1. Introduction
Within 10° latitude of the equator, organized atmospheric convection occurs across scales ranging from individual thunderstorm systems to planetary-scale disturbances such as the Madden–Julian oscillation (MJO). Understanding what drives these different scales is a key question for tropical weather prediction and accurate simulation of the atmospheric general circulation. At the cloud scale, we have a fairly good knowledge of convective system structure and the processes that determine how storms evolve (see, e.g., Rottuno et al. 1988). Our understanding decreases as convection aggregates from individual storms to mesoscale clusters, and eventually to larger-scale disturbances like the MJO. A theory connecting convection with large-scale dynamics for the MJO is notably absent, and was part of the motivation for the Dynamics of the Madden–Julian Oscillation (DYNAMO) field experiment conducted during the fall of 2011 in the central Indian Ocean. DYNAMO researchers observed the convective clouds and the convective population with detailed observations from the surface, aircraft, and satellites, including Doppler cloud and precipitation radar, as well as other in situ ocean and atmosphere measurements (Powell and Houze 2013; Moum et al. 2014).
One of the main goals of DYNAMO was to better describe the convective processes leading up to the formation of a MJO event. MJO events typically form in the central Indian Ocean and propagate eastward over the eastern Pacific. They affect weather, such as hurricane development in both the eastern Pacific and western Atlantic tropical warm pools (Maloney and Hartmann 2000; Camargo et al. 2009). Much of our understanding of the MJO structure is based on measurements taken in the western tropical Pacific Ocean as part of the extensive Tropical Ocean and Global Atmosphere Coupled Ocean–Atmosphere Response Experiment (TOGA COARE) field program in 1992–93 (Webster and Lukas 1992), which sampled three MJO events propagating over the western Pacific from the Indian Ocean. MJO events are usually fully developed with a large-scale atmospheric circulation that maintains strong convective activity by the time they reach the western Pacific.
A primary emphasis during the DYNAMO field program was to build an observational network similar to TOGA COARE for the genesis phase of the MJO over the central Indian Ocean, with the goal of understanding the transition from conditions of suppressed convection, to the active phase of the MJO characterized by numerous convective clusters and high average precipitation rates. Convection during the suppressed phase is typically limited by large-scale subsidence and characterized by trade wind cumulus and cumulus congestus systems with only occasional isolated deep convective activity (Johnson et al. 1999). These relatively shallow convective features are believed to moisten the lower troposphere preceding the active phase of the MJO (Kemball-Cook and Weare 2001). As the active phase develops, convective features increase in scale and organization, expanding from isolated cells to large clusters spanning hundreds of kilometers. Synoptic-scale moisture convergence is the dominant source of moisture during the active phase and explains a large fraction of the observed precipitation (Lin and Johnson 1996; Johnson and Ciesielski 2013; de Szoeke et al. 2015).
Convection and surface fluxes are also affected by convectively forced cold pools. Over the tropical ocean, convectively forced cold pools appear as cool and slightly drier air masses beneath convective systems, with wind gusts enhancing surface fluxes as they spread laterally. They have been observed in association with convection at many scales ranging from precipitating trade wind cumulus (Zuidema et al. 2012) to mesoscale convective complexes (Young et al. 1995). In midlatitudes, cold pool boundaries are known to trigger new convection and are often implicated in the development of severe storms (Rotunno et al. 1988; Weckwerth and Parsons 2006; Houston and Wilhelmson 2012). In the deep tropics, cold pools can cause a more than doubling of the local latent heat flux (Jabouille et al. 1996). Understanding how cold pools interact and potentially enhance tropical convection is important for accurate representation of convection in large-scale models where these processes are not well resolved.
On average, heating by deep convection through condensation is the primary mechanism that balances radiative cooling throughout the tropics, thereby maintaining a radiative-convective equilibrium (RCE) over long time scales. Latent heating is nearly balanced by adiabatic cooling of updrafts, and is redistributed by direct convective eddy transport and via adiabatic subsidence in regions of low convective activity. This results in a tropical atmosphere that maintains a steady temperature profile close to the moist adiabatic lapse rate as demonstrated in sounding data from DYNAMO (see Fig. 1). Moisture, on the other hand, varies significantly in the tropics (Fig. 1b) depending on the local conditions. Sources of tropical moisture can be divided into large-scale convergence and surface latent heat flux. We further divide surface latent heat flux into regional average values and anomalies generated by locally enhanced evaporation from surface wind variations in cold pools produced during convective events. We address the relative role of these different water sources on convection as the active phase of the MJO develops.
Skew T–logp temperature profile (solid) for the average DYNAMO conditions from the R/V Revelle along with a histogram of (a) observed temperature and (b) observed dewpoint temperature. The dashed line signifies the dewpoint temperature used in the model initial conditions.
Citation: Monthly Weather Review 143, 7; 10.1175/MWR-D-14-00247.1
In this study we employ a cloud-resolving large-eddy simulation (LES) model to examine how convection responds to external forcing from prescribed domain-scale moisture convergence and different background wind conditions. We test if convective heating is a linear function of increased moisture convergence and/or increased mean surface winds, and if convectively forced processes, such as the formation of cold pools or increased cirrus anvil coverage, generate significant feedbacks to convection through changes in surface fluxes. The main objective of this work is to determine the processes that change the column heat budget and to better understand how organized convective systems affect the environment leading up to the active phase of the MJO.
The paper begins with a description of the coupled large-eddy simulation model, its initial conditions, and its boundary conditions in section 2. In section 3, results from a basic case simulation are presented with an analysis of cold pool formation. Section 4 compares the heat and moisture budgets from the suite of experiments. The role of ocean surface properties is presented in section 5. Section 6 summarizes the conclusions.
2. Model description and experimental design
A basic set of experiments with three prescribed large-scale moisture convergence values are conducted. The first case represents the suppressed phase with low precipitation, and the second and third cases represent periods during the active phase of the MJO with moderate and strong precipitation rates, respectively. Prescribed winds in these cases are relatively light (<10 m s−1) with weak vertical shear. An additional experiment is conducted with higher surface winds and suppressed MJO phase moisture convergence to examine how increased surface fluxes from stronger winds affect convective activity versus externally forced moisture convergence.
Simulations are conducted using a version of the Skyllingstad and Edson (2009) LES model that includes parameterizations for the radiative transfer of infrared and solar radiation (Mlawer et al. 1997) along with a seven-component cloud microphysics scheme (Thompson et al. 2008). Model equations are based on Deardorff (1980) with a subgrid closure based on the filtered structure function method from Ducros et al. (1996). The momentum equations are expressed in flux form and solved using the third-order Adams–Bashforth–Moulton scheme. Scalar quantities are integrated with a conservative Van–Leer method based on Colella (1990). Pressure is solved using a split time compressible scheme based on Wicker (2009) with an implicit method in the vertical to allow for higher resolution near the model surface.
The LES model is coupled to an ocean model based on the K-profile parameterization (KPP) boundary layer scheme from Large et al. (1994) with momentum nonlocal terms defined as in Smyth et al. (2002). Each ocean grid point is treated as an independent column that interacts with the LES model grid point. Ocean columns do not exchange properties in the horizontal direction. As shown later in our analysis, lateral gradients of SST and salinity generated by surface flux variability are typically quite small. Coupling between the LES and ocean is accomplished using the COARE version 2.5 bulk flux algorithm described in Fairall et al. (1996) as implemented in Vickers and Mahrt (2006). For efficiency and because the ocean responds much slower than the atmosphere, KPP is evaluated once every 20 time steps (roughly 40–50 s) using instantaneous atmospheric fields.
Simulations are conducted over a horizontally periodic domain having dimensions of 537.6 km × 537.6 km and grid spacing of 300 m. Experiments are initialized by interpolating a coarse-resolution domain spinup integration of 4 days with grid spacing of 600 m, which reaches a quasi-equilibrium population of active convective clusters. Results are analyzed over the 3-day period from days 6–8 (hours 144–216).
The simulations use a stretched vertical grid with the lowest grid cell spacing of 10 m increasing to 150 m at approximately 600-m height. Constant 150-m spacing is then applied up to roughly 11 250 m, followed by a gradual increase in the grid spacing to 1000 m at the model top near 24 km. Vertical boundaries are rigid with Rayleigh velocity damping gradually applied in the upper ⅕ (5 km) of the domain for absorbing vertical propagating internal waves based on Durran and Klemp (1983).
Initial conditions for temperature and humidity are set to a mean, idealized profile from the Research Vessel (R/V) Revelle radiosondes as shown in Fig. 1. Free troposphere humidity varies considerably between convective and suppressed conditions in DYNAMO. In our simulations, specific humidity is reduced above 3000 m representing conditions during the suppressed stage of the MJO. A total of four scenarios are simulated by varying large-scale moisture convergence in the marine boundary layer to represent different phases within the MJO cycle as shown in Table 1. Surface wind speed and shear are varied as shown in Fig. 2. Initial velocity profiles of cases M100, M200, and M450 represent weak shear and low wind conditions near the beginning of the November MJO event with imposed boundary layer moisture convergence of 100, 200, and 450 W m−2, respectively. In the fourth case (M100W), winds are increased by 6 m s−1 throughout the lower atmosphere with 100 W m−2 moisture convergence (Fig. 2). The initial wind profile is used to define a weak background pressure gradient by assuming the flow is in geostrophic balance for a latitude of 2°N.
Model simulation parameters for conditions representing different phases of MJO development. Transition refers to conditions leading up to the active phase. Shear parameters are defined in text.
Initial wind profiles for cases M100, M200, and M450 (solid) and case M100W (dashed).
Citation: Monthly Weather Review 143, 7; 10.1175/MWR-D-14-00247.1
In addition to the standard model scalars (temperature and moisture variables), we also include a variable representing the origination height of air parcels. The value of the parcel height is reset every 8 h to the local gridpoint height. Advection and turbulent mixing are used to transport the parcel height as with other model scalars, such as the potential temperature and moisture parameters. As is shown below, the parcel height parameter provides a unique way to examine the process of vertical mixing generated by moist tropical convection, and the important role of downdrafts in forming cold pools and broad regions of midtropospheric vertical exchange.
Simulating the tropical atmosphere with a limited-domain model and prescribed external forcing typically produces an atmospheric vertical temperature structure that differs from the observed tropics. For example, simulations with both single-column and cloud-resolved models forced using observed heating and moistening rates from TOGA COARE produced reasonably accurate precipitation rates, but with temperature errors of ±2°–3°C distributed throughout troposphere (Woolnough et al. 2010). Most of these errors are attributed to missing large-scale circulation feedbacks that would act to disperse temperature anomalies through changes in vertical motion.
Methods for including these circulations typically prescribe a vertical velocity that is adjusted to enforce a known temperature profile (e.g., the weak temperature gradient approximation; Sobel and Bretherton 2000; Raymond and Zeng 2005). Another method for including large-scale forcing involves coupling convection to oscillatory low-mode gravity wave systems (Kuang 2008), yielding somewhat improved profiles of mean vertical velocity in comparison with the weak temperature gradient method (Wang et al. 2013). Such methods to include the effect of large-scale dynamical forcing on a limited-area domain are somewhat ad hoc. Though they succeed at preventing unrealistic drift of the average temperature profile, their effect on convective processes is not clear.
Rather than applying a prescribed temperature adjustment via vertical motion as in these long-term studies, we impose an idealized, fixed large-scale moisture flux and focus on the short-term processes that change the radiative balance and temperature structure during the initial convective adjustment. Simulations are short enough that we do not expect or require that they reach full equilibrium, and they do not drift too far from a realistic temperature profiles except when forced with very high moisture convergence. We consider idealized cases with wind shear similar to tropical conditions encountered during the DYNAMO experiment. A mixed-layer ocean model is used to simulate possible feedbacks from variable sea surface temperature. Simulations are limited to ~10 days, which is sufficient for establishing a quasi RCE with only small mean temperature and moisture trends, and allows for much higher resolution and model fidelity than used in previous cloud-resolving model (CRM) simulations.
3. Convective structure
a. Basic convective conditions
We begin our analysis with an overview of the basic convective structures that are present to some degree in all of the experiments. Plots of cloud cover and surface properties for case M100 are shown in Fig. 3. Convection typically consists of 5–10 isolated 10–20-km convective cells. At times they form clusters greater than 100 km across as shown by the centrally located cloud mass in Fig. 3. Simulated convective cells and clusters are most evident in the surface temperature and flux fields, which indicate where cold pools have expanded from convective downdrafts. Cold pool systems are characterized by air masses with cooler temperatures, higher surface winds, and corresponding increased surface latent heat flux. Initially, cold pools are drier than the surrounding boundary layer because they originate from higher, colder air that has much lower specific humidity. This is shown for example by the cluster of cold pools near the upper center of the domain at y = 400, x = 250 km where the lowest temperatures coincide with dry regions in the plot of specific humidity (Fig. 3c). Air in these cold pools originates from heights of 2000–3000 m as indicated by the parcel height (Fig. 3f).
Case M100 (a) cloud albedo, along with surface (b) temperature (K), (c) specific humidity (g kg−1), (d) surface wind speed (m s−1), (e) latent heat flux (W m−2), and (f) parcel height from day 8, hour 3 (tracer initialized 8 h earlier).
Citation: Monthly Weather Review 143, 7; 10.1175/MWR-D-14-00247.1
Cold pools are generated by negatively buoyant downdrafts that develop from evaporative cooling as rain and snow fall through unsaturated air beneath active convection. Eventually the cool air mass sinks to the surface, where increased surface pressure gradients from both dynamic and hydrostatic forcing generate strong outward-directed flows. Large cold pool regions are typically formed through the combination of multiple smaller cells, which then merge together as indicated by the patches of low specific humidity in Fig. 3 near x = 250, y = 400 km. As the cold pool expands, strong surface latent and sensible heat flux act to increase the specific humidity and gradually warm the cold pool. The highest fluxes are collocated with regions of strong winds that are produced along the leading edge of the cold pool. Fluxes over 300 W m−2 are produced in response to wind gusts up to 20 m s−1 in the developing system in Fig. 3.
Older cold pool regions persist for many hours as their temperature recovers, for example as seen in the slightly lower temperature and specific humidity in the old cold pool region evident near x = 350, y = 300 km. Because cold pools thin as they expand, the increasing surface moisture is confined near the surface and generates a ring of higher specific humidity around older cold pools (Fig. 3c). During the recovery phase of the cold pool, there are often small fronts in the moisture and temperature that are less obvious in the velocity field. For example, the surface temperature and specific humidity in Fig. 3 exhibits a sharp discontinuity that extends from about x = 300, y = 400 km southeastward to x = 500, y = 300 km, with almost no change in surface wind speed. Cold pool strength and life cycle displayed in Fig. 3 are consistent with previous LES experiments reported in Moeng et al. (2009) and Khairoutdinov et al. (2009).
b. Cold pool structure
The evolution of a cold pool cluster is presented in greater detail in Figs. 4–6. In this example, new convective cells are developing along a remnant cold pool boundary that is barely visible in the temperature and specific humidity at hour 15 on day 7. As convective cells evolve, a series of cold pools form and expand away from the initial convective core. The cold pool boundary produced by these cells develops well-defined arc cloud lines (Fig. 4c) in response to increased convergence along the expanding front. Arc clouds are often observed in satellite imagery of tropical convection and occasionally in radar. Arc clouds visible in Fig. 4c move in a direction opposite the mean winds and do not lead to more large-scale convection, but dissipate. Cell propagation over eastward-expanding cold pools is consistent with propagation of squall lines that reduce the mean environmental shear (Rotunno et al. 1988).
Surface plots of (a) potential temperature, (b) specific humidity, and (c) cloud albedo from case M100 over a subregion of the full domain on day 7. The dashed black line indicates the location of cross sections and fluxes presented in Figs. 5 and 6.
Citation: Monthly Weather Review 143, 7; 10.1175/MWR-D-14-00247.1
Cross-sectional plots presented in Fig. 5 show how evaporative cooling by precipitation generates a locally stratified air mass that has much lower specific humidity than the surroundings into which it descends. The cold pools have strong outflow winds up to about 300 m, with maximum velocities concentrated at the surface. Air parcels are drawn into the downdraft structure from heights greater than 3000 m, as indicated by the parcel height plot in Fig. 5d. The model suggests that much of the cold pool air mass originates from well above the boundary layer. The parcel height also shows how the expanding cold pool displaces air upward along the leading edge with transport of near-surface air up to near 300 m. This uplift may also promote new cloud development by lifting slightly higher specific humidity air from the surface (Feng et al. 2015). Uplift at cold pool outflow boundaries generates new convection in trade wind cumulus regimes (Li et al. 2014).
Vertical cross sections along x = 350 km of (a) potential temperature (K), (b) specific humidity (g kg−1), (c) meridional velocity (m s−1), and (d) parcel height parameter (m) taken from case M100 at hour 16 on day 7. Cross sections correspond to the black dashed line indicated in Fig. 4. The arc cloud is associated with the front at y = 190 km.
Citation: Monthly Weather Review 143, 7; 10.1175/MWR-D-14-00247.1
Surface fluxes are strongly influenced by the properties in the cold pool as demonstrated by a plot of the instantaneous fluxes in Fig. 6. Increased surface winds and reduced humidity generates a latent heat flux that peaks near 400 W m−2 at the nose of the cold pool front moving southward at y = 190 km and exceeds 200 W m−2 over 10 km traversing the cold pool. Wind stress and sensible heat flux are also much higher in the cold region. Precipitation rates of ~60 mm h−1 for this case are consistent with the brief, intense rain events commonly observed during the DYNAMO cruise (Moum et al. 2014). A more quantitative analysis of cold pool effects on surface wind and fluxes is presented below in section 4d.
Surface (a) rainfall over previous 10-min period, (b) friction velocity, (c) sensible heat flux, and (d) latent heat flux from day 7, hour 16. Cross sections correspond to the north–south line indicated in Fig. 4 at x = 350 km.
Citation: Monthly Weather Review 143, 7; 10.1175/MWR-D-14-00247.1
4. Sensitivity to moisture convergence
The convective population experiences and responds to the varying large-scale moisture convergence in different phases of the MJO. Previous CRM model–observation comparisons have shown the importance of moisture convergence in setting the strength of convection and average precipitation rates. Predicted rainfall amounts are consistent with measurements even though average model temperature and moisture content can drift significantly from observed profiles, for example as simulated by Woolnough et al. (2010) and Wang et al. (2013). Here we examine how convection responds to three different amounts of large-scale moisture convergence (cases M100, M200, and M450) that are representative of the transition from the suppressed stage to the active stage of the MJO.
a. Surface wind speed response
In each of the cases, the strength of moisture convergence changes convection frequency, strength, and overall behavior. We plot the surface zonal wind component u, using a time–longitude Hovmöller diagram along a line of latitude (Fig. 7), thereby representing both the eastward propagation of convective systems and their relative strength. This representation shows the acceleration of the surface wind by cold pools, which typically appears as a couplet of positive and negative velocity as a cold pool intersects the meridional location plotted. Zonal wind maxima indicate the leading edge of cold pool systems from recently active convective cells. In case M100, cold pools are generated by large, slow-moving systems that last ~6 h and occasionally generate a strong eastward wind gust, for example, on day 7 near x = 480 km (Fig. 7a). The number of cold pool systems is limited, with relatively few interactions between cold pool edges leading to new convective cells. Cold pool edge propagation in the zonal direction is sometimes ~5 m s−1, but does not show a well-organized structure over time.
Hovmöller diagrams of zonal velocity and specific humidity at y = 268.8 km for case (a),(b) M100; (c),(d) M200; and (e),(f) M450.
Citation: Monthly Weather Review 143, 7; 10.1175/MWR-D-14-00247.1
Case M200 (Figs. 7c,d) produces more convective cells that tend to propagate eastward as indicated by long-lived eastward gust fronts. Slow-moving, westward-propagating gust fronts are indicated by regions of negative velocity. Gust fronts are typically active for about 100–150 km with propagation speeds between 5 and 8 m s−1. Intersections between cold pools from different convective cells occasionally generate new convection, for example, as shown by the gust front intersection just after day 7 near x = 380 km. Animations of surface temperature and specific humidity suggest that many new cells are generated along old cold pool boundaries that have well-defined temperature and humidity gradients, but do not have significant wind variations.
Cell propagation in case M450 has stronger outflows both westward and eastward. Phase speeds again range between 5 and 8 m s−1, depending on the particular cell. Systems in this case are more vigorous, long lived, and tend to traverse the periodic domain multiple times. New cells form at the intersections of eastward- and westward-moving cold pools that are slightly displaced in time from the original cold pool boundaries, for example, on day 7 at x = 240 km, breaking the regularity of their propagation. The mean surface wind speed in cases M450 and M200 is about 3.5 m s−1, whereas in case M100 it is about 1.5 m s−1, indicating that stronger convection with increased moisture convergence generates more downward vertical momentum transport from the upper-level winds.
The importance of convergence generated by cold pools (dynamic forcing) versus concentration of moist static energy (thermodynamic forcing) has been investigated using both field data and modeling (Tompkins 2001; Zuidema et al. 2012; Li et al. 2014). Tompkins (2001) used an LES model for smaller-scale convection to show that increased near-surface moist static energy in cold pool outflows is responsible for most new cell development. Zuidema et al. (2012) did not find conclusive evidence for this hypothesis in precipitating trade cumulus systems. Li et al. (2014) used high-resolution simulations to show that lifting by cold pools was the most important mechanism for convective initiation. Simulations presented here suggest that both lifting and moisture enhancement are important with new cells concentrated primarily along old cold pool margins. In some cases, especially in the M450 case, the intersection of cold pool boundaries can lead to a new convective cell. Specific humidity concentrations are often higher preceding the edge of the cold pools, with the lowest values located in the core of new cold pools. High specific humidity alone does not predict new cell development, but seems to be a part of cell propagation.
b. Vertical temperature profiles
Increasing moisture convergence also has a significant effect on the atmospheric structure, demonstrated by the average vertical temperature and dewpoint temperature at the end of the simulation (Fig. 8). Our simulations are initialized with representative soundings, after which the sounding evolves due to the combined effects of the internal physics of the model. In all of the simulations, the initial 1–2 days (during the low-resolution spinup) are characterized by average cooling of the atmosphere from a net radiative flux to space. Convection typically requires 3–4 days to develop and balance this heat loss. After 9 days the moisture convergence has affected the thermal structure with larger moisture flux generating a warmer profile through greater latent heat release. Observed and modeled temperature profiles all essentially follow moist adiabats in the upper troposphere, but display steeper (less stable) lapse rates below the melting layer at ~600 hPa. The model overestimates the instability in this layer when compared to observations. To a smaller degree, observations also show an increased lapse rate at this level as shown, for example, in Mapes (2001).
Horizontally averaged (a) temperature and (b) dewpoint temperature from the initial condition (black dashed) and at hour 216 from Case M100 (blue), M200 (green), and M450 (red). Also shown are the histograms of DYNAMO radiosonde measurements from Fig. 1.
Citation: Monthly Weather Review 143, 7; 10.1175/MWR-D-14-00247.1
In all cases the stratosphere warms, lowering the height of the tropopause. This is most likely because shortwave radiative heating is too strong because the default RRTM ozone concentration is higher than that found in the tropical lower stratosphere. Greater convective mixing deepens the tropopause for cases with stronger moisture flux convergence. Case M100 shows that moisture convergence of 100 W m−2 is insufficient to balance the longwave radiative cooling, leading to a new equilibrium profile with tropospheric temperatures about 4° cooler than the original temperature profile. This case represents conditions during the suppressed phase of the MJO. At this time subsidence of the large-scale circulation warms the troposphere, to offset the effect of longwave radiative cooling. We also note that the simulated dewpoint in the upper troposphere is systematically higher than the observations. This discrepancy amounts to only ~0.002 g kg−1 because saturation-specific humidity at these low temperatures is very small.
Increasing the prescribed moisture convergence to 200 W m−2 (case M200) produces a temperature sounding almost identical to the initial state between ~600 and ~150 hPa indicating that for this case increased latent heat flux in the upper troposphere is appropriate for balancing longwave radiative cooling. Case M450 has a further increase in temperature and moisture where the entire troposphere temperature profiles exceeds the observed conditions by more than 5°C. Temperatures in M200 below 600 hPa are noticeably warmer and indicate higher moisture than the baseline DYNAMO averages, but are for the most part within the spread of observed radiosonde data. Cases M200 and M450 represent the active phase of the MJO and would be expected to have an average positive vertical velocity component and associated adiabatic cooling to compensate for the large-scale stratiform heating profile. Mean upward motion, as diagnosed from TOGA COARE and DYNAMO during active convective periods, would tend to both cool the sounding directly through adiabatic ascent and promote enhanced convection and vertical transport of low-level moisture.
The similarity between the M200 soundings and the observed profiles is reassuring given that the total moisture source (evaporation plus moisture convergence) in this case represents a rainfall rate consistent with the average DYNAMO rainfall (~9 mm day−1) (Johnson and Ciesielski 2013). This probably represents the phase of the MJO when large-scale moisture flux is increasing, but large-scale vertical motion is small and not affecting the temperature. If we view the MJO as a single cycle of a mode-1 baroclinic wave system, for example, as considered in Kuang (2008), then the 200 W m−2 case represents a nodal point where the wave-forced vertical velocity is near zero and the average temperature profile is controlled mostly by local convective processes.
c. Average heat and moisture budget


Heating rates averaged horizontally and over hours 144–216 for case M100 (solid), M200 (dashed), and M450 (dash–dot).
Citation: Monthly Weather Review 143, 7; 10.1175/MWR-D-14-00247.1
The overall heat budget is dominated by net radiative cooling and latent heat release. Above 10 km, vertical transport acts to move heat released in the lower troposphere to the tropopause at 12–14-km height. Heat is also moved downward from midlevels into the boundary layer below 1 km. Vertical eddy transport in the upper troposphere is thought to have a key role in the formation of stratiform anvil regions that define the mesoscale cloud mass in tropical clusters (Houze 2004; Mapes 2001; Houze 1997). Midlevels have very small heating from vertical transport suggesting that active updrafts are compensated by subsidence between convective cells.
As M increases, LW radiative divergence partially offsets greater LH near 7–10- and 13–15-km height, indicating a slight negative feedback that reduces warming of the profile. The response of the LW heating to increased LH is demonstrated more clearly by calculating the depth-integrated heating rates as shown in Table 2. As LH increases, cooling from LW net divergence also increases, but not enough to offset the added heating and consequently the profile warms slightly (Fig. 8). Increased total LW cooling is likely due to higher atmospheric temperature and humidity, which cause a greater emission of infrared radiation via Planck’s law. Enhanced low clouds could also promote greater longwave heat loss; however, we did not perform a detailed analysis of the cloud effects on the radiative budget that would be necessary to verify this effect.
Depth- and time-averaged heating rates for latent heat (LH), shortwave solar radiation (SW), and longwave infrared radiation (LW). Rates are in °C day−1.
The structure of the latent heating profile suggests various processes; for example, minimum values just below the freezing level at 5 km indicate the melting of frozen precipitation and maximum values at the boundary layer top near 1 km indicate the formation of trade wind cumulus. Cooling below the lifted condensation level is produced by the evaporation of rain, augmenting the formation of cold pools. Splitting the convective transport term between upward and downward motions (not shown) indicates that boundary layer cooling by evaporation of hydrometeors and adiabatic ascent of low θ boundary layer air under developing convective systems is balanced by adiabatic warming of descending air parcels.
The latent heating profile in each case has three distinct maxima that can be interpreted as the trimodal tropical cloud structure described by Johnson et al. (1999). The low-level peak represents boundary layer clouds, the midlevel maximum represents warm cumulus congestus, and the broad upper-level maximum signifies deep convection. Key to this structure is the maximum in net latent heating around 4–5-km height, which may also be responsible for the reduced lapse rate just below the modeled freezing level. Lapse rate anomalies at these heights are a distinct feature of the tropical atmosphere and are thought to result from melting of precipitation (Mapes 2001; Folkins 2013) and the large-scale circulation generated by the vertical latent heat distribution associated with synoptic-scale convective organization (Posselt et al. 2008). Reduced net latent heating in our simulation below the freezing level results primarily from melting of snow and graupel as demonstrated by the negative latent heating term from freezing and melting, in general agreement with cloud resolving modeling experiments by Yasunaga et al. (2008) on the formation of clouds in the melting layer.
As a test, we conducted a short, highly idealized experiment (using a coarse-resolution simulation) where terms responsible for heat loss from melting and sublimation of ice-phase particles were removed from the model. The resulting latent heating profile from this experiment (Fig. 10) shows the contribution of both melting and sublimation removing heat near 5 km. Without sublimation the simulation generates and maintains a much thicker upper-level cloud structure (not shown). This experiment suggests that accurate estimation of sublimation rates is crucial for a realistic representation of the effect of cirrus anvils and stratiform precipitation on the tropical convective heating rate and circulation.
Net latent heating rates for case M200 (coarse resolution) averaged horizontally and over the first 12 h of day 6 for experiments with heat flux removed for melting (small dash) and both melting and sublimation (large dash), along with the control experiment (solid).
Citation: Monthly Weather Review 143, 7; 10.1175/MWR-D-14-00247.1

Comparison of column-integrated moisture budget terms integrated over the model domain expressed as rainfall rate. Terms are water vapor storage Q, surface precipitation −P, large-scale imposed moisture flux M, and evaporation flux E. Data are from days 6 to 8.
Citation: Monthly Weather Review 143, 7; 10.1175/MWR-D-14-00247.1
Overall, the local evaporation flux is equivalent to 1–2 mm day−1 precipitable water in cases M100 and M200, and about 3 mm day−1 for case M100W, signifying the modest increase in latent heat flux from increased surface winds. For the M450 case, increased near-surface humidity limits the latent heat flux by lowering the difference in the air–sea humidity values (Δq). We note that the net storage of water vapor increases, consistent with the nonsteady simulation.
d. Surface flux
Surface evaporation increases 40% in the active phase of the MJO (de Szoeke et al. 2015), and previous modeling and observational studies (Wu and Guimond 2006; Jabouille et al. 1996) show that convection enhances surface fluxes through the action of increased winds from cold pool systems and squall lines. One might expect then that increasing the strength of convection through moisture convergence would result in stronger surface fluxes. However, except for the increased mean wind case (M100W), changes in average specific humidity and temperature actually decrease latent and sensible fluxes as moisture convergence increases (Fig. 12). Evaporation is strongly affected by the mean wind speed. The stronger average wind speed in case M100W increases latent heat flux by about 40%.
Latent and sensible heat flux (W m−2) for each case.
Citation: Monthly Weather Review 143, 7; 10.1175/MWR-D-14-00247.1
Similar fluxes in case M100 and M200 and the reduction in flux in case M450 imply that cloud-scale wind variations from increased convective activity do not have a significant effect on the average fluxes. Average simulated fluxes, however, may also be controlled by the changing average atmospheric structure forced by variations in M. For example, in the M450 case increased humidity is clearly leading to reduced latent heat flux as simulated in our limited-area model. Specific humidity was remarkably constant over the DYNAMO experiment (de Szoeke and Edson 2015), and the high humidity and low fluxes in case M450 are not representative conditions observed during the active phase of the MJO.
e. Cold pool fluxes
In an attempt to separate out the influence of cold pools on surface wind speed variability, we composited the average surface wind according to the surface air temperature distribution. We defined cold pool regions by considering the lowest tenth of the temperature distribution. Averages of the wind stress and latent heat flux were calculated for the cold pools and the area outside the cold pool regions. Results from this analysis are shown in Fig. 13 for each of the simulated cases with cold pool regions labeled “cp” and regions outside of cold pools labeled “o.” For each case the cold pool regions exhibit higher wind stress than the rest of the domain. For stronger moisture convergence, the wind stress is larger in response to more intense convective systems. Wind stress is also increased in case M100W in response to the overall increased surface wind speed. Case M450 generates organized, propagating convective lines that persist for many hours causing a substantial increase in cold pool wind stress, but they have little effect on the background wind stress value. Except for case M100W, for areas outside of cold pools the simulations show similar wind stress values indicating that nonconvective areas are not as strongly affected by the background moisture convergence.
Surface wind statistics for areas assigned as cold pools (cp) and background (o). Cold pool regions are defined as grid points having temperatures in the lowest 10% of the total distribution.
Citation: Monthly Weather Review 143, 7; 10.1175/MWR-D-14-00247.1
Performing a similar analysis on the latent heat flux also shows differences between the cold pool versus noncold pool regions with fluxes larger in the cold pool regions (Fig. 13b). Although the wind affects the magnitude of the latent heat flux, changes in the average water vapor content can have a stronger effect, reducing latent heat flux for stronger moisture convergence and background humidity. Background fluxes are relatively consistent between M100 and M200; however, in case M450 fluxes are relatively small outside of cold pools where the boundary layer is near saturation relative to the SST because of unrealistically high moisture accumulation.
5. Sea surface properties
On scales greater than 1000 km, variations in sea surface temperature (SST) have a significant effect on convection in the tropics by producing hydrostatic pressure gradients and driving large-scale circulations (e.g., Back and Bretherton 2009). Less is known about the impact of SST variations on convective activity at smaller scales. Li and Carbone (2012) found a systematic relationship between convection and SST variability over scales of 400–1000 km in the western Pacific warm pool indicating possible air–sea feedbacks. In the current simulations, rainfall and variations in cloud cover can generate variations of ~0.5°C in SST as the combined effects of surface buoyancy from reduced salinity and reduction in solar forcing from cloud shadows affect the surface energy budget. This is shown in Fig. 14 from case M200 corresponding to Fig. 3 where rain events reduce surface buoyancy and effectively prevent vertical mixing in shallow surface layers that cool more rapidly than surrounding waters during the night. A similar effect is active during the day (not shown) when salinity stratification limits convective mixing in freshwater pools leading to hot spots in the SST field. Also evident are larger regions of SST variability that are generated by cloud shadows that limit daytime solar heating of the surface.
SST and sea surface salinity from case M100, along with SST at night (day 8, hour 3).
Citation: Monthly Weather Review 143, 7; 10.1175/MWR-D-14-00247.1
One objective of this study was to determine if variations in SST from rainfall and cloud shading are significant enough to control the formation and evolution of convection. Studies of mesoscale SST variability have shown that stationary SST gradients correlate with local wind fields when averaged over periods of weeks to months (e.g., Chelton et al. 2004; Small et al. 2008). In the current study, surface fluxes are often strongly affected by convective activity implying that flux variations from SST differences of O(0.1°C) would be overwhelmed by convective wind-forced flux variability. To test this hypothesis, we conducted an experiment where coupling between the ocean and atmosphere was performed using the mean surface winds over the entire domain, rather than applying the local wind at each grid point. Consequently, variations in surface flux for this case were entirely controlled by the SST. Overall, the latent heat flux varied by about 5–10 W m−2 when using the average winds, which is much less than the typical latent heat flux variations of over 100 W m−2 when using gridpoint wind data (e.g., Fig. 3). We also conducted an experiment where all fluxes were calculated with the average SST rather than the point values. Neither convective features, such as cell propagation as shown in Fig. 7, nor average vertical profiles were strongly affected by the uniform SST forcing.
6. Summary and discussion
Simulations are conducted with a cloud-resolving LES model for a range of imposed moisture convergence representing conditions in the tropical equatorial Indian Ocean. Surface conditions generated by the model are constantly changing with numerous small fronts in momentum, temperature, and moisture that are closely tied to convectively generated cold pools. This variability generates complex spatial patterns in surface fluxes and often affects where new convection forms. Nevertheless, the most important factor determining the overall strength of convection is the prescribed moisture convergence generated by implied larger-scale circulations representing MJO conditions in suppressed and active phases. Simulations for three different moisture convergence scenarios yield precipitation amounts that are nearly equal to the prescribed excess moisture (evaporation + moisture convergence), indicating that increased convection may change local surface fluxes for individual convective systems but does not significantly change the overall water budget.
The simulations show that changes in radiative and surface fluxes due to differing ensembles of convection are unable to maintain the constant average temperature structure observed in the tropics for different large-scale moisture flux convergence. A stronger moisture flux increases the tropical sounding temperature. The closest match between model results and observations are attained using a combined large-scale and local-scale flux of about 250 W m−2 (case M200). An equivalent rainfall rate balancing this flux is
While case M200 establishes a nearly constant equilibrium state similar to observations, case M100 exhibits a very gradual decline in temperature and specific humidity. In contrast, case M450 is very far from an equilibrium state and has both warming and moistening that would likely lead to an unrealistic sounding if the simulation were to continue. Case M450 represents moisture convergence during the active phase of the MJO when the large-scale circulation is characterized by an average upward vertical velocity that would act to adiabatically cool the lower atmosphere. Consequently, excess heating and moistening could be rectified by adding a domain-scale upward vertical velocity, as is done when applying the weak-temperature gradient approximation (e.g., Raymond and Zeng 2005) or wave-induced motions (Wang et al. 2013), which would enhance precipitation and cool the entire domain via adiabatic expansion. The lack of these external effects limits the applicability of the results as the simulations drift away from the initial sounding. This is especially true for case M450 where the added moisture strongly affects surface fluxes. However, these methods of imposing large-scale forcing all make significant assumptions about the processes that act to maintain the nearly constant vertical temperature profile observed in the tropics. Other than moisture flux we did not impose large- scale forcing, because to do so would arbitrarily affect simulated convection in ways that are not easy to diagnose.
One goal of this study was to quantify how the convective ensemble affects local surface winds and fluxes with the idea of eventually representing these systems more accurately in bulk flux parameterizations. Many current bulk formulas (e.g., Fairall et al. 1996) include a “gustiness” parameter that accounts for wind speed enhancement of surface fluxes by turbulent eddies in the atmospheric boundary layer. For climate models having grid spacing of ~10–50 km or larger, an additional gustiness parameter could be included for adding the effects of convection on wind speed.
Cold pools due to evaporating hydrometeors in deep precipitating cumulus convection enhance surface heat fluxes moderately, chiefly by increasing the scalar wind speed at the surface in gusts at the front of the cold pools. Relatively cold dry air impinges on the surface, but air in the shallow spreading cold pools quickly becomes density stratified, reducing evaporation by reducing transfer of surface humidity across the surface boundary layer. Our analysis indicates that wind stress associated with cold pools gradually increases across a range of convective activity as measured by total precipitation. Wind stress in the undisturbed area not covered by cold pools was considerably lower in all of the cases, suggesting that outside of cold pools convection has a smaller effect on boundary layer wind variability. For background boundary layer wind speeds of 1 m s−1 latent heat flux increased about 20%–30% in cold pools. Simulation results indicate that adding the effects of cold pool flux enhancement to standard bulk flux algorithms could lead to heat and momentum flux changes of about 5%–10% in the area average.
Observations from the DYNAMO experiment reported in de Szoeke and Edson (2015) support the simulated increase of cold pool fluxes over the background values. Average DYNAMO all-cruise observed wind speed was 27% greater in cold pools resulting in a ~30 W m−2 increase in latent heat flux. Simulated wind speed enhancement presented here was typically much larger than these measurements because of our weak background wind speed; however, the simulated latent heat flux increase was also 20–40 W m−2. In general, observations suggest that enhanced average wind speed is responsible for much of the variability in latent heat flux during MJO events (as shown here by case M100W), with sea–air specific humidity differences having a lesser effect.
Interaction between convection and the upper ocean through rainfall effects on salinity and cloud shading was found to have a very small effect on the surface fluxes. Spatial variations of surface fluxes linked to SST patterns were almost always less than 10 W m−2 and usually did not persist for more than one day in the model. This variability is two orders of magnitude less than flux variations from wind gusts associated with cold pools.
The MJO and other synoptic-scale dynamical variability modulate convection. Tropical convection representative of that observed in the Indian and Pacific warm pools, even in the quiescent phase of the MJO, requires significant large-scale moisture convergence. With commonly available computing resources, simulating the (<100 m) scales of relevant processes in cumulus convection (e.g., updrafts, turbulence, and cloud macrophysics) precludes a domain large enough (>106 m) to simulate the wide areas of compensating subsidence and surface evaporation, whose lateral moisture convergence is required for the deep convecting regions and for the large-scale heat and moisture balance of the general circulation. Without external forcing, the model temperature structure in our limited-area LES gradually cools from the initial sounding by longwave radiative heat loss, approaching a new equilibrium temperature profile that depends on the net heat fluxes over the domain. These fluxes are insufficient to maintain the relatively warm temperature observed during DYNAMO. Larger-scale circulations not contained within our domain—such as the Walker and Hadley circulations, atmospheric equatorial waves, and the MJO—act to increase the net surface flux of moist static energy, thereby maintaining the temperature structure observed in DYNAMO. Adding a domain-averaged net moisture flux makes up for the energy radiated away and maintains the upper-tropospheric temperature.
Acknowledgments
This research was funded by the National Science Foundation Grant OCE-1129419, the Office of Naval Research Grant N00014-10-1-0299, and the National Oceanic and Atmospheric Administration Grants NA11OAR4310076 and NA13OAR4310157. We would like to acknowledge high-performance computing support from Yellowstone (ark:/85065/d7wd3xhc) provided by NCAR’s Computational and Information Systems Laboratory, sponsored by the National Science Foundation, and from the Department of Defense High Performance Computer Modernization Program. Comments from the two reviewers were very helpful in focusing the discussion and improving the overall paper.
REFERENCES
Back, L. E., and C. S. Bretherton, 2009: On the relationship between SST gradients, boundary layer winds, and convergence over the tropical oceans. J. Climate, 22, 4182–4196, doi:10.1175/2009JCLI2392.1.
Brown, P. J., and C. D. Kummerow, 2014: An assessment of atmospheric water budget components over tropical oceans. J. Climate, 27, 2054–2071, doi:10.1175/JCLI-D-13-00385.1.
Camargo, S. J., M. C. Wheeler, and A. H. Sobel, 2009: Diagnosis of the MJO modulation of tropical cyclogensis using an empirical index. J. Atmos. Sci., 66, 3061–3074, doi:10.1175/2009JAS3101.1.
Chelton, D. B., M. G. Schlax, M. H. Freilich, and R. F. Milliff, 2004: Satellite measurements reveal persistent small-scale features in ocean winds. Science, 303, 978–983, doi:10.1126/science.1091901.
Colella, P., 1990: Multidimensional upwind methods for hyperbolic conservation laws. J. Comput. Phys., 87, 171–200, doi:10.1016/0021-9991(90)90233-Q.
Deardorff, J. W., 1980: Stratocumulus-capped mixed layers derived from a three-dimensional model. Bound.-Layer Meteor., 18, 495–527, doi:10.1007/BF00119502.
de Szoeke, S. P., and J. B. Edson, 2015: Intraseasonal air-sea interaction and convection observed during DYNAMO/CINDY/AMIE and TOGA COARE. The Global Monsoon System: Research and Forecast, C. P. Chang, Ed., World Scientific, in press.
de Szoeke, S. P., J. B. Edson, J. R. Marion, C. W. Fairall, and L. Bariteau, 2015: The MJO and air-sea interaction in TOGA COARE and DYNAMO. J. Climate, 28, 597–622, doi:10.1175/JCLI-D-14-00477.1.
Ducros, F., P. Comte, and M. Lesieur, 1996: Large-eddy simulation of transition to turbulence in a boundary layer developing spatially over a flat plate. J. Fluid Mech., 326, 1–36, doi:10.1017/S0022112096008221.
Durran, D. R., and J. B. Klemp, 1983: A compressible model for the simulation of moist mountain waves. Mon. Wea. Rev., 111, 2341–2361, doi:10.1175/1520-0493(1983)111<2341:ACMFTS>2.0.CO;2.
Fairall, C. W., E. F. Bradley, D. P. Rogers, J. B. Edson, and G. S. Young, 1996: Bulk parameterization of air–sea fluxes for Tropical Ocean–Global Atmosphere Coupled Ocean Atmosphere Response Experiment. J. Geophys. Res., 101, 3747–3764, doi:10.1029/95JC03205.
Feng, Z., S. Hagos, A. K. Rowe, C. D. Burleyson, M. N. Martini, and S. P. de Szoeke, 2015: Mechanisms of convective cloud organization by cold pools over tropical warm ocean during the AMIE/DYNAMO field campaign. J. Adv. Model. Earth Syst., doi:10.1002/2014MS000384, in press.
Folkins, I., 2013: The melting level stability anomaly in the tropics. Atmos. Chem. Phys., 13, 1167–1176, doi:10.5194/acp-13-1167-2013.
Houston, A. L., and R. B. Wilhelmson, 2012: The impact of airmass boundaries on the propagation of deep convection: A modeling-based study in a high-CAPE, low-shear environment. Mon. Wea. Rev., 140, 167–183, doi:10.1175/MWR-D-10-05033.1.
Houze, R. A., 1997: Stratiform precipitation in regions of convection: A meteorological paradox? Bull. Amer. Meteor. Soc., 78, 2179–2196, doi:10.1175/1520-0477(1997)078<2179:SPIROC>2.0.CO;2.
Houze, R. A., 2004: Mesoscale convective systems. Rev. Geophys., 42, RG4003, doi:10.1029/2004RG000150.
Jabouille, P., J. L. Redelsperger, and J. P. Lafore, 1996: Modification of surface fluxes by atmospheric convection in the TOGA COARE region. Mon. Wea. Rev., 124, 816–837, doi:10.1175/1520-0493(1996)124<0816:MOSFBA>2.0.CO;2.
Johnson, R. H., and P. E. Ciesielski, 2013: Structure and properties of Madden–Julian Oscillations deduced from DYNAMO sounding arrays. J. Atmos. Sci., 70, 3157–3179, doi:10.1175/JAS-D-13-065.1.
Johnson, R. H., T. M. Rickenbach, S. A. Rutledge, P. E. Ciesielski, and W. H. Schubert, 1999: Trimodal characteristics of tropical convection. J. Climate, 12, 2397–2418, doi:10.1175/1520-0442(1999)012<2397:TCOTC>2.0.CO;2.
Kemball-Cook, S. R., and B. C. Weare, 2001: The onset of convection in the Madden–Julian oscillation. J. Climate, 14, 780–793, doi:10.1175/1520-0442(2001)014<0780:TOOCIT>2.0.CO;2.
Khairoutdinov, M. F., S. K. Krueger, C.-H. Moeng, P. Bogenschutz, and D. A. Randall, 2009: Large-eddy simulation of maritime deep tropical convection. J. Adv. Model. Earth Syst., 1 (15), doi:10.3894/JAMES.2009.1.15.
Kuang, Z., 2008: Modeling the interaction between cumulus convection and linear internal gravity waves using a limited-domain cloud system-resolving model. J. Atmos. Sci., 65, 576–591, doi:10.1175/2007JAS2399.1.
Large, W. G., J. C. McWilliams, and S. C. Doney, 1994: Oceanic vertical mixing: A review and a model with a nonlocal boundary layer parameterization. Rev. Geophys., 32, 363–403, doi:10.1029/94RG01872.
Li, Y., and R. E. Carbone, 2012: Excitation of rainfall over the tropical western Pacific. J. Atmos. Sci., 69, 2983–2994, doi:10.1175/JAS-D-11-0245.1.
Li, Z., P. Zuidema, and P. Zhu, 2014: Simulated convective invigoration processes at trade wind cumulus cold pool boundaries. J. Atmos. Sci., 71, 2823–2841, doi:10.1175/JAS-D-13-0184.1.
Lin, X., and R. H. Johnson, 1996: Heating, moistening, and rainfall over the western Pacific warm pool during TOGA COARE. J. Atmos. Sci., 53, 3367–3383, doi:10.1175/1520-0469(1996)053<3367:HMAROT>2.0.CO;2.
Maloney, E. D., and D. L. Hartmann, 2000: Modulation of hurricane activity in the Gulf of Mexico by the Madden–Julian Oscillation. Science, 287, 2002–2004, doi:10.1126/science.287.5460.2002.
Mapes, B. E., 2001: Water’s two height scales: The moist adiabat and the radiative troposphere. Quart. J. Roy. Meteor. Soc., 127, 2353–2366, doi:10.1002/qj.49712757708.
Mlawer, E. J., S. J. Taubman, P. D. Brown, M. J. Iacono, and S. A. Clough, 1997: Radiative transfer for inhomogeneous atmosphere: RRTM, a validated correlated-k model for the longwave. J. Geophys. Res., 102, 16 663–16 682, doi:10.1029/97JD00237.
Moeng, C.-H., M. A. LeMone, M. F. Khairoutdinov, S. K. Krueger, P. A. Bogenschutz, and D. A. Randall, 2009: The tropical marine boundary layer under a deep convection system: A large-eddy simulation study. J. Adv. Model. Earth Syst.,1 (16), doi:10.3894/JAMES.2009.1.16.
Moum, J. N., and Coauthors, 2014: Air–sea interactions from westerly wind bursts during the November 2011 MJO in the Indian Ocean. Bull. Amer. Meteor. Soc.,95, 1185–1199, doi:10.1175/BAMS-D-12-00225.1.
Posselt, D. J., S. C. van den Heever, and G. L. Stephens, 2008: Trimodal cloudiness and tropical stable layers in simulations of radiative convective equilibrium. Geophys. Res. Lett., 35, L08802, doi:10.1029/2007GL033029.
Powell, S. W., and R. A. Houze, 2013: The cloud population and onset of the Madden–Julian Oscillation over the Indian Ocean during DYNAMO-AMIE. J. Geophys. Res. Atmos.,118, 11 979–11 995, doi:10.1002/2013JD020421.
Raymond, D. J., and X. Zeng, 2005: Modelling tropical atmospheric convection in the context of the weak temperature approximation. Quart. J. Roy. Meteor. Soc., 131, 1301–1320, doi:10.1256/qj.03.97.
Rotunno, R., J. B. Klemp, and M. L. Weisman, 1988: A theory for strong, long-lived squall lines. J. Atmos. Sci., 45, 463–485, doi:10.1175/1520-0469(1988)045<0463:ATFSLL>2.0.CO;2.
Skyllingstad, E. D., and J. B. Edson, 2009: Large-eddy simulation of moist convection during a cold-air outbreak over the Gulf Stream. J. Atmos. Sci., 66, 1274–1293, doi:10.1175/2008JAS2755.1.
Small, R. J., and Coauthors, 2008: Air–sea interaction over ocean fronts and eddies. Dyn. Atmos. Oceans, 45, 274–319, doi:10.1016/j.dynatmoce.2008.01.001.
Smyth, W. D., E. D. Skyllingstad, G. Crawford, and H. Wijesekera, 2002: Nonlocal fluxes and Stokes drift effects in the K-profile parameterization. Ocean Dyn., 52, 104–115, doi:10.1007/s10236-002-0012-9.
Sobel, A. H., and C. S. Bretherton, 2000: Modeling tropical precipitation in a single column. J. Climate, 13, 4378–4392, doi:10.1175/1520-0442(2000)013<4378:MTPIAS>2.0.CO;2.
Thompson, G., P. R. Field, R. M. Rasmussen, and W. D. Hall, 2008: Explicit forecasts of winter precipitation using an improved bulk microphysics scheme. Part II: Implementation of a new snow parameterization. Mon. Wea. Rev., 136, 5095–5115, doi:10.1175/2008MWR2387.1.
Tompkins, A. M., 2001: Organization of tropical convection in low vertical wind shears: The role of water vapor. J. Atmos. Sci., 58, 529–545, doi:10.1175/1520-0469(2001)058<0529:OOTCIL>2.0.CO;2.
Vickers, D., and L. Mahrt, 2006: Evaluation of the air-sea bulk formula and sea-surface temperature variability from observations. J. Geophys. Res., 111, C05002, doi:10.1029/2005JC003323.
Wang, S., A. H. Sobel, and Z. Kuang, 2013: Cloud-resolving simulation of TOGA-COARE using parameterized large-scale dynamics. J. Geophys. Res. Atmos., 118, 6290–6301, doi:10.1002/jgrd.50510.
Webster, P. J., and R. Lukas, 1992: TOGA COARE: The Coupled Ocean–Atmosphere Response Experiment. Bull. Amer. Meteor. Soc., 73, 1377–1416, doi:10.1175/1520-0477(1992)073<1377:TCTCOR>2.0.CO;2.
Weckwerth, T. M., and D. B. Parsons, 2006: A review of convection initiation and motivation for IHOP_2002. Mon. Wea. Rev., 134, 5–22, doi:10.1175/MWR3067.1.
Wicker, L. J., 2009: A two-step Adams–Bashforth–Moulton split-explicit integrator for compressible atmospheric models. Mon. Wea. Rev., 137, 3588–3595, doi:10.1175/2009MWR2838.1.
Woolnough, S. J., and Coauthors, 2010: Modelling convective processes during the suppressed phase of a Madden–Julian oscillation: Comparing single-column models with cloud-resolving models. Quart. J. Roy. Meteor. Soc., 136, 333–353, doi:10.1002/qj.568.
Wu, X., and S. Guimond, 2006: Two- and three-dimensional cloud-resolving model simulations of the mesoscale enhancement of surface heat fluxes by precipitating deep convection. J. Climate, 19, 139–149, doi:10.1175/JCL3610.1.
Yasunaga, K., A. Hashimoto, and M. Yoshizaki, 2008: Numerical simulation of the formation of melting-layer cloud. Mon. Wea. Rev., 136, 223–241, doi:10.1175/2007MWR2012.1.
Young, G. S., S. M. Perugini, and C. W. Fairall, 1995: Convective wakes in the equatorial western Pacific during TOGA. Mon. Wea. Rev., 123, 110–123, doi:10.1175/1520-0493(1995)123<0110:CWITEW>2.0.CO;2.
Zuidema, P., and Coauthors, 2012: On trade wind cumulus cold pools. J. Atmos. Sci., 69, 258–280, doi:10.1175/JAS-D-11-0143.1.