1. Introduction

Marine boundary layer (MBL) clouds affect the local top-of-atmosphere (TOA) radiation budget mainly through reflection of solar radiation, covering a substantially less reflective ocean surface. Reflection increases with cloud cover, amount of cloud condensate, and the number of particles that the condensate is distributed over (Twomey 1974), and has also been shown to depend on cloud morphology (e.g., McCoy et al. 2017, 2023) and cloud phase (e.g., McCoy et al. 2014). Understanding cloud properties and faithfully representing them in Earth system models is paramount for reliable climate projections. Ample uncertainty in recent CMIP5 and CMIP6 model intercomparisons has been traced back to MBL clouds within the postfrontal sector of extratropical cyclones (e.g., Tselioudis et al. 2021), in particular to biases in MBL mixed-phase clouds at middle and high southern latitudes (e.g., Zelinka et al. 2020; Bodas-Salcedo et al. 2016).

Midlatitude marine cold-air outbreaks (CAOs) produce such mixed-phase clouds in the postfrontal sector of extratropical cyclones during off-summer months (Fletcher et al. 2016) and may also serve as a proxy for CAOs at higher latitudes. CAOs commonly form near-overcast clouds of roll-like structure in a rapidly deepening MBL in response to enormous turbulent surface heat fluxes where cold air first meets relatively warm waters (Atkinson and Zhang 1996; Brümmer 1997, 1999; Müller et al. 1999; Renfrew and Moore 1999). The progressive accumulation of cloud condensate as well as decrease of aerosol available as cloud condensation nuclei (CCN) during their evolution with fetch results in the formation of substantial rain, similar to subtropical stratocumuli (e.g., Comstock et al. 2004), which acts to stratify the MBL through subcloud rain evaporation and inhibits the vertical transport of heat and moisture (Abel et al. 2017). In the absence of efficient MBL mixing, the cloud deck transitions toward a broken, sometimes open-cellular cloud field (e.g., Brümmer 1999; McCoy et al. 2017). Such overcast to broken cloud transitions can be expected to occur earlier where cloud condensate accumulation and CCN loss is swifter (e.g., Yamaguchi et al. 2017), which promotes precipitation (e.g., Comstock et al. 2004) and results in reduced extents of overcast clouds within postfrontal sectors.

Midlatitude CAOs typically coincide with dry intrusions (DIs; e.g., Raveh-Rubin 2017). A DI is a dry and relatively cold free-tropospheric (FT) air mass that typically subsides from the upper troposphere into the postfrontal sector (Browning and Reynolds 1994; Browning 1997) as part of a synoptic-scale circulation in extratropical cyclones (Thorncroft et al. 1993). DIs affect the postfrontal sector nonhomogeneously and DI strength and pattern depend on the cold-front shape (Browning 1997) and intensity (Raveh-Rubin and Catto 2019). DIs can intensify fronts (Catto and Raveh-Rubin 2019) and modify the postfrontal MBL, as shown at a North Atlantic station far downwind from cloud transitions (Ilotoviz et al. 2021). Because CAOs rapidly entrain FT air into the MBL during their evolution, we are motivated to understand the impact of DIs on cloud transitions. The frequent formation of marine CAOs eastward of continents, where there are warm ocean boundary currents, makes the northwest Atlantic an ideal site for targeted flight campaigns, such as the ongoing, multiyear Aerosol Cloud Meteorology Interactions over the western Atlantic Experiment (ACTIVATE) project (Sorooshian et al. 2019).

Here we examine a precampaign CAO event in the northwest Atlantic that produced earlier cloud transitions closer to the low pressure system (Fig. 1a). We hypothesize that this changing timing of cloud transitions is facilitated by gradients in the meteorological impact of the prevailing DI. We compile reanalysis data, satellite observations, and Lagrangian large-eddy simulations along several MBL trajectories. Using budgets of the simulated total moisture mixing ratio, liquid–ice potential temperature, and CCN, we highlight processes that promote earlier transitions.

View Full Size

(left) Polar-orbiting imagery taken by MODIS on board Aqua and (right) geostationary imagery from GOES-12 that was downwind oriented along each trajectory. MODIS imagery is shown as true colors and overlaid cloud cover isolines of 0.75 (shown as orange lines; derived from regional fractions of MYD06 cloud optical depth greater than 2.5) in the postfrontal area (here defined as region of negative w₇₀₀ and surface pressure above 1010 hPa, cf. Fig. 2) and positions of the trajectories at time of image acquisition (yellow symbols; panels additionally highlight trajectories selected for simulations). Geostationary imagery is either shown as brightness temperature (BT; using band 4 with λ = 10.70 μm) when solar zenith angle SZA > 85° or else as reflectance at λ = 0.67 μm. Cloud-cover isolines of 0.75 for each trajectory indicate cloud formation and cloud breakup (both shown in orange-red) and are derived as fraction of pixels with reflectance greater than 0.11 or brightness temperature smaller than 277 K where albedo is unavailable.

Citation: Journal of the Atmospheric Sciences 80, 12; 10.1175/JAS-D-23-0040.1

• Download Figure
• Download figure as PowerPoint slide

2. Material and methods

a. The selected cold-air outbreak case

The CAO examined in this study occurred between 17 and 19 March 2008 in the postfrontal sector of an extratropical cyclone. As described in more detail in Tornow et al. (2021), this CAO event was selected according to a weather state analysis (Tselioudis et al. 2021) and geographically and seasonally matches the recent ACTIVATE field campaign. Moving off the eastern seaboard, the postfrontal sector passes over a strong gradient in sea surface temperature (SST; Fig. 2, left), associated with the Gulf Stream. Using MERRA-2 fields [section 2b(1)], we find a peak marine CAO index of 15 K, defined as MCAO = θ_SST − θ₈₅₀ (e.g., Papritz et al. 2015), a maximum subsiding motion at 700 hPa of 50 mm s⁻¹, and a core pressure of about 970 hPa (Fig. 2, right).

View Full Size

MERRA-2 fields of environmental and meteorological parameters extracted at 1800 UTC 17 Mar 2008: (left) sea surface temperature (SST), (top right) mean sea level pressure and horizontal wind speed at 950 hPa, and (bottom right) vertical wind speed at 700 hPa and the marine cold-air outbreak index (MCAO = θ_SST − θ₈₅₀). Trajectories (shown as lines) follow the 250 m horizontal wind and were launched simultaneously at 0600 UTC 17 Mar 2008. (left) Symbols highlight 3-hourly coordinates along those trajectories used for large-eddy simulations. (right) Their position at 1800 UTC is shown as black dots. Green rhombi indicate cities and their population (through symbol size).

Citation: Journal of the Atmospheric Sciences 80, 12; 10.1175/JAS-D-23-0040.1

• Download Figure
• Download figure as PowerPoint slide

b. Satellite data and reanalysis

1) MERRA-2

MERRA-2 fields are provided 3-hourly on a 0.5° × 0.625° latitude–longitude grid (Gelaro et al. 2017). Using the horizontal wind components at an altitude of 250 m and progressing in discrete time steps, we extracted nine trajectories along a geodetic line spanning 38.58°N, 74.52°W to 43.58°N, 69.91°W, launched forward in time at 0600 UTC 17 March 2008. At each reanalysis time, we also extracted profiles of meteorological fields (i.e., vertical and horizontal wind components, temperature, relative humidity) as well as instantaneous sea surface temperature to compile parameters used in section 3.

2) GOES imagery

The Geostationary Operational Environmental Satellite 12 (GOES-12) imagery was downloaded from the Space Science and Engineering Center (SSEC), University of Wisconsin–Madison, using McIDAS version 4. Imagery was available every 30 min at near-nadir footprints of 1 and 4 km for channels in the visible and near-infrared, respectively. In this study, we use raw counts from band 1 (λ = 0.67 μm) converted to reflectance, r. Raw counts from band 4 (λ = 10.7 μm) were converted to radiance in order to obtain brightness temperature (BT). For conversion, we used scaling factor and offset values from https://www.ospo.noaa.gov/Operations/GOES/calibration/gvar-conversion.html. Using the interpolated geolocation and local horizontal wind as track direction from trajectories, we extract imagery within 50 km cross-track direction and 25 km along-track direction and recast imagery into a downwind-oriented coordinate system. Using an approximate solar zenith angle (SZA), we plot BT whenever SZA > 85° and reflectance otherwise.

3) Satellite cloud retrievals

Products derived from multiple low-Earth-orbiting satellites were collected along trajectories following similar distance criteria as applied to GOES imagery in the previous section. Here, we extract footprints within 50 km of both along- and cross-track direction, effectively spanning an area of 100 × 100 km². We rely on products obtained from multispectral imagery on board four satellite platforms (Table 1) and microwave radiometry (MWR) on board five satellite platforms (Table 2). We use instantaneous spaceborne MWR retrievals of the sum of both cloud water and rain (referred to simply as LWP throughout this work), with clear sky also taken into account to arrive at spatial medians. From the Multisensor Advanced Climatology of Liquid Water Path algorithm (MAC-LWP; Elsaesser et al. 2017), more commonly known for providing monthly products spanning several decades, LWP is here binned hourly using all available spaceborne platforms overpassing the CAO event. The MAC algorithm uses the latest Remote Sensing Systems cloud, water vapor, surface wind, surface rainfall, and SST retrieval inputs (Hilburn and Wentz 2008), and applies a number of bias corrections and a cloud to total liquid reverse-engineering approach to derive LWP (full details in Elsaesser et al. 2017). Suspended liquid water in its entirety interacts with radiation, lending an obvious reason to focus on total liquid, but another advantage of working in total versus cloud water path is that large sources of model–observational discrepancy, artificially arising from differences in how a model versus observational product defines cloud versus precipitating liquid, are avoided. For the same five satellite platforms, we use surface precipitation retrievals from Integrated Multi-satellitE Retrievals for GPM (IMERG). Consistent with simulations (section 2c), we compute cloud cover as fraction of imager-based optical thickness values greater than 2.5 (Wyant et al. 1997). For all other quantities, we determine domainwide median values and the interquartile range.

Table 1.

Retrievals from imagers aboard polar-orbiting satellites.

View Table

Table 2.

Retrievals from orbiting passive microwave imagers.

View Table

3. Results

While SST evolution and aerosol advecting from the continent set the stage for all CAO trajectories, we hypothesize that DIs and their meteorological pattern affect the cloudy MBL of cold-air outbreaks such that overcast to broken cloud transitions closer to the low pressure system occur earlier. Figure 3a is an extension of Tselioudis and Grise (2020) and presents a conceptual sketch that places a DI aloft in the postfrontal sector, the typical location of midlatitude cold-air outbreaks.

View Full Size

(left) Conceptual sketch of atmospheric flows within an extratropical cyclone, updated from Tselioudis and Grise (2020), showing positions and directions of dry intrusion (DI), dry conveyer belt (DCB), cold conveyer belt (CCB) curling around the low pressure center (L), and warm conveyer belt (WCB). Blue and red colors indicate relatively warm and cold temperatures, respectively. Numbers (1 and 2) in the left panel relate to (right) conceptual cartoons that distinguish the range in conditions within the postfrontal sector that can be characterized as follows: 1) enhanced subsidence, milder wind speed, and relatively dry and warm FT, leading to slowed MBL deepening and a delay in cloud transitions (i.e., and extended period of overcast conditions), and 2) reduced subsidence, elevated lower-tropospheric wind speed, and a relatively humid and cool FT, leading to rapid MBL deepening at great entrainment rates and causing a relatively early transitions.

Citation: Journal of the Atmospheric Sciences 80, 12; 10.1175/JAS-D-23-0040.1

• Download Figure
• Download figure as PowerPoint slide

To study the meteorological pattern of a dry intrusion and the associated cloud history, we produce Lagrangian MBL trajectories from MERRA-2 fields, which we number from starting points going from south to north. As seen in Fig. 2, these trajectories depart from the eastern seaboard simultaneously and are nearly parallel along a southeast direction. Along each trajectory, we extract profiles from MERRA-2 as well as geostationary imagery from near-infrared and visible channels.

Both the snapshot from a polar-orbiting satellite (Fig. 1a) and continuous geostationary imagery (Fig. 1b) show that trajectories farther north experience delayed cloud formation and earlier transitions toward a broken cloud field. For example, N-8, starting relatively far north, forms an overcast cloud deck after 5 h that then transitions to a broken cloud field at 10 h, resulting in an overcast period of about 5 h. In N-3, starting closer to the southern end, the cloud deck becomes overcast after 2 h and the cloud transition occurs at 12 h, resulting in 10 overcast hours.

Derived from extracted reanalysis profiles, Fig. 4 shows selected meteorological parameters describing the MBL and FT air. Trajectories starting farther north (compared to ones starting farther south) are characterized as follows:

1. a delayed rise in SST, facilitated by the Gulf stream progressively peeling off the coast going north (Fig. 2, left),

2. a more humid FT (Fig. 4c); for example, reaching a relative humidity at 700 hPa RH₇₀₀ ≈ 50% for N-9 in the first few hours, while for most other trajectories RH₇₀₀ < 10%,

3. smaller FT large-scale subsidence at 700 hPa (Fig. 4d), with values as low as w₇₀₀ ∼ 10–20 mm s⁻¹ compared to values of 40–60 mm s⁻¹ farther south,

4. greater horizontal MBL wind speed at 250 m altitude (Fig. 4e); reaching |υ_250m| ∼ 15–20 m s⁻¹ in contrast to 10–15 m s⁻¹ farther south,

5. a colder MBL (not shown) that is reflected in the marine cold air outbreak index (MCAO = θ_SST − θ₈₅₀), roughly leading to similar peaks of MCAO = 15 K values across trajectories (approximately when exceeding SST = 290 K) but remaining at elevated MCAO values for larger periods compared to trajectories farther south (Fig. 4f), and

6. reduced lower-tropospheric stability (here LTS_CAO = θ_700hPa − θ_100m, explained below).

View Full Size

Meteorological variables from MERRA-2 along trajectories. (top to bottom) Mean sea level pressure (MSLP), sea surface temperature (SST), relative humidity at 700 hPa (RH₇₀₀), large-scale vertical wind at 700 hPa (w₇₀₀), 250 m wind speed, the potential temperature difference between the near-surface layer and 850 hPa (MCAO = θ_SST − θ₈₅₀), and the lower-tropospheric stability, LTS_CAO = θ_700hPa − θ_100m, redefined to characterize purely atmospheric stability. Trajectory 3 is highlighted in bold black line and symbols.

Citation: Journal of the Atmospheric Sciences 80, 12; 10.1175/JAS-D-23-0040.1

• Download Figure
• Download figure as PowerPoint slide

SST control on cloudiness (e.g., Norris and Leovy 1994; Klein et al. 1995) is apparent from the timing of cloud formation across trajectories (Fig. 1) that is aligned with the shape of the warm edge (Fig. 2, left). Once clouds are formed, all other characteristics would collectively be expected to correspond to earlier transitions farther north because they are generally more conducive to cloud condensate growth, such that larger LWP/N_d ratios are reached earlier, associated with greater cloud-base precipitation in warm stratocumulus (e.g., Comstock et al. 2004). Following, we briefly survey earlier studies that mostly focused on subtropical clouds. For instance, a more humid FT promotes cloud thickening (e.g., Wood 2007) and increased cloudiness (e.g., Scott et al. 2020) via reduced entrainment drying, where any negative impact on cloudiness from humidity-driven increases in downwelling radiation (e.g., Sandu et al. 2010) should be weak for surface-driven convection in CAOs. Reduced subsidence promotes MBL deepening, which furthers condensate growth and has been shown to promote cloudiness (e.g., Bretherton et al. 2013; Myers and Norris 2013; Scott et al. 2020; Eastman et al. 2021). Greater wind speeds enhance surface fluxes, which may promote shallow convection and, in turn, aid aforementioned MBL deepening; earlier studies also note the connection of increased wind speed as an indicator of cloud transitions (e.g., Wang et al. 2016; Eastman et al. 2022). A colder MBL enhances surface fluxes, too, and is also expected to increase primary ice formation (e.g., Kanji et al. 2017) that can then be amplified by secondary ice production (e.g., Field et al. 2017), processes shown to accelerate cloud transitions in large-eddy simulations (e.g., Eirund et al. 2019; Tornow et al. 2021). Reduced lower tropospheric stability, here shown reversely through the CAO parameter MCAO (e.g., Naud et al. 2020, which used 800 instead of 850 hPa), indicates imminent cloud breakups (Eastman et al. 2022) and is generally associated with diminished cloud fraction (Klein and Hartmann 1993) and optically thinner clouds (Scott et al. 2020).

Aside, we note that the LTS definition conceived by Klein and Hartmann uses surface air temperature at an unspecified height corresponding to measurements from a variety of buoys and ships (Woodruff et al. 1987). Assuming that height is less than 10 m, it lies within a sharp temperature gradient between the surface and the more well-mixed portion of the subcloud layer in CAO conditions. To avoid complications in interpretation of MBL stability, we use a near-surface potential temperature at 100 m altitude to better characterize MBL stratification above a sharp near-surface gradient. We also note that while LTS_CAO does not consistently anticorrelate with MCAO, a conventionally defined LTS does (Fig. S3).

The pattern just described generally agrees with composite analysis of many extratropical cyclones: closer to the low pressure system Field and Wood (2007) found greater column-averaged relative humidity and greater near-surface wind speed, and Naud et al. (2016) reported reduced subsidence. In brief, trajectories farther north exhibit meteorological MBL and DI-affected FT properties that favor earlier transitions.

For four selected trajectories (denoted by symbols in the left panel of Fig. 1), we perform Lagrangian large-eddy simulations. In general, the trajectories can be grouped: N-1 and N-3 are late-transitioning cases, whereas N-6 and N-9 are early transitioning ones. Conceptually, these two groups correspond to the bottom- and top-right panels of Fig. 3, respectively.

Next, we compare simulation output with the sometimes sparsely available satellite retrievals in Fig. 5. To increase the number of samples, we incorporate overpasses that intersected the trajectory up to an hour before and after (shown as smaller symbols). Also, MAC-LWP and IMERG were available during day and night, roughly doubling the number of samples. In baseline simulations of the early transitioning cases (N-6 and N-9), simulations generally appear to reproduce an observed minimum in cloud-top temperature (Figs. 5c,d, solid lines versus symbols) and an observed maximum in cloud cover (Figs. 5g,h), which precedes substantial rain onset in the simulations. Thereafter, the simulations and observations both indicate a downward trend in cloud optical thickness (Figs. 5k,l) and LWP (Figs. 5o,p). In baseline simulations of the late-transitioning cases, by contrast, substantial rain formation is delayed by several hours (N-3) or does not occur at all (N-1). A sensitivity test with subsidence obtained from ERA5 reduced by a factor of 2 along trajectory N-1 offers improved agreement with the observed evolution in cloud cover (Fig. 5e, dash–dotted gold line) and other quantities, whereas a sensitivity test with greater initial MBL aerosol to N-3 better matches retrieved N_d after substantial rain formation (using 400 mg⁻¹; Fig. 5r, black dashed line) but also somewhat worsens agreement with other observations. For N-9, a lower N_INP of 0.1 L⁻¹ also improves agreement with N_d (magenta dotted line in Fig. 5e) owing to reduced microphysical riming-related loss. Surface precipitation rates from IMERG, a product that we consider of great absolute uncertainty under mixed-phase conditions, suggest generally greater rates for N-6 and N-9 compared to N-1 and N-3, qualitatively in line with simulations. As an indicator of precipitation onset, we extract each trajectory’s local time when rates start exceeding 0.01 mm h⁻¹ and find that N-6 (4.5 h) onset occurs earlier than N-9 (6.0 h) and N-3 (9.0 h), lining up well with the simulations. We defer more precise observational evaluation and consideration of retrieval uncertainties, for example, through the use of satellite instrument simulators, to future work where in situ data will also be available, but conclude here that simulations and sensitivity tests spanning N-1 and N-3 exhibit a longer buildup of LWP prior to substantial rain formation than those spanning N-6 and N-9. Another variation, which switches off shortwave and longwave radiation, was applied to N-6 and will be discussed in section 4.

View Full Size

For selected trajectories (shown panel by panel in horizontal direction and distinguished by color) highlighted in Fig. 1 and several meteorological variables (shown panel by panel in vertical direction) we compare large-eddy simulations (lines) and coincident products from several instruments (symbol shapes) on board satellite platforms. The x axis shows hours since the trajectory start (i.e., 0600 UTC 17 Mar 2008). Domain averages: (a)–(d) cloud-top temperature, (e)–(h) cloud cover (defined as fraction of grid columns or retrieved pixels with optical thickness > 2.5; thicker lines mark values > 0.75), (i)–(l) cloud optical depth, (m)–(p) liquid water path, (q)–(t) cloud droplet number concentration (in simulations defined as domain average weighted by gridcell cloud condensate mass mixing ratios), and (u)–(x) surface precipitation rate. Solid lines mark the baseline setup and other line types denote additional simulations that were performed for selected trajectories (see main text for explanation). Smaller symbols were obtained by assuming a stationarity over ±1 h, enabling to sample farther along each trajectory. Thicker lines in (e)–(h) and gray shades both mark the period between cloud formation and rain onset.

Citation: Journal of the Atmospheric Sciences 80, 12; 10.1175/JAS-D-23-0040.1

• Download Figure
• Download figure as PowerPoint slide

The simulations and plausible sensitivity tests also reproduce some aspects of the observed timing of cloud formation and transition, which can be linked to simulated precipitation quantities (Fig. 6). For example, N-3 is the first to form the cloud deck and among the last to transition toward the broken cloud field as seen in satellite imagery (cf. Fig. 1b). Looking at Fig. 5, N-6 and N-9, on the other hand, experience markedly shorter overcast times because of a stronger decline in N_d paired with a steeper growth in total condensate that leads to relatively early and substantial rain (Fig. 6), which in all simulations corresponds to the LWP maximum. Also seen in Fig. 6, elevated IWP found in N-6 and N-9 accelerate transitions through an intensified MBL preconditioning via riming-related effects (Tornow et al. 2021).

View Full Size

As in Fig. 5. Only simulations are shown. The listed variables: (a)–(d) lifting condensation level and inversion height, (e)–(h) cloud water path, (i)–(l) ice water path, and (m)–(p) rainwater path.

Citation: Journal of the Atmospheric Sciences 80, 12; 10.1175/JAS-D-23-0040.1

• Download Figure
• Download figure as PowerPoint slide

Which processes expedite the rain onset that is potentially driven by faster rates of 1) N_d decline and 2) LWP growth? Question 1 can be answered through a CCN budget that has been applied to simulation output (e.g., Tornow et al. 2021) and field observations (Tornow et al. 2022). To answer question 2, we further examine temperature and moisture budgets as well as changes in cloud geometric thickness ΔH. We note that budgets consider an MBL top as H = z_inv + 200 m. The model-diagnosed z_inv corresponds to the center of an inversion layer that is typically 400 m thick; alternative values for H fail to close the budgets that we present below (a thicker inversion layer over CAOs may be caused by strong variability in cloud tops and smaller buoyancy jumps between MBL and FT). Time averages are provided analogously to Tornow et al. (2021): for the time frame starting with cloud formation (defined as time when cloud cover exceeds 75%) and ending with rain onset (defined as time when RWP exceeds 25 g m⁻², approximately coincident with the LWP maximum); we note that including times outside this range could substantially alter averages.

Up until rain onset, the CCN decline is dominated by two terms (Fig. 7a): dilution from FT entrainment and collisional loss between hydrometeors of mixed phase. The surface influx of sea spray, counteracting both terms, is an order of magnitude smaller. Collisional loss increases with IWP as a result of riming (Tornow et al. 2021). Where IWP is low (e.g., N-1), entrainment dilution can dominate as inferred from recent CAO field observations (Tornow et al. 2022). Across trajectories, the sum of both dominant terms produces the greatest CCN loss rates in N-6 and N-9, in N-6 mainly through great collisional loss, and in N-9 more equally via collisional loss and entrainment dilution.

View Full Size

MBL (a) CCN budget and (b) decomposed entrainment rate averaged over the period from cloud formation (i.e., CC > 75%) until the onset of substantial precipitation (i.e., RWP = 25 g m⁻²) of baseline simulations; trajectory-individual periods are marked in Figs. 5e–h through thicker lines and gray shading throughout Figs. 5 and 6. Colors explain relevant budget source terms that are diagnosed from LES output. In the CCN budget, black dots show the sum of diagnosed budget terms, and gray dots show the actual average change rate. Both temporal change of boundary layer height H and subsidence rate w_sub = −w_LS at cloud top contribute to entrainment. In addition, we show the 5th–95th percentiles and mean of w_sub-700 at 700 hPa (bars and dots).

Citation: Journal of the Atmospheric Sciences 80, 12; 10.1175/JAS-D-23-0040.1

• Download Figure
• Download figure as PowerPoint slide

LWP growth occurs where cloud geometric thickness, which we define here as ΔH = H − z_LCL, increases. We next utilize simulation output to better understand tendencies in ΔH. First, we examine the role of MBL temperature and moisture in shaping geometric thickness. For an adiabatic subcloud parcel, greater moisture lowers its lifting condensation level (LCL) whereas greater temperature elevates it, leading to greater and smaller ΔH, respectively. Between cloud formation and rain onset, the MBL-average temperature increases in all cases (Fig. 8, top; here using liquid–ice potential temperature, a variable conserved for moist adiabatic motion) and MBL-average moisture also increases in all except N-9 (Fig. 8, bottom; here using total water mixing ratio, another conserved variable). Averages of both are mainly driven by entrainment, providing relatively warm and dry air, and by turbulent surface fluxes, which warm and moisten the MBL. However, substantially greater entrainment rates in N-6 and N-9 (Fig. 7b) do not lead to substantially greater drying and warming effects (Fig. 8) owing to relatively more humid FT air (Fig. 4) that is also cooler (not shown).

View Full Size

As in Fig. 7, the (a) liquid–ice potential temperature and (b) total moisture mixing ratio budget is shown for the period up until rain onset.

Citation: Journal of the Atmospheric Sciences 80, 12; 10.1175/JAS-D-23-0040.1

• Download Figure
• Download figure as PowerPoint slide

To quantify the effects of temperature and moisture changes on the LCL, we consider quasi-adiabatic clouds, as a simple extension of the adiabatic assumption. For each time step, we adopt H as cloud-top height and diagnose adiabaticity f_ad to match simulated LWP, as ${\text{LWP}}_{\text{sim}}=f_{\text{ad}}{\int }_{\text{LCL}}^H(c_p/L_{\upsilon })({\Gamma }_d-{\Gamma }_m)\rho (h-\text{LCL})dh$, where c_p is specific heat at constant pressure, L_υ is latent heat of vaporization, Γ_d and Γ_m are dry and moist adiabatic lapse rates, respectively, and ρ is air density. In all cases except N-1 f_ad exhibits steady or decreasing values in time, where a decrease in f_ad corresponds to less LWP in the absence of geometric thickness tendencies. Next, we fix selected aspects in time to explore the sensitivity of LCL of quasi-adiabatic clouds. As presented in Figs. 9b–d, allowing MBL-average moisture (blue) or temperature (red) to evolve (as analyzed in Fig. 8) induces opposing effects on LCL, subsequently impacting ΔH, which changes LWP nonlinearly. Changes in the LCL (black in Fig. 9b) emerge as a superposition of both tendencies and range between rates of about 2 mm s⁻¹ for N-1 and N-3, implying a near cancellation of temperature and moisture impacts on LCL, and rates of 15 and 8 mm s⁻¹ for N-6 and N-9, respectively, suggesting a dominant effect of increasing temperature (N-6 and N-9) and decreasing moisture (N-9) on LCL that should tend to reduce LWP. Simulated cloud-base height increases prior to rain (Figs. 6a–d), corroborating the quasi-adiabatic analysis above. Thus, the LCL evolution cannot explain ΔH increases that correspond to greater LWPs.

View Full Size

Per trajectory, bars from left to right in each panel show the change in (a) inversion height, (b) lifting condensation level (LCL), (c) resulting geometric thickness, and (d) liquid water path. Changes are shown for simulations (black) and three quasi-adiabatic calculations that deviate from calculations that were fitted via f_ad to match the simulated LWP (not shown). All else being equal, we fix near-surface temperature (blue) and near-surface water vapor mixing ratio (red) to understand changes in LCL. We also fix inversion height (orange) to demonstrate the importance of a rising cloud top for geometric thickness and LWP tendencies.

Citation: Journal of the Atmospheric Sciences 80, 12; 10.1175/JAS-D-23-0040.1

• Download Figure
• Download figure as PowerPoint slide

We can therefore infer that LWP increase must be driven by MBL deepening. As shown in Fig. 9a, the inversion height in N-6 and N-9 rises at rates of about 70 and 110 mm s⁻¹, respectively, outweighing the LCL rise by about an order of magnitude, and far exceeding the inversion height rise of N-1 and N-3 (about 3 and 20 mm s⁻¹, respectively). Thus, cloud thickening in late-transitioning cases is driven primarily by MBL deepening.

In general, MBL deepening is attributable to enormous surface heat fluxes that we find in all trajectories, usually exceeding 500 W m⁻² (Fig. S2, lumping sensible and latent heat fluxes together). However, across trajectories deepening is only weakly proportional to these flux values (not shown), indicating other factors at play. Greater increases in inversion height instead coincide with weaker subsidence and lower tropospheric stability (Fig. 4), both of which are associated with the overlying DI meteorological pattern. Weaker subsidence allows swifter deepening and LWP buildup. For example, halving subsidence rates in N-1 (Fig. 5, left column) leads to lower cloud-top temperatures, earlier cloud formation, and faster rates of LWP buildup and N_d reduction. Despite weaker subsidence, entrainment rates are increased (Fig. S2) as a result of even faster MBL deepening. In line with earlier analysis, increased entrainment-driven warming and drying and its effects on LCL remain well exceeded by cloud-top deepening that is substantially increased compared to the baseline setup (i.e., about 30 instead of 3 mm s⁻¹). We also tested an increased subsidence rate on trajectory N-9 that shows a slower rate of LWP buildup and N_d reduction as well as reduced IWP as a result of a shallower boundary layer (Fig. 10). To test the role of FT moisture gradient that we noted above, we increase q_υ in the driest case, N-1, from about 0.1 to 1.0 g kg⁻¹ (i.e., FT moisture found in between N-3 and N-6). We find prolonged LWP buildup that leads to an earlier rain onset and reduction in cloud cover (Fig. 10).

View Full Size

As in Fig. 5, we show baseline simulations (identical to Fig. 5) and two different variations: (left) greater FT water vapor mixing ratio (N-1) and (right) increased large-scale subsidence (N-9).

Citation: Journal of the Atmospheric Sciences 80, 12; 10.1175/JAS-D-23-0040.1

• Download Figure
• Download figure as PowerPoint slide

In summary, taken together, the large-eddy simulations offer plausible explanations for the observed spatial gradient in cloud regime transition to substantial rain onset. Analysis of the simulations also provides a route to assessing the role of processes that produce this gradient through combined effects on CCN and LWP.

4. Discussion

The analysis presented above suggests extensions to the conceptual picture of extratropical cyclones. Figure 3 is a variation of Tselioudis and Grise (2020) that now includes the DI that, as shown above, imposes gradients in meteorological parameters in the atmospheric column directly overlying the postfrontal sector. Initial MBL conditions can be considered partly DI independent (e.g., underlying SST pattern and meridional air temperature gradient) and partly dependent (e.g., varying inversion strength). The reduced subsidence that drives MBL deepening exhibits a north–south gradient that qualitatively matches the idealized DI concept presented in Browning (1997), showing a downward funnel of dry air that fans out in the lower troposphere: near the low pressure system the FT air subsides less (or even lifts) to overrun the cold front, and farther south its descent continues, curling anticyclonically (not illustrated here). North–south gradients in FT temperature and humidity coincide with a gradient in entrainment rates such that entrainment warming and drying effects are roughly similar across the sector. However, entrainment proportionally reduces CCN concentrations within the MBL; here we assume uniform CCN-poor conditions in overlying FT lacking evidence to the contrary (Tornow et al. 2022). New particle formation may induce FT variability, as found above summertime Southern Ocean clouds (McCoy et al. 2021). Upwind aerosol sources, here indicated through population number in Fig. 1a, may create downwind gradients in available CCN and INP that we largely ignore in this work but have studied in the past on a single trajectory (Tornow et al. 2021). For this case, N_d retrievals shown in Fig. 5 broadly indicate weak aerosol gradients. All of these gradients substantially shape the evolution of CCN and LWP, and thereby ultimately drive rain-initiated cloud regime transitions.

Which mechanism facilitates greater entrainment rates? We extracted factors related to the entrainment parameterization that was used in Chun et al. (2023). According to this analysis, entrainment is expected to be greater where there is 1) greater entrainment efficiency (A), 2) greater buoyancy production B), and 3) smaller buoyancy jump between MBL and FT (Δb). Shown in Fig. 11, buoyancy production scales with surface fluxes, steadily rising from south to north, except for N-9. Instead, N-9 shows a reduced buoyancy jump, potentially compensating the lack of aforementioned buoyancy production. We do find increased entrainment efficiency in trajectories going from south to north. However, in contrast to Chun et al. here efficiency inversely scales with cloud water mixing ratio at cloud top, leaving room for another mechanism driving efficiency. As already argued throughout this study, we consider subsidence as a mechanistic inhibition of MBL deepening, here enhancing efficiency where subsidence is reduced. Compared to subtropical clouds, B is an order of magnitude greater while A and Δb are an order of magnitude smaller.

View Full Size

Following Chun et al. (2023), we extract terms of an entrainment parameterization and average for each baseline simulation over the period from cloud formation to substantial rain onset: (a) entrainment rate w_e, (b) entrainment efficiency A, (c) buoyancy production B, (d) buoyancy jump between MBL and FT Δb, and (e) cloud water mixing ratio at cloud top q_c-top.

Citation: Journal of the Atmospheric Sciences 80, 12; 10.1175/JAS-D-23-0040.1

• Download Figure
• Download figure as PowerPoint slide

Greater wind speed has been shown to initiate substantial rain on a multiday time scale, leading to subsequent closed to open cloud regime transitions in subtropical stratocumuli (Eastman et al. 2022). In their analysis wind speed correlates with surface moisture fluxes and no apparent correlation with SST or large-scale subsidence was found. For the cold-air outbreak in this study, greater MBL wind speed also coincides with faster formation of substantial rain and subsequent cloud transitions. However, wind speed only shows a weak correlation with a generally large surface moisture flux. Here, fluxes are primarily shaped by the underlying SST pattern that shows a displaced warm edge in cases of greater wind speed and that appears to mainly regulate cloud formation and is likewise unassociated with cloud breakup. Instead of wind speed, we link reduced subsidence to an earlier rain-initiated transition, which is promoted by cloud thickening that is primarily attributable to MBL deepening. Subsidence is negatively correlated with MBL wind speed in this case, and could create noncausal correlations of cloud breakup to wind speed in CAOs. Compared to Eastman et al. (2022) the dynamic range of most variables in CAOs are at least twice as high (e.g., divergence, wind speed) or even a magnitude or so greater (e.g., SST, surface fluxes), possibly amplifying boundary layer dynamical aspects that are muted in the subtropics. We recommend that future efforts explore regimes that respond to wind speed and large-scale subsidence in shallow convection.

Trajectories of greater wind speed are located farther poleward, therefore show lower MBL temperature, and as demonstrated here, reach greater cloud-top heights (thus lower cloud-top temperatures) enabled by reduced subsidence rates. In effect, these trajectories have a greater portion of their MBL submerged in supercooled condition, potentially leading to more frozen condensate that has been shown to cause earlier transitions. Such a poleward gradient in cloud-top temperature can generally be expected for extratropical cyclones in either hemisphere and further extends the existing picture of typical cyclone characteristics (Field and Wood 2007; Naud et al. 2016).

Results present a contrast to subtropical stratocumulus clouds, where LWP budgets indicate a near balance between cloud-top radiative cooling and FT entrainment (van der Dussen et al. 2014; Hoffmann et al. 2020). Namely, CAO conditions in the present study exhibit a strong imbalance as evidenced by the rapid LWP growth. Switching off both short- and longwave radiation (and thereby radiative warming and cooling) in simulation N-6 (Fig. 5) acts to only mildly slow this growth rate, indicating the dominant role of surface fluxes $\mathcal{O}({10}^2\text{–}{10}^3){\mathrm{W}\hspace{0.17em}\mathrm{m}}^{-2}$ in driving convective mixing of heat and moisture and MBL deepening. Such MBL deepening in a subsiding environment together results in great entrainment rates $\mathcal{O}(10){\text{cm}\hspace{0.17em}\mathrm{s}}^{-1}$, as diagnosed from observations for a CAO case (Tornow et al. 2022). Entrainment of relatively warm and dry FT air tends to reduce LWP, but in contrast to stratocumulus clouds fails to establish a quasi-steady state, instead transitioning at variable rates to a substantially precipitating state in most simulations and all observed trajectories.

Simulations in this study largely neglect microphysical uncertainties arising from 1) the parameterization of rain formation, 2) the parameterization of primary and secondary ice production, and 3) the possible presence of smaller and larger aerosol modes. Future efforts will pursue 1 by considering other formulations as well as bin microphysics. Uncertainties from 2 have been shown to be substantial (Field et al. 2017; Kanji et al. 2017) and may explain the discrepancies found between satellite-observed and simulated N_d (Figs. 5q–t). With respect to 3, in situ observations of CAOs in the ACTIVATE campaign find smaller aerosol modes present (L. Ziemba 2023, personal communication; E. Crosbie 2023, personal communication); current efforts are examining the potential to activate smaller aerosol beyond the onset of substantial rain when peak supersaturation tends to increase. So-called giant CCN, on other hand, may also be relevant and overlap with 1. Uncertainties from surface flux parameterizations can influence simulated cloud transitions, further exacerbated by uncertainties in boundary conditions (e.g., Seethala et al. 2021). A brief comparison suggests comparable moisture and temperature profiles from ERA5 in the FT, but highlights greater uncertainty for dynamical aspects, such as large-scale vertical wind (Fig. S4), which we have identified as one leading factor in the timing of cloud regime transitions.

This study examined a single CAO event. While this case features archetypal characteristics of a CAO event, it is unclear whether our findings apply to a wider range of events. To explore general cloud regime transitions and their modulation through DIs, a broader survey is needed that considers CAOs under a range of conditions, ideally including events in other regions and seasons.

5. Summary

This study examines cloud transitions in a cold-air outbreak (CAO) and investigates the impact of a meteorological pattern influenced by a prevailing dry intrusion. Trajectories farther north experience earlier transitions owing to the combined effect of reduced subsidence, greater free-tropospheric (FT) humidity, a colder marine boundary layer (MBL) with an elevated MCAO index for a prolonged duration, and less lower-tropospheric stability compared to trajectories farther south. Sea surface temperature, which increases away from the coast at varying rates across the trajectories, appears to mainly regulate cloud formation. Large-eddy simulations (LES) generally reproduce observed cloud evolution features and an associated north–south gradient in cloud-regime transition timing. Simulated transitions are all triggered by the onset of substantial rain and our analysis indicates that earlier breakups farther north coincide with more rapid N_d loss and LWP increase, both contributing to earlier warm rain formation. A budget analysis of CCN indicates that N_d reduction is mainly driven by entrainment of low-CCN FT air and through loss from hydrometeor collision, including riming (Tornow et al. 2021). We find that increases in cloud geometric thickness, a proxy for LWP, is greater where subsidence is weaker. This finding is corroborated by a sensitivity test with subsidence halved that substantially speeds up an unrealistically late-transitioning case. By contrast, evolving MBL-average temperature and moisture, mainly shaped by surface fluxes and FT entrainment, conspire to produce weaker impacts that typically oppose LWP growth by raising the lifting condensation level. Last, we identify areas of study that we consider worthy of future work, such as examining a greater number of observed CAO cases and improving the representation of frozen hydrometeor production and multimodal aerosol in LES.

Data availability statement.

Reanalysis data can be downloaded at https://gmao.gsfc.nasa.gov/reanalysis/MERRA-2/data_access/, geostationary imagery at https://www.ssec.wisc.edu/datacenter/goes-archive/, and nongeostationary satellite retrievals at https://www.earthdata.nasa.gov/. MAC-LWP hourly retrievals and large-eddy simulation output are available upon request.