Diurnally forced convection was observed over the Tiwi Islands, north of the Australian continent, as part of the Maritime Continent Thunderstorm Experiment. Immature peninsula-scale (5–15 km) sea breezes were observed to initiate moist convection early each day, principally through convergence that results from the confluence or collision of peninsula breeze fronts. Convection initiated by peninsula-scale breezes usually fails to organize beyond a small cluster of cells and dissipates as a local event. Mature island-scale (∼100 km) breezes develop by late morning and subsequently play a pivotal role in the forcing and evolution of organized convection.
The initiation of mesoscale convective systems (MCSs) is observed to be a direct consequence of breeze front collisions for only ∼20% of the days on which organized convection develops. This is referred to as “type A” forcing and it occurs when normal convective development is delayed or otherwise suppressed. Type A forcing is nature’s backup mechanism and it is less likely to produce large or strong mesoscale convective systems when compared to the general population of events.
On approximately 80% of days during which organized convection develops, a multiple-stage forcing process evolves through complex interactions between preferred sea breezes and convectively generated cold pools. So-called type B forcing emerges 1–3 h before penetration of the sea-breeze fronts to the interior island. Type B evolution has at least four stages: 1) leeward- or other preferred-coast sea-breeze showers that develop small cold pools, 2) showers that travel inland when their cold pools become denser than the marine boundary layer, 3) westward propagation of squalls that result from a merge or maturation of small cold pools, and 4) interaction between a gust front and a zonally oriented sea-breeze front of island scale (∼100 km). A collision of gust fronts, emanating from separate convective areas over Bathurst and Melville Islands, can excite a fifth stage of development associated with many of the strongest systems.
A principal finding of this study is that all MCSs over the Tiwi Islands can be traced backward in time to the initiation of convection by island-scale sea breezes, usually of type B near leeward coasts. Subsequent convective evolution is characteristic of traveling free convection elsewhere in that it organizes according to cold pool, shear balance, and mean flow factors. The presence of a critical level in the lower troposphere is a unique aspect of the theoretical “optimal condition” associated with island convection in a low-level jet regime; however, the data presented here suggest that the effects of surface layer stagnation may be of greater practical importance.
Since the aforestated conclusions are based on time series of rather limited duration, the reader is cautioned as to uncertainty associated with the climatological frequency of events as described herein. Furthermore, the authors have not examined external forcings, which may be associated with large-scale circulations.
Tropical islands are among the rainiest places on Earth, however, the reasons for this are often poorly understood. Quantitative prediction of tropical island convection and its larger-scale effects remain problematic at all timescales. Radiatively, hydrologically, and from a momentum and chemical redistribution standpoint, island convection represents a significant component of tropical weather systems within the global climate system.
Studies of convection over islands with steep orography are fairly numerous and the forcing of rainfall in such circumstances is increasingly understood, even when orography is sufficient to block flow (e.g., Smolarkiewicz et al. 1988; Carbone et al. 1995, 1998). Conversely, studies of convection over flat islands that are large enough to sustain convection for an appreciable fraction of the diurnal cycle are quite rare. A notable contribution to this topic is by Simpson and colleagues (Simpson and Bunker 1952; Simpson and Stern 1953; Stern and Simpson 1953) who observed boundary layer heating over Nantucket Island and applied linear theory to describe island boundary layer evolution. Broadly speaking, however, mechanisms associated with the evolution of deep convection over flat islands have been inadequately identified and largely unquantified.
Our purpose is to report an improved definition of the convective life cycle as may be generally applicable to flat tropical islands in the easterly jet regime. The Tiwi Islands fit this description and these are the site of legendary Hector. Hector is a vigorous mesoscale convective system (MCS) that develops daily during the transition and break seasons in response to diurnal heating (Keenan et al. 1990, 2000).
a. Maritime Continent Thunderstorm Experiment and local environment
The Maritime Continent Thunderstorm Experiment (MCTEX) was conducted in late 1995 on the Tiwi Islands, 50–100 km north and west of the Australian mainland, at latitude 11°S (Keenan et al. 2000). The combined areal extent of the islands is 150 km zonally and 50 km meridionally (Fig. 1). A narrow (1–7 km) strait separates the islands, with Bathurst Island to the west and Melville Island to the east. Most of the area is eucalypt forest with mangroves and the islands are nearly flat (120-m maximum elevation). During the transition seasons (November–December, February–March) and intermittent monsoon break periods (mid-December–mid-February), an ESE tradewind regime prevails, including a 70-kPa jet (5–15 m s−1) and weak surface winds (Keenan and Carbone 1992; Keenan et al. 2000). These features, along with others that are characteristic of the Tropics, are illustrated (Fig. 2a) in the 1130 local time (0200 UTC) sounding from Maxwell Creek on 28 November. Figure 2b shows the lower troposphere at 1330 local time (0400 UTC), including a mature cloud-capped convective boundary layer, at the time of rapidly developing deep convection. Hereinafter, first-listed times will be local time, UTC+9.5, since the phenomena are strongly controlled by the diurnal cycle.
b. Literature review
Convection within the tropical easterlies has been studied for decades over continents and oceans (e.g., Houze 1977; Barnes and Seickman 1984; Leary and Houze 1979; Simpson et al. 1980; LeMone et al. 1984;Chong et al. 1987; Velasco and Fritsch 1987). Similarly, sea breezes have been the subject of many investigations (e.g., Pielke 1974; Physick 1980; Rotunno 1983; Simpson 1987; Kraus et al. 1990; Simpson 1994). Studies specific to the Tiwi Islands are relatively few. Keenan et al. (1990) observed Hector by means of radar from the mainland of Australia, near Darwin (Fig. 1), about 70–120 km south. The observations were highly suggestive of center-island confluence of sea breezes and cumulus “mergers” described by Simpson et al. (1980, 1993) and Keenan et al. (1994). Simpson et al. (1993) identified first-order mergers that took the form of a zonal “cloud bridge” across the island complex aided by gust front outflows between neighboring cumuli (Tao and Simpson 1984, 1989). They described a second-order merger process that led to a solid line of precipitation radar echo extending zonally across the Tiwi Islands, presumably at or near the confluence of zonally oriented sea-breeze fronts. Significantly, Simpson et al. (1993) associated suppressed convection with “fast” surface layer flow (5–7 m s−1).
Keenan and Carbone (1992) described convection in the region of northern Australia, including the Tiwi Islands, where sea-breeze collisions were suspected as a main initiation mechanism. They concluded that long-lived squall lines and other convective systems evolved and propagated in a manner generally consistent with other mesoscale convective systems observed in an easterly jet regime (Leary and Houze 1979; Barnes and Sieckman 1984). Keenan and Carbone found that typical squall systems truly propagated in the sense that westward motion was at a speed slightly faster than the strongest easterly momentum in the troposphere. There was consistent evidence for redistribution of easterly jet (70 kPa) momentum to the surface layer, thus accounting for the convective system propagation. They also observed that squall systems behaved in a manner consistent with established bulk Richardson number (Weisman and Klemp 1982) and cold pool balance concepts (e.g., Thorpe et al. 1982; Rotunno et al. 1988).
Golding (1993) numerically simulated Tiwi Island convection with a U.K. Met. Office model. The simulations showed an island-scale confluence of sea breezes, downward transport of easterly momentum, and cold pool interactions among breeze-forced storms. Crook (1997) performed both idealized and MCTEX case-specific simulations with a cloud model (Clark 1977). He found preferred convective boundary layer development toward the lee coast associated with linear flow past an idealized elliptical heat source. Crook concluded that collision of zonal sea-breeze fronts routinely initiates or amplifies deep convection. Recent simulations by Saito et al. (2000, manuscript submitted to Mon. Wea. Rev., hereafter SKHP) are broadly consistent with these findings, including extensive sea breeze–cold pool interactions.
Studies conducted in Florida as part of the Convection and Precipitation/Electrification (CaPE) Experiment are germane to the evolution of Tiwi Island convection, both for the similarities and differences in environmental conditions. In comparison, the Florida peninsula is a very large landmass that has a meridional orientation and similarly aligned sea-breeze fronts. The vertical shear over Florida is most often westerly, with westerly flow aloft. Several publications, including Wakimoto and Atkins (1994), Atkins et al. (1995), Kingsmill (1995), and Weckwerth et al. (1997), lend insight to factors associated with the convection that is forced by Florida sea breezes. From the CaPE data it has been determined that longitudinal rolls in the convective boundary layer strongly modulate the moist convection along sea-breeze fronts (e.g., Weckwerth et al.). Sea-breeze propagation inland is largely consistent with gravity current propagation, but this is quite slow with respect to the width of the Florida peninsula (Wakimoto and Atkins). Strong gust fronts propagate eastward from Florida’s west coast and often interact with the east coast sea breeze. While this usually leads to increased convection, it does not always strengthen the net forcing associated with the sea breeze alone (Kingsmill).
The very high frequency of convection on the Tiwi Islands and the relative uniformity of conditions facilitate the detection of local environmental influences that may prescribe the evolution of Hector. Among such influences are the initiation of thunderstorms along boundary layer convergence lines and intersections of convergence lines (Wilson and Schreiber 1986; Atkins et al. 1995). The initiation of particularly intense storms is often associated with the confluence or collision of convergence lines (Purdom 1976; Wilson and Schreiber 1986; Carbone et al. 1990; Fankhauser et al. 1995). Convective lines have been shown to exhibit sensitivity to the relative motion between gust fronts and storm cells (Weisman and Klemp 1986; Wilson and Megenhardt 1997). Finally, Crook (1996) has demonstrated the strong sensitivity of convection intensity to small-amplitude and small-scale changes in boundary layer water vapor and temperature.
c. Data and data analysis issues
Herein, the MCTEX analyses utilize data from the Flinders Cessna 340 aircraft, the Aerosonde remotely piloted aircraft, the C-POL Doppler polarimetric radar, a mesonet of 15 automatic weather stations, Geostationary Meteorological Satellite (GMS) imagery, and serial soundings taken from Maxwell Creek (Fig. 1) as described in Keenan et al. (2000). The aircraft cross-sectional analyses (e.g., Figs. 6a,b) are produced from time series that are averaged to 1 Hz. These data are interpolated, using a Laplacian-based algorithm (Young and van Woert 1992), onto a rectangular spatial grid with approximately 2-km and 50-m spacing in horizontal and vertical directions, respectively. A Laplacian smoothing process and a contouring routine are subsequently applied (Young and van Woert 1992). When cross-sectional analyses span a coastline we make small adjustments to datum locations as a consequence of differences in sampling time. By referencing all data to the coastline, a linear dilation is applied to the along-track position of data according to the propagation of the sea-breeze front (typically less than 1-km movement).
All radar data presented are from the Bureau of Meteorology Research Center’s C-pol radar, which was located at Nguiu (Fig. 1). This radar is both Doppler and polarimetric in its capabilities; however, the polarimetric data are not used herein. Radar reflectivity data shown herein are from conical scans at the lowest elevation angle feasible (0.5°–2.5°) and thereby represent conditions in the lowest 2 km, depending upon radar range. Both radar reflectivity and Doppler data are used in conjunction with aircraft data to determine the positions of sea-breeze fronts, gust fronts, and their respective motions. The radar methods employed are similar to those described by Wilson and Schreiber (1986) and further validated by Wilson et al. (1994). All satellite data presented herein are visible imagery from Japan’s GMS. This imagery is received and stored at somewhat degraded resolution (4 km) in Australia by the Bureau of Meteorology. Neither the resolution nor the navigation information associated with these data support precise location of convergence lines, including breeze fronts.
All sounding data herein are from the MCTEX site at Maxwell Creek, located in west-central Melville Island (Fig. 1), or special soundings acquired from ascents and descents in the Cessna 340. Convective boundary layer depths are clearly identified from inspection of surface-based mixed layers in these soundings. It has been determined that the rawinsondes of Vaisala manufacture had a (∼5%) low humidity bias that was in itself a function of instrument age, temperature, and relative humidity. Standard corrections, developed and approved by Vaisala Corporation and the National Center for Atmospheric Research (NCAR), were applied to these data by NCAR.
As will be discussed in section 2b and elsewhere, it was essential to determine the regional mean flow in the lower boundary layer and surface layer. With few exceptions, this determination was straightforward given the corroborating data from island surface stations, 920-MHz profiler and rawinsonde soundings at Maxwell Creek, Doppler radar, and the Cessna 340 missions. In estimating this flow, highest weight was given to the aircraft, profiler, and radar data when available, and on radiosonde data thereafter. Surface station data, while nearly always consistent with other data sources, were relied upon least, owing to exposure uncertainties, breeze-affected sites at coastal locations, and a tendency toward stagnation at the 10-m height over the islands’ interiors. One day (28 November) presented serious difficulty in categorizing the direction of lower boundary layer flow because no consistent mean flow was indicated from any source.
d. Brief summary of events and the environment in MCTEX
The primary MCTEX datasets were acquired from 20 November through 4 December 1995. Organized deep convection occurred on each of these 14 days. Table 1 summarizes many environmental factors that are known or suspected to influence the initiation, organization, propagation, and intensity of deep convection. These include both thermodynamic and kinematic quantities as well as the identification and location of active sea breezes, gust fronts, and radar echo characteristics. Quantities derived from rawinsonde data are generally representative of the period 1030–1230 (0100–0300) whenever possible. Throughout this manuscript, numerous references will be made to the contents of Table 1 in conjunction with discussion of processes and events.
Note in Table 1 that convective available potential energy (CAPE), as estimated from rawinsonde soundings, varies from 3000 J kg−1 to conditional neutrality. These values are weakly correlated with measures of convective intensity on any given day and are often energetically inconsistent with the intensity of convection observed (e.g., as estimated by area and time of radar reflectivity, dBZ). Convective inhibition (CIN, J kg−1) is similarly unreliable as a predictor variable but it is weakly correlated with convection that is delayed or suppressed. Apparent inconsistencies between CAPE/CIN and convective activity may be related to heterogeneous lower-tropospheric humidity conditions, as was observed by aircraft and surface stations in the normal course of island boundary layer and sea-breeze evolution. Weckwerth et al. (1996) reported similar inconsistencies in Florida and attributed these to the modulation of water vapor by horizontal convective rolls.
Table 1 also provides wind-related information, including “low level” winds (below 500 m), the easterly jet maximum wind, and “steering level” (2–4-km average) winds. Low-level winds provide important information on the residence time available for sensible heating over various parts of the islands. Table 1 indicates that 28 November is the best MCTEX case to exhibit the effects of boundary layer flow stagnation. So-called steering-level winds are characteristic of individual convective cell motions. Average low-level shear (below the jet maximum) is provided as a measure of ambient horizontal vorticity in the cold pool layer. The vector subtraction of gust front motion and jet maximum wind allows a determination of MCS “propagation,” which is defined as westward convective line motion faster than any easterly momentum in the lower troposphere (Barnes and Seikman 1984; Keenan and Carbone 1992). As seen from Table 1, MCSs in MCTEX typically propagated 1–2 m s−1.
e. Outline of contents
Section 2 describes the early sea breezes, cumulus distributions, and the basic island boundary layer (IBL) structure. Section 3 examines the transition from precipitating sea-breeze cumulus to traveling convection. Section 4 describes the mature MCS stage of convection (Hector) for the range of conditions discussed in section 3. Section 5 places the principal findings in a broader conceptual framework, and section 6 summarizes the principal conclusions.
2. Breeze and island boundary layer development
Based upon the literature and other expectations, MCTEX was designed, in part, to confirm a “collision of sea breezes” hypothesis associated with MCS initiation. The sea breeze initially develops from a shallow land-breeze condition as described by Keenan et al. (1990). The land breeze often extends well into Beagle Gulf south of the islands where it can interact with the mainland land breeze to force weak oceanic convection in the predawn hours. Except under synoptically disturbed conditions and unlike the westerly monsoon, there is relatively little deep convection over the surrounding ocean during the transition season. Solar heating establishes a well-developed sea breeze by midmorning (Fig. 3) that hugs the island periphery. Surface winds over the island are usually light (<4 m s−1) and sometimes highly variable in direction. During MCTEX the surface winds blew from all quadrants with similar frequency (Table 1). Lower-tropospheric (1–4 km) winds were always easterly from 4 to 14 m s−1, thus establishing a persistent easterly shear condition from the surface through 2–4 km. From a meteorological perspective, sea surface temperature is uniformly warm around the islands, averaging approximately 31°C.
a. Immature breezes and peninsula convection
We define breeze development through 1100 (0130) as “immature.” The immature breeze is characterized by a growing IBL with a steadily increasing density discontinuity between land and sea, of order 0.2%–0.5% in the lowest few hundred meters as measured by aircraft penetrations (Fig. 4). Preferred cumulus development is on or near peninsulas, and many of these peninsulas produce the earliest scattered showers along the sea-breeze front. Figure 3 illustrates an immature breeze that is in transition to a mature breeze and will eventually force a mesoscale convective system. Coastal flow is directed landward, approximately normal to the local topography. A shallow cloud line over western Melville Island (Fig. 3) is suggestive of a weak internal breeze associated with Apsley Strait, which separates Bathurst and Melville Islands. The interior island flow is weak and variable with a SE tendency, the prevailing wind direction over the ocean in the marine boundary layer (MBL). With minor exceptions in Fig. 3, convection over the Tiwi Islands is firmly tied to the sea-breeze fronts and does not travel inland. Also evident in Fig. 3 (upper right) is the presence of westward-traveling convection E of Melville Island, which originated over the mainland near Cape Don (Fig. 1).
Figure 4a illustrates a similar immature breeze on 23 November when light WSW flow prevailed at the surface together with the easterly jet aloft. Convection is relatively well developed at the eastern peninsulas of Melville Island, as is common on days with a westerly component in the surface flow. In its initial flight leg of the day, the Cessna 340 sampled the western peninsula breeze over Bathurst Island at 70-m altitude along the track indicated in Fig. 4a. Data from the SE–NW segment of this track (Fig. 4b) depict virtual potential temperature (θυ), water vapor mixing ratio (q), equivalent potential temperature (θe), wind direction, and infrared (IR) radiative surface temperature (°C) across the peninsula. The change, Δθυ, is approximately 1°C, a mere 0.3% density deficit over the island. Owing to this thermal forcing, confluence of the wind across the peninsula is approximately 3 × 10−4 s−1. With growth of the IBL, vertical entrainment of dry air depletes water vapor by up to 2 g kg−1 toward the northern shore of the peninsula. Here we observe a corresponding decrease in θe associated with the IBL entrainment, a result that is noteworthy for all convective IBLs observed in MCTEX. While strong positive land–atmosphere sensible and latent heat fluxes are present, there is a cumulative decrease of 3–5 g kg−1 in the IBL specific humidity by midafternoon. One consequence of IBL drying is to have heated (but potentially less energetic) IBL air surrounded by dense (but potentially the more energetic) modified maritime air, which is trapped within the breeze circulation near shore. The breeze fronts in Fig. 4b are located at slight θυ depressions at approximately 2 and 9 km N of the S shore. These are approaching one another at a rate of approximately 1 m s−1, and converge or “collide” to amplify precipitating convection over the peninsula. On some days peninsula convection can be quite vigorous but it is localized, apparently because the quantity of heated IBL air over a small peninsula is limited.
b. Island-scale boundary layer structure
1) Zonal structure in a developing IBL
The IBL structure on 23 November (Fig. 5) is consistent with a slow mean WSW flow across a heated surface with dry air above the boundary layer inversion. Figure 5a shows the Cessna time series of IBL zonal variation over a 180-km distance at 100-m altitude from west of Bathurst (left) to the east coast of Melville (right). The IBL at this time is relatively deep (500–800 m), but considerably shallower than its mature depth (1.2 km). A west to east trend in the boundary layer temperature (and depth) is indicated by the steady increase in θυ from 32.5° to 34°C. The windward breeze front is located near the west coast of Bathurst Island, approximately 5 km inland. This breeze front is similar to an early peninsula sea breeze in that the θυ discontinuity is <1°C (<0.3%) and there is a landward reduction in q of nearly 2 g kg−1. The eastern Melville breeze front is located near shore and is characterized by a θυ discontinuity of 2°C (0.7%). Figure 5a shows nearly 3 g kg−1 reduction in q from the MBL to the IBL. On the island scale, there is a weak trend for reduced q with increasing boundary layer depth. The pattern in θe follows that of q. As will be discussed in the next section, the fully developed IBL (approximately 2 h hence) is 1.2 km deep at Maxwell Creek and it is 2.5°–3°C (θυ) warmer than the MBL at 100–300-m altitude near the leeward shores.
2) Meridional structure in a mature IBL
Figure 5b shows the Cessna time series at 300-m altitude of a mature IBL in meridional cross section about 2 h later than Fig. 5a. The IBL exhibits a clear warming/deepening trend from the windward south shore to the leeward north shore, where Δθυ is 3°C (1%). Another common pattern of q and θe is illustrated on this flight segment, namely the pronounced q minima (14 g kg−1) inland of each breeze front in comparison either to the MBL (18 g kg−1) or to the island interior (16 g kg−1). As will be shown in section 3b, there is increased vertical mixing near breeze fronts. Note that q and θe reach a local maximum near the leeward breeze front where confluent flow (1.5 × 10−3) and vertical air motions of 1–2 m s−1 are observed (above 500-m altitude). We discuss the significance of these and other sea-breeze structures in section 3b.
3. Transition to traveling convection
a. Typical conditions under slow mean flow
It is unusual for low-level flow to exceed 4 m s−1 (Table 1) during the transition season. It follows that days on which the low-level flow is weak are likely to exhibit typical convective developments. It has been shown that periods of stronger flow may be associated with suppression of convective activity (Simpson et al. 1993). On 23 November, 1–4 m s−1 WSW flow conditions favor development of a mature breeze front (Fig. 6a) along the eastern shore of Melville Island. One hour after the peninsula-scale convective activity (Fig. 4a) begins to subside, more vigorous convection initiates on mature leeward coast breeze fronts. Deeper clouds initially develop, along the east coast sea-breeze front (hereinafter ECBF) and subsequently along the north coast breeze front (NCBF).
Figure 6b shows the first radar echoes located near the ECBF. These echoes are the seeds of organized convection that eventually lead to an MCS. Figures 6c,d show vertical cross sections synthesized from a vertical stack of three Cessna penetrations between 100 and 900 m just before convection undergoes rapid evolution. The erect ECBF is 2 km inland as defined by the leading edge of the θυ gradient (Fig. 6c), and extends upward to approximately 900 m. The density difference across the breeze front averages about 1°C (0.3%) throughout its depth, being twice that value near the surface and less than half that near the top. Gravity current propagation rate, V, is defined by Benjamin (1968) and Simpson and Britter (1980) as
where k is a dimensionless internal Froude number of order unity, g is gravitational acceleration (m s−2), h is the depth of the current (m), U is the environmental headwind (m s−1), and Δθυ/θυ is the fractional density difference. Unlike the finding of Wakimoto and Atkins (1994), where k was 1.07 in a Florida sea breeze, the propagation of this breeze front (1.7 m s−1) and others in the MCTEX dataset is little more than half that rate (k = 0.5–0.6). These data are reasonably consistent with the findings of Simpson (1969), who found k = 0.62 in laboratory experiments. However, the observed propagation speed is at the extreme low end of that which could attributed to a pure gravity current mechanism. As will be illustrated in section 3b, a major difference between sea breezes and the classical “dam break” problem is the strong diabatic heating in breeze flow, as air passes over a monotonically increasing fetch of heated land.
Updrafts of 2–5 m s−1 (Fig. 6d) are observed from 1 to 7 km on the IBL side of the breeze front. Flow extends rearward (eastward) over the head of marine air (Fig. 6c). Specific humidity and θe are enhanced on the MBL side of the frontal interface. While the cloud updrafts are located on the warm/dry side of the breeze front, rain will soon fall on both sides of the ECBF.
Transport associated with convection and subcloud precipitation evaporation leads to a more complicated thermal structure than had previously existed along the ECBF (Fig. 7). At 1238 (0308) the Cessna penetration of this developing shower reveals the formation of the first convectively produced cold pool. Divergent winds sampled by the Cessna are shown to coincide with the highest radar reflectivity in the developing storm (Fig. 7a). Easterly momentum is consistent both with the ECBF and prevailing winds above the lower IBL. Time series from this location are shown in Fig. 7b and reveal a small cold pool (<2 km) under the reflectivity core. It is located 7 km inland, just on the IBL side of the ECBF. This cold pool is relatively dry and θυ is 1.2°C colder than the nearby MBL air. A conservative thermodynamic property of this air, wet-bulb potential temperature (θw), requires it to have been transported downward to the 500-m level from 88 to 89 kPa, a level that is slightly above the IBL. Approximately 1.5 g kg−1 of moistening occurred, thus cooling this air by 3.5°–4°C.
An observation of general significance in the MCTEX dataset is that a “chain reaction” of localized IBL convergence zones appears to be triggered by the development of numerous early cold pools near breeze fronts, similar to the one sampled by the Cessna on 23 November. While qualitatively similar in concept to first-order mergers reported by Simpson et al. (1993), these preferentially occur in regions not far removed from leeward coast breeze fronts. Figure 8 shows the development of radar echoes over the 2-h period following Fig. 7a. On 23 November there is a period of organizational malaise (Figs. 8a,b) where numerous small echoes are initiated at the western edge of the original cold pool. From this case and numerous others, it is known that these echoes produce cold pools of their own, which subsequently interact with other nearby cold pools. When this quasi-chaotic stage of development is complete (e.g., Fig. 8c herein), eastern Melville Island is covered with air that has been cooled 3°–6°C from the heated interior and is 1°–3°C cooler than marine air, which surrounds the islands. This stage of development is crucial, representing a transition from breeze-forced and breeze-maintained convection to self-organized and self-sustained traveling squalls. Convection frees itself from the ECBF and begins to travel westward as an organized small squall (Fig. 8c). In the particular instance of 23 November, subsequent westward propagation is along the leeward NCBF, thus forming the basis for initiation of an MCS.
b. Delayed or suppressed conditions
1) Convective development
Pure sea breezes, unmodified by convectively generated cold pools, are usually short lived in the Tiwi Islands, owing to rapid convective development and the early disruption caused by precipitation (Golding 1993, SKHP). For this reason, deeply penetrative breeze fronts (>20 km inland) occur on a minority of days, typically when moist convection is suppressed or delayed. These deeply penetrative breeze fronts can be associated with the forcing of Hectors in a zonal region over the interior island (e.g., 26, 27, 29 November). The convective systems that result from this type of forcing tend to be smaller in size compared to the larger population (Keenan et al. 2000; Table 1 herein), and growth to the MCS stage is delayed by 1–3 h from the typical diurnal evolution. Soundings (Table 1) suggest below average CAPE for these events; however, the Maxwell Creek data are less than fully consistent with this characterization. November 26 is the purest example of late-day organized convection that is associated with deeply penetrative zonal sea breezes. An afternoon sounding was taken from Maxwell Creek near the time of NCBF passage (Fig. 9). Conditions departed significantly from the norm in that IBL humidity is 2–4 g kg−1 below average;it is very dry aloft; the level of free convection is high (80 kPa); and westerly shear above the easterly jet is above average in strength (5 × 10−3 s−1). Organized deep convection was delayed from an average onset time of 1247 (0317) to 1516 (0546), as defined by Keenan et al. (2000) and, similarly, in Table 1 (column 12) herein. Extensive airborne observations of the south coast breeze front (SCBF) were performed throughout the genesis period under prevailing NE flow conditions.
Northeasterly surface flow favors the development of convection along southern and western coasts as depicted in Fig. 10a at 1200 (0230). Convection is suppressed, in part, because average IBL water vapor (prior to breeze front confluence) is only 12–13 g kg−1, about 4 g kg−1 below the norm. Following the breeze front confluence, q increases by 2 g kg−1. Typically, free convection is in progress by 1400 (0430), but Fig. 10b shows a suppressed condition where only shallow breeze cumuli are positioned along breeze fronts at that time. Inland penetration of the breezes, uncomplicated by precipitating convection, proceeds until 1500 (0530) when deep convection finally develops. The confluence or collision of island breeze fronts is evident in Fig. 10b, save for a small pocket of unmodified IBL near the center of Melville Island. The continuous zonal line of cloudiness is reminiscent of the “cloud bridge” associated with a “first-order merge” described by Simpson et al. (1993). Deep convection is subsequently forced along the central island convergence zone (Fig. 10c) at 1515 (0545). While delayed, the conditions are not permanently suppressed since a Hector of below average strength develops later on this day. Maximum areally integrated rainfall rate occurs at 1634 (0704) and the Hector rain volume ranks 17th in magnitude out of 20 events during MCTEX as calculated by Keenan et al. (2000).
2) Breeze circulation structure and propagation
We observed the SCBF for a 2-h period between 1200 (0230) and 1400 (0430) by means of multiple passes of the Cessna, nominally at the 70-, 300-, and 980-m levels. We tracked the breeze front using radar to a position 22 km inland, near the midisland collision at approximately 1500 (0530). Cessna cross sections of the evolving SCBF are shown in Fig. 11. The locations of these data have been adjusted for propagation within each aircraft-sampling period as described in section 1c. Figure 11 illustrates the salient features of the breeze front at two times separated by approximately 1 h. The top two panels of Fig. 11 show the wind and thermodynamic structures circa 1230 (0300). The breeze front (indicated by arrow) is 6.5 km inland and extends through 1-km altitude. The density difference (Fig. 11a) between the MBL and IBL is 2.2°C (0.7%) near the surface, decreasing to near zero at 1 km. The circulation is clearly solenoidal from 6 km inland to well offshore. Vertical air motions (Fig. 11b) are strongly positive near the breeze front with a 3 m s−1 updraft in the thermal gradient region at 300 m and a 2 m s−1 updraft near 1 km, slightly forward of the surface breeze front. Specific humidity structure is strongly suggestive of a gravity current “head” (Simpson 1987, 1994), reflecting the deepening of water vapor as the marine air passes over land. While solenoidal lifting occurs, sensible and latent surface fluxes also modify the MBL properties, once the breeze front has propagated a sufficient distance onshore.
Figures 11c,d show progression of the breeze front 1 h later. Movement of the front is 2 m s−1, which, according to (1), is the propagation speed of a pure gravity current if k = 0.55. As evidenced in (Figs. 11c,d) the marine air over land is heavily modified at this time. This results in a dual-frontal structure where the density difference (Δθυ) is stronger near the coast than the “primary” breeze front located 14 km inland. Since the breeze front continually propagates farther from its source of cool air, sensible heating over the land surface, together with vertical entrainment of potentially warmer air, have more time to decrease the MBL–IBL density difference. There is strong evidence of vertical entrainment over the ocean in Fig. 11d. Specific humidity falls below 12 g kg−1 at 650 m, which is the driest air in the entire breeze circulation. This feature is consistent with the “wake region” of some gravity currents (e.g., Simpson 1987, 1994). Finally, the updraft (Fig. 11d) is consistent with the observed thermodynamic structure, having a maximum vertical velocity slightly to the IBL side of the breeze front, and tilted seaward.
Interpretations from Fig. 11 are confirmed upon inspection of the individual time series (Fig. 12). The upper panels (Figs. 12a,b) are 980-m legs circa 1230 (0300) and 1330 (0400) that clearly reveal the overshooting updraft at the breeze front (indicated by arrows). In both passes, at the position of the breeze front, there is a spike of negative buoyancy (Δθυ = −1°C) and enhanced specific humidity (q = 14 g kg−1) that is consistent with the Maxwell Creek sounding (Fig. 9). The radiative surface temperature also indicates a cloudy sky condition. The lower panels (Figs. 12c,d) show conditions at 70-m altitude. The aforestated updraft properties are uniquely satisfied at the primary breeze front (indicated by arrows) in both cases. The updraft air is approximately 1°C cooler and 1 g kg−1 moister than the IBL, and 3.5°C warmer and 2 g kg−1 drier than the MBL. In any particular pass of the aircraft through various breeze fronts, θe may be elevated or relatively constant at this interface, but in both of these instances it is strongly elevated (4°C, not shown). Growth of the IBL eventually conquers inhibition on 26 November. The breezes elevate both the amount and depth of the specific humidity, thereby locally “manufacturing” CAPE (Crook 1997; SKHP). The result is a Hector, somewhat delayed, displaced, and of below average strength.
3) Development under conditions of near stagnation
On 28 November, island surface layer mean flow was near stagnation (Figs. 2 and 3). Owing to the long residence time of surface layer air over the islands, IBL properties are expected to be relatively uniform under these circumstances. Figure 13 depicts convection associated with the ECBF near local noon. Initially (Fig. 13a) echoes are scattered along three Melville breeze fronts with the most vigorous developments along the ECBF and the SCBF. As evidenced by the visible cloud tops in Fig. 13d, two showers enter the traveling convection stage at approximately 1200 (0230), one along the SCBF, which travels WNW toward GOOS (Fig. 1) and the other along the ECBF positioned similarly to the typical case described in section 3a. The GOOS echo produces a relatively minor cold pool in south central Melville while the ECBF echo exhibits stronger growth. Figure 13b shows this ECBF radar echo at 1152 (0222) and the location of a developing cold pool with thermodynamic characteristics that are summarized in Fig. 13c. There is a sharp wind reversal associated with a developing gust front at the western edge of the radar echo. The Cessna time series (Fig. 13c) reveal the subcloud condition at 100-m altitude. A θυ depression of 3°C (relative to the undisturbed IBL) is located at the dot in Fig. 13b and the strongly cooled region is >10 km wide. This event is 1°C cooler than the nearby MBL and q is elevated by 1.4 g kg−1 with respect to the mean IBL condition. According to these properties (and unlike the case in section 3a), the cooled air is essentially a wet-bulb moistening of the IBL mixed layer. This cold pool establishes a major gust front along its western boundary, which subsequently forces one of the strongest MCSs in MCTEX. Immediately following, section 4 discusses the evolution of this gust front and its dynamical interaction with the NCBF under conditions of near stagnation.
4. Transition to Hector
a. Gust front–breeze front interaction
As illustrated in section 3, the transition to traveling squalls is accompanied by a westward propagating gust front (hereinafter, GF). All mesoscale convective systems observed during MCTEX were the consequence of a GF that subsequently amplified convection along the full extent of a zonal breeze front (ZBF). An annotated GMS image of a typical GF–ZBF configuration is shown in Fig. 14 for 25 November, when the GF propagated westward along the SCBF under N flow conditions. Preexisting sea-breeze convection is amplified forced by the GF at the westward-propagating intersection of both fronts.
Herein we present quantitative evidence of strong forcing at the intersection between a westward-propagating GF and the NCBF on 28 November. The Cessna sampled this intersection at the time of rapid evolution from a small squall to the MCS stage of development. Vertical velocity up to 5 m s−1 was measured below cloud base by the Cessna along the GF associated with convection that was traveling westward from the ECBF in Fig. 13d. Vigorous cumulus congestus was observed visually for ∼20 km along the GF, southward from the original radar echo. At 1200 (0230) this storm and the shower near GOOS are the only precipitating clouds to separate from their respective breeze fronts and enter the interior of the island. A mere 30 min later (Fig. 15), the convectively generated cold pool is mature (Δθυ = −9°C) and it propagates W and SW. The GF location in Fig. 15 shows the outflow boundary where the leading edge is defined as the analyzed position of Δθυ = −1°C (depression relative to the undisturbed IBL) at the Cessna flight level (700–900 m). As evidenced in Fig. 15, 2–7 m s−1 updrafts are forced along the GF, thus triggering new convective cells that soon lead to Hector. The strongest updraft in Fig. 15 is located at the intersection of the GF and the NCBF where the most vigorous convective development is evident (Fig. 16b). At this stage of development, both the thermodynamic and momentum properties of the outflow are consistent with air that had its origin near the easterly jet maximum (73 kPa), thus signaling a transition to the mature MCS stage. This finding is also fully consistent with momentum transport in long-lived squall lines over the continent (Keenan and Carbone 1992) and over the tropical North Atlantic (Barnes and Seikman 1984).
b. Examples of mature MCSs
The GF–ZBF interaction is the defining process associated with MCS formation. We believe this interaction is an explicit description of the Simpson et al. (1993) second-order merge. Numerous variations of this MCS forcing were observed on a daily basis during MCTEX, three of which are illustrated in Figs. 16, 17, and 18. Owing to limitations in the MCTEX dataset, however, none other than the 28 November case could be quantified by means of in situ aircraft data at the critical stage of convective amplification (Fig. 15).
1) 28 November 1995
As discussed in sections 1c, 3c, and 4a, the near-stagnation conditions over the island on this day are accompanied by a tendency for 1–2 m s−1 SE flow over the ocean. In principal, this slight flow might favor the north and west breeze fronts. There is evidence in Fig. 16 for preferred convective development along the NCBF. However, contrary to leeward coast tendencies, the ECBF is more active early in the development of organized convection. Simpson et al. (1993) observed that convective intensity tends to be negatively correlated with surface-layer wind speed, where suppression is evident when speeds reach 5–7 m s−1. The extensive convective activity on 28 November (e.g., Fig. 16d) may be causally related to an especially well-developed IBL, which can result from exceptionally long residence time of air over land. Given the absence of any other distinguishing characteristic on 28 November, we favor the hypothesis that surface layer stagnation is part of an “optimal” environmental condition.
A noteworthy fifth stage of convective development on 28 November is the collision of GFs (Figs. 16c,d) from separate convective areas over Bathurst and Melville Islands at approximately 1400. Based on established findings (e.g., Wilson and Schreiber 1986; Atkins et al. 1995), this collision appears to have a causal relationship to the amplification of a meridional squall line as shown in Fig. 16e. Collision of mature GFs occurs on a minority of days and is associated with several of the strongest MCSs (Keenan et al. 2000; Table 1 herein).
2) 20 November 1995
MCS forcing under SW flow begins with a small squall and attendant GF at the leeward ECBF (Fig. 17b). These travel westward along the leeward NCBF (Figs. 17d,e,f). The strongest radar echo is located at the NCBF–GF intersection (Fig. 17d). The strongest radar echoes are in excess of 55 dBZe and constitute the leading convective line in an otherwise disorganized cluster of convection (Fig. 17e). Some of the high-reflectivity cores are located slightly N of the NCBF, suggesting that relatively unmodified MBL air may have become actively involved in deep convection.
3) 3 December 1995
Figure 18 illustrates MCS forcing under NE flow. A zonally oriented squall and attendant cold pool form over SW Melville Island along the SCBF and Apsley Strait. The primary GF propagates westward (Fig. 18d), eventually forcing a meridional orientation to the leading convective line (Figs. 18e,f). Prior to MCTEX, this distribution of radar echoes (when viewed in less detail from the continent) may have falsely conveyed a sense of zonal initiation mechanisms, including the collision of ZBFs. While a single ZBF is indeed a strong forcing, the initiation of strong convection is commonly accomplished via a meridional GF or a collision of two such GFs. We believe that so-called “reorientations” (e.g., Keenan et al. 1994) of Hector, as viewed from the continent, are nearly always associated with vigorous GFs or collisions of these.
As daily observations routinely confirm (Table 1 herein; Keenan et al. 2000), convection is more likely to organize in a squall-like manner on the west edge of a cold pool in the Tiwi Islands. Apparently, this is because the ambient horizontal vorticity (4 × 10−3 s−1) in easterly shear is of opposite sign to the horizontal vorticity at the western boundary of a cold pool, thus favoring more erect updrafts (e.g., Rotunno et al. 1988) and the continual regeneration of convection.
A flat, elliptical island of order 100-km dimension can act as an oriented heat source and create an optimal condition both for the initiation of convection as well as its subsequent organization and propagation (Fig. 19). Hypothetically this requires the following:
a vertical shear vector aligned with the island’s major axis,
a surface layer flow that is in directional opposition to the shear vector,
a lower-tropospheric critical level (flow direction reversal), and
a free-tropospheric condition that can sustain deep convection.
Alignment of the surface flow with an island’s major axis allows for increased diabatic heating and IBL growth as air traverses the heat source, thereby facilitating leeward coast initiation of convection. Directional opposition of the vertical shear vector, together with the presence of a critical level (Moncrieff and Liu 1999), provides steering winds aloft, which are in opposition to the surface flow. Given the above, horizontal vorticity balance considerations (Rotunno et al. 1988) generally will favor the growth of new convection on the downshear/landward side of convectively produced cold pools near the leeward coast. Given the steering winds, this allows propagation of convection across the full extent of the potentially energetic IBL. Other combinations of island orientation, vertical shear, and flow direction may support strong convection, but these are nonoptimal (Moncrieff and Liu 1999) for long-lived organized convection on a relatively small island.
Increased inhibition or delay in the onset of free convection alters the above arguments to the extent that a confluence or collision of sea breezes is possible and may be responsible for direct initiation of deep convection. If the meridional wind is small and time is sufficient, the confluence will occur near the islands’ major axis. This reduces leeward coast tendencies and results in the abrupt initiation of organized convection interior to the island. There may be a decrease in the duration of convection as a consequence of reduced storm residence time over land, given that its initiation is, broadly speaking, over the interior island (as opposed to the optimal “end” of an island).
Delayed onset of organized convection, which can lead to a confluence or collision of sea breezes, suggests a high sensitivity to island size. Other factors being equal, one might expect a smaller island to routinely initiate convection through the collision of sea breezes and, conversely, for this to be a rare event on larger islands. However, the sensitivity to island size is unclear because a smaller island will limit the residence time of surface layer air for a given flow speed. Except for the limiting case of stagnation, a smaller island will yield reduced heating, thereby leading to a slower breeze front propagation rate. In the limiting case of stagnation the IBL will be uniformly developed to its full potential, in which case the duration of heating and the latent/sensible partitioning become the limiting factors. Detailed sensitivity tests with an appropriate PBL model, including rigorous treatments of surface physics are needed to address this question quantitatively. The prima facia evidence may suggest that the Tiwi Islands are optimally sized for convection to develop, primarily by a proliferation of cold pools and secondarily via the collision of zonal breezes.
In the particular case of MCTEX, optimal conditions are often satisfied by easterly shear over a zonal island heat source, and the regular occurrence of surface stagnation or slow westerly flow. On many days, the surface flow direction in MCTEX does not satisfy (or clearly satisfy) the theoretical optimal condition, which includes an IBL critical level. In those instances, strong MCSs of long duration can occur over the islands if the surface flow is very slow and the negative buoyancy production capacity of the troposphere is well matched to the lower-tropospheric shear. For these reasons and others, the MCTEX dataset exhibits somewhat less sensitivity to the theoretical optimum than one might anticipate. Nevertheless, very strong Hectors are produced on days that closely fit the idealized concept of optimal conditions (Keenan et al. 2000; Table 1 herein).
Sea breezes and convection were observed over the Tiwi Islands, north of the Australian continent, as part of the Maritime Continent Thunderstorm Experiment. There are two distinct forcing regimes leading to Hector:type A—direct, resulting from a confluence of zonally oriented sea breeze fronts and type B—indirect, resulting from a multistage, chaotic interaction among sea breezes and convectively generated cold pools, usually near a leeward coast. In the MCTEX dataset, type B forcing was operative on at least 11 of 14 days for which supporting data are available. Type A forcing appears to have been in effect on the remaining three days. There is additional evidence of type A evolution in the literature (Keenan et al. 1990; Keenan and Carbone 1992; Simpson et al. 1993; SKHP; Crook 1997). Type A forcing may be viewed as nature’s backup mechanism when type B forcing fails. A longer time series from the region may alter these statistics significantly, but it is our expectation that a decided majority of days will prove to be of type B during the unstable transition and break seasons. For type B forcing, there are at least four distinct stages of precipitating convection: 1) showers that are forced and maintained at the sea-breeze front, 2) traveling showers that move to the interior once they produce air colder than the MBL, 3) westward-propagating squalls resulting from the maturation of cold pools in easterly shear, and 4) a GF–ZBF interaction that forces the most intense convection at the frontal intersection. An optional fifth stage is defined by a collision of GFs from coexistent MCSs, which tend to yield the most intense MCSs.
Immature peninsula scale (5–15 km) breezes are observed to initiate the first precipitating convection, principally through convergence that results from the confluence or collision of opposing-coast breeze fronts. Such breezes develop in the morning of each day and have a virtual temperature discontinuity of approximately 1°C. Convection initiated by peninsula breezes usually fails to organize beyond a small cluster of cells and dissipates as a local event.
On days when deep convection is delayed or suppressed, so-called type A evolution prevails. One or both sea-breeze fronts may penetrate more than 20 km inland from the north and south coasts, often forcing convection well inland. Owing to an increased depth of moisture from these solenoidal circulations and heat fluxes from the island’s surface, conditions are created that can overcome thermodynamic inhibition and serve to increase the amplitude of convection (over that which a morning sounding might suggest is possible). However, in our limited dataset, late-day initiation of convection over the interior seems less likely to produce large convective systems that are sometimes associated with type B forcing.
Owing to air parcel residence time over the land, the island boundary layer is deepest and warmest near leeward coasts where a mature sea breeze can maintain a 2°–3°C discontinuity. Island-scale sea breezes from opposing coasts (separated by 50 km or more) typically do not collide because these interact with convectively produced cold pools 1–3 h before the inland progression of breeze fronts would permit an island-scale collision. So-called type B complexity, associated with multiple breeze–cold pool interactions, is at the heart of convective evolution in the Tiwi Islands and is the daily norm.
A principal finding of this study is that mesoscale convective systems (Hectors) can be traced backward in time to the convection that is initiated by sea breezes, usually along leeward coasts (Fig. 20). When evaporatively produced cold pools become cooler than the nearby sea breeze, convection frees itself from the breeze maintenance mechanism, allowing it to travel and feed on the heated island boundary layer. Subsequent convective evolution is characteristic of traveling free convection elsewhere in that it organizes mainly according to cold pool, shear balance, and mean flow factors (Moncrieff and Liu 1999; Rotunno et al. 1988).
The westward-propagating squalls subsequently force very strong convection along a dominant sea-breeze front that is oriented zonally near the north or south coast of the island complex. The ZBF is preconditioned to transition into strong convection, since the front is heavily populated with cumulus congestus clouds prior to arrival of the squall GF. The dominant ZBF is determined by the meridional component of ambient flow in the vicinity of the islands. Under conditions of near stagnation, convection is widespread throughout the islands owing to an exceptionally long residence time of air over land and the resulting well-developed IBL at practically all locations. Climatologically, westward-propagating MCSs are favored given the ubiquitous easterly shear and easterly winds in the low to midtroposphere (Table 1).
Given that an easterly shear condition prevails in the Tiwi Islands, an optimal condition for strong convective development is often achieved when surface layer winds are westerly with easterlies aloft (Moncrieff and Liu 1999), conditional instability is substantial (500–2500 J kg−1), and convective inhibition is weak. Under these circumstances, precipitating convection is initiated at the ECBF and organizes into a small meridional squall once an adequate cold pool develops. Easterly winds favor subsequent westward propagation and, in the words of Simpson et al. (1993), “mine” the island boundary layer across the full zonal extent of the islands.
In general, we found the effects of continental Australia to be subtle, except for the apparent advection of dry continental air under southerly flow conditions. Convection over the mainland at Cape Don (Fig. 1) may have exerted an external influence in some cases, since this is a favored area for convective development and it is close to Melville Island. For example, the ECBF convection on 28 November may have been influenced by convection that traveled westward from Cape Don.
At least two major questions could not be resolved with the MCTEX dataset. Which environmental factors most often control type A versus type B forcing? Improved understanding of the environmental controls is essential to the short-term prediction of storms as well as to quantitative precipitation forecasts. Our future work will report on factors that affect daily predictions of island thunderstorm activity as well as nowcasts of specific storms on a 0–2-h timescale.
Does the conditionally energetic air within the MBL routinely undergo convective overturning of the diurnal type as observed in the Tiwi Islands? Unlike the westerly monsoon regime (Keenan and Carbone 1992), we are unable to confirm an instance of sustained deep convection that is fed by unmodified MBL air in the easterly trade wind regime. If MBL air is routinely trapped, this may explain why daytime convection occurs almost exclusively over land, except when large-scale disturbances enter the Maritime Continent region. This may also help to explain the vigor of tropical island convection, owing to regional mass-balance considerations, if convective overturning is mainly focused over islands in suitable oceanic regions.
MCTEX was conducted with the permission of the Tiwi Land Council. The authors are indebted to the efforts of K. Glasson, N. Roediger, and P. May for the acquisition of radar, Cessna, and profiler datasets, respectively. Postprocessing of these data would not have been possible without the dedicated assistance and skills of P. Powers, B. Sanjeev, and R. Schaffer at the BMRC. Additional assistance with sounding data and figure preparation was provided by E. Miller, D. Megenhart, and M. Crown of NCAR. The lead author is deeply appreciative for the discussions and insights provided by Joanne Simpson. Thanks are also due to N. A. Crook, T. Weckwerth, and M. Moncrieff for their insightful reviews of an earlier manuscript, to the anonymous reviewers of this manuscript, and to C. Bousquet for final preparation of materials.
* The National Center for Atmospheric Research is sponsored by the National Science Foundation.
Corresponding author address: Dr. Richard E. Carbone, Mesoscale and Microscale Meteorology Division, National Center for Atmospheric Research, P.O. Box 3000, Boulder, CO 80307.