Clouds in Tropical Cyclones

Robert A. Houze Jr. Department of Atmospheric Sciences, University of Washington, Seattle, Washington

Search for other papers by Robert A. Houze Jr. in
Current site
Google Scholar
Full access


Clouds within the inner regions of tropical cyclones are unlike those anywhere else in the atmosphere. Convective clouds contributing to cyclogenesis have rotational and deep intense updrafts but tend to have relatively weak downdrafts. Within the eyes of mature tropical cyclones, stratus clouds top a boundary layer capped by subsidence. An outward-sloping eyewall cloud is controlled by adjustment of the vortex toward gradient-wind balance, which is maintained by a slantwise current transporting boundary layer air upward in a nearly conditionally symmetric neutral state. This balance is intermittently upset by buoyancy arising from high-moist-static-energy air entering the base of the eyewall because of the radial influx of low-level air from the far environment, supergradient wind in the eyewall zone, and/or small-scale intense subvortices. The latter contain strong, erect updrafts. Graupel particles and large raindrops produced in the eyewall fall out relatively quickly while ice splinters left aloft surround the eyewall, and aggregates are advected radially outward and azimuthally up to 1.5 times around the cyclone before melting and falling as stratiform precipitation. Electrification of the eyewall cloud is controlled by its outward-sloping circulation. Outside the eyewall, a quasi-stationary principal rainband contains convective cells with overturning updrafts and two types of downdrafts, including a deep downdraft on the band’s inner edge. Transient secondary rainbands exhibit propagation characteristics of vortex Rossby waves. Rainbands can coalesce into a secondary eyewall separated from the primary eyewall by a moat that takes on the structure of an eye. Distant rainbands, outside the region dominated by vortex dynamics, consist of cumulonimbus clouds similar to non–tropical storm convection.

Corresponding author address: Robert A. Houze Jr., Dept. of Atmospheric Sciences, University of Washington, Box 351640, 610 ATG, Seattle, WA 98195. Email:


Clouds within the inner regions of tropical cyclones are unlike those anywhere else in the atmosphere. Convective clouds contributing to cyclogenesis have rotational and deep intense updrafts but tend to have relatively weak downdrafts. Within the eyes of mature tropical cyclones, stratus clouds top a boundary layer capped by subsidence. An outward-sloping eyewall cloud is controlled by adjustment of the vortex toward gradient-wind balance, which is maintained by a slantwise current transporting boundary layer air upward in a nearly conditionally symmetric neutral state. This balance is intermittently upset by buoyancy arising from high-moist-static-energy air entering the base of the eyewall because of the radial influx of low-level air from the far environment, supergradient wind in the eyewall zone, and/or small-scale intense subvortices. The latter contain strong, erect updrafts. Graupel particles and large raindrops produced in the eyewall fall out relatively quickly while ice splinters left aloft surround the eyewall, and aggregates are advected radially outward and azimuthally up to 1.5 times around the cyclone before melting and falling as stratiform precipitation. Electrification of the eyewall cloud is controlled by its outward-sloping circulation. Outside the eyewall, a quasi-stationary principal rainband contains convective cells with overturning updrafts and two types of downdrafts, including a deep downdraft on the band’s inner edge. Transient secondary rainbands exhibit propagation characteristics of vortex Rossby waves. Rainbands can coalesce into a secondary eyewall separated from the primary eyewall by a moat that takes on the structure of an eye. Distant rainbands, outside the region dominated by vortex dynamics, consist of cumulonimbus clouds similar to non–tropical storm convection.

Corresponding author address: Robert A. Houze Jr., Dept. of Atmospheric Sciences, University of Washington, Box 351640, 610 ATG, Seattle, WA 98195. Email:

1. Introduction

It is tempting to think of clouds in the inner regions of tropical cyclones as cumulonimbus that just happen to be located in a spinning vortex. However, this view is oversimplified, as the clouds in a tropical cyclone are intricately connected with the dynamics of the cyclone itself. Perhaps up to now it has not been important to know the detailed inner workings of tropical cyclone clouds. However, as high-resolution full-physics models become more widely used, forecasting the probabilities of extreme weather and heavy precipitation at specific times and locations will become an increasingly feasible goal for landfalling cyclones. In addition, prediction of sudden and rapid changes in storm intensity remains perhaps the greatest challenge in tropical cyclone preparation and warning. Rapid intensity changes often occur in conjunction with rapid reorganization of the cyclone’s mesoscale cloud and precipitation structures. Realistic simulation of the clouds and precipitation, therefore, is a particularly important aspect of the models’ ability to make these forecasts accurately and for the right physical and dynamical reasons.

Accurate representation of tropical cyclone clouds in numerical models is also important from a climate dynamics standpoint. Tropical cyclones have been at the center of discussions of global warming effects. Although the issue remains unresolved, tropical cyclones may increase in number and/or intensity as the earth warms. Since tropical cyclones are major producers of both cloud cover and precipitation in the tropics and subtropics, the correct prediction of tropical cyclone behavior in global climate models will depend ultimately on the accuracy with which tropical cyclone clouds are represented.

For these reasons, it seems appropriate to synthesize and organize the available information on the diverse cloud processes within tropical cyclones. This review presents a dynamical and physical description of the main types of clouds in a tropical cyclone with an emphasis on how they relate to the dynamics of the cyclone itself. The paper will proceed by first examining the cloud structures associated with the formation of a tropical cyclone. Then the review will consider in detail how the mean symmetric structure of the vortex affects cloud structure. Following that will be a discussion of how cloud structures are affected and controlled by storm asymmetries, including those brought on by the storm’s translation, the large-scale environmental wind shear, the dynamics of rainbands (which may evolve into a new eyewall) and embedded convection. The review will extend to an even smaller scale by examining the microphysics and electrification of clouds within tropical cyclones.

2. Definitions, climatology, and the synoptic-scale contexts of tropical cyclones

According to the Glossary of Meteorology (Glickman 2000), a tropical cyclone is any low pressure system having a closed circulation and originating over a tropical ocean. It is further categorized according to its peak wind speed: a tropical depression has winds of <17 m s−1, a tropical storm has a peak wind of 18–32 m s−1, and a severe tropical cyclone has a peak wind of 33 m s−1 or more.1 In this article, we will use the term tropical cyclone to refer primarily to the stronger tropical storms and severe tropical cyclones. Severe tropical cyclones have local names. They are called hurricanes in the Atlantic and eastern North Pacific Oceans, typhoons in the western North Pacific Ocean, and cyclones in the South Pacific and Indian Oceans. They are all the same phenomenon.

Gray (1968) identified most of the environmental factors favoring tropical cyclones. They originate over oceans, as their primary energy source is the latent heat of water vapor in the atmospheric boundary layer. They nearly always form over regions where the sea surface temperature (SST) exceeds 26.5°C (Fig. 1a). Once tropical cyclones form, they tend to be advected by the large-scale wind. Figure 1b shows how the tracks are generally westward at lower latitudes, where easterlies dominate the large-scale flow. The storms often “recurve” toward the east, when they move poleward into the midlatitude westerlies. When they encounter land or colder water, they die out or (more rarely) transition into extratropical cyclones (Jones et al. 2003).

Although tropical cyclones form at low latitudes, generally between 5° and 20° (Fig. 1a), they rarely form within ±5° of the equator because the Coriolis force is too weak for low-level convergence to be able to generate relative vorticity.2 Gray (1968) found that genesis regions not only are located off the equator, where the Coriolis force is nonzero, but also have higher-than-average relative vorticity of the surface wind field and temperature stratification that is at least moderately conditionally unstable, and weak vertical shear of the horizontal wind. The strong background positive vorticity helps to trap energy released in convection so that it contributes via generation of potential vorticity to strengthening of the positive vorticity. Weak shear of the background flow allows the cyclone to develop vertical coherence. The climatological presence of strong shear over the South Atlantic, plus the lack of synoptic precursor disturbances, accounts for the extremely rare occurrence of tropical cyclones in that region. Besides the factors highlighted by Gray (1968), the relative humidity of the environment must not be too low (DeMaria et al. 2001). Otherwise, clouds cannot fully develop, and without clouds to transport moist static energy upward from the boundary layer in contact with the ocean, there can be no tropical cyclone.

The large-scale preconditions for tropical cyclogenesis noted by Gray (1968) and DeMaria et al. (2001) may be set up by a variety synoptic-scale processes. Bracken and Bosart (2000), Davis and Bosart (2001), and McTaggart-Cowan et al. (2008) have shown that about half of Atlantic tropical cyclones occur when African easterly waves (Burpee 1972; Reed et al. 1977, 1988; Thorncroft and Hodges 2001) move westward off the African continent, and a tropical cyclone spins up in the trough of the wave, provided the wave trough does not ingest dry midlevel air from the Saharan region (Dunion and Velden 2004). Other Atlantic tropical cyclones develop when midlatitude synoptic-scale dynamics operating over a warm ocean surface set up conditions of large-scale shear and vorticity that lead to tropical cyclogenesis. According to Bracken and Bosart (2000), Davis and Bosart (2001), and McTaggart-Cowan et al. (2008), these midlatitude conditions involve midlatitude troughs and/or fronts extending into low latitudes.

The ultimate formation of a tropical cyclone in a favorable synoptic-scale environment entails the concentration of perturbation kinetic energy into a low pressure system with a horizontal scale of a few hundred kilometers. The concentration of energy on this scale may be partly a downscale process, whereby a developing storm extracts energy from larger scales of motion. One idea suggested by Webster and Chang (1997) is that when a synoptic-scale wave in the easterlies enters a region of large-scale negative stretching deformation (∂u/∂x < 0, where u is the large-scale mean zonal wind component) it will become more energetic, shrink in scale, and be confined to the lower troposphere. However, tropical cyclone development also certainly involves strong upscale feedback in which convective clouds connect the intensifying cyclonic circulation with the boundary layer and focus and intensify vorticity locally. Dunkerton et al. (2008) have suggested a mechanism whereby the parent synoptic disturbance contains a subsynoptic-scale subregion within which smaller-scale phenomena can flourish and efficiently accomplish the upscale feedback. They call this the “marsupial theory,” where the favored subregion for development is analogous to the pouch of a mother kangaroo. Their theory applies specifically to the case in which the parent synoptic system (i.e., the kangaroo) is a tropical easterly wave. In the wave’s critical layer (where the speeds of the wave and mean flow are equal), a “pouch” region develops where parcel trajectories form closed cyclonic loops. Clouds forming in this closed-off region are protected from deleterious large-scale environmental influences such as the intrusion of dry air. Upscale cyclogenesis processes connected with clouds can therefore flourish in this synoptically protected zone, where they are confined and combine to strengthen the cyclonic circulation. If this protected region (or “sweet spot”) of the wave is located in a region of warm SST and minimal large-scale vertical shear, and preexisting low-level positive relative vorticity, convective clouds and precipitation in this region will be especially conducive to concentrating low-level convergence and positive vorticity and strengthening the low. While this marsupial theory of cyclogenesis has been developed in the context of tropical easterly waves, it may be that the nontropical precursor synoptic-scale features identified by McTaggart-Cowan et al. (2008) also form protected zones and/or regions of negative stretching deformation in ways that have not yet been investigated.

3. Clouds involved in tropical cyclogenesis

a. Idealization of the clouds in an intensifying depression

As noted in the previous section, tropical cyclogenesis is partly an upscale process, with convective-scale dynamics locally adding energy and vorticity to a preexisting cyclonic disturbance of the large-scale flow, in a region synoptically predisposed to development. Convective clouds are central in the upscale positive feedback to tropical cyclogenesis. Steranka et al. (1986) noted from satellite data that prolonged “convective bursts” occurred “in the near region surrounding the depression centers before the maximum winds initially increased.” These bursts of cirrus outflow from convective clouds stand out in satellite imagery and suggest that a very special type of cumulonimbus cloud is involved in tropical cyclogenesis.

Figure 2 illustrates conceptually the variety of convective entities that may exist in the region where a convective burst presages intensification of a preexisting vortex to tropical storm status. Some of these convective entities are individual towers, while others take the form of a mesoscale convective system (MCS), which contains both deep convective cells and stratiform clouds and precipitation (see Houze 2004 for a review of MCSs). The convective entities in Fig. 2d are depicted within a low-level preexisting cyclonic circulation. From Gray’s (1968) climatology we know that cyclogenesis occurs in an environment rich in vorticity at low levels, and the cloud-scale processes depicted in Fig. 2 can only operate effectively to enhance cyclogenesis in a preexisting region of low pressure and maximum vorticity such as indicated by the L and the dashed cyclonic streamline in the figure. As discussed in section 2, the work of Dunkerton et al. (2008) suggests further that cyclogenesis may be especially effective in a closed cyclonic circulation protected from the larger-scale environment.

Figures 2a–c give a simplified depiction of the life cycle of the cloud-scale MCS as it would occur within an assumed larger-scale environment rich in vorticity at lower levels. The general characteristics of the schematic life cycle are an adaptation of the generic MCS life cycle described by Houze (1982), but with an emphasis on the way vorticity might develop within the MCS. The MCS begins as one or more isolated deep convective towers (Fig. 2a). The vorticity of the low-level environment is stretched by convergence at the base of the buoyant convective updrafts and advected upward. The updrafts thus become upward extending centers of high positive vorticity, which Hendricks et al. (2004) called vortical hot towers (VHTs; see Fig. 2), as a special case of the “hot tower” terminology introduced by Riehl and Malkus (1958) to describe deep tropical convection in general. As an individual convective tower dies off, it weakens and becomes part of a precipitating stratiform cloud (Houze 1997). New towers form adjacent to the stratiform region, so that at its mature stage of development the MCS has both convective and stratiform components (Fig. 2b). The base of the stratiform cloud deck is in the midtroposphere, as it is mostly composed of the material of the upper portions of convective cells. The vertical profile of heating in the stratiform region leads to the development of a mesoscale convective vortex (MCV) in midlevels in association with the stratiform cloud deck (see Houze 2004). In the case depicted in Fig. 2, the stratiform region is composed of the remnants of convective cells, which themselves have considerable positive vorticity. These vortical hot tower remnants add more vorticity to the MCV. During the middle life cycle state in Fig. 2b, the MCS contains vorticity structures both in the form of convective-scale VHTs and in the form of the wider stratiform-region MCV. In the late stages of the MCS life cycle, the hot towers cease forming, but the stratiform cloud region containing MCV vorticity remains for some hours (Fig. 2c).

b. Example of a vortical hot tower

An example of vortical hot tower has been documented with airborne Doppler radar by Houze et al. (2009) in an MCS located in the depression that became Hurricane Ophelia (2005). The basic structure of the updraft is illustrated in Fig. 3. Doppler radar showed a deep, wide, intense convective cell of a type that has been previously thought to occur in intensifying tropical depressions (Zipser and Gautier 1978) but had not been documented in detail until high-resolution airborne Doppler radar data were collected in the pre-Ophelia depression. The observed updraft of the cell was 10 km wide and 17 km deep and had updrafts of 10–20 m s−1 throughout its mid- to upper levels (Figs. 3a,b). It contained a cyclonic vorticity maximum (shown at 8-km altitude in Fig. 3c). This vorticity maximum, centered on the updraft, confirms that the cell was a vortical hot tower. The massive convective updraft was maintained by strong positive buoyancy (virtual potential temperature perturbation >5°C at 10 km; Fig. 3a), probably aided by latent heat of freezing at higher altitudes. The convective updraft was fed by a layer of strong inflow several kilometers deep. Wind-induced turbulence, just above the ocean surface, enriched the equivalent potential temperature of the boundary layer of the inflow air, thus creating an unstable layer with little convective inhibition. This air was raised to its level of free convection when it encountered the denser air located at low levels in the rainy core of the convection. Although evaporative cooling and precipitation drag occurred in the rain shower of the cell (see the negative temperature perturbation below 2 km in Fig. 3a), the negative buoyancy was insufficient to produce a strong downdraft or gust front outflow to force the updraft. Instead the updraft air was forced initially upward when the ambient low-level wind of the developing depression encountered the density gradient at the edge of the rainshower. This type of updraft formation without the aid of a gravity current outflow from a downdraft is not common, but its possibility of occurrence was pointed out in numerical simulations by Crook and Moncrieff (1988). The lack of a divergent spreading cool-air pool made this intense convection an extremely effective feedback mechanism for cyclogenesis in the pre-Ophelia depression. The production of low-level mass convergence, and associated stretching and upward advection of vorticity in the lower troposphere of the region where cyclogenesis was occurring, was not offset by strong downdraft divergence, and air of low equivalent potential temperature was not transported into the boundary layer to offset the gain of moist static energy from the sea surface.

The dominance of updraft motion and relative lack of downdraft motions in the convective burst of convection in the intensifying depression was also noted by Zipser and Gautier (1978) in their analysis of flight-level data. Nolan (2007) conducted idealized model simulations of convection in a tropical cyclogenesis environment and found that updraft mass transport dominated as the depression approached tropical storm stage. Zipser and Gautier (1978) further noted that the mesoscale convergence feeding the massive convective updraft present in the MCS they investigated was sufficient to account for the intensification of the synoptic-scale depression vortex in which the MCS was located. A sharp increase of upward mass flux with height occurred throughout the lower portion of the updraft in the pre-Ophelia convection (see the isolines of the vertical derivative of the vertical mass flux in Fig. 3b). The concomitant stretching of vorticity and generation of potential vorticity3 in the lower portion of the updraft, and the strong vertical motions of the updraft advecting the concentrated vorticity vertically, created a deep convective-scale vorticity perturbation manifested as a convective-scale cyclonic vortex—that is, a vortical hot tower. This vortex is evident in Fig. 3c.

The vorticity field shown in Fig. 3c shows both negative and positive vorticity centers near the center of the updraft. This structure is consistent with the updraft tilting the environmental horizontal vorticity into the vertical to produce a vortex couplet in midlevels, as described by Rotunno (1981) and Montgomery et al. (2006). For simplicity, this tilting-produced couplet is not shown in Fig. 2a since the positive member of the tilting-produced couplet dominates, probably because it combines with the positive vorticity of the stretched and upward-advected boundary layer vorticity to produce deep updrafts that are centers of cyclonic rotation. Moreover, numerical experiments suggest that the negative vorticity anomalies tend to be expelled from the developing tropical cyclone vortex while the positive anomalies are retained (Montgomery and Enagonio 1998). Thus, the negative anomalies are not relevant to the developing cyclone as positive feedback elements.

c. Ensemble of clouds in a developing storm

The ensemble of clouds in the intensifying depression shown in idealized form in Fig. 2d is based on real examples. Simpson et al. (1997) and Ritchie et al. (2003) showed satellite imagery of MCSs rotating around the center of the developing Southern Hemisphere Tropical Storm Oliver (1993). Sippel et al. (2006) described a similar set of MCSs seen in both satellite imagery and coastal radar observations as the newly formed Tropical Storm Allison (2001) made landfall in Texas. The MCSs in Allison were distributed around the developing cyclone, and the coastal radar showed positive vorticity anomalies in the active convective elements of the MCSs, probably similar to the vortical hot tower illustrated in Fig. 3.

Convective-ensemble behavior similar to Allison and Oliver occurred during the genesis stage of Hurricane Ophelia (2005). Figure 4 shows infrared satellite and coastal radar imagery just prior to the pre-Ophelia tropical depression intensifying to tropical storm intensity. A day before the depression reached tropical storm intensity (Figs. 4a–d), strong convection was prevalent over a wide area, but the clouds and precipitation exhibited no tropical storm–like structure. At the time of Figs. 4c,d, the convection seen on radar had grouped into three MCSs ∼200 km in horizontal dimension, each having both active convective cells and areas of stratiform precipitation. The vortical hot tower cell shown in Fig. 3 was located in the MCS located northeast of the Melbourne, Florida, coastal radar. Thus, the pre-Ophelia depression had a cloud population containing intense rotational convective cells and MCSs scattered about in the low-pressure area as depicted schematically in Fig. 2d.

By the time of Figs. 4e,f, the overall area covered by deep convection in the pre-Ophelia depression had decreased and became focused on a single very intense MCS. In the next several hours, this particular mesoscale precipitation area radically changed its shape and took on the structure of a tropical storm (Figs. 4g,h). Nolan’s (2007) idealized model simulations suggest that the long period of moistening by prior convection (e.g., Figs. 4a–d) may facilitate conversion from MCS to tropical storm structure. The radar echo exhibited an incipient eyewall and a well-defined principal rainband (as defined by Willoughby et al. 1984b; Willoughby 1988) extending from south to east to north of the storm and having a convective structure upwind and a stratiform structure downwind, as described in papers on mature hurricane rainbands (Atlas et al. 1963; Barnes et al. 1983; Hence and Houze 2008). Details of principal rainbands will be discussed in section 8c.

d. Cloud feedback in cyclogenesis

Figure 2d depicts an idealized scenario in which a population of clouds occurs in a region where large-scale conditions have dictated the presence of a preexisting low-level weak cyclonic circulation. The cloud population within this region consists of a combination of isolated deep convective cells with cyclonic vorticity maxima in the updrafts (as in Fig. 2a), one or more mature MCSs with both cyclonically rotating convective cells and MCVs (as in Fig. 2b), and older MCSs with residual MCVs (as in Fig. 2c). Each of the clouds contains significant vorticity perturbations in the form of convective-scale vortical hot towers and/or stratiform-region MCVs. These in-cloud vorticity perturbations can feed back positively to the larger-scale cyclonic vorticity. The deep convection generates potential vorticity. Collectively, the clouds accumulate and distribute vorticity derived from the boundary layer and low levels of the free environment throughout a deep layer. By these means they help to organize the preexisting weaker synoptic-scale cyclonic circulation into a tropical storm. The background positive vorticity helps to reduce the Rossby radius (Houze 1993, p. 53) within which the effect of the clouds is combined.

Tropical cyclogenesis is not just a question of strengthening the preexisting vortex. The vortex must reorganize to become a tropical cyclone. An important question is how a cloud population such as that depicted conceptually in Fig. 2d, or shown by example in Fig. 4, helps convert a preexisting benign synoptic-scale cyclonic circulation to a structure having the specific configuration of a tropical cyclone. One of the distinctive features of a tropical cyclone is a narrow annulus of maximum wind at some distance (usually between 10 and 100 km) from the storm center. This zone is referred to as the radius of maximum wind (RMW), and it is kept in near–thermal wind balance by a strong secondary circulation, which in turn produces an eyewall cloud. Outside the eyewall precipitation occurs in mesoscale spiral rainbands. These cloud and precipitation features of the mature tropical cyclone will be discussed in detail in subsequent sections of this paper. As a cyclonic disturbance reaches tropical storm strength, it also develops an RMW and incipient eye, eyewall, and rainbands.

The foregoing question then reduces more specifically to how upscale feedback by the cloud population depicted in Fig. 2d helps to convert the preexisting cyclonic circulation to a cyclonic disturbance exhibiting an RMW, eye, eyewall, and rainbands. One idea about how this conversion occurs is that the convective and mesoscale vorticity perturbations contained within the members of the cloud population, as depicted in Fig. 2d, are eventually sheared apart by the radial gradient of the vortex wind and mixed into the mean flow around the low in a process of axisymmetrization.4 Hendricks et al. (2004) and Montgomery et al. (2006) present evidence from a model that the axisymmetrization process distributes subsynoptic-scale vorticity perturbations, such as those contained in the vortical hot towers and MCVs, into a ring of high vorticity at a specific distance from the storm center, giving the enhanced low pressure system an RMW and thus a structure like that of a tropical cyclone.

While axisymmetrization must play a role, a remaining curious fact is that the tropical cyclone center has been observed to occur definitively within a particular member of the cloud population rather than as a collective smearing of the effects of the whole population around the preexisting low (Fig. 4h). Bister and Emanuel (1997) have suggested that cooling due to melting and evaporation of precipitation below the base of the stratiform cloud is involved in the extension of the middle level MCV downward. However, their idea requires a particularly strong stratiform-region downdraft, and observations of MCSs in intensifying tropical depressions indicate that these MCSs have relatively weak downdrafts (Zipser and Gautier 1978; Houze et al. 2009). Another way for one MCS to become a storm center was demonstrated in a numerical model of a midlatitude MCS by Chen and Frank (1993). In their simulation, the MCV strengthened and built downward as a result of humidification in the region of the MCV; the humidification lowered the buoyancy frequency and thus led to a reduction in the Rossby radius. This occurred in the absence of a lasting downdraft. Finally, there may be a stochastic aspect to the process. As the larger depression gathers strength from all the ongoing convective activity and gradually takes on tropical storm structure in its wind field, the probability increases that one of the MCSs with rotational convective cells and/or MCV will occur in the ideal spot (presumably the exact center of the depression) where the MCS’s cloud structure interacts with the vortex center to metamorphose into a structure that has an incipient eye, eyewall, and spiral rainbands.

4. Overview of the mature tropical cyclone

a. Visible clouds

In satellite pictures, the clouds in a mature tropical cyclone are dominated by upper-level cirrus and cirrostratus spiraling anticyclonically outward. In the visible image of Hurricane Katrina (2005) in Fig. 5a, the rough tops of convective clouds penetrate through the smooth cirrus outflow. The most striking feature is the open eye of the storm located in the middle of the spiraling cloud pattern. A zoomed-in satellite view of the eye in Fig. 5b shows that the cloud top surrounding the eye slopes downward toward the ocean surface. Seen from an aircraft flying inside the eye (Fig. 5c), the cloud surface bounding the eye region and sloping at about a 45° angle gives an observer on the plane the impression of being inside a giant circular sports stadium with the grandstand banking upward and outward. This huge sloping cloud bank, called the eyewall, is highlighted by the sunlight on the east side of the eye in all three panels of Fig. 5. The ocean surface is not visible in Figs. 5b,c. Rather it is obscured by low stratus or stratocumulus clouds. This low cloud cover is common in the eye of a strong tropical cyclone. The dynamics of clouds in the eye and eyewall regions will be discussed further in sections 57.

b. Three-dimensional wind field

The visible cloud pattern seen in Fig. 5 is determined by the wind and thermodynamic structure of the cyclone. An example of the low-level wind field in a tropical cyclone (Gloria 1985) is shown in Figs. 6a,b. The winds were constructed from rawinsonde, dropsonde, and aircraft Doppler radar measurements. In Fig. 6a, the wind data have been filtered to remove wavelengths less than about 150 km near the center of the storm and 440 km in the outer portions of the figure. This view emphasizes the large-scale flow pattern in which the storm is embedded. In Fig. 6b, the analysis retains wavelengths down to about 16 km in the center of the figure and down to about 44 km in the outer part of the figure. In this higher-resolution analysis, the tropical cyclone vortex itself is highlighted. The streamlines show boundary layer air spiraling inward toward the center of the storm. The shaded isotachs (Fig. 6b) show the roughly annular zone corresponding to RMW at ∼20 km (0.18° latitude) from the storm center. The air parcels spiraling inward at 900 hPa tend to increase their angular momentum, and the RMW occurs where the radial inflow rate abruptly slows down. Radial convergence, increased tangential wind speed, and sudden upward turning of the air current occur at this location and produce the eyewall cloud. Inside the RMW is the eye, where the wind speeds drop off almost immediately to nearly zero, and the vertical air motion is downward, producing the eye of the storm by suppressing clouds, except for the low-level stratus capping the mixed layer (Fig. 5). The dynamics of the eye are discussed further in section 6.

The 200-hPa wind pattern (Figs. 6c,d) illustrates that although the tropical cyclone’s outflow is strong, it is generally asymmetric. In this case, the outflow is concentrated in a channel northeast of the storm center. The depth of the cyclone core is also apparent; even at this high altitude, the flow is cyclonic near the storm center but changes to anticyclonic ∼100 km from the storm center. Details of the cyclonic flow near the center of the storm are illustrated by the high-resolution analysis in Fig. 6d.

Vertical cross sections of the broadscale mean radial u and tangential υ components of the wind of a tropical cyclone are shown in Figs. 7a,b, which are composites of data collected in many storms. Most evident in the radial wind field (Fig. 7a) is the increase in the inward directed component at low levels as one approaches the storm center. Strong radial outflow is evident at the top of the storm, at about the 200-hPa level. An important feature missed by these old, low-resolution composite observations is the occurrence of a shallow radial outflow layer at the top of the boundary layer, which is related to the fact that the strong winds near the center of the vortex become supergradient (Smith et al. 2008; this will be discussed in section 6). The strongest tangential wind field (Fig. 7b), defining the RMW, is at very low levels. The average dropsonde profile obtained on aircraft penetrations of tropical cyclones is characterized by a broad maximum centered ∼500 m above the surface (Franklin et al. 2003). The strong convergence in the eyewall at low levels and the strong outflow aloft must be balanced by strong upward motion. The broadscale pattern of vertical motion in a tropical cyclone is shown in Fig. 7c. The overall cloud and precipitation amounts are determined by this vertical mass transport, which on average is upward over the region located within 400 km of the storm center. However, this large-scale mean vertical motion pattern does not have the spatial resolution to provide much insight into cloud structures, nor does it show the downward motion in the eye. To see these details, special aircraft observations and radar instrumentation are required.

c. Equivalent potential temperature and angular momentum in relation to the eye and eyewall

In the large-scale environment, far from the center of a tropical cyclone, the typical sounding (Jordan 1958) shows that stratification of equivalent potential temperature θe is dominated by potential instability (decrease of θe with height) in the lower troposphere, with θe reaching a minimum at about the 650-hPa level. Above that level, the air is potentially stable. The pattern of θe changes markedly as one proceeds inward toward the center of a tropical cyclone (Bogner et al. 2000). The typical pattern of θe within a tropical cyclone is indicated by the example in Fig. 8a. In the low levels, the values of θe increase steadily to a maximum in the eye of the storm. In the vicinity of the eyewall (∼10–40 km from the storm center), the gradient of θe is the greatest, and the isotherms of θe rise nearly vertically through the lower troposphere, then flare outward as they extend into the upper troposphere. Since above the boundary layer θe is nearly conserved following a parcel, these contours reflect the flow of air upward and outward in the eyewall. In the very center of the storm, θe decreases strongly with height. The center of low θe at 500 hPa is evidence of subsidence concentrated in the eye of the storm.

The vertical circulation of the tropical cyclone in relation to θe is illustrated qualitatively in Fig. 8b, where the low-level radial flow is depicted as converging into the center of the storm (consistent with Fig. 7a) in the boundary layer below cloud base. As it flows inward, turbulence produces a well-mixed boundary layer of high θe. When this air enters cloud in the eyewall zone, it ascends to the upper troposphere approximately along the lines of constant θe. The θe lines also reflect a concentrated descent of air in the eye of the storm, which will be discussed further in section 5.

The dynamical necessity of the funnel-like shape of the outward-sloping flow lines in the eyewall region was reasoned out by early meteorologists. Haurwitz (1935) noted that a vanishing pressure gradient at some high level indicates that the strong low-level pressure gradient in a tropical cyclone must be associated with an outward slope of the boundary of the core of warm air in the center of the storm. Durst and Sutcliffe (1938) argued that rising rings of air in a warm-core storm in which the radial pressure gradient decreases upward must move outward for the centrifugal and Coriolis forces (dictated by conservation of their surface angular momentum) to balance the weaker pressure gradients aloft. This reasoning foreshadowed today’s view of the inner core of the storm (Emanuel 1986; Smith et al. 2008).

The contemporary view expresses the eyewall circulation in terms of parcels of air rising out of the boundary layer and subsequently conserving both angular momentum and θe as they ascend in the free atmosphere. The angular momentum m about the central axis of the storm is given by
where r is the radial coordinate measured from the eye of the storm, υ is the tangential wind component, and f is the Coriolis parameter. Above the boundary layer, where frictional effects are small, m is approximately conserved following a parcel, as is θe. The isopleths of m and θe are therefore congruent, implying that above the boundary layer the eyewall cloud is in a state of approximate conditional symmetric neutrality (Houze 1993, 55–56).

Although the outward spreading of the m and θe lines is consistent with the theory of a balance vortex and is a typically observed characteristic of tropical cyclones, there are nonetheless observations and model simulations that indicate that vertical convection is often embedded in an eyewall cloud, modifies its structure, and significantly affects the overall intensity of the tropical cyclone. To fully examine the eye and eyewall dynamics, sections 5 and 6 will discuss how the basic or mean structure of the eye and eyewall cloud can be thought of in terms of the slantwise conditionally symmetrically neutral component of motion in a thermal wind–balanced tropical cyclone vortex. Section 7 will then examine the buoyancy driven vertical drafts superimposed on the eyewall cloud.

5. The eye

As noted in section 4a and illustrated by Fig. 5, the eye is a region of unique and strong dynamics that involves two distinct and different types of clouds, a layer of low-topped stratus and/or stratocumulus clouds extending horizontally over the eye region and an outward-sloping eyewall cloud. These two types of clouds are depicted schematically in Fig. 9 in relation to the wind field in the tropical cyclone. Both are related to the dynamics of the tropical cyclone vortex. The depiction in Fig. 9 is from a thorough examination of observations in the eye region of hurricanes by Willoughby (1998).

The conditions in the eye region are ideal for the formation of a cloud-topped mixed layer (Lilly 1968; Stevens et al. 2003; Bretherton et al. 2004) because subsidence in the eye produces a stable layer at the top of a vigorous mixed layer over a warm ocean surface. The top of the stratiform cloud with a few cumuliform turrets appears below the aircraft in Fig. 5c at the top of the mixed layer cloud, near the upper extent of the subsidence-induced stable layer.

The sloping eyewall cloud highlighted by sunlight in Fig. 5 is primarily a manifestation of the secondary circulation, which, as shown in Fig. 9, is a radial-vertical overturning transverse to the cyclone’s primary circulation, which refers to the general cyclonic rotation of the wind around the eye. In a balanced tropical cyclone, the secondary circulation consisting of the radial u and vertical w components of the wind keeps the primary circulation (i.e., the υ component of the wind) in gradient-wind equilibrium. The low-level radial inflow branch of the secondary circulation gains latent heat energy via upward turbulent flux of sensible and (primarily) latent heat (i.e., moist enthalpy) as it flows over the ocean surface toward the eyewall. Latent heat released in the vertical component of the eyewall circulation provides the energy that powers the vertical circulation and maintains the tropical cyclone strength and structure.

The secondary circulation is characterized both by rising motion in the eyewall zone and compensating subsidence, occurring partly concentrated in the eye region and partly more diffusely in the far field radially well outside the eyewall. Figure 10 shows the secondary circulation computed by considering point sources of heat and momentum placed near the RMW in an idealized tropical cyclone–like vortex (Shapiro and Willoughby 1982). From this basic result, it is evident that the concentrated subsidence in the eye region is a consequence of the heating in the eyewall cloud. The subsidence in the eye, conceptualized and labeled as “forced dry descent” in Fig. 9, keeps the eye cloud free except for the mixed-layer stratus/stratocumulus at the top of the boundary layer. Pendergrass and Willoughby (2009) and Willoughby (2009) have shown that variations in hypothesized heating patterns associated with the eyewall and different environmental conditions lead to a variety of variations. The latter paper indicates that transient bursts of heating in the eyewall, for example from embedded buoyant convective towers, might produce significant alterations in the storm structure. The subsidence computed for the eye region is accompanied by divergence at low levels within the eye region. Figure 9 indicates that the eyewall is fed in part by the low-level radial outflow from the eye region as well as by radial inflow from outside the eyewall. Important implications of this ingestion by the eyewall of air from the boundary layer of the eye region are discussed in section 7.

Shapiro and Willoughby’s (1982) idealized tropical cyclone circulation shown in Fig. 10 is not in a steady state because the RMW contracts over time. Since the RMW is collocated with the eyewall, this behavior is referred to as eyewall contraction. The nonlinear response to the eyewall heat and momentum source has the maximum rate of increase of tangential wind velocity located slightly radially inward of the RMW. This result implies that the RMW (Fig. 8b) of a storm in thermal wind balance should decrease in time, and observations and numerical models confirm that many eyewalls contract (Willoughby et al. 1982; Willoughby 1990, 1998; Houze et al. 2006, 2007). The dashed lines in Fig. 9 indicate an earlier position of the inner edge of the eyewall cloud. As the eyewall contracts, the winds in the eyewall zone increase according to conservation of angular momentum and other factors, and the central pressure of the storm decreases as the pressure gradient associated with the new location and strength of the maximum wind continually adjusts toward gradient-wind balance.

The downward air motion in the eye region and divergence of mass away from the eye at low levels seen in Fig. 9 can be thought of partly as a compensation for the gradual contraction of the eyewall boundary over time. Also contributing to the subsidence in the eye is turbulent mixing of angular momentum at the interface of the rapidly rotating eyewall and more quiescent eye region. This mixing is suggested by the entrainment arrows at the inner edge of the eyewall cloud in Fig. 9. To maintain thermal wind balance in the eye region the increase of tangential wind velocity and relative change in the vertical shear of the horizontal wind in the eye needs to be matched by a stronger horizontal temperature gradient, which can only be provided by adiabatic subsidence warming in the center of the vortex (Emanuel 1997).

Figure 9 includes two types of downdrafts based on close-up observations of tropical cyclones during aircraft penetrations. On the outer side of the eyewall, negatively buoyant precipitation driven downdrafts (see further discussion in section 7d) occur at lower levels. Below cloud base, they merge with the frictional inflow beneath the eyewall and become part of the boundary layer air spreading into the eye region. At higher levels along the inner edge of the eyewall, a thin layer of moist air cascades down the eyewall (Willoughby 1998). Mixing of the dry air in the eye and moist air in the eyewall cloud occurs in this thin layer. Evaporation of cloud particles cools the air and produces a cascade of negatively buoyant air down the side of the eyewall cloud. This cascade is visible as a mist of particles mixed into the dry air adjacent to the eyewall cloud (Fig. 11, from Willoughby 1998). This downdraft on the inner side of the eyewall has also been well documented by aircraft measurements of the vertical velocity (Jorgensen 1984, see also footnote in section 7c).

6. Dynamics of the mean eyewall cloud

a. Sloping angular momentum surfaces

The eyewall surrounding the eye of the storm is like no other cloud. Although it is often described as a “convective” cloud, the eyewall does not simply follow the dynamics of ordinary cumulus and cumulonimbus, which arise from purely vertical accelerations releasing buoyant (conditional or potential) instability. Rather, as noted in section 4c, the eyewall cloud is shaped to great measure by slantwise convective motions, where the combination of vertical and horizontal accelerations (responding to both buoyancy and inertial forces) maintain a state of near–conditional symmetric neutrality. In this section, we summarize the dynamics of the slantwise component of the eyewall circulation. Section 7 will describe important buoyant convective perturbations of this structure.

The essential dynamics of the slantwise circulation of the sloping eyewall cloud can be inferred from Emanuel (1986) and Smith et al. (2008), who provide an interconnected theoretical linkage of air–sea interaction at the surface, atmospheric boundary layer structure, the secondary circulation of the balanced vortex, and the moist conditionally symmetrically neutral upflow in the eyewall. Emanuel’s (1986) approach is often parameterized in terms of an approximate Carnot cycle (section 3 of his paper), which gives insight into tropical cyclone intensity. However, the focus of the present review is cloud dynamics, not storm intensity. To understand the air motions giving rise to the eyewall cloud, the following discussion makes use of the more explicit version of Emanuel’s approach rather than its Carnot parameterization.5

The degree to which an m surface in the eyewall cloud of a balanced tropical cyclone slopes outward can be obtained from the basic equations for a two-dimensional, axisymmetric vortex that is in hydrostatic and gradient-wind balance (see appendix A). For such a vortex, conservation of m requires that the slope of a particular m surface be
where r is radius from the storm center and p is pressure. Thus, the slope of an m surface is determined by the ratio of the vertical to the horizontal shear of m. Note also that the right-hand side of (2) is the ratio of the shear of the swirling wind to the local vertical vorticity. In sections 6bd, it will be shown that with substitution from the thermal-wind equation for a balanced vortex [(A3)], (2) implies that the slope of the m surfaces with height must be outward, as depicted in Fig. 8b.

Equation (2) provides a basis for obtaining mathematical expressions that describe the dynamics of the eyewall cloud. An expression for m(r, p) is substituted into (2) to determine the geometry of the streamlines. With the help of the basic equations of a symmetric vortex, one can obtain the magnitudes of the wind components, and the thermodynamic properties of the circulation of an idealized tropical cyclone vortex containing the eyewall cloud. The solution is based on the observation that the tropical cyclone circulation in the eyewall region tends to be conditionally symmetrically neutral [∂θes/∂z = 0 on a surface of constant m, where z is height and θes is the saturation equivalent potential temperature defined by (B1)], but not especially strongly conditionally unstable (∂θes/∂z usually only slightly negative in the vertical direction). Consistent with these observations, the eyewall cloud is assumed to be undergoing slantwise, nearly neutral vertical motions above the boundary layer. The ascending air is assumed further to be emanating from a boundary layer whose value of θe has been determined by upward flux of sensible and latent heat from the ocean surface. For an idealized circularly symmetric tropical cyclone vortex, this approach yields both the primary and secondary circulation by connecting the eyewall dynamics directly to the air–sea interaction that occurs in the boundary layer over the warm ocean.

b. Boundary layer assumptions and implications

Emanuel (1986) divides the boundary layer of the idealized two-dimensional axisymmetric vortex into the regions I, II, and III shown in Fig. 12a. All three regions are assumed to be well mixed by turbulence and to be of constant depth h. The production of turbulent kinetic energy in the boundary layer by shear is substantial because of the strong low-level winds circulating around the center of the storm, while the generation of turbulence by buoyancy is strongly promoted by the warm ocean surface. The turbulence produces strong upward eddy flux of latent heat (and to a lesser degree sensible heat) everywhere in the lower part of the boundary layer. The flux at the top of the boundary layer, however, is assumed to vary significantly from one region of the storm to the next. In region III, rainbands (to be discussed in section 8) are expected to be important contributors to the turbulent flux at the top of the boundary layer, with convective updrafts and downdrafts transporting high and low θe out of the boundary layer, respectively. Region II is the region of the eyewall cloud, and the flux at the top of the boundary layer there is assumed to be entirely positive and dominated by the mean upward flux associated with the secondary circulation of the tropical cyclone vortex. Turbulent flux at the top of the boundary layer here is considered to be very small relative to this mean upward flux. Region I is the eye of the storm (section 5).

Emanuel’s (1986) simplified treatment of the basics of eyewall dynamics (Fig. 12a) does not consider any interaction between the eyewall and region I. However, in real storms the air diverging out of the eye region at low levels (Fig. 9) can and does feed extremely high-θe air into the base of the eyewall cloud. The eyewall cloud vertical motions are intensified by the entrainment of high-θe air from the boundary layer of the eye region, and local convective cells of buoyant updraft motion can form within the eyewall in locations where the entrainment particularly strongly enhances the buoyancy. To take the entrainment of high-θe air from the eye into account, Smith et al. (2008) replaced Emanuel’s (1986) boundary layer model with one that allows for air from the eye region to enter the base of the eyewall cloud (Fig. 12b). Implications for enhancing buoyancy within the eyewall in this manner will be discussed further in sections 6e and 7ac.

A subtle but significant feature of Emanuel’s (1986) boundary layer treatment is that the wind at the top of the boundary layer is assumed to be in gradient balance. Smith et al. (2008) employed a boundary layer model that allows for frictionally induced cross-isobaric flow in the boundary layer. In their model, the frictionally enhanced radial inflow in the boundary layer at the boundary of regions A and B (Fig. 12b) is much stronger than the radial inflow in Emanuel’s version. One result is that in region A the tangential wind (into the plane of the cross section) becomes supergradient (as a consequence of the Coriolis force acting on the radial inflow). Outward-directed Coriolis and centrifugal forces acting on the supergradient tangential wind in turn slow the boundary layer inflow. Convergence in region A, resulting in part from this effect, feeds the eyewall cloud with higher-θe air than would be supplied from directly beneath the eyewall as in Emanuel’s (1986) model. But the rising supergradient flow turns radially outward in a layer atop the inflow layer as it seeks gradient balance. This low-level radial flow reversal allows air to flow out of the eye region into the base of the eyewall cloud.

c. Connecting the balanced vortex with a simplified boundary layer

Although Emanuel’s (1986) model underestimates the θe of the air entering the base of the eyewall (cf. Figs. 12a,b), the overall effect of this underestimate is slight on the mean structure of the tropical cyclone. Bryan and Rotunno (2009) show that the parcels of air entering from the eye make the eye only about 0.3 K warmer on average and compose only about 8% of the air rising in the eyewall. The individual parcels entering from the eye do have increased in buoyancy and, as will be discussed in later sections, may account for locally intense updrafts embedded within the eyewall. But since the mean structure of the eyewall cloud is largely accounted for by Emanuel’s approach, it remains useful and instructive to utilize his approach to illustrate the basic sloping structure of the streamlines (i.e., the m and θe surfaces) in the eyewall cloud. This slantwise circulation remains the primary determinant of the sloping nature of the eyewall cloud, and following this simplified approach we can obtain a view of how the overturning of air required to balance the tropical cyclone vortex takes on the structure made visible by the outward-sloping eyewall cloud (Figs. 5c and 11).

The secondary circulation consistent with the boundary layer regions II and III in Fig. 12a is obtained by considering the equations for a vortex in both hydrostatic and gradient-wind balance (appendix A). The secondary circulation is indicated by mapping the isolines of angular momentum m in the rp plane; since m is conserved, the m surfaces are also the trajectories of the mean air motion. Following Emanuel (1986), we obtain the surfaces of angular momentum m(r, p) implied by the conservation of m by substituting in (2) and integrating upward, using the value of m at the top of the boundary layer as a boundary condition and assuming that the mature vortex has adjusted to a state of conditional symmetric neutrality everywhere above the boundary layer.

Since the region above the boundary layer is assumed to have adjusted to conditional symmetric neutrality, the saturation equivalent potential temperature θes (appendix B) is uniform along surfaces of constant m above height z = h. The value of θes on the m surface is assumed to be equal to the value of the boundary layer θe at the point where the m surface intersects the top of the boundary layer; that is,
This assumption assures that a parcel that becomes saturated in the well-mixed boundary layer of the tropical cyclone will experience neutral buoyancy when displaced along an angular momentum surface extending above the top of the boundary layer—that is, the condition of conditional symmetric neutrality will be maintained in the vortex above the boundary layer, particularly in the eyewall cloud. Emanuel’s (1986) assumed relationship of m and θes above the boundary layer to θe in the boundary layer is illustrated by Fig. 12a.
Since under conditionally symmetrically neutral conditions there is a 1-to-1 correspondence between thermodynamic variable θes and the angular momentum m, thermodynamic relations can be employed to relate the radial gradient of m to thermodynamic variables. The result (see Houze 1993, 418–419, for the derivation) is
where α is specific volume, T is temperature, and cp is the specific heat at constant pressure. With ∂m/∂r given by (4) and ∂m/∂p given by the thermal-wind equation in (A6), the slope of the m surface (2) becomes
Integrating (5) along an m surface from some arbitrary radius r to r = ∞ leads to
where Tm(p) is the temperature on the m surface at pressure p and To is the outflow temperature on the m surface (i.e., the temperature at r = ∞), and we have made use of (3).
Since it is assumed that the tropical cyclone has adjusted to a state of conditional symmetric neutrality, Tm(p) in (6) is given by the temperature along the saturation moist adiabat corresponding to θe(h). Equations (3) and (6) then allow construction of the fields of m and θes throughout the region above the boundary layer, provided To and the radial distributions of p, m, and θe at the top of the boundary layer (z = h) are known. To make the calculations tractable, the outflow temperature To is assumed to be a constant, with a value representative of the upper levels of the storm. The rationale for this simplifying assumption is that the streamlines emanating from the eyewall cloud all exit the storm at upper levels (see Emanuel 1986 for further discussion). The radial distributions of p, m, and θe at z = h are determined, following Emanuel (1986), by assuming that the temperature at the top of the boundary layer is a constant TB. Then (3), (6), and the gradient-wind equation applied at the top of the boundary layer [the primary assumption removed and revised by Smith et al. (2008)] lead to
where the subscript a indicates evaluation at a radius far distant from the storm center and Rd is the gas constant for dry air. This e