Abstract

Observations and cloud-resolving simulations of elongated cloud plumes (or “cloud trails”) past the Lesser Antilles islands in the Caribbean Sea are presented. Analysis of one year of visible satellite images reveals that each island forms cloud trails on 30%–40% of days, typically in the afternoon in response to diurnal island heating. On around 10% of days the cloud bands are very well organized, with lengths of and durations of min. Radiosonde analysis suggests that the well-organized events are favored by moderate-to-strong trade winds (6–10 m s−1) and stronger trade inversions. The simulated cloud trails, which are consistent with observations in their morphology and diurnal cycle, are organized by quasi-linear bands of thermally forced convergence within the heated island wake. They are sensitive to overland surface fluxes, inversion strength and height, terrain height, and trade-wind speed. While surface fluxes control the strength of the wake thermal circulations, the inversion controls precipitation and the disruption of cloud trails by subcloud cold pools. The impacts of terrain height and wind speed are multifaceted, including control over (i) the mechanical flow regime, (ii) the intensity of wake turbulence, (iii) the cloud-trail length, (iv) the wake thermal anomaly, and (v) elevated-heating effects (which strengthen the thermal convergence). Dimensional analysis is used to develop empirical scalings for the wake thermal circulation, which describe the suite of numerical sensitivity tests reasonably well.

1. Introduction

Tropical islands are hotspots for cumulus convection due to three main processes: (i) mechanical lifting of impinging airflow by island orography and/or frictional convergence, (ii) thermal circulations driven by differential surface heating between land and water, and (iii) local static destabilization due to daytime surface heating. Such convection influences the local climate by altering the radiation balance and producing reliable precipitation over favored locations. It also impacts the regional climate by growing upscale into organized convective systems and by transforming the impinging air mass through water removal and vertical mixing (e.g., Smith et al. 1997; Kirshbaum and Smith 2009). Because of its local and regional impacts, as well as its value as a prototype for clouds in other environments, island-forced convection merits a sound conceptual understanding and accurate prediction.

Mechanically forced island convection forms as potentially unstable airflow is forced to ascend or divert around the island terrain. The processes underlying such convection are similar to those over continental mountains except for modulations associated with the land–sea contrast (see below) and coastal convergence. When the flow contains sufficient horizontal momentum to directly ascend the orography, cumuli tend to form over the windward slopes where upward displacement is maximized (e.g., Kirshbaum and Smith 2009; Minder et al. 2013). By contrast, when the flow lacks the momentum to ascend the terrain it tends to detour around it, causing clouds to form in convergence zones upwind and downwind (e.g., Smolarkiewicz et al. 1988; Yang and Chen 2008).

Thermally forced island convection stems from the low heat capacity of land (relative to water), which leads to an amplified diurnal temperature cycle that contrasts sharply with the surrounding ocean. The associated buoyancy gradients drive thermally direct circulations that, under suitable environmental conditions, may initiate moist convection. The most common thermally forced island circulations are land–sea breezes, which produce strong ascent along and ahead of their frontal surfaces (e.g., Carbone et al. 2000; Fovell 2005). Whereas land breezes propagate offshore at night, sea breezes propagate onshore during the day. Generally, sea-breeze-forced convection is more intense because the convecting air has been destabilized by surface heating over land. Island thermal circulations may be enhanced by the presence of orography, which acts as an elevated heat source that strengthens horizontal buoyancy gradients (e.g., Crook and Tucker 2005; Kirshbaum and Wang 2014). However, orography may also disrupt island convection by retarding the inland propagation of sea breezes (Barthlott and Kirshbaum 2013).

In contrast to sea-breeze convection directly over islands, we focus on the lesser studied problem of thermally forced convection downwind of islands. When the ambient wind is sufficiently strong (or the island sufficiently narrow) to inhibit sea breezes, elongated cloud plumes (or “cloud trails”) may form in the lee. These features have been observed over Anegada, British Virgin Islands (Malkus 1963); Nantucket, Massachusetts (Malkus and Bunker 1952); Barbados (Souza 1972); the Pacific island of Guadalupe (Dorman 1994); Nauru (Nordeen et al. 2001; McFarlane et al. 2005; Matthews et al. 2007); and Hawaii (Yang et al. 2008a,b). Altogether, these studies suggest that cloud trails develop in a broad range of climates. Because cloud trails tend to form over both flat and mountainous islands, and because they tend to prevail in the afternoon, their underlying mechanisms are likely predominantly thermal (rather than mechanical) in nature. However, mechanical forcing may still influence their location, intensity, and organization.

The most detailed cloud-trail studies to date were conducted by Yang et al. (2008a, hereafter Y08a) and Yang et al. (2008b, hereafter Y08b), who examined cloud bands past the islands of Hawaii and Kauai/Niihau using satellite data and numerical simulations, respectively. The afternoon cloud frequency in August was enhanced past each of these islands, but for different reasons. For Kauai and Niihau, which lie below the trade-wind inversion, the clouds were attributed to a combination of mechanical leeside convergence of island-deflected airstreams and thermally driven convergence within the heated island wake (Y08b). For the island of Hawaii, which easily pierces the trade-wind inversion at a height of 4 km, the trade winds tend to divert around it, preventing the island-heated air from traveling downwind. However, afternoon cloud frequency was still enhanced in the lee, which Y08a attributed to convergence between decelerated wake flow and accelerated marine flow rounding the island edges, along with coincident collisions of multiple sea breezes.

While the above studies offer useful insights on cloud trails, a thorough understanding is still lacking. Currently, the bulk of cloud-trail research has focused on the Hawaiian islands, so the generality of the mechanisms of Y08a and Y08b is unclear. There has also been limited study of the sensitivities of cloud trails to topographic forcing and environmental conditions. Because Y08a and Y08b used full-physics real-case simulations, they could not systematically examine the impacts of such parameters. Moreover, although the horizontal grid spacing () used in their simulations was sufficient to represent cloud trails without a cumulus parameterization scheme, it was insufficient to resolve in-cloud turbulence or the narrow boundary layer convergence lines (e.g., Bryan et al. 2003; Warren et al. 2014), both of which are fundamental to cloud trails. Finally, while the qualitative mechanisms proposed by Y08a and Y08b are physically plausible, a more quantitative understanding is required to improve their prediction.

To advance the understanding of island cloud trails, we investigate their occurrence, dynamics, and sensitivities over the Lesser Antilles islands in the eastern Caribbean Sea. With heights of 1.5 km or less, these islands typically do not pierce the trade-wind inversion, rendering them more similar to Kauai than to Hawaii. We focus on Dominica (Fig. 1), a mountainous island that has received intensive recent study because of its persistent convective precipitation and pronounced dynamical wakes (Smith et al. 2009a,b, 2012). Section 2 analyzes one year of visible satellite images that were archived for the Dominica Experiment (DOMEX) field campaign in 2011 (Smith et al. 2012). Section 3 presents cloud-resolving numerical simulations to study cloud-trail dynamics and sensitivities, and section 4 uses dimensional analysis to scale the thermal-circulation strength. Section 5 provides further discussion, and section 6 provides the conclusions.

Fig. 1.

Maps of the (left) western Atlantic Ocean and Caribbean Sea, (middle) the Lesser Antilles region under consideration, and (right) Dominica. The filled contours of the Dominica map represent terrain heights in km.

Fig. 1.

Maps of the (left) western Atlantic Ocean and Caribbean Sea, (middle) the Lesser Antilles region under consideration, and (right) Dominica. The filled contours of the Dominica map represent terrain heights in km.

2. Observations

To establish cloud-trail frequency, variability, and environmental conditions over the Lesser Antilles, we analyzed visible satellite data from the National Oceanic and Atmospheric Administration (NOAA) Geostationary Operational Environmental Satellite 13 (GOES-13) with a nominal space–time resolution of 1 km/30 min. We considered one year of data, from 9 June 2010 to 8 June 2011 (this period was chosen arbitrarily). Data were available in the form of image files archived for DOMEX as well as raw satellite data downloaded from NOAA’s Comprehensive Large Array-Data Stewardship System (CLASS). Because of missing data, the satellite images were only available on 352 of the 365 days.

We focused on the 13.5°–17.5°N latitude band centered on Dominica, including (from south to north) St. Lucia, Martinique, Dominica, Guadeloupe, Montserrat, Antigua, Nevis, and St. Kitts (Fig. 1). Environmental conditions were represented by 1200 UTC atmospheric soundings from Guadeloupe (TFFR), which were available on all but 14 of the 352 satellite days. Because of their suboptimal timing (6–8 h prior to peak cloud-trail occurrence) and their modification over land, these soundings do not perfectly represent the upstream conditions during cloud-trail events. Nonetheless, they are still useful for broadly contrasting a wide variety of cases.

We first manually classified the occurrence and organization of cloud trails past each of the seven islands using image animations for each of the 352 satellite days (because of their close proximity, St. Kitts and Nevis were considered to be a single island). Although an automated detection algorithm would have facilitated a more objective and longer climatology, such an algorithm was considered unnecessary for this preliminary analysis. The wake clouds were classified into one of three categories: (i) no organized convection (null), (ii) weakly organized and/or short-lived cloud trails (type I), or (iii) well-organized and quasi-stationary cloud trails (type II). For a type-II classification, a cloud trail needed to exhibit a quasi-linear structure with a length of 100 km, a width of 10 km, and a duration of 90 min. To obtain a single regional classification for each event, we assigned each island an integer corresponding to its individual classification (0: null, 1: type I, and 2: type II) and averaged this integer over all islands (λ). The event was classified as type II if , type I if , and null if .

To demonstrate the classification scheme, Fig. 2 shows satellite imagery from three representative events at around 1900 UTC [1500 local solar time (LST)]. In the null event (18 April 2011; Fig. 2a), clouds developed directly over the islands rather than downwind. Cloud trails were evident in the type-I event (29 January 2011; Fig. 2b), but were not elongated or persistent enough to satisfy the type-II criteria. In contrast, the cloud trails in the type-II event (29 November 2010; Fig. 2c) did satisfy these criteria. The diurnal cycle of cloud trails is illustrated by the time evolution of the best-organized type-II event in the database (26 February 2011) (Fig. 3), where elongated cloud trails persisted throughout the afternoon. Although the longest bands formed past Montserrat (~220 km) and Guadeloupe (~160 km), most other islands formed type-II cloud trails as well.

Fig. 2.

Examples of visible satellite images for each event type. (a) Null event at 1539 LST 18 Apr 2011, (b) type-I event at 1539 LST 29 Jan 2011, and (c) type-II event at 1509 LST 29 Nov 2010. The arrow on each panel indicates the mass-weighted boundary layer wind direction over 0–1500 m from the corresponding TFFR sounding.

Fig. 2.

Examples of visible satellite images for each event type. (a) Null event at 1539 LST 18 Apr 2011, (b) type-I event at 1539 LST 29 Jan 2011, and (c) type-II event at 1509 LST 29 Nov 2010. The arrow on each panel indicates the mass-weighted boundary layer wind direction over 0–1500 m from the corresponding TFFR sounding.

Fig. 3.

Visible satellite images of the Lesser Antilles (at 1-km resolution) from 26 Feb 2011 at (a) 1009, (b) 1509, (c) 1609, and (d) 1709 LST. The arrow in (a) indicates the mass-weighted boundary layer wind direction over 0–1500 m from the TFFR sounding.

Fig. 3.

Visible satellite images of the Lesser Antilles (at 1-km resolution) from 26 Feb 2011 at (a) 1009, (b) 1509, (c) 1609, and (d) 1709 LST. The arrow in (a) indicates the mass-weighted boundary layer wind direction over 0–1500 m from the TFFR sounding.

Because cloud trails are promoted by island heating (Matthews et al. 2007, Y08b), they tend to be suppressed by island clouds. Such clouds may develop because of (i) local mesoscale forcing (see section 1), (ii) landfalling trade-wind cumuli, and/or (iii) larger-scale disturbances that overspread the region (e.g., easterly waves and tropical cyclones). While the clouds in categories (i) and (ii) are cumuliform with gaps that permit full insolation, those in category (iii) are largely stratiform and block insolation more broadly (as well as masking the cloud trails). Hence, we exclude “disturbed” cases where an island was obscured by larger-scale clouds for at least 3 h over 0600–1800 LST. All seven islands were disturbed on 102 days, and some (but not all) were disturbed on another 135 days. Table 1 provides cloud-trail statistics for the seven islands. The frequency of disturbed days is similar for all islands (133–139 days). Each island developed noticeable cloud trails on 33%–40% of all days and 56%–67% of undisturbed days. However, type-II cloud trails developed relatively rarely (7%–21% overall or 11%–31% of undisturbed days).

Table 1.

Statistics from the observational analysis. Description of event types in the top column are provided in the text. All entries are percentages.

Statistics from the observational analysis. Description of event types in the top column are provided in the text. All entries are percentages.
Statistics from the observational analysis. Description of event types in the top column are provided in the text. All entries are percentages.

The cloud diurnal cycle is illustrated by cloud-frequency statistics, which are found by converting raw satellite counts to effective albedo, then applying an albedo threshold of 0.15 to identify clouds (following Y08b). Although such cloud masks can be uncertain near coastlines, our interest in oceanic clouds renders this issue irrelevant. The cloud frequency for the full year (excluding the 102 disturbed days) is compared for the morning (0800–1100 LST) and afternoon (1400–1700 LST) hours in Figs. 4a and 4b. This frequency is generally low over the ocean (0%–10%) and higher over the islands (%), with a suppression downwind (west) of the larger and taller islands of Guadeloupe, Martinique, and Dominica. This suppression arises mainly from island gravity wave breaking, which dries the wake boundary layer (e.g., Kirshbaum and Smith 2009). The cloud frequency is broadly enhanced in the afternoon, possibly due to cloud-albedo sensitivity to solar zenith angle and/or deep convection over surrounding landmasses (see Fig. 1). Despite the high frequency of cloud-trail events in Table 1, the afternoon cloud-frequency enhancement only extends about ~50 km to the west of most islands (Fig. 4b), consistent with Fig. 3b of Y08b. These enhancements diminish downwind due to averaging over many nonoverlapping and short-lived events.

Fig. 4.

Cloud-mask analysis over the Lesser Antilles, excluding disturbed days with extensive large-scale cloudiness. (a) Year-long morning (0800–1100 LST) and (b) afternoon (1400–1700 LST) cloud frequency. (c) Morning and (d) afternoon cloud frequency for type-II events with easterly winds (β = 75°–105°). (e) Morning and (f) afternoon cloud frequency for type-II events with east-northeasterly winds (β = 45°–75°).

Fig. 4.

Cloud-mask analysis over the Lesser Antilles, excluding disturbed days with extensive large-scale cloudiness. (a) Year-long morning (0800–1100 LST) and (b) afternoon (1400–1700 LST) cloud frequency. (c) Morning and (d) afternoon cloud frequency for type-II events with easterly winds (β = 75°–105°). (e) Morning and (f) afternoon cloud frequency for type-II events with east-northeasterly winds (β = 45°–75°).

A more prominent cloud-trail signal emerges by restricting consideration to type-II events and grouping them according to upstream wind direction. Of the 73 type-II events, 90% exhibited easterly (β = 75°–105°) or east-northeasterly (β = 45°–75°) winds, consistent with the Dominica climatology of Smith et al. (2009b). Figures 4c,d and 4e,f show the diurnal evolution of the easterly (E) and east-northeasterly (ENE) events, respectively, illustrating that the bands tend to develop in the afternoon and align with the background wind. Because of the narrowness of the cloud trails and their varying alignments within each wind-direction bin, the time-averaging tends to produce spokes of enhancement radiating away from each island. Weaker bands in between some islands are also apparent in the mornings, suggesting a reversal of the wake circulations at night.

To study the environmental conditions of cloud trails, we compare composite TFFR 1200 UTC soundings of null and type-II events in Fig. 5. Despite obvious similarities (humid, conditionally unstable boundary layers, dry free tropospheres, and easterly winds up to 500 hPa), the type-II profiles differ in several key ways. First, consistent with Y08b, their winds are stronger, with mass-weighted 0–1500-m speeds of 7.8 m s−1 (type-II E), 6.3 m s−1 (type-II ENE), and 4.5 m s−1 (null). The differences between null and type-II events are significant at the 95% confidence level. Second, the trade-wind inversion is stronger, with a ~0.002 K hPa−1 decrease in the temperature lapse rate over 800–700 hPa. Third, the free troposphere is markedly drier, with dewpoint depressions 10°–20°C larger than in the null events. The latter two points suggest stronger large-scale subsidence, which causes adiabatic warming that dries the free troposphere and strengthens the trade-wind inversion. Finally, the surface pressure is about 2 hPa higher in type-II events (~1014 vs ~1012 hPa).

Fig. 5.

Skew T–logp plot of composite radiosondes for null events (black), type-II events with easterly winds (blue), and type-II events with east-northeasterly winds (green). Temperature profiles are in solid lines and dewpoints are dashed. For the winds, half and full barbs correspond to increments of 2.5 and 5 m s−1, respectively.

Fig. 5.

Skew T–logp plot of composite radiosondes for null events (black), type-II events with easterly winds (blue), and type-II events with east-northeasterly winds (green). Temperature profiles are in solid lines and dewpoints are dashed. For the winds, half and full barbs correspond to increments of 2.5 and 5 m s−1, respectively.

Because of surface friction over land, the 0–1500-m TFFR wind speeds quoted above are likely weaker than the (unobserved) winds in the impinging marine flow. Taking averages over 600–1500 m, which omits much of the decelerated subcloud layer, gives mean wind speeds of 9.6 (easterly), 8.0 (east-northeasterly), and 5.7 m s−1 (null), with the differences between type-II and null events again significant at the 95% confidence level. Note also that the trade-wind inversions in Fig. 5 are poorly defined due to the vertical averaging over many cases with different inversion strengths and heights.

3. Numerical simulations

To interpret the observations and improve the understanding of cloud trails, we conduct cloud-resolving simulations with the Advanced Research Weather Research and Forecasting (WRF-ARW, hereafter WRF) Model version 3.5 (Skamarock et al. 2008). The objective of these experiments is not to reproduce a specific cloud-trail event, but rather to represent cloud trails with an acceptable level of realism to facilitate examination of their dynamics and sensitivities. Thus, we avoid the complexity of horizontally inhomogeneous, time-varying background states or long island chains. Instead, we focus on adequately representing cloud trails past a single island and conducting additional experiments to isolate their sensitivities.

a. Setup

We use a quasi-idealized configuration wherein the topographic forcing and background flow are both realistic but highly simplified. The domain size is by by in the x, y, and z directions. The terrain-following vertical grid uses 71 levels with spacings that increase from at the surface to 240 m at . We use three different horizontal grid spacings (): 1000, 500, and 250 m. The lateral boundaries are open (radiative) and the upper boundary is rigid with a 5-km Rayleigh-damping layer below it to absorb gravity wave energy. While the flow dynamics are thus distorted above 5 km, the simulated cumuli and gravity waves rarely extend that high.

For the island topography we use Dominica, which contains two volcanic peaks nearly 1.5-km high with a saddle in between (Fig. 1). We place the island in the upper-right part of the domain (, ) to minimize lateral-boundary interference downwind. To constrain the overland heat fluxes, we performed a separate full-physics WRF simulation of the 26 February 2011 event (not shown). The maximum sensible (H) and latent heat () fluxes in that simulation both exceeded 250 W m−2 locally, with cloud shading over the windward slopes limiting the island-averaged H and to 140–150 W m−2. To represent this heating simply, we prescribe sinusoidal surface-flux functions with amplitudes of uniformly over the island surface. The fluxes are initially zero (at 0600 LST), reach their maximum at 1200 LST, and return to zero by 1800 LST (the end of the simulation). The implications of using prescribed (and not interactive) fluxes are discussed in section 6. The island surface is no slip with a bulk aerodynamic drag coefficient of .

The initial flow is horizontally homogeneous and based on a single sounding profile. Neither TFFR soundings (due to flow modifications over land) nor special DOMEX soundings (which did not sample a type-II event) were used for this profile. Rather, we use a sounding from the large-eddy simulations of Kirshbaum and Grant (2012), which was based on data from the Rain in Cumulus over the Ocean (RICO) field campaign over the western Atlantic in 2004–05 (Stevens and Seifert 2008). This profile (RICO-LES) contains a subcloud layer from the surface up to the cloud base () with mean winds of 8 m s−1, a cloud layer up to , and an inversion up to (Fig. 6). Because this sounding was only defined up to 3.5 km, it is extended to using a pseudoadiabatic lapse rate, constant relative humidity, and vertically extrapolated winds (based on velocity gradients over the upper 500 m).

Fig. 6.

Skew T–logp plot of the RICO sounding. Temperature is in black and dewpoint is in gray. For the winds, half and full barbs correspond to increments of 2.5 and 5 m s−1, respectively.

Fig. 6.

Skew T–logp plot of the RICO sounding. Temperature is in black and dewpoint is in gray. For the winds, half and full barbs correspond to increments of 2.5 and 5 m s−1, respectively.

To ease the interpretation of the island flow dynamics, we enforce that the background oceanic flow in Fig. 6 is maintained throughout the simulation. Thus, we use a free-slip, unheated ocean surface and assume the initial state to be in geostrophic balance. The Coriolis force is applied only to flow perturbations using an f-plane approximation with a latitude of . While this setup is highly simplified, it is preferable to a more complicated one with full model physics, which requires compensating (and poorly constrained) large-scale forcings to maintain a steady state. Two potential disadvantages of this setup are its inability to capture the “seeding” of island convection by impinging flow inhomogeneities (Kirshbaum and Smith 2009; Smith et al. 2012) and a failure to account for wake boundary layer moisture recovery after island convection and wave breaking depletes it. However, the simulated wake moisture depletion is only ~1 g kg−1 and the wake flow is quickly replaced by converging marine air (as will be seen). Thus, wake moisture depletion likely has a minimal impact on the simulated cloud-trail dynamics.

The only subgrid parameterizations in the simulations are the Morrison two-moment cloud microphysics and a 1.5-order turbulent kinetic energy (TKE) turbulence closure (with TKE = 0 initially). All simulations use the Morrison scheme’s default cloud-droplet concentration of . Because this value is more representative of continental than marine air, additional simulations with (representing marine air) were performed. Because of increased droplet radius and precipitation efficiency, the domain-wide rainfall increased by ~25% (not shown). However, the cloud-trail morphology was largely unaffected, which suggested only weak sensitivity to . To represent turbulent mixing, the simulations rely on the combination of resolved dynamics, numerical diffusion, and the subgrid turbulence closure rather than a boundary layer scheme. Again, this approach is followed to preserve the prescribed upstream flow throughout the simulation. It is better justified at smaller grid spacings where explicit turbulence is able to develop.

b. Control simulations

We begin with the RICO simulation, which uses the above configuration with . This grid spacing balances the needs of reasonably resolving the cloud trails while limiting computational expense. The liquid water path (LWP) from this simulation at 1500 and 1700 LST (Figs. 7a,d) reveals a cloud cluster that forms past Dominica and propagates downwind during the afternoon. The convection is noticeably less organized than the satellite-observed cloud trails in Figs. 2 and 3, which stems from the relatively deep cloud layer in Fig. 6 and heavy rainfall. The maximum wake rainfall accumulation is over the 12-h simulation. This rainfall creates strong evaporative cold pools that disrupt the linear cloud organization in the wake. Note that in the satellite climatology, such downwind cloud clusters were typically classified as type-I cloud trails and constituted about 30% of such events.

Fig. 7.

Liquid water path (LWP) and terrain height (contoured every 200 m) for (a),(d) RICO; (b),(e) NORN; and (c),(f) REF simulations: (a)–(c) 1500 and (d)–(f) 1700 LST.

Fig. 7.

Liquid water path (LWP) and terrain height (contoured every 200 m) for (a),(d) RICO; (b),(e) NORN; and (c),(f) REF simulations: (a)–(c) 1500 and (d)–(f) 1700 LST.

Two additional simulations illustrate the pronounced impacts of rainfall on the cloud trails. In the first (NORN), microphysical autoconversion is switched off, eliminating rain entirely. In the second (REF), the inversion (and, hence, and ) is lowered by 800 m, which halves the depth of the cloud-bearing layer and reduces by 79%. Figure 8 compares surface winds, perturbation θ(), and vertical velocity (w) midway through the subcloud layer (), for the RICO, NORN, and REF cases at 1500 LST. All three cases develop a thermal wake with extending up to 50 km downwind (Fig. 8). Convergent marine flow driven by the wake buoyancy gradient progressively narrows the thermal wake, forming an elongated band of low-level convergence and ascent along the wake centerline. In the RICO case, the convergence shifts to the edges of the cold pool, causing the cloud trail to broaden downwind (Figs. 7a,d). By contrast, the central convergence band extends much farther downstream in the NORN case, creating a longer and narrower cloud trail (Figs. 7b,e). The REF case lies between these two extremes, with a much weaker cold pool that broadens the cloud trail slightly (Figs. 7c,f).

Fig. 8.

Surface potential temperature perturbations, wind vectors, subcloud vertical velocity at , and terrain height at 1500 LST for (a) RICO, (b) NORN, and (c) REF simulations. The contour of vertical velocity is 0.5 m s−1 and terrain height is contoured every 200 m.

Fig. 8.

Surface potential temperature perturbations, wind vectors, subcloud vertical velocity at , and terrain height at 1500 LST for (a) RICO, (b) NORN, and (c) REF simulations. The contour of vertical velocity is 0.5 m s−1 and terrain height is contoured every 200 m.

Further insight into the REF flow dynamics is provided by surface winds and w at at 1500 LST (Fig. 9a). The impinging flow ascends the island terrain and forms a weakly turbulent wake with three updraft bands streaming downwind. Two bands are anchored to the island’s two main peaks, suggesting that they are owing to elevated heating effects (e.g., Kirshbaum and Wang 2014) and/or leeside convergence (e.g., Banta 1990) (these two effects are difficult to disentangle). The updraft bands merge about 50 km downwind of the coastline to form a dominant band that organizes the cloud trail. To investigate the role of Dominica’s dual-peak terrain on the cloud trail, we performed an additional simulation where the terrain saddle was filled in with a Gaussian ridge (not shown). In this case, a single dominant updraft band formed in the immediate wake, causing the cloud trail to shift upwind by ~30 km. Thus, the complexity of Dominica’s terrain acts to suppress the cloud trail in the immediate wake.

Fig. 9.

(a) Vertical velocity w at , liquid water path (filled in gray shade), surface wind vectors, and terrain height (contoured every 200 m) from the REF simulation at 1500 LST. (b) Vertical velocity (filled in red/blue shades and contoured in black), cloud liquid water (filled in gray shade), plane-parallel wind velocities (arrows), and potential temperature θ (lines), averaged in space over the length of the rectangle from (a) and in time over 1500–1700 LST. No negative w contours appear because the subsidence is too weak.

Fig. 9.

(a) Vertical velocity w at , liquid water path (filled in gray shade), surface wind vectors, and terrain height (contoured every 200 m) from the REF simulation at 1500 LST. (b) Vertical velocity (filled in red/blue shades and contoured in black), cloud liquid water (filled in gray shade), plane-parallel wind velocities (arrows), and potential temperature θ (lines), averaged in space over the length of the rectangle from (a) and in time over 1500–1700 LST. No negative w contours appear because the subsidence is too weak.

Figure 9b shows a vertical cross section of w, cross-band wind component (), , and θ across the REF cloud trail, which is created by defining a rotated rectangle aligned with the subcloud updraft band at an angle from the horizontal (Fig. 9a). The long () and short () axes have lengths of and , which encompasses the nonprecipitating section of the band (see Fig. 9a). We interpolate the model fields onto the rotated coordinate system and average them in (over ) and in time (over 1500–1700 LST) to yield a smoothed cross section in and z. This cross section indicates strong convergence at the surface and ascent through the subcloud layer. The updraft splits into two outflow branches at and an updraft that continues through the cloud layer. The mean cloud-layer ascent is weaker than the subcloud layer ascent despite stronger instantaneous updrafts (up to 5 vs ~1 m s−1; not shown). The former is weaker because it is averaged over both updrafts and downdrafts while the latter is averaged over a continuous updraft. A second outflow layer is found just below cloud top (1600–1800 m), above which isentropes bulge upward into the inversion.

c. Sensitivity tests

Of the three simulations above, the one that exhibits the most similarity with type-II observations (e.g., Fig. 3) is REF. Whereas intense rainfall in RICO severely degrades the cloud-trail organization, NORN excludes precipitation and is thus fundamentally unrealistic. Thus, REF is chosen as the reference for a series of sensitivity experiments. Although not comprehensive, these tests isolate several key processes controlling cloud-trail dynamics (for reference, settings for the various sensitivity tests are summarized in Table 2). To compare the simulations, Table 3 presents various output metrics, which are calculated over the portion of the domain lying downwind of the island ( and ):

  1. Updraft mass fluxes near cloud base (MFcb, ) and just below the inversion base (MFinv, ), averaged over 1200–1800 LST. While MFcb includes both cloudy and clear air, MFinv includes only cloudy air. A threshold of is imposed to exclude weaker updrafts.

  2. Conditionally averaged cloud-core properties over 1200–1800 LST, at : area fraction , buoyancy , vertical velocity , and liquid water mixing ratio . “Core” grid points satisfy , , and , where b is defined as 
    formula
    g is the acceleration of gravity; 
    formula
    is density potential temperature; , and are gas constants for dry air and water vapor; is total water mixing ratio; and is the initial profile.
  3. Maximum cloud-top height () over 1200–1800 LST.

  4. Maximum surface precipitation () and total precipitation mass at 1800 LST.

Table 2.

Parameter choices for the sensitivity simulations. The italics for the ZCB0.5 and ZCB2.0 simulations denote that only dry versions of these simulations are performed.

Parameter choices for the sensitivity simulations. The italics for the ZCB0.5 and ZCB2.0 simulations denote that only dry versions of these simulations are performed.
Parameter choices for the sensitivity simulations. The italics for the ZCB0.5 and ZCB2.0 simulations denote that only dry versions of these simulations are performed.
Table 3.

Diagnostics from the numerical sensitivity tests, including mass fluxes at cloud base MFcb and inversion base MFinv, average core fraction , buoyancy , vertical velocity , and liquid water content at , cloud-top height , maximum precipitation accumulation , and precipitation mass . All quantities are computed over the “downwind” region (, ) over the 1200–1800 LST period.

Diagnostics from the numerical sensitivity tests, including mass fluxes at cloud base MFcb and inversion base MFinv, average core fraction , buoyancy , vertical velocity , and liquid water content  at , cloud-top height , maximum precipitation accumulation , and precipitation mass . All quantities are computed over the “downwind” region (, ) over the 1200–1800 LST period.
Diagnostics from the numerical sensitivity tests, including mass fluxes at cloud base MFcb and inversion base MFinv, average core fraction , buoyancy , vertical velocity , and liquid water content  at , cloud-top height , maximum precipitation accumulation , and precipitation mass . All quantities are computed over the “downwind” region (, ) over the 1200–1800 LST period.

1) Horizontal resolution

The choice of for the REF case, which is based on practical considerations, is insufficient to properly resolve deep convection (e.g., Bryan et al. 2003), let alone the shallow convection within cloud trails. As a rule of thumb, adequate numerical resolution of a physical process requires to be about one-tenth its characteristic scale. Satellite analysis suggests that cloud trails exhibit cross-stream horizontal scales of 1–2 km before broadening because of cold-pool dynamics or cloud outflow. Thus, may be needed, which is impractical for the numerous large-domain experiments performed herein. Nevertheless, to gain insight into the sensitivity to , we perform two simulations that are identical to REF except that one uses (DX0250) and the other uses (DX1000).

Figures 10a–c indicate a few subtle differences between the LWP and surface winds of the three simulations. First, whereas the cloud trail forms along the wake centerline in the REF and DX0250 cases, it forms along the southern wake edge in the DX1000 case. This positional shift stems from reduced turbulent vertical mixing of easterly momentum within the subcloud layer of the DX1000 case: averaged subcloud total (resolved plus subgrid) momentum flux (calculated over a 20 km 20 km box centered within the immediate wake at ) is only half as large in the DX1000 case (0.036 kg m2 s−2) as in the REF case (0.074 kg m2 s−2), causing a slower recovery of near-surface wind speeds. The weaker wake winds in the DX1000 case undergo more frictional backing, which imparts a stronger northerly wind component that shifts the convergence line southward.

Fig. 10.

LWP, surface wind vectors, and terrain height (contoured every 200 m) at 1500 LST for all of the sensitivity simulations of section 3c. Simulation names are provided in the bottom-right portion of every panel.

Fig. 10.

LWP, surface wind vectors, and terrain height (contoured every 200 m) at 1500 LST for all of the sensitivity simulations of section 3c. Simulation names are provided in the bottom-right portion of every panel.

In addition, as decreases the cumuli over and downwind of the island become more cellular and the cloud trails become drier. Because smaller permits faster-growing convective cells (e.g., Kirshbaum et al. 2007), it allows for stronger convection over the short time that air parcels traverse the island. Combined with increased mountain wave turbulence due to the taller and steeper terrain at smaller , this produces a stabler immediate wake that locally inhibits convection [an additional simulation with no latent heating (not shown) suggests that the wave breaking dominates the wake stabilization]. The surface-based convective available potential energy (CAPE) along the wake centerline decreases from around 150 J kg−1 (DX1000) to 120 J kg−1 (REF) to 70 J kg−1 (DX0250) (Figs. 11a–c). Note that around 20 J kg−1 of the CAPE reduction in the DX0250 case arises from weak domain-wide resolved mixing in the subcloud layer. The increased wake stability at smaller contributes to a reduction of most core-vigor metrics and rainfall (Table 3).

Fig. 11.

Surface-based CAPE, surface wind vectors, and terrain height (contoured every 200 m) at 1500 LST for (a) DX1000, (b) DX0500, (c) DX0250, (d) HM0.0, (e) HM0.5, and (f) HM2.0 simulations.

Fig. 11.

Surface-based CAPE, surface wind vectors, and terrain height (contoured every 200 m) at 1500 LST for (a) DX1000, (b) DX0500, (c) DX0250, (d) HM0.0, (e) HM0.5, and (f) HM2.0 simulations.

Another potential contributor to the weakened convection at smaller is a transition from mostly laminar (DX1000) to partially turbulent (DX0250) convective overturning within the cloud trail. The stronger explicit turbulence at smaller , which is reflected by increased MFcb (due to enhanced turbulent transport across cloud base) and (due to faster cell growth), modifies important processes within the cloud trail. In particular, the entrainment of environmental air is carried out implicitly (through subgrid mixing and numerical diffusion) in laminar clouds but explicitly in turbulent clouds (e.g., Bryan et al. 2003), which may render the cloud dilution sensitive to . The reduced and at smaller is consistent with the notion that the more turbulent clouds are more strongly diluted by entrainment. However, because this effect cannot be easily separated from the wake-stability effect above, we defer a more detailed analysis of it to future work.

While the above variations in strongly impact the character of turbulence in unstable portions of the flow (e.g., windward cumuli, island wave breaking, and the cloud trail), the sensitivity to is still weak: the cloud trail in all three cases is similar in morphology, intensity, and precipitation. Thus, despite its deficiencies, the choice of = 500 m appears sufficient to capture the essential aspects of cloud-trail dynamics.

2) Surface heat fluxes

To examine the sensitivity of cloud trails to surface heat fluxes, we conduct three simulations where H and LE are scaled by factors of 0 (FLX0.0), 0.5 (FLX0.5), and 2 (FLX2.0) relative to the REF case (Table 2). In the FLX0.0 case a long, weakly convergent, and decelerated wake develops with little to no cloud formation (Fig. 10d), which contrasts with the strongly convergent and accelerated wake in the REF case (Fig. 10b). All measures of subcloud forcing and cloud vigor drop precipitously in the FLX0.0 case: MFcb and MFinv decrease by one to two orders of magnitude, the convective cores are less numerous and vigorous, is shallower, and rainfall is minimal (Table 3). Thus, for the environmental conditions considered herein, surface heating is essential for cloud trails to develop.

As the fluxes increase, the wake convergence strengthens and the cloud trail intensifies. Because of the modest thermal forcing in the FLX0.5 simulation, a weak cloud trail develops far downwind (Fig. 10e). By contrast, intense convergence in the FLX2.0 simulation produces a vigorous cloud trail with heavy rainfall (Fig. 10f), which creates a strong cold pool that rapidly widens the cloud trail downwind. The updraft mass fluxes, core area and buoyancy, and rainfall increase monotonically with H and LE in the FLX0.5, REF, and FLX2.0 sequence (Table 3). At the same time, and both decrease, likely due to stronger compensating subsidence among the closely packed cloud cores and the depletion of by precipitation. Thus, while the cloud-trail intensity increases with heating rate (FLX2.0), the cloud-trail organization is maximized under moderate heating (REF).

3) Terrain height

The mechanical response of trade-wind flow to island terrain may be described by the regime diagram of Schär and Smith (1993, hereafter SS93), who considered a single layer of shallow-water fluid beneath a density discontinuity. The regimes depend on two parameters: and the upstream Froude number , where is the terrain height, D is the layer depth, U is the wind speed, is the reduced gravity, is the density of the layer, and is the density change across the discontinuity. As and/or increases, the regime transitions from 1) inviscid flow over the terrain, to 2) flow over with a leeside hydraulic jump and decelerated wake, and then to 3) flow separation around the terrain and wake flow reversal (possibly with lee vortices). Regime 2 has two subcategories: 2a (downwind wake flow) and 2b (reversed wake flow).

Because SS93’s theory was developed for dry, unheated flows, it cannot fully describe the moist and heated flows considered herein. Nonetheless, it provides a useful reference point to aid the model interpretation. To apply this theory to our experiments, we follow Smith et al. (1997) by choosing D as the midpoint of the combined cloud and inversion layer (1250 m), as the mean cross-barrier wind component below D, as the mean terrain height along the ridge axis, and , where at the surface and across the cloud and inversion layers. For the REF case we obtain and (regime 2a), the mechanical dynamics of which are illustrated by the unheated FLX0.0 simulation, where a decelerated and weakly convergent wake stretches over 100 km downwind (Fig. 10d). This case sharply contrasts with the wake acceleration and quasi-linear convergence of the REF case (Fig. 10b), reinforcing that thermal forcing drives the cloud-trail formation. To more thoroughly assess the impacts of mechanical forcing on cloud trails, we conduct three simulations where is scaled by factors of 0 (HM0.0), 0.5 (HM0.5), and 2 (HM2.0) relative to the REF case (Table 2). Whereas the HM0.0 and HM0.5 simulations both lie within regime 1 of SS93, the HM2.0 simulation falls within regime 3.

The cloud trails in the HM0.0–HM0.5 cases are similar to that in the REF case (Figs. 10g,h) except for being slightly weaker with lower and (Table 3). This weakening arises in part from reduced elevated heating and/or mechanical convergence past flatter terrain, which will be examined further in section 4. In addition, the lack of island convection or wave breaking in the HM0.0 and HM0.5 cases leads to enhanced CAPE in the immediate wake (Figs. 11b,d,e), which promotes stronger convection prior to the organized thermally forced convergence farther downwind (Figs. 10d,e). This convection stabilizes the flow and ventilates the wake boundary layer heat anomaly, the latter of which weakens the thermal convergence band feeding the cloud trail. The comparison in Figs. 12a, 12b, and 8c suggests that the wake is progressively diminished by subcloud vertical motion in the HM0.5 and HM0.0 cases. Note that the weakened convergence in these cases is not reflected by MFcb in Table 3 because the updrafts in the immediate wake are included in the calculation. However, if the eastern boundary of this calculation is shifted to 200 km, MFcb in the HM0.0 case becomes 12% lower than that in the REF case.

Fig. 12.

As in Fig. 8, but for the (a) HM0.0, (b) HM0.5, (c) HM2.0, (d) U0.5, (e) U1.5, and (f) U2.0 simulations.

Fig. 12.

As in Fig. 8, but for the (a) HM0.0, (b) HM0.5, (c) HM2.0, (d) U0.5, (e) U1.5, and (f) U2.0 simulations.

A major change in flow dynamics occurs in the HM2.0 case, where the taller mountain (2 km) pierces the inversion. As expected in regime 3, the island flow consists of upstream flow deflection, wake flow reversal, and lee vortices (Fig. 10i). In contrast to the quasi-linear cloud-trail organization past shorter islands, the lee vortices cause the cloud trails to meander wildly. Moreover, stronger island convection and intense gravity wave breaking deplete most of the moist instability (Fig. 11f), leaving the wake less favorable for cloud trails. The wake thermal anomaly is the strongest in this case (Fig. 12c) due to intense turbulence over the island and the trapping of island-heated air in the stagnant immediate wake.

4) Boundary layer winds

To isolate the cloud-trail sensitivities to trade-wind speed, we conduct three simulations where the initial subcloud-layer winds () are scaled by 0.5 (U0.5), 1.5 (U1.5), and 2 (U2.0) relative to REF (Table 2). The winds are linearly relaxed back to the REF profile from to . While the U0.5 case lies in regime I of SS93, the U1.5 and U2.0 cases lie in regime 2b. As seen in Figs. 10b and 10j, the U0.5 cloud trail is shorter and less organized than that in REF. These changes arise from diminished thermal advection and stronger upstream blocking, which trap the island-heated air in the immediate wake and limit its ability to organize the cloud trail downwind (Fig. 12d). Flow blocking is increased when the nondimensional mountain height , where N and U are the mean Brunt–Väisälä frequency and wind speed in the subcrest layer (e.g., Smith 1989), exceeds unity. Herein, M doubles from 1 in the REF case to 2 in the U0.5 case, causing increased flow deflection around the island and a more stagnant immediate wake (Figs. 10j and 12d).

As increases, the wake thermal anomaly is weakened (due to decreased residence time of air parcels over the heated island) and is carried farther downwind (Figs. 12d–f). Contrary to the expectation of regime 2b of SS93, no wake flow reversal occurs—the wake flow remains consistent with regime 2a. The more elongated thermal anomaly gives rise to long and well-organized cloud trails in the REF and U1.5 cases (Figs. 10b and 10k). However, this organization degrades in the U2.0 case (Fig. 10l) due to a diminished wake thermal anomaly and increased turbulence production within the hydraulic jump (SS93). This increased turbulence is apparent from the surface winds and w at in Fig. 13. The w variance in the boxed wake region increases from 0.025 (REF) to 0.031 (U1.5) to 0.054 m2 s−2 (U2.0). This turbulence ventilates the wake thermal anomaly and disrupts the quasi-linear convergence zone organizing the convection.

Fig. 13.

Vertical velocity w at and surface wind perturbation vectors at 1500 LST for the (a) REF, (b) U1.5, and (c) U2.0 simulations. The boxed area is used for the vertical velocity variance comparison in section 3c.

Fig. 13.

Vertical velocity w at and surface wind perturbation vectors at 1500 LST for the (a) REF, (b) U1.5, and (c) U2.0 simulations. The boxed area is used for the vertical velocity variance comparison in section 3c.

While MFcb progressively increases in the REF, U1.5, and U2.0 cases due to stronger mechanical turbulence, MFinv and most other cloud diagnostics reach their maxima in the REF case (Table 3). Thus, moderate-to-strong subcloud winds appear most favorable for organized cloud trails past Dominica due to their combination of sufficient diabatic thermal anomalies, strong downwind heat advection, and modest wake turbulence. This finding is consistent with section 2, where observed type-II cloud trails preferred moderate but not excessive wind speeds over the 0–1500-m layer (6–9 m s−1).

5) Inversion strength

As found in section 3b, the inversion height influences cloud-trail organization through its control over precipitation. The inversion strength might be expected to have similar effects, in that stronger inversions may inhibit rainfall and thus enhance cloud-trail organization. Based on the trade-wind climatology of Davison et al. (2013), the REF inversion [with a lapse rate of 0.1 K (50 m)−1 over 1600–1800 m] is slightly stronger than average. To evaluate a broad range of inversion strengths, we perform three simulations where the θ increase across the inversion layer is scaled by 0.5 (INV0.5), 1.5 (INV1.5), and 2.0 (INV2.0) (Table 2). While the associated modifications to the upstream are insufficient to shift the flows out of regime 2a of SS93, they clearly impact the cloud trail, which becomes longer and wider with increasing (Figs. 10m–o). A secondary cloud trail also develops about 30 km north of the primary one in the INV1.5 and INV2.0 cases.

The weaker inversion in the INV0.5 case increases the initial CAPE from ~120 J kg−1 in the REF case to over 600 J kg−1. While the inversion remains largely effective at capping convection below 2.5 km, deeper cells develop near the leading edge of the cloud trail where the heated wake air is abruptly lifted. These cells create heavy rain and strong cold pools that enhance the low-level convergence (and thus MFcb). However, they also obliterate the wake thermal anomaly and cut off the inflow to the cloud trail, which inhibits convection farther downwind. Thus, despite generating the largest MFcb, highest cloud tops, and heaviest rainfall, the INV0.5 case produces a relatively short cloud trail. By contrast, the sharper inversions in the INV1.5 and INV2.0 cases effectively cap convection at the inversion, as reflected by progressively reduced and rainfall (Table 3). Whereas the REF clouds breach the inversion and then evaporate, those in the INV1.5–INV2.0 cases spread laterally at the inversion base, which widens the cloud-trail LWP signature. Similarly, the secondary cloud trail in these cases arises from the advection of cloud generated in the immediate wake that fails to evaporate. The suppressed rainfall in these cases lengthens the cloud trail by allowing the quasi-linear wake convergence to remain intact.

4. Scaling

In this section, we investigate whether the subcloud thermal circulations that drive island cloud trails may be quantified using a simple scaling that incorporates key environmental and topographic parameters. To start, we assume that the thermal circulations are controlled by five external parameters: , , , , and , where is a reference temperature and and are the along-flow and cross-flow island dimensions. In this analysis we neglect mechanical wake effects, which renders only important for its control of the residence time of air parcels over the island and h only important for elevated-heating effects (which will be incorporated into H).

We first analyze the horizontal component of the thermal wake circulation, which we denote by . Given dimensional variables ( and the five listed above) and independent dimensions (length, time, and temperature), Buckingham’s Pi theorem implies that three nondimensional numbers are available to describe the system, two of which are independent (e.g., Kundu and Cohen 2002). From dimensional analysis we find these parameters to be

 
formula
 
formula
 
formula

The two independent nondimensional parameters ( and ) represent the vertical aspect ratio of the circulation and the nondimensional buoyancy gradient in the cross-flow direction (which drives the circulation). The island aspect ratio (), which appears in , controls the projection of the island heating onto the cross-flow wake thermal circulation. We solve for using

 
formula

Because G is a power-law function of each nondimensional parameter,

 
formula

where , , and are empirical constants. A similar expression is obtained for the vertical velocity scale , but with different constants , , and .

To consider elevated mountain heating, we replace H with , where is the elevated-heating contribution. Crook and Tucker (2005) represented elevated heating by assuming an exponentially decaying surface-based heating function and isolating (and then linearizing) the terrain-induced baroclinicity. Although their heating function was highly simplified, it still provides a useful first-order quantification of elevated-heating effects. Thus, following Crook and Tucker (2005), we set , where is the vertical heating scale. To eliminate the spatial dependence of h, we replace it with its approximate average over the island () to give

 
formula

We estimate the exponents in (7) by diagnosing the simulated convergent wind speeds and vertical motion from a series of 12 “dry” numerical simulations that are identical to those of section 3, but that and are set to zero. These simulations enable quantification of subcloud circulations in the absence of latent-heat feedbacks. The simulations, which are denoted by adding “-DRY” to the moist simulation name, include REF-DRY, three surface-flux experiments (FLX0.0-DRY, FLX0.5-DRY, FLX2.0-DRY), two mountain-height experiments (HM0.0-DRY and HM0.5-DRY), two wind speed experiments (U0.5-HM0.0-DRY and U1.5-HM0.0-DRY), two inversion-strength experiments (INV0.5-DRY and INV2.0-DRY), and two simulations where the subcloud-layer depth is halved and doubled, with no changes to the winds or overlying atmosphere (ZCB0.5-HM0.0-DRY and ZCB2.0-HM0.0-DRY). To minimize the impacts of mechanical effects that are neglected by the scaling, we use flat terrain for the wind speed and subcloud-layer-depth experiments.

To quantify the simulated cross-flow winds and subcloud ascent, we follow a similar procedure to Fig. 9a by creating a rotated rectangle over the strongest section of the wake updraft at . The rectangle length and width are and , with the long axis again aligned with the updraft band at an angle θ from the horizontal ( is halved in the U0.5-HM0.0-DRY simulation due to the shorter updraft-band length). After interpolating the data onto the new grid, we find the maximum (at the surface) and w (at ) at each to get and . We then average these values in and in time (over 1500–1700 LST) to obtain and .

The scalings of and in (7) are compared to the simulated and in Fig. 14 (except for the unheated FLX0.0-DRY case). For these calculations we set , , and while varying , , H, and according to their values in each simulation. A good fit to the data is found with , = = = 0.5, , and , for which the coefficient of determination () is 0.92 for and 0.93 for . This gives the following empirical scalings:

 
formula
 
formula

The simulated data collapses well onto these lines, suggesting that (9) and (10) capture the key sensitivities of the wake thermal circulations. For reference, Fig. 14 overlays the scaling results if is neglected in (9) and (10), indicating that the flat and mountainous cases no longer fall onto the same line. Thus, consideration of elevated heating appears necessary to capture the terrain-height sensitivities.

Fig. 14.

Comparison of (a) scaled with simulated and (b) scaled with simulated . Plus symbols indicate flat-terrain simulations and triangles indicate mountain simulations. Blue symbols use the full scaling with elevated heating from (8) in (9)(10) and green symbols neglect the elevated-heating contribution. A good linear fit to the scaling is shown in dashed lines.

Fig. 14.

Comparison of (a) scaled with simulated and (b) scaled with simulated . Plus symbols indicate flat-terrain simulations and triangles indicate mountain simulations. Blue symbols use the full scaling with elevated heating from (8) in (9)(10) and green symbols neglect the elevated-heating contribution. A good linear fit to the scaling is shown in dashed lines.

Provided the mechanical response is not strongly nonlinear, we apparently can neglect it in the scaling. In the unheated FLX0.0-DRY case (Fig. 10d), purely mechanical leeside convergence produces and , which are modest but significant. In principle, this convergence may superpose with thermal convergence to enhance the circulation strength (as suggested by Y08b). However, the difference between in the REF-DRY and HM0.0-DRY simulations (0.7 m s−1), which differ only in their terrain height, is less than half of the above-mentioned in the FLX0.0-DRY simulation. As most of this difference can already be accounted for by elevated-heating effects, the superposition of mechanical and thermal circulations appears small enough to neglect.

Because this scaling only addresses the subcloud circulation strength and neglects mechanical effects, it can only be used to predict the initiation (rather than the organization) of cloud trails past shorter islands. Consider, for example, the U0.5-DRY and FLX2.0-DRY cases, for which the scaling predicts and to be identical. However, the U0.5 and FLX2.0 simulations develop very different cloud trails, with enhanced organization in the latter (Figs. 10f and 10l). These differences arise from factors that are omitted in the scaling. In particular, upstream flow blocking in the U0.5 case prevents the island-heated air from traveling downstream and organizing the cloud trail. Also, the stronger winds in the FLX2.0 case double the length of the wake thermal circulation and the associated cloud trail.

5. Further discussion

a. On the wind speed sensitivity

To explain the enhanced afternoon cloud frequency and organization past Kauai and Niihau under stronger trade winds, Y08b argued that the stronger trade winds led to stronger thermal advection, which, in turn, strengthened wake convergence and subcloud ascent. While our short climatology of the Lesser Antilles in section 2 reveals a similar sensitivity to trade-wind speed, we question Y08b’s hypothesis on two fronts. First, it is the temperature gradient, not the temperature advection that controls thermal-circulation strength (e.g., Crook 2001; Kirshbaum 2013). Although stronger winds increase thermal advection, they also reduce the integrated sensible heating over the island, which weakens the downwind temperature gradients. Second, support for Y08b’s hypothesis was provided by a wake cross section at a fixed downwind location (see their Fig. 17). However, because cloud-trail positions are sensitive to wind speed, these cross sections may have sampled the bands at different phases of their development.

For the above reasons it is not surprising that Y08b’s hypothesis does not completely hold under closer inspection. In particular, the subcloud updrafts in the simulations of section 4 actually strengthen as decreases. This is consistent with the scaling in (10), which suggests an inverse dependence of on . Thus, the degradation of cloud-trail organization at smaller does not arise from weaker subcloud updrafts. Based on section 3c, we argue that this degradation is owing to the combination of weaker advection and stronger flow blocking, which trap the island heat anomaly in the immediate wake.

A new finding from the wind speed experiments is that very strong trade winds can disrupt cloud-trail organization by diminishing the wake thermal anomaly and enhancing wake turbulence. Because of physical limits on subtropical pressure gradients, along with surface friction over Guadeloupe, the TFFR-observed subcloud winds never approached those in the U1.5 or U2.0 simulations (12–16 m s−1). Nevertheless, to demonstrate that stronger winds can indeed have a suppressive effect on cloud trails, we examine the strongest-wind event of our database (excluding tropical cyclones): 7 February 2011, for which the mass-weighted subcloud (0–600 m) and boundary layer (0–1500 m) wind speeds from the TFFR sounding were 9.3 and 12.5 m s−1 (the true impinging winds were likely stronger). Martinique, Dominica, and Guadeloupe received ample sunshine, the TFFR cloud base was located at ~750 m and the cloud-layer depth was less than 800 m. Although sufficient low-level moisture, strong diabatic heating, and a shallow cloud layer all favored organized cloud trails, the observed cloud plumes were faint and poorly organized (Fig. 15), rendering this a type-I event. As in the U2.0 case, the strong impinging winds may have acted to degrade the cloud-trail organization. By contrast, in the nine cases where the TFFR 0–600-m winds were more akin to the REF case (), seven were classified as type II.

Fig. 15.

Visible satellite images from (a) 1439 and (b) 1609 LST 7 Feb 2011, indicating poor cloud-trail organization under strong boundary layer winds. The arrow in (a) indicates the mass-weighted boundary layer wind direction over 0–1500 m from the corresponding TFFR sounding.

Fig. 15.

Visible satellite images from (a) 1439 and (b) 1609 LST 7 Feb 2011, indicating poor cloud-trail organization under strong boundary layer winds. The arrow in (a) indicates the mass-weighted boundary layer wind direction over 0–1500 m from the corresponding TFFR sounding.

b. Limitations of the experiments

Although the findings herein provide useful insights into the cloud-trail dynamics, limitations of the observational and numerical analyses introduce some uncertainties that merit discussion. Observationally, our brief subjective analysis is prone to error and unable to distinguish subtle differences between cloud trails past the different islands. To enable a longer, more objective observational analysis that can be easily extended to other areas, an automated cloud-trail detection algorithm is needed. Furthermore, our analysis of visible imagery provides no information about nighttime cloud patterns. Because visible sensors cannot see clouds at night, and because infrared sensors have too coarse a grid resolution to detect most cloud trails, there is no obvious solution to this issue.

The simulations contained various idealizations including a free-slip, unheated ocean surface and the neglect of cloud–radiative feedbacks. While these idealizations were necessary to isolate and quantify processes of interest, they may also influence some of the simulated trends. Over mountainous islands in particular, the reduction in insolation associated with mechanically forced clouds may counteract the enhancements in wake thermal convergence associated with elevated heating. This may help to explain why the observed cloud-trail frequency is lower past Dominica than past the shorter islands (Table 1). Finally, the horizontal grid spacing () of the simulations was too coarse to fully resolve the moist and dry turbulence in the flows considered. Although preliminary results suggested only weak sensitivity to , further simulations with large-eddy resolution () would be useful to evaluate the robustness of this result.

Because of space constraints we did not comprehensively study the sensitivities of cloud trails in section 3c. One potentially important parameter that was neglected is cloud-layer humidity, which Davison et al. (2013) found to be highly variable in their RICO-based trade-wind climatology. Because the cloud-layer humidity controls the suppressive impacts of entrainment on cloud buoyancy, liquid water, and depth, it may strongly regulate rainfall and, in turn, cloud-trail organization. We thus hypothesize that drier cloud layers are more favorable for well-organized cloud trails.

6. Conclusions

We have presented observations and numerical simulations of elongated cloud bands (or “cloud trails”) past the Lesser Antilles, a mountainous Caribbean island chain subject to persistent trade-wind flow. Manual analysis of one year of visible satellite data revealed that, for all the islands considered, cloud trails developed around 30%–40% of days and around 60% of “undisturbed” days without larger-scale cloudiness. On about 20% of the days, well-organized and quasi-stationary bands with lengths and durations min formed past multiple islands. Consistent with a recent study of cloud trails past Kauai and Niihau (Y08b), the cloud trails developed in the afternoon due to diurnal heating over their parent islands. Radiosonde analysis suggested that the steadiest and most elongated cloud trails formed under the combination of moderate-to-strong boundary layer winds, strong trade-wind inversions, high pressure, and dry free tropospheres, the latter three consistent with stronger large-scale subsidence.

Cloud-resolving simulations with the WRF Model provided insights into the dynamics and sensitivities of the cloud trails. Using a representative trade-wind sounding as the initial state, the island of Dominica as the topographic forcing, and a reasonable diurnal cycle of island surface fluxes, the simulations produced realistic afternoon cloud trails. These clouds formed within the turbulent thermal wake, where cross-stream buoyancy gradients generated narrow, quasi-linear subcloud convergence bands aligned with the ambient winds. These updraft bands initiated cloud trails with lengths of up to 200 km and widths of 1–3 km, consistent with the satellite images.

As in the observations, the best-organized simulated cloud trails formed under moderate-to-strong boundary layer winds (6–12 m s−1), shallow cloud layers ( deep), strong trade-wind inversions, and medium terrain heights (500–1000 m). The fact that such events require both strong winds and shallow inversions may help to explain their infrequency. Based on the theoretical arguments of Nuijens and Stevens (2012), stronger trade winds tend to be associated with deeper (rather than shallower) trade-wind inversions. The simulated cloud trails were only weakly sensitive to the horizontal grid spacing but strongly sensitive to environmental and terrain-related parameters. Deeper cloud layers and/or weaker inversions led to heavy rainfall and strong subcloud cold pools that interfered with the delicate quasi-linear cloud-trail organization. By contrast, stronger inversions produced longer cloud trails with broad anvils due to saturated outflow at the inversion base. Stronger island heating increased the wake thermal anomaly, which enhanced the subcloud updrafts but also increased the wake moist instability, leading to more vigorous moist convection and stronger evaporative cold pools. When the surface fluxes were eliminated, the cloud trails completely vanished despite weak mechanical convergence in the wake.

The simulated cloud trails were also sensitive to island terrain height, which controlled the mechanical flow response and elevated-heating effects. For cloud trails to form, the flow needed to ascend the terrain (rather than detour around it) and lie within regimes 1–2 of the shallow-water regime diagram of SS93. Within these regimes the impinging airflow was heated by the island surface and then carried into the wake. In regime 3, the island surface pierced the inversion and the impinging flow detoured laterally around it, leaving the island-heated air trapped in the wake while lee vortices disrupted the cloud-trail organization. The cloud trails were the best organized for intermediate terrain heights, where moist convection was suppressed in the immediate wake due to over-island convection and gravity wave turbulence, preserving the wake thermal anomaly until organized thermally driven convergence developed farther downwind. The orography also strengthened the wake thermal convergence (due to elevated heating) and mechanical convergence (due to lateral flow deflection).

The subcloud winds controlled the simulated cloud-trail organization through their impacts on subcloud wake turbulence, thermal advection, and flow blocking. While weaker winds (~4 m s−1) strengthened the wake thermal anomaly by increasing the residence time of air parcels over the heated island, they were unable to advect the heat far downwind (due to diminished advection and stronger flow blocking). As a result, the wake convergence band was shorter and readily eliminated by precipitating convection. Moderate-to-strong winds (8–12 m s−1), by contrast, carried the wake thermal perturbation farther downwind through a moderately turbulent wake, which was more favorable for elongated cloud trails. Very strong subcloud winds (~16 m s−1) both diminished the wake thermal perturbation and intensified the wake turbulence, which together degraded the cloud-trail coherence.

Dimensional analysis was used to develop empirical scalings for the thermally forced subcloud circulations in the heated wake. The scalings used external parameters including the island dimensions, subcloud-layer winds and depth, and island sensible heat flux. Provided that elevated-heating effects were considered, the scalings captured the simulated sensitivities of the wake circulations with good accuracy, which is impressive given the complexity and nonlinearity of the flows under consideration. Because all of the input parameters are either known or available from coarse-resolution global forecasts, the scaling may provide a framework for efficiently predicting cloud-trail initiation.

Acknowledgments

We thank Ron Smith, Justin Minder, and Dave Schultz for helpful discussions during this study. Raw satellite data were downloaded from NOAA’s Comprehensive Large Array-Data Stewardship System (CLASS) and radiosonde data were obtained from the University of Wyoming. The first author was funded by the Natural Science and Engineering Research Council (NSERC) Discovery Grant NSERC/RGPIN 418372-12, and the simulations were performed on the Guillimin supercomputer at McGill University, under the auspices of Calcul Québec and Compute Canada. The second author was partially funded by the U.K. Natural Environment Research Council (NERC) Precipitation Structures over Orography (PRESTO) project at the University of Manchester (Grant NE/1024984/1).

REFERENCES

REFERENCES
Banta
,
R. M.
,
1990
: Atmospheric Processes over Complex Terrain. Meteor. Monogr., No. 45, Amer. Meteor. Soc., 323 pp.
Barthlott
,
C.
, and
D. J.
Kirshbaum
,
2013
:
Sensitivity of deep convection to terrain forcing over Mediterranean islands
.
Quart. J. Roy. Meteor. Soc.
,
139
,
1762
1779
, doi:.
Bryan
,
G. H.
,
J. C.
Wyngaard
, and
J. M.
Fritsch
,
2003
:
Resolution requirements for the simulation of deep moist convection
.
Mon. Wea. Rev.
,
131
,
2394
2416
, doi:.
Carbone
,
R. E.
,
J. W.
Wilson
,
T. D.
Keenan
, and
J. M.
Hacker
,
2000
:
Tropical island convection in the absence of significant topography. Part I: Life cycle of diurnally forced convection
.
Mon. Wea. Rev.
,
128
,
3459
3480
, doi:.
Crook
,
N. A.
,
2001
:
Understanding Hector: The dynamics of island thunderstorms
.
Mon. Wea. Rev.
,
129
,
1550
1563
, doi:.
Crook
,
N. A.
, and
D. F.
Tucker
,
2005
:
Flow over heated terrain. Part I: Linear theory and idealized numerical simulations
.
Mon. Wea. Rev.
,
133
,
2552
2564
, doi:.
Davison
,
J. L.
,
R. M.
Rauber
,
L.
DiGirolamo
, and
M. A.
LeMone
,
2013
:
A revised conceptual model of the tropical marine boundary layer. Part I: Statistical characterization of the variability inherent in the wintertime trade wind regime over the western tropical Atlantic
.
J. Atmos. Sci.
,
70
,
3005
3024
, doi:.
Dorman
,
C. E.
,
1994
:
Guadalupe Island cloud trail
.
Mon. Wea. Rev.
,
122
,
235
242
, doi:.
Fovell
,
R. G.
,
2005
:
Convective initiation ahead of the sea-breeze front
.
Mon. Wea. Rev.
,
133
,
264
278
, doi:.
Kirshbaum
,
D. J.
,
2013
:
On thermally forced circulations over heated terrain
.
J. Atmos. Sci.
,
70
,
1690
1709
, doi:.
Kirshbaum
,
D. J.
, and
R. B.
Smith
,
2009
:
Orographic precipitation in the tropics: Large-eddy simulations and theory
.
J. Atmos. Sci.
,
66
,
2559
2578
, doi:.
Kirshbaum
,
D. J.
, and
A. L. M.
Grant
,
2012
:
Invigoration of cumulus cloud fields by mesoscale ascent
.
Quart. J. Roy. Meteor. Soc.
,
138
,
2136
2150
, doi:.
Kirshbaum
,
D. J.
, and
C.-C.
Wang
,
2014
:
Boundary layer updrafts driven by airflow over heated terrain
.
J. Atmos. Sci.
,
71
,
1425
1442
, doi:.
Kirshbaum
,
D. J.
,
R.
Rotunno
, and
G. H.
Bryan
,
2007
:
The spacing of orographic rainbands triggered by small-scale topography
.
J. Atmos. Sci.
,
64
,
4222
4245
, doi:.
Kundu
,
P. M.
, and
I. M.
Cohen
,
2002
: Fluid Mechanics. 2nd ed. Elsevier Science, 730 pp.
Malkus
,
J. S.
,
1963
:
Tropical rain induced by a small heat source
.
J. Appl. Meteor.
,
2
,
547
556
, doi:.
Malkus
,
J. S.
, and
A. E.
Bunker
,
1952
: Observational studies of the airflow over Nantucket island during the summer of 1950. Papers in Physical Oceanography and Meteorology, Vol. XII, No. 2, Massachusetts Institute of Technology and Woods Hole Oceanographic Institution, 50 pp.
Matthews
,
S.
,
J. M.
Hacker
,
J.
Cole
,
J.
Hare
,
C. N.
Long
, and
R. M.
Reynolds
,
2007
:
Modification of the atmospheric boundary layer by a small island: Observations from Nauru
.
Mon. Wea. Rev.
,
135
,
891
905
, doi:.
McFarlane
,
S. A.
,
C. N.
Long
, and
D. M.
Flynn
,
2005
:
Impact of island-induced clouds on surface measurements: Analysis of the ARM Nauru Island effect study data
.
J. Appl. Meteor.
,
44
,
1045
1065
, doi:.
Minder
,
J.
,
R. B.
Smith
, and
A. D.
Nugent
,
2013
:
The dynamics of ascent-forced orographic convection in the tropics: Results from Dominica
.
J. Atmos. Sci.
,
70
,
4067
4088
, doi:.
Nordeen
,
M. L.
,
P.
Minnis
,
D. R.
Doelling
,
D.
Pethick
, and
L.
Nguyen
,
2001
:
Satellite observations of cloud plumes generated by Nauru
.
Geophys. Res. Lett.
,
28
,
631
634
, doi:.
Nuijens
,
L.
, and
B.
Stevens
,
2012
:
The influence of wind speed on shallow marine cumulus convection
.
J. Atmos. Sci.
,
69
,
168
184
, doi:.
Schär
,
C.
, and
R. B.
Smith
,
1993
:
Shallow-water flow past isolated topography. Part I: Vorticity production and wake formation
.
J. Atmos. Sci.
,
50
,
1373
1400
, doi:.
Skamarock
,
W. C.
, and Coauthors
,
2008
: A description of the Advanced Research WRF version 3. NCAR Tech. Note NCAR/TN-475+STR, 113 pp. [Available online at http://www.mmm.ucar.edu/wrf/users/docs/arw_v3_bw.pdf.]
Smith
,
R. B.
,
1989
:
Hydrostatic airflow over mountains
. Advances in Geophysics, Vol. 31, Academic Press,
1
41
, doi:.
Smith
,
R. B.
,
A. C.
Gleason
,
P. A.
Gluhosky
, and
V.
Grubiŝić
,
1997
:
The wake of St. Vincent
.
J. Atmos. Sci.
,
54
,
606
623
, doi:.
Smith
,
R. B.
,
P.
Schafer
,
D. J.
Kirshbaum
, and
E.
Regina
,
2009a
:
Orographic enhancement of precipitation inside Hurricane Dean
.
J. Hydrometeor.
,
10
,
820
831
, doi:.
Smith
,
R. B.
,
P.
Schafer
,
D. J.
Kirshbaum
, and
E.
Regina
,
2009b
:
Orographic precipitation in the tropics: Experiments in Dominica
.
J. Atmos. Sci.
,
66
,
1698
1716
, doi:.
Smith
,
R. B.
, and Coauthors
,
2012
:
Orographic precipitation in the tropics: The Dominica experiment
.
Bull. Amer. Meteor. Soc.
,
93
,
1567
1579
, doi:.
Smolarkiewicz
,
P. K.
,
R. M.
Rasmussen
, and
T. L.
Clark
,
1988
:
On the dynamics of Hawaiian cloud bands: Island forcing
.
J. Atmos. Sci.
,
45
,
1872
1905
, doi:.
Souza
,
R. L. D.
,
1972
: A study of atmospheric flow over a tropical island. Ph.D. thesis, Florida State University, 203 pp.
Stevens
,
B.
, and
A.
Seifert
,
2008
:
Understanding macrophysical outcomes of microphysical choices in simulations of shallow cumulus convection
.
J. Meteor. Soc. Japan
,
86A
,
143
162
, doi:.
Warren
,
R. A.
,
D. J.
Kirshbaum
,
R. S.
Plant
, and
H. W.
Lean
,
2014
:
A ‘Boscastle-type’ quasi-stationary convective system over the UK southwest peninsula
.
Quart. J. Roy. Meteor. Soc.
,
140
,
240
257
, doi:.
Yang
,
Y.
, and
Y.-L.
Chen
,
2008
:
Effects of terrain heights and sizes on island-scale circulations and rainfall for the island of Hawaii during HaRP
.
Mon. Wea. Rev.
,
136
,
120
146
, doi:.
Yang
,
Y.
,
S.-P.
Xie
, and
J.
Hafner
,
2008a
:
Cloud patterns lee of Hawaii island: A synthesis of satellite observations and numerical simulation
.
J. Geophys. Res.
,
113
, D15126, doi:.
Yang
,
Y.
,
S.-P.
Xie
, and
J.
Hafner
,
2008b
:
The thermal wake of Kauai island: Satellite observations and numerical simulations
.
J. Climate
,
21
,
4568
4586
, doi:.