they respond to changes in specified (observed) large-scale tendencies and sea surface temperature. In this case, observational data on the scale of the entire TOGA COARE Intensive Flux Array are used in model evaluations. Results and recommendations from the first model intercomparison workshop, as well as the next steps in the intercomparison, are summarized. In the longer term, cloud system research in Working Group 4 will contribute to the Climate Variability and Predictability Global Ocean Atmosphere Land System program, as regards the large-scale effects of cloud systems up to intraseasonal timescales. Another contribution will be to space-borne measurements; for example, cloud-profiling capability will provide data critical to the comprehensive evaluation of upper-tropospheric moisture distribution in cloud-resolving models. Besides additional studies in tropical cloud systems, convection in cold air outbreaks and convection over continents have a high priority. 1. Introduction *National Center for Atmospheric Research, Boulder, Colorado. + University of Utah, Salt Lake City, Utah. # European Centre for Medium-Range Weather Forecasts, Reading, United Centre Nationale de la Recherche Meteorologique, Meteo- France, Toulouse, France. & National Aeronautics and Space Administration, Goddard Space Flight Center, Greenbelt, Maryland. Corresponding author address: Dr. Mitchell W. Moncrieff, National Center for Atmospheric Research, Mesoscale and Microscale Meteorology Division, P.O. Box 3000, Boulder, CO moncrief@ncar.ucar.edu In final form 15 November American Meteorological Society The Global Energy and Water-Cycle Experiment (GEWEX) Cloud System Study (GCSS) was established by the World Meteorological Organization World Climate Research Programme (WCRP). The parent program is described in the GEWEX Science Plan (1990). GCSS has the challenging task of developing physically based parameterizations of cloudrelated processes for both general circulation models (GCMs) and global numerical weather prediction (NWP) models a problem of long standing so these models can be as realistic as possible. There is also a need to understand the role of clouds in the climate system at a basic level, another long-standing aspect. Note that many climate problems are so complicated that modeling is the only way they can be fully addressed. Examples of particularly difficult problems are the role of precipitating clouds in climate (Browning 1990), in the global hydrologic cycle (Chahine 1992), and in atmosphere ocean coupling (CLIVAR Science Plan 1995). From an even broader 831

2 perspective, IPCC (1995) recognizes that the role of clouds is one of the key uncertainties in climate change scenarios. There are four working groups within the GCSS, namely, those concerned with boundary layer clouds, cirrus cloud systems, extratropical layer cloud systems, and precipitating convective cloud systems (Working Group 4). In scientific terms, these groups are not mutually exclusive. For example, precipitating convective cloud systems strongly interact with the boundary layer and often generate extensive cirrus decks, especially in a sheared environment. The core science objectives are described in GCSS (1993) and the GCSS Science Plan (1994), while recent progress is summarized in WCRP (1996). Here we focus on the objectives of Working Group 4, with particular attention to the Tropical Oceans Global Atmosphere (TOGA) Coupled Ocean Atmosphere Response Experiment (COARE), although other longer-term objectives are also summarized. The main tool of GCSS Working Group 4 is the cloud-resolving model (CRM), which is a numerical model that resolves cloud-scale circulations in either two or three spatial dimensions. For example, in simulating a precipitating convective cloud system, a CRM can resolve individual convective cells while its domain encompasses the entire cloud system. In contrast, even high-resolution global NWP models, with a typical grid size of less than 100 km, cannot resolve the individual convective cells or even the accompanying mesoscale circulations. These are subgrid-scale processes, whose collective effects on the largescale fields of the GCM must somehow be determined or parameterized. A CRM is able to determine these collective effects directly, to the extent that the parameterizations of its own subgrid processes are accurate. CRMs are effective tools for the simulation of cloud systems. They have been in existence for more than two decades but have yet to be fully exploited. With few exceptions, they have been used for cloud and mesoscale process research rather than for parameterization development (Moncrieff 1995). One of the first tasks of the GCSS is to clearly demonstrate that CRMs can quantify the large-scale effects of cloud systems. To this end, much progress has been made. An example of cloud system evolution in a two-dimensional CRM simulation is shown in Fig. 1. This simulation used the interactive cloud-radiation interaction scheme developed for climate models and tested against satellite observations (Kiehl et al. 1994). The different types of system occur in response to the evolution of the large-scale forcing. Observational estimates of forcing averaged over the TOGA COARE Intensive Flux Array (IFA) were provided by Lin and Johnson (1996), although this is not the final COARE dataset. (Improved datasets will be used by the working group as and when they become available.) Bulk properties, such as mass and momentum fluxes, which are impossible to accurately obtain from observations, are readily obtained from these simulations. For example, Fig. 2 (from X. Wu et al. 1997, manuscript submitted to J. Atmos. Sci.) shows the convective mass flux derived from a 39-day TOGA COARE simulation using a CRM. The decomposition of mass flux into updraft and downdraft components can be used to evaluate and improve mass-flux-based parameterization schemes. FIG. 1. Snapshots of the total consensate (cloud water, ice, rain) at 1230 UTC from cloud-resolving model simulations of convection in TOGA COARE, forced by large-scale observational analysis: (a) 22 December 1992, (b) 23 December 1992, and (c) 24 December Units are g kg Vol. 78, No. 5, May 1997

3 A problem faced by GCMs involves not only determining how cloud systems affect the large-scale fields but also how formation, maintenance, dissipation, and structure of cloud systems relate to largescale variables. Cloud system formation and structure involve highly nonlinear interactions among the various physical processes that affect cloud formation (e.g., large-scale ascent, in-cloud convective circulations, boundary layer turbulence, surface fluxes, microphysics, and radiative transfer). Cloud-forming processes that are directly related to the large-scale circulation, such as large-scale ascent or advection, are usually termed large-scale forcing or large-scale tendencies. These can be observationally measured and resolved in GCMs but have to be specified in present-day CRMs. This is necessary because, even with the most powerful computers, CRM domains are not big enough to simulate the large-scale fields and resolve the clouds, at least in three spatial dimensions. Once the forcing is specified, a CRM can calculate cloud formation, structure, and transports. Of course, the predictability of cloud systems is another matter. While the cloud-scale and mesoscale circulations that dynamically couple the physical processes are explicitly resolved and accurately treated in CRMs, these models have their own inherent uncertainties. For example, a CRM with a 1-km grid can adequately resolve deep precipitating systems but not small, shallow cumulus clouds. Two-dimensional models can simulate the evolution of precipitating convective cloud systems in domains spanning several thousand kilometers and up to intraseasonal timescales. Three-dimensional model domains are typically a few hundred kilometers on a side and can be run for a week or so. The minimum spatial resolution required to realistically approximate cloud systems, which typically contain both deep and shallow convection, needs to be quantified. Similar uncertainties exist in the representation of boundary layer turbulence and in-cloud turbulence, cloud microphysics, and radiative transfer. These small-scale processes are subgrid scale and must be parameterized, even in CRMs. In particular, it is important to determine the level of sophistication required of microphysical parameterizations in CRMs. (Note that, in the prognostic treatment of clouds in GCMs and NWP models, microphysical processes are represented in a much simpler way than in CRMs.) Because of these inherent uncertainties, CRMs need to be intercompared and extensively evaluated against observations. GCSS Working Group 4 is conducting a series of basic studies and model intercomparisions to evaluate CRMs and use the synthetic datasets thereby provided to develop better cloud-related parameterizations. The adopted forum is a series of specialized meetings organized by the GCSS. The inaugural model intercomparison meeting of Working Group 4 was Evaluation and Intercomparison of GCSS Cloud- FIG. 2. Mass fluxes computed from a two-dimensional 39-day TOGA COARE cloudresolving model simulation: (a) updraft, (b) downdraft, and (c) total in units of hpa h 1. [From X. Wu et al. (1997, manuscript submitted to J. Atmos. Sci.).] 833

4 Resolving Models using TOGA COARE Observations held October 1996 in Annapolis, Maryland. It was cohosted by the National Aeronautics and Space Agency s (NASA) Goddard Space Flight Center and the National Center for Atmospheric Research (NCAR) Clouds in Climate Program (CCP). To complement the model intercomparison and evaluation work (presently based on TOGA COARE data), the GCSS Working Group 4 has cohosted two other scientific meetings. The WCRP Workshop on Cloud Microphysics Parameterizations in Global Atmospheric Circulation Models, held May 1995, in Kananaskis, Alberta, Canada, is reported in WCRP (1995). The European Centre for Medium- Range Weather Forecasts (ECMWF) Workshop on New Insights and Approaches to Convective Parameterization, was held 4 7 November 1996; the results of this meeting will be published as an ECMWF Workshop Proceedings. Summarizing their use in climate-related research, the most valuable aspects of CRMs are their large computational domains (which encompass entire cloud systems) and the coupling of physical processes through explicitly resolved cloud-scale and mesoscale circulations (which minimizes uncertainties associated with the treatment of cloud-scale and mesoscale dynamics). For these reasons alone, we anticipate that CRMs will be used much more extensively during the next decade than in the past for cloud system parameterization. The remainder of this paper has eight sections. In the next section we describe how cloud-resolving models can be used to develop parameterizations, followed by a selection of science questions in section 3. We then set certain TOGA COARE objectives into a GCSS context, followed, in section 5, by a summary of the model intercomparison and evaluation procedure. Section 6 summarizes the key results from the Annapolis meeting, followed by a summary of longer-term objectives. Finally, conclusions are drawn in section 8. 2.Using cloud-resolving models to develop physically based parameterizations A general approach for using CRMs in parameterization studies is now sketched. First, consider the process of developing or improving a parameterization. A key step is its testing. This can be accomplished in several ways, the most obvious of which is to perform climate simulations including the parameterization, which has the advantage that the parameterization is tested in the setting where it is used. However, there are disadvantages. First, it is usually very difficult to attribute particular deficiencies to particular aspects of the GCM s formulation. Second, climate simulations are computationally expensive and elaborate, so only a limited number of model runs can be made. Third, the weather systems simulated by climate models do not accurately represent those in the real world, so only statistical comparisons with observations are meaningful. Another approach is to use the parameterization in an NWP model, then compare the forecasts against observations. Unlike climate simulations, this approach allows detailed comparisons with individual weather events. However, like climate simulations, it can be very difficult to link particular deficiencies of the forecasts to particular aspects of the model s formulation. An additional complication is the elaborate data-ingest procedure (i.e., assimilation and analysis) required for NWP models, which can affect the interpretation of model results (e.g., spinup). There are two approaches for testing parameterizations external to climate models. The first is the semiprognostic test, in which a parameterization or group of parameterizations is applied in an atmospheric column, which can be thought of as a single column of a GCM. In a GCM, adjacent grid columns provide fluxes of mass, heat, and water vapor but can only interact on resolved scales. In the semiprognostic approach, the vertical motion and the horizontal advective tendencies of temperature and water vapor mixing ratio that would have been produced by the fluxes from adjacent columns are instead specified from observations. (In some instances, idealized data may be used in place of observations.) However, errors in the computed local time rates of change must be due to either measurement errors or deficiencies in the parameterization in question. This approach isolates the parameterization being tested from all other components of the GCM an advantage of the method. Note that a semiprognostic test requires minimal computional resources. A semiprognostic test can be applied over a sequence of observation times. However, errors in the computed tendencies at one time have no effect on subsequent tendencies; in other words, there is no feedback. On one hand, this is an advantage because the time-averaged tendencies can be very flawed; on the other hand, it is a disadvantage because parameterizations can produce errors that arise directly from 834 Vol. 78, No. 5, May 1997

5 feedback. Such problems cannot be detected with semiprognostic tests. Another problem with semiprognostic tests is that everything that is not parameterized must be accurately measured. The second approach for testing climate model parameterizations uses single-column models (SCMs). In some ways this approach is similar to the semiprognostic test. In an SCM, however, the computed tendencies are used to predict new values of the prognostic variables in the atmospheric column. In other words, feedback within the column is allowed. A problem with SCMs is that the time-averaged tendencies typically have to be accurate because of the various types of feedback. A second problem is that feedback involving the large-scale circulation cannot be included; this requires a full NWP model or GCM. How can CRMs be used to test and develop parameterizations? A CRM domain can be considered to represent a GCM grid volume; in this sense, it is a physically sophisticated (two- or three-dimensional) explicit dynamical integrator of processes that are represented (one-dimensionally) by an SCM. The explicit treatment of nonlinear dynamics that fundamentally couple process interactions is the unique feature of CRMs. The large-scale fields and tendencies required by both models, as well as the observations used to evaluate them, are essentially the same. Can CRMs be used in place of observations for testing SCMs? CRMs are designed to simulate many of the cloud-scale processes that must be parameterized in either an SCM or a GCM. They can certainly provide quantities that are difficult to observe, such as the vertical distributions of liquid water, ice, and cloud mass flux. Moreover, multiscale dynamical interactions are a key aspect in fundamental processes such as convective momentum transport (Moncrieff 1981, 1992), which explicit modeling can address. However, as already noted, CRMs do require parameterizations of microphysics, turbulence, and radiative transfer, which can introduce uncertainties. Nevertheless, used judiciously and properly evaluated, synthetic CRM results can diagnose and alleviate problems inherent to the design of SCMs. Once deficiencies in an existing parameterization have been quantified, how can the parameterization be improved? Selected aspects of a parameterization can be studied in detail using a CRM. For example, it is assumed in many cumulus parameterizations that the subcloud layer air that forms the cumulus updrafts has the mean properties of the subcloud layer. The validity of this assumption (and alternatives) can be examined using a CRM. (This approach should also help in general by increasing our basic understanding of cloud systems.) Idealized and/or observed largescale conditions can be used in such studies. The advantages of physically based (as opposed to empirical) parameterizations were described by Randall (1989). Examples of using observations, SCMs, and CRMs to test cloud parameterizations for climate models were described by Randall et al. (1996). SCMs and CRMs are being increasingly used to develop physically based parameterizations. For example, observational datasets from the 1974 Global Atmosphere Research Program (GARP) Atlantic Tropical Experiment (GATE) were used in conjunction with an SCM to develop a convective parameterization scheme for NWP models by Tiedtke (1989). This approach is illustrated by path A in Fig. 3. Gregory and Rowntree (1990) used a similar method to develop a convection scheme for both NWP and climate models. Datasets generated by CRMs using idealized observed large-scale conditions have also been used for parameterization purposes (see path B in Fig. 3). Xu and Krueger (1991) evaluated diagnostically based cloudiness schemes, while Xu and Randall (1996) developed a cloud amount parameterization based on the condensate mixing ratio; Gregory and Miller (1989) quantified mass fluxes, entrainment, and detrainment; and Xu and Arakawa (1992) semiprognostically tested cumulus parameterizations as a function of GCM horizontal grid size. FIG. 3. Two routes to physically based convective parameterizations. Path A uses observations and SCMs, while in Path B synthetic data from a CRM forced by large-scale observations are added. [Adapted from Randall et al. (1996).] 835

6 CRMs have also recently been used to test methods for parameterizing momentum flux in GCMs (Gregory et al. 1997). They have also been used to evaluate idealized dynamical models of organized convection transport. For example, the momentum flux by squall-line convective systems, formulated by Moncrieff (1992), was evaluated in numerical experiments by Wu and Moncrieff (1996). Redelsperger and Lafore (1988) used a threedimensional CRM to examine the thermodynamic transport by organized convection and to make comparisons with radar data. Tao et al. (1993a) used a CRM to develop retrieval methods of convective heating profiles in the Tropical Rainfall Measurement Mission (TRMM). On a global basis, the radiative effects of clouds have a key role in determining the earth s energy balance. Upper-tropospheric (anvil) clouds associated with tropical deep convection are especially significant. The direct effects of radiative heating on individual precipitating convective cloud systems is less evident, and a number of mechanisms have been proposed. CRMs have been extensively used to investigate these mechanisms (e.g., Tao et al. 1993b; Sui et al. 1994; Chin 1994; Xu and Randall 1995; Fu et al. 1995). Convective clouds (e.g., downdrafts) are well known to affect surface properties and the energy budget. The effect of surface fluxes on the tropical circulation was shown by Miller et al. (1992). Note that convective precipitation affects the thermal stability of the oceanic mixed layer and its response to solar heating (Anderson et al. 1996). This suggests that two-way explicit coupling of CRMs to upper-ocean models is timely. The role of environmental shear is fundamentally important because it affects the selection of specific regimes of convection and, thereby, their transport properties (Moncrieff 1981). Yet shear is not contained in any existing parameterization scheme. Its most noticeable effect is on momentum fluxes by organized convection, a problem that is beginning to be quantified (Moncrieff 1992, 1995). The inclusion of environmental shear is argued to be necessary in a universal treatment of convection for large-scale models. This is not only because it affects momentum transport, but also because it affects what type of convection actually occurs. The diversity of the above examples attests to the value of CRMs not only in parameterization development, but also to advance our basic understanding of the large-scale effects of cloud systems. The scene is set for moving forward with a comprehensive series of studies in an approach that can be used in a range of cloud system types. 3.Science questions Precipitating convective cloud systems are, by far, the most complex forms of atmospheric convection. They strongly interact with physical processes that operate on spatial scales ranging from microns to thousands of kilometers, and on timescales from fractions of a second to months. Following is a (far from exhaustive) list of specific questions that needs to be answered in order to improve the parameterization of precipitating convective cloud systems, not to say quantify their broader role in climate research and prediction. a. Cloud formation, microphysics, and water vapor How do convective cloud systems affect water vapor distribution? How is the formation of tropical cirrus clouds coupled to deep convection? What is the minimum level of sophistication in microphysical parameterization required to model the bulk properties of precipitating cloud systems in both CRMs and in GCMs? b. Organized convection How does the organization of convection on mesoscales affect the net heating, drying, and horizontal momentum, and is it important on a global scale? Since vertical shear has a primary role in organization, how can it be incorporated into cumulus parameterizations? What dynamical factors control the initiation of convection and how can these be used to define physically based activation methods in parameterization schemes? c. Boundary layer and surface fluxes Are present surface flux schemes and boundarylayer models, used in combination with convective downdraft parameterizations, adequate in convectively disturbed conditions? Is the 1-km horizontal resolution typically used to model precipitating cloud systems adequate, considering that these larger systems coexist with small-scale and shallow clouds? 836 Vol. 78, No. 5, May 1997

7 d. Closure How well do the different types of closure used in cumulus parameterizations [e.g., quasiequilibrium, moisture convergence, and convective available potential energy (CAPE) tendency] perform when tested using observational and CRM datasets? 4.TOGA COARE in the GCSS context To establish scientific priorities, science teams within Working Group 4 were charged to select key issues involving the physics and parameterization of precipitating convective cloud systems (Fig. 4). Consideration of the team reports (GCSS Science Plan 1994) led to tropical cloud systems being selected as the top priority. This choice was based on the importance of deep convection in both weather forecasting and climate models, on the comprehensive manner in which scientific issues involving these cloud systems can be studied using CRMs, and on the availability of GATE and TOGA COARE datasets with which to comprehensively evaluate the models. Data from these two large field experiments are very useful for parameterization testing and development because the spatial and temporal scales of measurement are compatible with the resolution of climate models as well as with the domain size of CRMs. Nevertheless, data from small (focused) field campaigns are still required to evaluate how well CRMs approximate cloud systems per se; specifically, their small-scale turbulent, microphysical, and radiative characteristics. The scientific priorities of Working Group 4 reflect both TOGA COARE and GCSS objectives. The principal objectives of TOGA COARE, quoted from Webster and Lukas (1992), are to study the principal processes responsible for the coupling of the ocean and atmosphere in the western Pacific warm pool system, and the principal atmospheric processes that organize convection in the warm pool region. As stated in the GCSS Science Plan (1994), the principal GCSS objective is to develop better parameterizations of cloud systems for climate and numerical weather prediction models via an improved understanding of coupled physical processes. One of the conclusions of TCIPO (1994) was the need to conduct modeling activities to address scientific issues relating to convection in TOGA COARE. One area where cloud-scale and mesoscale models were recognized to be important was in regard to the FIG. 4. Summary of scientific and programmatic priorities of GCSS Working Group 4. large-scale role of convective organization and in addressing the structure and modulation of convection over the broad range of scales from about 100 km to synoptic scales. Cloud systems in the western Pacific region are organized on a hierarchy of scales ranging from about 10 km to more than 1000 km. In fact, the largest cloud systems on earth ( superclusters ), which occur in the western Pacific and Indian Oceans, are partly resolved in operational global weather forecasting models at a T213 spectral truncation (or about 90 km in physical space). This has been found to cause model errors; for example, the interaction of explicitly resolved (albeit surrogate) diabatic heating with large-scale wind shear causes an incorrect vertical transport of horizontal momentum (Moncrieff and Klinker 1997). Methodology Working Group 4 activities are organized by a steering committee (consisting of the authors of this paper). This committee selects specific studies and observational data with which to initialize and evaluate the CRMs, sets out strategies for analyzing the CRM data in a parameterization context, organizes 837

8 meetings, and establishes scientific conclusions and recommendations. The principal objectives of the GCSS TOGA COARE studies are to address deficiencies in the parameterization of physical processes by using TOGA COARE observations to initialize, force, and evaluate CRMs and SCMs. For example, path B in Fig. 3 consists of four steps (involving observations, CRMs, and SCMs) to generate physically based parameterizations. Summarized in Fig. 5, these steps use observations to evaluate parameterizations of subgrid-scale processes in CRMs and identify weaknesses that limit their ability to adequately simulate cloud systems; use CRMs to simulate precipitating convective cloud systems subjected to observed large-scale forcing; evaluate CRM predictions of the large-scale effects of convection against observational datasets; and evaluate and improve SCMs by comparing them to observations and to domain-averaged CRM diagnostics. 5.Model intercomparison and evaluation using TOGA COARE data To make a start on the intercomparison project, the steering committee opted for two complementary studies: (i) Two- and three-dimensional modeling of a squall line on a timescale of many hours. The initial conditions and evaluation datasets were obtained during intensive observations in TOGA COARE. (ii) Two-dimensional modeling of the evolution of convection subjected to observed, time-dependent, largescale forcing on a timescale of a week. Initial conditions for these simulations, the evolving largescale forcing and the data used in model evaluation involve averages over the TOGA COARE IFA, in keeping with the nature of parameterization research. a. Squall line (case 1) During recent years, a great deal of modeling and observational attention has been devoted to squall lines. Studies have ranged from idealized dynamical models to numerical simulations containing reasonably sophisticated parameterizations of microphysics. However, intercomparisons of numerical simulations have yet to be comprehensively conducted, and simulations seldom have been evaluated against field data. Simulations of a squall line observed in the Intensive Observation Period on 22 February 1993 are being undertaken. This includes sensitivity studies of subgrid-scale parameterizations used in cloudresolving models; for example, different microphysical parameterization schemes, downdraft and updraft mass fluxes, effects of convection on surface fluxes, momentum fluxes, and interaction between convective and stratiform regions. Observations used in evaluations include those obtained by airborne Doppler radar, from which fields of wind and reflectivity can be derived. The initial conditions used the sounding analyses of LeMone et al. (1994). The simulations will determine to what extent three-dimensional CRMs can reproduce the structure and evolution of the observed mesoscale convective systems. b. Cloud system evolution in time-dependent largescale forcing (case 2) This study is designed to evaluate how CRMs simulate the interaction of precipitating convective cloud systems with observed large-scale fields, for example, to quantify the response of cloud systems to largescale forcing and the large-scale effects of cloud systems through many episodes of deep convection. The results are evaluated using the observed large-scale properties averaged over the entire TOGA COARE IFA. A multiday period during part of the December 1992 westerly wind burst (Eldin et al. 1994) was simulated by two-dimensional models (one three-dimensional experiment was conducted) using a domain size of about 500 km and horizontal grid sizes of 1 3 km. The models were forced by the observed evolving large-scale horizontal winds and advective tendencies of temperature and moisture averaged over the IFA (Lin and Johnson 1996). The feasibility and value of even longer integrations calculation was demonstrated by X. Wu et al. (1997, manuscript submitted to J. Atmos. Sci.) in a two-dimensional, 39-day simulation. The mathematical formulation of the model used was similar to Grabowski et al. (1996), in which the life cycle of convective systems during several easterly wave episodes in GATE was successfully simulated. (Threedimensional GATE simulations have recently been completed and three-dimensional TOGA COARE simulations are ongoing). As an example, Fig. 2 showed the convective mass fluxes obtained from the 39-day simulation. Single- 838 Vol. 78, No. 5, May 1997

9 column models were also run and compared against CRMs and observations. This demonstrates the GCSS strategy for achieving physically based parameterizations in Fig Key results from the model intercomparison The results described below are illustrative rather than comprehensive. a. Case 1 Only one simulation was submitted in time for the October meeting but others are in various stages of completion. This simulation showed that (i) ice physics was important to achieving realistic system, (ii) two- and three-dimensional model results differed substantially in the convective region but better agreement was found in the stratiform region, and (iii) twoand three-dimensional momentum fluxes were very different. The latter result contrasts with findings of simulated cloud systems in other parts of the world, notably squall lines in West Africa. In view of the need for detailed evaluation of CRMs (see step 1 in Fig. 5), future studies on case 1 will focus on the momentum transport properties of organized convection and surface fluxes, because these are poorly understood. Of intrinsic interest are differences between momentum fluxes in two and three spatial dimensions. This will involve evaluation of models against quad-doppler radar data from the two National Oceanic and Atmospheric Administration P3 aircraft. Future evaluations should include radar data retrieval in order to compare models and observations [cf. analysis of Convection Profonde Tropical squall lines in West Africa conducted by Redelsperger and Lafore (1988)]. b. Case 2 The second case concerned cloud system evolution within the IFA between 20 and 26 December 1992, which is the mature part of the December westerly wind burst. Results from simulations by seven models were submitted for intercomparison at the Annapolis meeting (21 23 October 1996). The models included five two-dimensional CRMs, one three-dimensional mesoscale model usinf parameterized convection, and one SCM. (Only four CRMs are included in the following results because the SST was coded 6 K too low in the fifth model.) Results of two FIG. 5. Four-step approach to physically based parameterization, which combines observations, cloud-resolving models, and single-column models, detailing path B indicated in Fig. 3. other SCMs were discussed during the meeting and will be added to the intercomparison dataset. The simulated temperature and water vapor fields can be compared to the observed fields discussed in Lin and Johnson (1996). Figure 6a shows the time series of the temperature error (the difference between the simulated and the observed IFA-averaged temperature) at 500 mb for four CRMs (thin lines), a mesoscale model (thick dot dash line), and an SCM (thick solid line). All models developed a cold bias of about 2 K with a systematic time variation. To put this error in perspective, recall that the latent heat released by condensation of 4 mm of rain is enough to warm an atmospheric column by 1 K. About 120 mm of rain fell during the 6-day period, enough to warm the atmosphere by about 30 K. This may be due to the the spinup of the model being greater than anticipated because the first day (20 December) was, in reality, convectively active; this would exacerbate spinup effects. The next intercomparison simulations 839

10 FIG. 6. Preliminary results from the Working Group 4 model intercomparison project (case 2) presented as time series of domain-averaged or IFA-averaged quantities. (a) Temperature at 500 mb; (b) precipitable water; (c) top-ofatmosphere upwelling infrared flux or outgoing longwave radiation; (d) cloud ice water path; and (e) cloud mass flux at 500 mb. Results are from four CRMs (thin lines), a mesoscale model (thick dot dash line), and a SCM (thick solid line). will be started during a less-disturbed period (e.g., 19 December 1992). Figure 6b shows the time series of the precipitable water (vertically integrated water vapor) for the same six models, along with the IFA observations (squares) and precipitable water (among other quantities). Figure 6c shows the time series of the top-ofatmosphere upwelling infrared flux (or OLR, outgoing longwave radiation) for the six models and confirms that the bulk characteristics of convection for the most part follow the observed values. After the first day (when results were affected by model spinup), most models reproduce the overall observed time variation of the OLR. Figure 6d is the time series of simulated cloud ice water path (IWP). The low OLR values generally correspond to large values of the IWP. Again, most models reproduce the overall observed time water or cloud ice or their vertical integrals after the first day. Figure 6e shows the time series of the simulated cloud mass flux at 500 mb. The time variation is very similar to that of the IWP; it is strongly modulated in a consistent pattern in all the simulations. Note that a high correlation between IWP and cloud mass flux was noted in a CRM study by Xu and Krueger (1991). Figure 6e also shows that the cloud mass flux simulated by the SCM is too large compared to the CRM values. This is because the SCM underestimates the downdraft mass flux. Model deficiencies are clearly apparent and are related to the predicted cloud fields. Unfortunately, observations of the IFA-averaged profiles of cloud liquid, the liquid water path, variation of the OLR, and ice water path for evaluating the CRM simulations are not available. The consistency among the models (especially among the CRMs) confirms that the bulk characteristics of convection are largely determined (in a diagnostic sense) by the large-scale advective tendencies, together with wind shear. Note that the differences between the simulations and observations is often larger than the differences between the models. This implies that any one model can be meaningfully used for sensitivity studies. An important future study will be to determine to what extent errors in observations (e.g., soundings) translate into uncertainties in the CRM results; this means that sensitivity studies are required. Two of the CRMs have been previously tested against GATE datasets. The temperature errors for these models were somewhat less for the GATE simulations than for Case 2. The IFA winds have been reanalyzed recently by merging profiler and radiosonde measurements (P. Ciesielski et al. 1997, manuscript submitted to J. Atmos. Oceanic Technol.). Simulations using the re- 840 Vol. 78, No. 5, May 1997

11 vised IFA analyses will gauge the impact of the merged winds on the simulated temperature and water vapor fields. The case 2 intercomparison should be considered as a prototype study that revealed unanticipated problems. (This is not unusual in intercomparison studies, especially when complex nonlinear interactions are involved.) In the follow-up intercomparison (case 3) spinup may be less of an issue; the merged profiler and radiosonde dataset of P. Ciesielski et al. (1997, manuscript submitted to J. Atmos. Oceanic Technol.) will be used to define the large-scale tendencies, the CRMs will be run a few days longer, and more SCMs and CRMs are expected to participate. The second GCSS Working Group 4 model intercomparison meeting will be held at NCAR May 1997, jointly with the COARE Flux Group and COARE Oceans Group. c. Recommendations The recommendations from the Annapolis meeting are as follows. Encourage full participation of the SCM community in the intercomparisons in order to facilitate the development of physically based parameterizations. Encourage collaboration with surface-based, aircraft-based, and space-based active and passive remote sensing to develop common interests in cloud system properties. Encourage the TOGA COARE community to provide definitive, well-tested datasets (cf. GATE). 7.Longer-term objectives Work on the aforementioned objectives will continue for the foreseeable future. Although in a more formative stage, longer-term objectives of Working Group 4 are now presented. a. Other convective cloud systems As noted in section 4, science teams within Working Group 4 selected key issues involving the physics and parameterization of precipitating convective cloud systems. Consideration of the team reports led to oceanic tropical cloud systems being selected as the top priority. The following three cloud systems were also ranked as a priority, to be studied as soon as resources become available. Marine and subtropical cloud systems. Certain systematic errors in the lower-tropospheric regions of NWP models are thought to be due to these cloud systems. These systems also have a large impact on the radiative balance of the atmosphere, for example, as measured by cloud radiative forcing. Of special interest is the Lagrangian evolution of boundary layer clouds into various regimes of organization in cold air outbreaks behind midlatitude depressions. Note that these cloud systems principally derive their energy from surface fluxes rather than from destabilization due to large-scale ascent, at least immediately downstream of the ice sheets where the surface fluxes are largest. Both CRMs and SCMs need to properly represent transitions among cloud systems in response to changing large-scale forcing, surface fluxes, and shear. Due to the dearth of observational datasets with which to evaluate the cloud-resolving models, it is anticipated that a future field program will be proposed, subsequent to testable scientific hypotheses being defined. Continental precipitating cloud systems. Precipitating cloud systems in both Tropics and midlatitudes also need to be studied in the climate context. In particular, the organized systems common over the central United States during the warm season are a good test bed for parameterization of convection over continents. These systems are quite different from oceanic or coastal convection and should be studied at the earliest opportunity. Some progress can be made using standard observations, in view of their high density compared to what is available over the tropical oceans. A more complete study could, for example, make use of the SCM datasets from the Department of Energy Atmospheric Radiation Measurement program, together with data from the GEWEX Continental International Project. Convective cloud systems over the Maritime Continent. Detailed data are available from the Maritime Continent Thunderstorm Experiment (MCTEX), which ran from mid-november through mid-december 1995 on Bathurst and Melville Islands near Darwin, Australia. This is an example of how coastal (land/sea breeze) effects influence the initiation of convection organized by various shear regimes. Although in situ high-altitude measurements were not made, comprehensive sets of remotely sensed data on tropical deep convection were obtained using surface-based instrumentation (e.g., profilers, lidar, millimeter wavelength radar, 841

12 and dual-polarization 5-cm wavelength Doppler radar) in objectives relating to convection initiation, cloud-scale dynamics, and convectively generated cirrus. These data will help evaluate the microphysical dynamical interactions explicitly modeled in three-dimensional CRMs. b. Plans involving CLIVAR-GOALS For the foreseeable future, datasets from GATE and TOGA COARE will be the most comprehensive available regarding tropical oceanic convection and its large-scale effects. Although the TOGA program ended in 1995, the follow-on program is the WCRP Climate Variability and Predictability (CLIVAR). The subprogram most relevant to GCSS is the Global Ocean Atmosphere Land System (GOALS), which is planned to run from 1995 to The scientific basis of these programs can be found in GOALS (1994) and the CLIVAR Science Plan (1995). The bridge that GCSS Working Group 4 is building between the convective mesoscale and the largescale communities will assist the scientific transition between COARE and CLIVAR GOALS. Future modeling studies undertaken by GCSS will help define convection-related objectives, for example, to determine why precipitating convection becomes organized into large systems and how these affect the atmospheric and oceanic circulations at large scales. Purely on computational grounds, two-dimensional CRMs could be integrated out to seasonal timescales. However, the logical extension of the present approach, albeit a challenging one, is to quantify the effect of resolved convection on intraseasonal timescales. This could involve, for example, the explicit role of clouds in the Madden Julian Oscillation (MJO; Madden and Julian 1971). In an idealized framework, Yano et al. (1995) showed how different parameterization schemes affect MJO-like structures. It would be possible to explicitly resolve the (planetary scale) MJO and attendant cloud systems using CRMs. This is tractable with modern computers, at least in two spatial dimensions. In particular, interactively (two way) nested models are a way to bridge the inherent gap in scale between explicit cloud-scale response and large-scale influence/control on convection. The CLIVAR Scientific Steering Group recently endorsed an international meeting (COARE98) to be held in Colorado in late June This will involve the COARE scientific community, the CLIVAR largescale numerical experimentation groups, and GCSS Working Group 4. As a leadup, the GCSS Working Group meeting in May 1997 will be a collaboration with the COARE Flux Group and the COARE Oceans Group. In summary, the GCSS, together with its supporting research programs, will help facilitate the transition between TOGA COARE and CLIVAR GOALS science objectives, especially those concerning the physics and role of multiscale cloud systems. It is anticipated that prototype numerical experiments conducted by GCSS are likely to inspire new scientific activities. c. CloudSat The GCSS Science Panel recently endorsed an initiative to use space-based millimeter radar to measure cloud system properties. A potential future NASA mission (called CloudSat) is led by G. Stephens (Colorado State University). It is a focused program aimed at providing new observational data that will enable the research community to advance the understanding of cloud- and aerosol-related processes and their effects on radiative fluxes. A unique aspect is the vertical profiling of cloud properties. Measurements obtained from the proposed CloudSat instruments (e.g., millimeter radar, microwave radiometer, and lidar), during the surface- and aircraft-based development stage or during the ultimate deployment of these instruments in space, will help evaluate the upper-tropospheric regions of modeled cirrus-generating cloud systems. In particular, simulations of convection in GATE by Grabowski et al. (1996) and in TOGA COARE by X. Wu et al. (1997, manuscript submitted to J. Atmos. Sci.) suggest that the so-called large-scale forcing of condensed water and ice was an important quantity affecting upper-tropospheric humidity. This is notoriously difficult to accurately represent in CRMs and GCMs because it is affected by far-field processes (e.g., large-scale horizontal advection) as much as by more localized cloud-radiative interactions. In turn, the CRM data on tropical cloud systems from X. Wu et al. (1997, manuscript submitted to J. Atmos. Sci.) have shown how CloudSat instruments see these cloud systems. Since it is essential to get cloud-radiation interaction right in order to achieve an accurate surface energy budget, the combination of CloudSat and CRMs has far-reaching implications. This is an example of how the coupling among the physical processes by explicitly resolved CRM dynamics is producing syn- 842 Vol. 78, No. 5, May 1997

13 thetic realizations of cloud systems whose large-scale evaluation requires new types of observations. d. Tropical Rainfall Measurement Mission (TRMM) A joint United States Japan space project, to be launched in late 1997, will provide measurement of rainfall over the Tropics, including precipitation radar (14 GHz) and microwave radiometers, among other instruments. A main TRMM objective is to advance our understanding of the global energy and water cycle by providing four-dimensional distributions of rainfall and inferred heating over the globe (Simpson et al. 1988). To help achieve this objective, CRMs provide modeled three-dimensional, timedependent synthetic datasets (ice/water cloud structure and melting level) as a test bed for satellite remote sensing missions. Two CRM applications on remote sensing, namely, formulating the latent heating profile retrieval and deriving/validating surface rainfall retrieval algorithms have been completed (e.g., Adler et al. 1991; Prasad et al. 1995; Yeh et al. 1995). In the future, the most important evaluation for TRMM products will come from simultaneous, colocated comparisons among space-based, ground-based retrievals, observations, and CRM simulations. The TOGA COARE dataset, particularly those cases where low-elevation radar reflectivities are available to estimate surface rain, as well as the aircraft measurements, provide an opportunity to use the CRM for evaluating TRMM measurements. Since conventional ground truth will hardly be available, the CRMs will be a valuable test of the consistency of precipitation distributions. The CRM microwave radiative model coupling can also provide a new way of testing cloud microphysical models, compared with the existing methods using penetrative aircraft and the slanting beams from surface radars. Comparing the model-derived brightness temperatures with those observed from multichannel microwave-equipped overflights permits a new, stringent test of vertically integrated hydrometeor profiles, which also comprises a test of model dynamics and microphysics (Simpson et al. 1998). 8.Conclusions GCSS was established to quantify the large-scale role of cloud systems and, in particular, to develop physically based parameterizations of various cloud systems for GCMs. As far as the parameterization of convection is concerned, a four-step approach involving observations, CRMs, and SCMs is adopted (Fig. 5). This involves a range of expertise, namely cloud-scale modeling, large-scale modeling, and evaluations against observational datasets. The execution of this complex task will be facilitated by three new initiatives. The NCAR Clouds in Climate Program (CCP) melds the cloud system research conducted in the Mesoscale and Microscale Meteorology Division with climate modeling in the Climate and Global Dynamics Division. Ongoing CCP studies are aligned with GCSS Working Group 4 objectives. The European Cloud-Resolving Modeling (EUCREM) is a collaborative research effort involving several European universities and institutions. Five observed convective occurrences (from a cold-air outbreak to a tropical convective complex) are being used to evaluate cloud-resolving models and hence to guide the development of improved convective parameterizations. The Atmospheric Radiation Measurement (ARM) Program SCM effort has an ongoing campaign at the Southern Great Plains Cloud and Radiation Testbed (CART) site in Oklahoma to make the observations required for the parameterization testing approach described in section 2. Cloud-resolving modeling, together with contributing research, is a comprehensive way to address the role of tropical cloud systems up to (at least) intraseasonal climatic timescales and to move forward with scientific objectives in common with CLIVAR GOALS, realizing that a better understanding of cloud-related processes, which are difficult or even impossible to accurately measure, are required. In the first instance, Working Group 4 will help organize the COARE98 international meeting and the agenda on cloud-resolving modeling and contributing research. Like the parameterization problem per se, the largescale aspects of precipitating cloud systems must be represented as simply as possible. This is a fundamental challenge, especially considering the physical complexity and nonlinearity of cloud system interactions with the large-scale environment. However, it one that can be scientifically addressed through cloudresolving modeling and contributing research. 843