Sensitivity of climate simulations to radiative effects of tropical anvil structure

. Climate sensitivity to the representation of tropical anvil is investigated in a version of the National Center for Atmospheric Research Community Climate Model. Common features of tropical anvil generation and structure, consistent with observations and cloud resolving models, are incorporated into a simple prognostic anvil parameterization. These features include anvil convective origin, vertical profile, phase, areal extent, and life span. Two numerical climate integrations are forced by 1985-1989 sea surface temperature (SST): the control, with simple diagnostic anvil, and the experiment, which simulates tropical anvil structure prognostically. The prognostic anvil formulation enhances ice and reduces liquid in the tropics. Increase in hydrometeor size associated with anvil weakens cloud radiative extinction per unit mass by factors of 1-3. The weaker mass extinction efficiency approximately balances enhanced ice amount so that anvil ice mass quadruples without biasing the mean radiative energy balance but significantly alters the vertical distribution of radiative effects. Enhanced anvil perturbs the tropical upper troposphere temperature structure more strongly in winter, when the column is clearer and anvil radiatively heats the troposphere above 200 mbar. In the summer tropics, enhanced anvil reduces radiative cooling up to 200 mbar and enhances cooling above that. The prognostic anvil formulation improves longwave cloud radiative response to SST cooling but worsens response to warming >2øC. The net response of convection is a shift toward the winter hemisphere in solstice months. These changes lead to a significant response in the extratropical height field in January. These results emphasize the importance of representing tropical anvil structure in climate simulations.


Introduction
Radiative forcing from the extended tropical upper tropospheric cloud known as cirrus anvil plays a dominant role in determining the diabatic heating which drives the general circulation. Tropical cirrus anvil originates in the complex interaction of a mesoscale convective system (MCS) with the environment. A general circulation model (GCM) does not resolve this interaction and must rely on sub-grid-scale methods to diagnose or predict anvil cloud. This study combines distinctive features of tropical anvil structure into a parameterization suitable for GCMs and then examines the impact of accounting for the radiative effects of tropical anvil on the climate system. The radiative effects of anvil depend on its distinctive life cycle and structure, which may be briefly summarized as follows: Deep convection is the ultimate source of tropical upper tropospheric extended clouds, i.e., tropical anvils [e.g., Webster and Stephens, 1980]. The relative area of convecting cores to the associated anvil is 10-20% [Leafy and Houze, 1980;Fu et al., 1990]. Anvil lifetime, typically 6-12 hours [Ackerrnan et al., 1988;Leafy and Houze, 1980], exceeds the duration of deep convection by many hours [Houze, 1989]. Thus although the cumulonimbus core produces the strongest radiative impact per unit area, the anvil region dominates the radiative impacts of the convective cluster as a whole [Machado and Rossow, 1993;Wong et al., 1993b]. Leary and Houze [1980] and Gamache and Houze [1983] inferred the hydrologic budgets of trop-Copyright 1997 by the American Geophysical Union.
Paper number 97JD02009. 0148-0227/97/97JD-02009509.00 ical convective systems from observations: Roughly 60-75 % of anq, il condensate is detrained from deep convective updrafts. The remainder, roughly 25-40% of anvil mass, is generated by circulations outside the deep convective core, i.e., in the anvil itself. Approximately 40% of MCS precipitation comes from the stratiform region. Observations and numerical simulations [e.g., Wong et al., 1993a;Sui et al., 1994;Grabowski et al., 1996;Heymsfeld, 1996, 1997] show that time mean condensate mixing ratio qc in tropical convective systems does not decrease significantly (but can increase) from the freezing level to --•300 mbar, above which qc decreases rapidly. The intrinsically mesoscale nature of these anvil features has hindered their representation in most GCM moist convection schemes [Donner, 1993]. How explicitly should these features of anvil'structure and lifecycle be represented in GCMs? The answer depends on climate sensitivity to tropical anvil.
Many previous GCM studies have advanced understanding of climate sensitivity to tropical anvils by modeling climate sensitivity to anvil representation. Ramanathan et al. [1983] showed nonblack cirrus is crucial to maintaining the observed tropical upper troposphere temperature structure and meridional temperature gradient. Charlock and Ramanathan [1985] showed simulated zonal average upper troposphere temperature increased significantly when treating frozen cloud particles as 20 /.•m larger than liquid. Slingo and Slingo [1988] showed tropical anvil not only warms the tropical upper troposphere but also accelerates the subtropical jets and excites responses in the northern hemisphere winter height field. Ramaswamy and Ramanathan [1989] showed shortwave heating in tropical anvil is a significant fraction of total diabatic heating above 300 mbar and plays an important role in the maintenance of the upper tropospheric temperature structure. Furthermore, they suggested the relative abundance of detached cirrus anvil to anvil embedded in deep convective systems determines the sign of net radiative heating above 300 mbar. Ramanathan and Collins [1991] hypothesized cirrus radiative forcing can be an important negative feedback for stabilizing column energy changes induced by local positive SST anomalies over warm ocean. Senior and Mitchell [1993] showed representing ice cloud prognostically rather than diagnostically could substantially alter modeled climate sensitivity to CO2 doubling and SST change. Sherwood et al. [1994] examined climate sensitivity to anvil radiative forcing over the west Pacific warm pool. They showed atmospheric heating by tropical cirrus is primarily balanced by vertical advective cooling, in accord with Ackerman et al. [1988]. Lohmann and Roeckner [1995] showed climate response to blackbody cirrus forcing resembles the response to increased SST forcing even though the direct heating mechanisms, enhanced cirrus radiative heating, and surface evaporation, respectively, are vertically distinct.
These GCM studies employed a variety of methods to determine anvil condensate amount. Most of the aforementioned studies diagnose anvil cloud from atmospheric thermodynamic properties (e.g., relative humidity, column vapor, and stability). However, prognostic schemes are more physically based than diagnostic. Senior and Mitchell [1993] showed radiative feedbacks from prognostic cloud could be significantly different than feedbacks from cloud diagnosed from relative humidity. Newer GCMs [e.g., Tiedtke, 1993;Del Genio et al., 1996] prognose stratiform anvil cloud from bulk microphysics. These schemes detrain condensate predicted by the moist convection parameterization into the stratiform anvil.
Our motivation in the present study is to examine the role of tropical anvil radiative forcing in a climate where the representation of tropical anvil agrees with the gross behavior of tropical mesoscale convective systems, summarized above. We replace a representation of tropical cloud which diagnoses cloud mass from column vapor with a prognostic representation which forecasts anvil generation from the vertical profile of convective mass flux and anvil precipitation from mesoscale budget estimates. Thus the forcing in the experiment is an integrated set of constraints (ratio of convective condensate generation to vertical mass flux profile, horizontal extent, vertical condensate profile, ice fraction, ratio of mesoscale precipitation to sublimation, and lifetime) consistent with tropical anvil structure but not present in most current GCM cloud parameterizations. By comparing the simulated climates to each other, we can deduce climate sensitivity to the representation of tropical anvil structure.
The parameterization of tropical anvil used in the numerical climate experiment is developed in section 2. Section 3 presents the mean climate response to the tropical anvil representation. Section 4 examines anvil response to SST forcing in the 1987 E1 Nifio. Section 5 contains the conclusions.

GCM Anvil Parameterization
Spatiotemporal scale mismatch between convective and stratiform processes and GCM resolution makes anvil parameterization difficult. GCMs currently employ two common methods to represent these processes: (1) diagnosing anvil cloud from column thermodynamic properties (e.g., relative We use a cloud resolving, cumulus ensemble model (CEM) to provide a high spatial and temporal resolution data set which spans the range of MCS activity from the convective to the GCM scale. A comprehensive review of our CEM simulation is presented by Grabowski et al. [1996]. Figure 2 (top) shows the variation of M c and anvil mass growth rate IWP through the first day of the CEM simulation. The initial anvil formation, lasting about 6 hours, occurred as a prescribed thermal instability-triggered concentrated convective updrafts (cores) which detrained frozen condensate into a cirrus anvil above 500 mbar. The variation of IWP is strongest during convectively active periods and subdued during the convectively quiescent period (400 <t <900 min) between the first and the second generation anvils. Nonconvective formation of anvil ice is also evident during the quiescent period. Periods when IWP is negative occur when net anvil dissipation (due to precipitation and sublimation) exceeds production.
In order to isolate the processes controlling ice generation from destruction we focused on the initial hours of anvil formation, when a single convective tower dominated the mass budget of the entire domain. Figure  (2) The experiment, denoted ANV, restricts diagnostic liquid condensate (and hence mixed phase cloud) to a narrower and warmer range than the control model according to Since CCM classifies some condensate as cold as -30øC as liquid (equation (2)), while ANV has no liquid colder than -5øC (equation ( of hydrometeor size and its radiative treatment is as described by Kiehl et al. [1996]: Model cloud droplet effective radius tel is fixed at 10/am over ocean and sea ice but varies from 5 to 10 /am over land. Ice crystal effective radius r ei represents an equivalent surface area sphere and varies linearly with a normalized pressure coordinate. Over ocean this results in rei = 10 /am for p >800 mbar to rei = 30 /am for p <400 mbar. Solar single-scattering and longwave emissivity properties are

from Slingo [1989] (liquid) and Ebert and Curry [1992] (ice).
Since ice hydrometeors are prescribed to be larger than liquid, classifying more condensate as ice (equation (3)) also decreases the extinction efficiency per unit mass of anvil [e.g., Zender and Kiehl, 1994]. This proves to be an important factor in diagnosing the cause of change in climatological cloud radiative properties in this sensitivity study.

Diagnostic Condensate Formation
For convenience this section summarizes the diagnostic treatment of condensate employed in the control model CCM. For a fuller description the reader is referred to Kiehl et al. [1996]. In CCM, grid box average qc is logarithmically proportional to total column vapor Qv and linearly proportional to local cloud fraction A through qc =APtip (4)

QI = P• e(-z/hl) (5) h l = 810 In Qv (6)
where p is density, p• ---0.18 g m -3 is "in-cloud" condensed water density at the surface, z is height, and h l is the scale height of condensed water. Note the strong coupling of q c to local surface temperature by the Clausius-Clapeyron relationship implicit in (6). Inserting tropical values for Q v, we find h l •<4 km, so qc monotonically decreases from the surface (for fixed A). A drawback to this procedure is that the upper tropospheric peak in q• profile of an anvil system must be captured by significantly modulating A across the anvil deck (equation (4)).

Sensitivity Study Results
To assess climate sensitivity to the features of convectively generated anvil described above, we compare the results of two numerical climate integrations forced with observed 1985-1989 SST. The control, denoted CCM, uses a diagnostic cloud scheme with no special provisions for anvil . The experiment, denoted ANV, forecasts anvil ice from (1), which incorporates modeled and observed characteristics of tropical anvil production and structure (i.e., explicitly linking anvil condensate generation to anvil base convective mass flux, strong vertical profiles of condensate up to 300 mbar, and increased ice fraction). The focus of the present study is on the radiative effects of the anvil condensate. Thus we restrict the effects of q i in (1) to radiative heating alone. The results focus on tropical climate, where anvil forcing is greatest.

Condensate Distribution
The climate response to anvil representation is driven by radiative forcing resulting from the distribution, partitioning (ice or liquid), and size of cloud condensate. The direct effect of representing the structure and convective production of anvils is to sequester more condensate in the upper troposphere, a larger fraction of which is ice. Figure 4 separates modeled January and July total condensed water path (CWP) by phase (CWP -= LWP + IWP). The refined anvil representation increases tropical IWP by factors of 2-4 and reduces tropical LWP by 20%. The net increase in tropical CWP is 10-50%, comprising a mean increase of upper tropospheric condensate with a repartitioning of condensate from liquid to ice due to (3). The mean increase is partly due to implicitly accotinting for anvil formed in mesoscale circulations. The tropical response to the experiment in July is similar to January in the preponderance of the results. Thus for economy we omit showing July results in most of the following fields.

alone (equation (3)) imposes a + 10 W m -2 shortwave forcing and a -9 W m -2 longwave forcing in the tropic s, for a net radiative forcing of +1 W m -2. Thus the increased upper tropospheric ice fractiQn due to (2) and shown in Figures 3 and 4 acts to significant!y weaken tropical cloud radiative forcing.
As mentioned in 'section 2.2, this is largely due to the larger hydrometeor size associated with ice. The magni•tude of the tropical radiative forcing by. the increased uppe? tropospheric condensate is 2-3 times the magnitude of the forcing due to the . increased ice fraction. "'

Tropical Upp•er Tropospheric Heating
In this experiment the changes in tropical anvil structure force the circulation by altering total radiative heating QR.  Figure 6 shows the change in zonal average QR and Qr in the tropics. Changes above 200 mbar, where condensation is weak, are due to the radiative heating perturbation induced by the enhanced anvil. Enhanced anvil perturbs tropical upper troposphere heating more strongly in winter, when the column is clearer and anvil radiatively heats the troposphere above 200 mbar. In the summer tropics, enhanced anvil occurs in a cloudier environment, reducing radiative cooling up to 200 mbar, and enhancing cooling above that. Thus winter and summer tropics fall, respectively, into the "anvil" and "deep" cloud scenarios of Ramaswamy and Ramanathan [1989]. Reduced optical depth keeps the intrinsically greater solar absorption of ice (relative to liquid) from causing a ubiquitous heating increase above 600 mbar.

Radiative Forcing
We present the radiative results of the experiment in terms of top-of-atmosphere (TEA) cloud forcing, observed by the Earth Radiation Budget Experiment (ERBE) satellite system from 1985 to 1989 [Hurrell and Campbell, 1992]. Shortwave cloud forcing (SWCF) is defined as the net increase in reflected shortwave (SW) flux at TeA due to cloud scattering and absorption. Figure 8 shows zonal average SWCF for January and July. Dramatic changes seen in condensate distribution and phase (Figures 4 and 5) are not seen in zonal average cloud forcing. The models predict similar equatorial LWP beneath 600 mbar but ANV has up to 5 times more equatorial The effect of cloud on terrestrial or longwave (LW) radiation, i.e., the reduction in outgoing longwave radiation (OLR) due to cloud condensate, is called longwave cloud forcing (LWCF). LWCF is a radiative proxy for tropical anvil. Figure  9 shows the geographic variation of change in January tropical LWCF due to convectively generated anvils. July results (not shown) confirm LWCF generally increased in the winter hemisphere and decreased in the summer. The strongest bias of the ANV prognostic anvil scheme is an overestimate of cloud forcing over wintertime desert, due to weak sublimation in subsidence regimes. Usually enhanced ice amount and fraction in the prognostic anvil balances the weaker mass extinction efficiency of large ice crystals because hydrometeor size is specified to increase with hydrometeor height ]. However, LWCF significantly decreases (and improves) over the central Indian Ocean in January despite a ubiquitous increase in anvil mass in the tropics (compare Figure 5). This is due to reduced upper level divergence over the central Indian Ocean, a region where prognostic anvil significantly alters the vertical distribution of cloud radiative effects.

Tropical Circulation
The change in large-scale divergent motion in the tropics due to prognostic anvil representation is shown in Figure 10, which depicts the geographic response of the January 200 mbar velocity potential X-Deep convection shifts toward the winter hemisphere (compare Figure 6). ECMWF analyses confirm that strong maxima (reduced divergence) over the central Indian Ocean eliminates a persistent convective bias. The decreased Indian Ocean convection also decreased subsidence over African and Arabian deserts, allowing too much high cloud to form there (Figure 9).
As seen above, TOA cloud forcing does not reveal the full extent of circulation change due to anvil representation. Tropical circulation is sensitive to the specific vertical (and horizontal) location of anvil heating [Ramaswamy and Ramanathan, 1989;Sherwood et al., 1994]. Figure 11 shows the vertical profile of simulated diabatic heating components over the central Indian Ocean for January conditions. The CCM heating profile, Figure 11 (

Cloud Response to SST Forcing in Equatorial Pacific
The equatorial Pacific SST anomaly associated with the 1987 E1 Ni•o provides a stringent test of model ability to mimic observed changes in convective patterns and associated anvil cloud. During the 1987 E1 Ni•o the center of deep convection, accompanying a large positive SST anomaly, shifted from the west to the central equatorial Pacific. Cloud forcing responded by increasing in the central and east equatorial Pacific through much of 1987, while cooler SST reduced cloud forcing in the west. Hartmann and Michelsen [1993] and Chou [1994] emphasize cloud enhancement from 10øS to 10øN was largely compensated by clearer sky from 10 ø to 30 ø in both hemispheres. We will use the strong SST anomaly in the equatorial Pacific region from 10øS to 10øN to test the deep convective response of the differing anvil representations to transient SST forcing. We focus on springtime behavior because equatorial SST peaks in April (when the seasonal cycle peaks), and proximity to the equinox maximizes the hemispheric symmetry of solar forcing [Ramanathan and Collins, 1991]. The spring SST of the entire equatorial Pacific (10øS-10øN, 140øE-90øW) warmed 0.9øK from 1985 to 1987, while the central equatorial Pacific alone (10øS-10øN, 180-130øW) warmed 1.2øK. Differencing the cold year (1985) spring months from the warm (1987) removes the mean cloud forcing state and isolates the cloud forcing sensitivity (which implicitly includes any reorganization of convection patterns) to SST change. Note that atmospheric response could not feed back to SST (a prescribed boundary condition). Figure 13 shows change in modeled cloud condensate 15qc for 1987-1985 over the equatorial Pacific averaged for spring. Note that (5 refers to 1987-1985 temporal change for a given model, not to intermodel change. High cloud increases in both models over the central equatorial Pacific (where (SSST peaks) in 1987 and decreases over the western equatorial Pacific. ANV predicts 15qc extrema at the same longitude as CCM (145øE, 175øW) but roughly 100 mbar higher and 2-4 times stronger. Yearly ERBE LWCF (not shown) suggests q c should increase from 140 ø to 110øW, as ANV predicts. Thus prognostic anvil representation strengthens and appears to improve CWP response to SST, but this cannot be demonstrated conclusively without reliable observational estimates of CWP.  <-20 W m -2 are more copious in ANV but extend unrealistically beyond &LWCF <-20 W m -2. This response and the weaker response of the prognostic anvil respresentation to &SST •>2øC (&LWCF >50 W m -2) stem from the prognostic formulation of cloud mass. The diagnostic cloud mass, which determines CCM cloud forcing, varies approximately exponentially with SST during convection ]. In contrast, ANV cloud mass (hence cloud forcing) peaks with max-imum 500 mbar convective intensity, which is not necessarily collocated with SST maxima [Hack, 1994]. By decoupling cloud mass from SST, the prognostic anvil representation couples cloud forcing more tightly to other factors determining convective intensity, e.g., atmospheric instability, evaporation, and surface level wind [Fu et al., 1990]. In summary, the prognostic representation of anvil production and structure has generally improved LWCF response for &SST < 0øC and worsened LWCF response for &SST > 2øC.

Conclusions
Radiative forcing from cirrus anvil plays a dominant role in determining the diabatic heating which drives the general circulation, yet many features of tropical anvil structure and life cycle are not represented in current GCM anvil parameterizations. Five year integrations of the National Center for Atmospheric Research Community Climate Model (CCM2)were used to elucidate climate features sensitive to the representation of anvil production and structure. Anvil features, emphasized in the sensitivity study, included the direct relationship between anvil growth and anvil-base convective mass flux, mesoscale condensate formation, the excess of ice over supercooled liquid, and the vertical distribution of condensate.
The direct effect of improving anvil representation is to sequester more condensate in the upper troposphere, a larger fraction of which is ice. The radiative effects of enhancing ice amount and fraction were approximately balanced by weaker radiative extinction per unit mass because anvil vertical location was tied to larger hydrometeor size. Thus top-ofatmosphere climatological radiative measures such as cloud forcing were not biased by two to fourfold increases in tropical anvil mass. The thermal structure and circulation of the tropics were more dramatically affected due to a vertical shift in heating.
Enhanced anvil perturbs the tropical upper troposphere temperature structure more strongly in winter, when the column is clearer and anvil radiatively heats the troposphere above 200 mbar. In the summer tropics, enhanced anvil occurs in a "deep cloud" environment, reducing radiative cooling up to 200 mbar and enhancing cooling above that. Reduced optical depth keeps the intrinsically greater solar absorption of ice relative to liquid from causing a ubiquitous heating increase above 600 mbar. Radiative heating contributes to warming the region just below the tropical tropopause 2ø-3øK.
On the basis of the 1987 E1 Nifio the prognostic anvil formulation improves longwave cloud radiative response to SST cooling but worsens response to warming >2øC. The net response of convection is a shift toward the winter hemisphere in solstice months. This convective reorganization reduced Hadley cell strength and eliminated a persistent convective bias in the central Indian Ocean. Moreover, the increased convection and high cloud north of the equator in January propagate Rossby waves to the extratropics. This causes significant ridging in the 500 mbar height field over the west coasts of North America and Europe, which substantially improves agreement with analysis.
In summary, climate features sensitive to anvil representation include tropical upper troposphere temperature structure, Hadley cell strength, tropical deep convection, and the northern hemisphere wintertime flow field. Many of these responses improved the climate simulation. Thus our study isolates some fundamental climate statistics in the tropics and extratropics that are partially controlled by features of tropical anvil not represented in most current GCM anvil parameterizations. Accounting for these features should be a high priority for future GCM cloud parameterizations.