The Advanced Baseline Imager (ABI) aboard the 16th Geostationary Operational Environment Satellite (GOES-16) is used to retrieve cloud properties based on the NASA Satellite ClOud and Radiation Property System (SatCORPS) algorithms, which primarily rely on the 0.65, 3.9, and 11.2 µm channels for derivation of cloud microphysical properties . This study uses the SatCORPS produced cloud retrievals over a domain (29–46° N, 78–60° W) covering the ACTIVATE deployment region . Cloud variables, including cloud mask and thermodynamic phase, temperature, height, and pressure, particle effective radius (ice and liquid), cloud water path (ice and liquid), optical depth, and broadband shortwave albedo, are produced at the native resolution of the infrared channels, which is 2 km at nadir. Nd is calculated using the (sub)adiabatic assumption as 1Nd=52πk(fadcw(T,P)τcQextρwre5)1/2 where τc is the cloud optical depth and re is the cloud (water) droplet effective radius. A value of 0.8 is assumed for k, the inverse of the width of the modified gamma droplet distribution . The temperature-pressure-dependent condensation rate (cw) is calculated based on GOES-16 retrieved cloud temperature (T) and pressure (P) . The adiabatic fraction (fad) is assumed to be 0.8 . An extinction efficiency (Qext) of 2 and a liquid water density (ρw) of 997 kgm-3 are used in the calculations. In order to minimize retrieval biases, additional filtering of re>1µm, τc>1, cloud phase identified as either “water” or “suspected water” (i.e. Nd is only calculated for liquid cloud pixels), and solar zenith angle (SZA)<65° is applied . We tested a stricter filtering of re>4µm and τc>4 e.g., which does not alter the resulting evolutionary characteristics in the LWP-Nd space. Therefore, to better capture the optically thin roll clouds at the western edge of the deck during cloud emergence, we adopt the relatively liberal thresholds for re and τc in Nd calculations.

For LWP, microwave imagery retrievals e.g. provide an independent constraint on visible imagery retrievals, which are known to be biased under high SZA . By excluding pixels with SZA>65° and restricting our analysis to 09:00–15:00 LT, we largely remove the nonlinear biases associated with high SZA. The remaining biases at lower SZA appear to be systematic when evaluated against independent microwave imagery retrievals . Because the primary quantities of interest for process inference in this study are the spatiotemporal gradients and tendencies of cloud properties, rather than their absolute magnitudes, the impact of the remaining systematic biases in LWP on identifying process fingerprints is minimal.

The (sub)adiabaticity assumption used in Nd retrievals has been widely applied to global marine single-level clouds e.g. and has been shown to perform well for stratiform clouds, though less so for more convective and broken cloud fields . Importantly, this retrieval method has been extended to MCAO clouds, particularly in the polar regions e.g. and mid-latitude regions including the North Atlantic e.g.. An adiabatic cloud model assumes that cloud liquid water content (LWC) increases linearly with height above cloud base, modified by an adiabatic fraction factor (i.e. fad) to account for entrainment-mixing. In-situ LWC and re profiles measured during the ACTIVATE campaign e.g. in generally support the validity of this assumption. To assess the robustness of our results to the subadiabatic assumption, we conducted sensitivity tests using different fad values in Nd calculations. Figure S1 in the Supplement shows that varying cloud subadiabaticity, representing different degrees of entrainment-mixing, quantitatively affects the cloud evolution in LWP-Nd space, as expected. However, the qualitative characteristics of these evolutions, which reflect underlying cloud and boundary-layer processes, remain robust. Thus, variations in the adiabatic assumption primarily influence the absolute magnitude of retrieved Nd, while preserving the spatial and temporal patterns that are the focus of this study.

A limitation in SatCORPS pixel-based thermodynamic phase classification is that mixed-phase clouds are not reported as a distinct class. The classification identifies a single, radiatively dominant cloud phase (liquid or ice) that best explains the observed top-of-atmosphere radiances, using iterative model-observation matching which is further constrained by retrieved cloud-top temperature through a series of logical tests . Therefore, the GOES-16 cloud phase classification used in this work should be interpreted as “the radiatively dominant cloud phase that best explains the observed multispectral radiances at the top of the atmosphere”. Although to the best of our knowledge no validation work has specifically targeted MCAO clouds, this retrieval algorithm has been validated against active sensors e.g. and in-situ measurements e.g. under broad range of cloud regimes. The cloudy scene type investigated in this study, i.e. non-polar, snow-/ice-free scenes, exhibit the highest hit rate (0.971) and the highest Hanssen–Kuipers' skill score (HKSS; 0.941) across all scene types when compared to phase classifications based on an active sensor . For IWP retrievals, a cloud reflectance and emittance model based on adding-doubling radiative transfer is used to compute reflectance lookup tables for ice crystals with effective diameters varying from 6 to 135 µm . This model assumes randomly distributed hexagonal ice columns with roughened surfaces having the normalized roughness parameter set equal to 1.0 and asymmetry factors g between 0.77–0.81 at 0.65 µm. As a result, the satellite-retrieved IWP is not able to resolve different morphologies of frozen hydrometeors.

Given the limitations of satellite retrieved cloud properties, we characterize the cloud system evolution, including its hydrometeor mixing states, from a domain-mean perspective through spatial aggregation. This approach focuses on the characteristic system evolution, rather than on detailed spatial structure, as appropriate for the satellite view.

Meteorological reanalysis fields, including SST, temperature and humidity profiles, surface latent and sensible heat fluxes (LHF and SHF), horizontal winds at 1000 hPa, and vertical velocity at 700 hPa, are obtained hourly at 0.25° from the European Centre for Medium-Range Weather Forecasts (ECMWF) fifth-generation atmospheric reanalysis ERA5;. The marine cold-air outbreak index M-index;, a measure of the strength of the MCAO event and the stability of the marine boundary layer, is calculated as the difference between the potential temperature at 800 hPa and the sea surface temperature . Buoyancy fluxes (QB) are calculated following as 2QB=SHF×(1+0.6q2 m)+0.6LHFcpLvT2 m where q2 m and T2 m are the specific humidity and temperature at a height of 2 m, and values of 1004 and 2.5×106Jkg-1K-1 are used for the specific heat of air at constant pressure (cp) and latent heat of vaporization (Lv), respectively, in the calculations.

ACTIVATE featured two spatially coordinated aircraft, a low-flying HU-25 Falcon and high-flying King Air, that conducted 162 joint flights across multiple seasons in each year between 2020–2022. In-situ measurements of liquid and frozen hydrometeors in clouds are collected by a fast cloud droplet probe FCDP; and a 2-dimensional stereo imager 2D-S;, both manufactured by SPEC Inc. and operated by the Deutsches Zentrum für Luft- und Raumfahrt (DLR). The FCDP covers hydrometeors with diameter size spanning 3–50 µm, and the 2D-S covers diameter size spanning 11.4–1465 µm. Together, Liquid water content (LWC) and ice water content (IWC) are derived from these two probes by assuming particles <100µm are liquid droplets and by discriminating ice from liquid particles based on particle asphericity for hydrometeors with diameter great than 100 µm . An area-to-mass parameterization in is used to calculate IWC. Corrections are applied for 2D-S image distortion, sample area, and shattering. A deeper discussion of the ACTIVATE campaign's flight and instrument details is provided in .

A key focus of this work is to characterize the driving processes for the morphological transition from closed-cellular, overcast to open-cellular, broken cloud fields during mid-latitude MCAO events. Therefore, it is crucial to capture the complete cloud transition considering the limitation that satellite microphysical properties are only derived during daytime due to the algorithm reliance on visible and near-infrared channels. However, given a typical boundary layer wind speed of 15 ms-1 during a mid-latitude MCAO event , Lagrangian trajectories typically travel less than 500 km in fetch during sunlit hours, which covers the cloud street and closed-cell stages well, but often falls short for the open-cell stage. For example, on 24 January 2021, the fetch of the Lagrangian trajectories (solid red lines), initialized in the morning from the western edge of the cloud street, reaches the open-cell stage during the night (Fig. b). Therefore, a real Lagrangian approach would limit our capability in characterizing the full transition due to the reliance of cloud microphysical retrievals on shortwave channels.

A unique feature of MCAO is the persistence of the boundary layer wind field during the event, evident in the overlapping of time-evolving wind vectors in Fig. a. Taking advantage of this feature, we examine the idea of representing cloud transition by sampling cloud variables from a single GOES-16 snapshot along forward trajectories advected by the wind field present at that time. Here, two types of forward, isobaric trajectories were generated for comparison using 1000 hPa winds for advection, following and based on their success in tracking the movement of low-level clouds. (i) Lagrangian trajectory: initialized at the western edge of the cloud street at 13:00 UTC and subsequently advected by time-evolving wind fields based on the hourly ERA5 wind data (e.g. red lines in Fig. ). (ii) Instantaneous trajectory: initialized at the western edge of the cloud street and subsequently advected by a time-invariant ERA5 wind field fixed at the time of the GOES-16 snapshot (e.g. black lines in Fig. ). As evident during the MCAO event on 24 January 2021 (Fig. ), the close proximity between the Lagrangian trajectories and the instantaneous trajectories supports the validity of this “space-time exchange” approach, in which a time-evolving evolution is represented by a trace in the spatial dimension at a given time.

To further demonstrate the validity of the “space-time exchange”, we examine gradients in additional large-scale meteorological conditions beyond surface winds, including SST, LHF, SHF, buoyancy flux, the M-index, and subsidence and RH at 700 hPa, along instantaneous trajectories for each MCAO event (Fig. ). Diurnal variations (between 14:00–20:00 UTC; denoted by colors in Fig. ) are minimal for most meteorological conditions, with subsidence and RH at 700 hPa exhibiting the strongest diurnal variability. The steady gradients in surface buoyancy flux and temperature support the applicability of the “space-time exchange”, while diurnal variations in the dynamical environment are used to interpret the observed diurnal variations in process fingerprints discussed in Sect. 3.6.

The validity of “space-time exchange” hinges on whether the timescale of the process of investigation (τproc) is much shorter than the timescale of large-scale meteorology evolution (τmet) . In our case, the slowly-evolving, relatively persistent boundary layer wind field and spatial gradients in large-scale meteorological conditions (i.e. SST, surface fluxes, and free-tropospheric subsidence) throughout the MCAO event are key underlying conditions that satisfy this requirement. Based on this prerequisite of the “space-time exchange”, we screen MCAO events that occurred during ACTIVATE winter deployments between 2020–2022 for these large-scale conditions, yielding five events for further investigation (Table 1). These cases are also included in the library of MCAO events surveyed by and are identified as strong cases for Lagrangian modeling case studies. All five events show the presence of liquid and frozen hydrometeors, as well as the presence of drizzle- and rain-sized particles . Further contextual large-scale meteorological conditions during these MCAO events are described Sect. 3.1, as well as in Table 2 of and in .

For each event, seven GOES-16 snapshots at the top of each hour between 14:00–20:00 UTC (09:00–15:00 LT) are analyzed to investigate the cloud evolution (Fig. ). To characterize the mean transition within each snapshot, five closely aligned (separated by a 0.1° increment in both latitude and longitude) 12 h instantaneous trajectories are generated, initialized at the western edge of the cloud deck with their locations manually selected to overlap with ACTIVATE flight paths (locations detailed in Table 1; see also Fig. S2 in the Supplement for flight path and instantaneous trajectories overlaid on GOES-16 snapshots). Figure  shows the GOES-16 0.65 µm reflectance along the instantaneous trajectories at the top of each hour between 14:00–20:00 UTC for the five MCAO events (7 snapshots per event). Rather persistent patterns of cloud evolution throughout sunlit hours are observed, except for the 11 January 2022 case where an extended overcast cloud deck is evident later in the day (Fig. ), consistent with the increasing subsidence over time (Fig. ). GOES-16 retrieved cloud properties and ERA5 meteorological fields within a 1°×1° area centered at each point along the trajectory are averaged across the 5 spatially separated starting points and across the 7 snapshots at the top of each hour during daytime to characterize the mean evolution of each event. Cloud properties are averaged over cloudy pixels (a cloudy average), except when examining the mixed-phase evolution of the cloud field, for which LWP and IWP are averaged over 1°×1° areas (a domain average). Water (ice) cloud fraction is calculated as the fraction of pixels within the 1°×1° area that are identified as either “water (ice)” or “suspected water (ice)”.

The clouds embedded in mid-latitude MCAO events off the east coast of North America typically undergo substantial boundary layer deepening as air masses move offshore across the Gulf Stream (GS). This evolution is marked by increasing SST, surface buoyancy fluxes, cloud LWP, and cloud-top height (Figs.  and ), as well as the emergence of frozen hydrometeors e.g.. Across the five MCAO events examined here, large-scale meteorological conditions along the instantaneous trajectories exhibit similar evolutionary patterns, albeit with different magnitudes, as air masses advect offshore and downstream across the GS (Fig. ). This behavior is consistent with the canonical evolution of boundary-layer thermodynamic and dynamical structure under MCAO conditions .

Specifically, this evolution is characterized by an initial increase in SST and M-index, which is associated with enhanced buoyancy fluxes and boundary-layer deepening that lift cloud tops to subfreezing temperatures (Fig. ). As the transition progresses toward a broken cloud field, mean cloud tops subsequently become shallower and warmer. This shallowing and warming of cloud tops results from a combination of factors, including weakened buoyancy fluxes (Figs.  and ) and cloud thinning induced by entrainment-mixing and precipitation. The nearly invariant cloud top temperature during the cloud-thickening stage on 29 January 2021 is likely due to trajectories being very close to land, where clouds abruptly dissipate. Large-scale subsidence and free-tropospheric RH are the key variables that distinguishing the five MCAO cases, with RH700 ranging from less than 10 %–∼40 % and ω700 spanning ∼5 to ∼15 hPah-1 (Fig. ). Both RH700 and ω700 remain relatively invariant as SST rapidly increases along the trajectories.

With the spatial patterns in meteorological conditions establishing the large-scale environment for cloud field evolution (Fig. ), we next examine traces in LWP-Nd space (Fig. ), domain-mean LWP-IWP space (Fig. ), and the albedo vs. cloud fraction space (Fig. ) to infer the underlying boundary layer and microphysical (liquid and mixed-phase) processes governing the cloud evolutions. We note that, in this study, LWP denotes cloud liquid water path.

A common feature among the five MCAO cases is the cloud thickening stage supported by the strengthening of buoyancy flux as the air moves across the GS to warmer ocean surfaces. This stage is marked by increases in both cloud LWP and droplet number concentration (Fig. b). A key feature of the LWP-Nd phase space is the inference of cloud microphysical processes using isolines of re calculated by invoking the (sub)adiabatic assumption for stratiform warm clouds e.g.. In particular, the 11 and 15 µm re isolines approximately delineate the transition from non-precipitating to precipitating regimes governed by collision-coalescence processes . Furthermore, one can underpin cloud processes based on the directionality of a trace in this space. These include microphysical processes: (i) droplet activation, marked by increases in LWP and Nd at constant re (i.e. along an re isoline; arrow 1 in Fig. a); (ii) condensational growth, marked by an increase in LWP at constant Nd (arrow 2); (iii) collision-coalescence, marked by a decrease in Nd at constant LWP (arrow 3); (iv) precipitation and evaporation, marked by decreases in both Nd and LWP (arrow 4); and (v) entrainment-mixing, also marked by decreases in both Nd and LWP (arrow 5). We note that these directionalities (arrows) are intended to conceptually illustrate the characteristic effects of a given process; other processes may also contribute to or amplify the observed behavior. For example, during MCAO events, rapid expansion of the MBL can lead to dilution of liquid condensate through entrainment of drier air, reducing both LWP and Nd. In addition, mixing between the MBL and free-tropospheric air masses can further dilute cloud condensation nuclei concentrations within the MBL, thereby limiting cloud droplet formation e.g.. It is evident that based on the directionality of the traces, these clouds have gone through a combination of droplet activation and condensational growth during the cloud thickening stage, as indicated by the gray symbols in Fig. b. These two processes (arrow 1 and 2), driven by the increasing SST and buoyancy flux, together allow the clouds to grow to their maxima in water condensate and shortwave albedo (Fig. b and c).

The process of water condensing onto existing droplets leads to larger cloud droplets (arrow 2), which increases the likelihood of collision-coalescence that will decrease Nd (arrow 3). This microphysical transition is evident in all 5 MCAO cases investigated here (Fig. b). Larger droplet sizes reduce the total water surface area available to reflect incoming photons, making clouds less reflective of shortwave radiation at a given LWP and thereby causing a steeper decline in liquid cloud albedo as LWP reduces (Fig. c). The breakup of overcast cloud fields, defined as cloud fraction <0.99 and indicated by the transition from filled to open markers in Fig. b, occurs at re between 11–15 µm for all cases, suggesting a precipitation-driven cloud breakup, except on 29 March 2022 (star markers). The event on 29 March 2022 is marked by the highest peak-Nd and the lowest peak-LWP among the 5 cases (Fig. b and c). The suppressed cloud thickening is in line with cloud-top entrainment feedbacks, whereby smaller cloud droplets promote entrainment-mixing by evaporating more rapidly and remaining longer within the entrainment interface layer due to reduced gravitational sedimentation, ultimately leading to reduced LWP . Meanwhile, a clear delay in cloud breakup is evident on this day (Fig. b, star markers), suggesting a shift from precipitation-driven to entrainment-driven cloud breakup mechanisms e.g..

As the MBL deepens and clouds thicken, cloud top height rises and temperatures drop below freezing. This often leads to the formation of frozen hydrometeors, occurring approximately when the cloud attains its maximum LWP (colored symbols in Fig. b). Given the formation of ice, mixed-phase processes such as the growth of ice particles at the expense of liquid condensate complicate the interpretation of trace-directionality in liquid-only phase spaces (i.e. LWP-Nd and albedo-LWP). Therefore, we further investigate the directionality of traces in the domain-LWP vs. domain-IWP space to infer mixed-phase processes. Satellite-based IWP retrievals are inherently challenging due to assumptions regarding ice particle shape and their associated radiative properties. We therefore circumvent the reliance on absolute IWP magnitudes by extracting physical information on process signatures through the “rate” of change, using the “space-time exchange” approach that leverages the broad spatial coverage of geostationary satellites at a given instance. Given that the rates at which liquid and ice evolve also depend on height, updraft speed, and ice-liquid partitioning , instead of parsing individual processes, we focus on identifying fingerprints of the prevailing processes based on trace directionality in satellite-derived GV spaces.

Figure  zooms into the mixed-phase stage of cloud evolution, depicting traces in LWP-IWP space beginning with the first appearance of frozen hydrometeors (domain IWP>1 gm-2; colored symbols in Fig. b). Because GOES-16 retrievals provide only a single radiatively dominant cloud phase for each pixel, we use the combination of domain-mean LWP and IWP to characterize the “mixing state” in a spatial, rather than a column, sense. This approach exploits the rich spatial information contained in a satellite snapshot while circumventing the need for a vertically resolved mixing state. Thus, cloud evolution in domain-mean LWP-IWP GV space is interpreted in a qualitative manner, focusing on characteristic behaviors. For readability and clarity, the ranges of instantaneous trajectories (see Sect. 2.5) are shown as gray bars in Fig. S3 in the Supplement. Figures  and share the same symbol conventions and symbol size represents Nd and color denotes cloud top temperature in Fig. . Among the five MCAO cases, two distinct types of traces emerge (magenta vs. black in Fig. ): (a) in the magenta group, ice forms at the expense of liquid, while (b) in the black group, ice and liquid grow simultaneously at first, after which ice continues to grow while liquid begins to decline rapidly. A key distinction between the two groups is the ratio between changes in LWP and IWP (i.e. d(LWP)d(IWP)) as domain IWP increases. While ice emergence begins at different LWP values, which reflects variations in large-scale meteorological conditions (Fig. ), the inferred d(LWP)d(IWP) is nonetheless similar across the magenta group. Here, the less negative d(LWP)d(IWP) suggests the prevalence of a diffusional process where water vapor migrates from droplets to ice through evaporation and deposition, known as the Wegener–Bergeron–Findeisen (WBF) process . In contrast, the black group is characterized by rapid liquid depletion (i.e. a more negative d(LWP)d(IWP)) preceded by the co-growth of ice and liquid, a signature of riming where existing ice particles collect droplets through collisional freezing . In this regime, liquid loss is further accelerated by precipitation fallout associated with the fast sedimentation of rimed ice, as well as by dynamical feedbacks whereby latent heat release from freezing promotes cloud-top entrainment-mixing. While riming- and precipitation-driven liquid depletion likely occur in both cases of the black group, as evident by the large effective radius (Fig. b), dynamical feedbacks further amplify liquid loss during the 11 January 2021 event (the most negative d(LWP)d(IWP); downward triangles) as a result of the dry free troposphere (Fig. c). The existence of riming in the black group is further supported by the coldest cloud-top temperatures (symbol colors in Figs.  and b), which favor the presence of abundant supercooled liquid water, and by the highest LWP (Figs. b and ), which accelerates the riming rate . Furthermore, cases in the black group are the cleanest (lowest Nd) among the five (Fig. ), which facilitates maintenance of a relatively high supersaturation, a condition that favors the co-growth of liquid and ice.

To evaluate the process fingerprints inferred from satellite snapshots, we seek in-situ evidence for distinct microphysical signatures between the black and magenta groups. A useful metric for distinguishing rapid collisional growth from slower diffusional growth of ice particles is the duration of ice-liquid coexistence, quantified as the fraction of in-cloud sampling time during which both LWC and IWC exceed 0.01 gm-3 . Falcon in-cloud sampling includes above-cloud-base (ACB), ascent, below-cloud-top (BCT), and descent legs, and the data are binned by longitude as a proxy for the stage of cloud evolution during MCAO events. Figure  shows that as MCAO clouds evolve farther downstream from the coast, the magenta group exhibits systematically longer ice-liquid coexistence durations than the black group, consistent with slower diffusional growth. In contrast, the shorter coexistence times in the black group suggest cloud evolution dominated by riming and rapid precipitation fallout of rimed ice.

We note that frozen hydrometeors do not always emerge (defined as domain IWP>1 gm-2) when cloud LWP is at its highest (indicated by the change of color from gray to violet in Fig. b). This behavior can be partly explained by the minimum cloud-top temperature (CTT) reached as clouds deepen along their trajectories, such that the coldest minimum-CTT (∼-15 °C; 11 January 2022; Fig. ) is associated with the earliest ice emergence, occurring while liquid mass is still growing, whereas the warmest minimum-CTT (∼-9 °C; 29 January 2021) corresponds to the latest onset of ice formation, when liquid condensate has already begun to deplete through warm precipitation and entrainment. In general, for mixed-phase clouds governed either by the slower diffusional growth process (magenta group) or the nonlinear, collisional growth process (black group), a higher LWP at the onset of ice formation typically leads to a subsequent higher IWP (Fig. ).

Previous studies have demonstrated that secondary ice production can play a non-negligible role in shaping the evolution of MCAO clouds e.g.. However, because space-retrieved ice particle number concentrations are highly uncertain, we do not attempt to infer this process here. Importantly, the occurrence of secondary ice production does not affect the trace evolution in the mass-centric LWP-IWP GV space (Fig. ). In addition, uncertainties in IWP retrievals arising from the assumed ice crystal shapes and related radiative transfer parameters prevent us from resolving small-scale structures and variations in frozen hydrometeor morphology, which may also exhibit distinct signatures across these MCAO events e.g.. In-situ aircraft measurements are clearly the best tool for that. Nevertheless, we do maintain that the domain-mean perspective (i.e. the 1° areal mean) still allows us to extract distinct characteristics of system evolution from the satellite view. We do concede, however, that our interpretations are contingent on the realism of the IWP retrievals.

The shortwave albedo of a cloudy scene scales almost linearly with cloud fraction for warm clouds, especially warm stratiform clouds . It becomes increasingly nonlinear when the convective cloud regime is included . For mixed-phase clouds, at a given cloud fraction, the scene-albedo is governed by the liquid-ice partition of the cloud as a result of the contrasting shortwave (SW) transmissivity between ice and liquid. Therefore, the albedo-CF scaling in mixed-phase clouds is expected to be nonlinear as well. Indeed we observe nonlinear behavior as the overcast cloud field breaks up in Fig. , which zooms into the broken stage of MCAO evolution (beginning from the open markers). In particular, distinguishable albedo-CF scaling is evident between the magenta (triangle, star, and square markers) and black (upside-down triangle and circle markers) groups (as categorized in Fig. ), with the black group showing a steeper scaling between albedo and cloud fraction. This behavior is consistent with the mixed-phase fingerprints identified in Fig. , where inferred collisional freezing in the black group helps accelerate liquid condensate depletion (the key lever on cloud albedo), leading to a rapid decline in scene albedo. In addition, the continued growth of frozen hydrometeors after breakup, evident on 1 March 2020 (Fig. , circle marker), further contributes to the decline in scene albedo.

Contrary to the Twomey effect, where higher Nd leads to brighter clouds , we observe the opposite behavior: scene albedo is negatively correlated with Nd, with the highest peak-Nd coinciding with the lowest albedo at cloud breakup (Fig. ). This arises because clouds are thinner at breakup in cases with the highest peak-Nd, as reflected by the negative correlation between peak-LWP and peak-Nd (Fig. b). Overall, for a given cloud fraction, we find a large spread in scene albedo during the transition from an overcast cloud deck to a broken cloud field (Fig. ). This spread is consistent with LWP acting as the primary control on cloud albedo, with additional contributions arising from differences in the amount of ice present within the cloud. As a result, the albedo-CF scaling in the presence of mixed-phase clouds appears highly nonlinear.

The cloud radiative effect is closely tied to cloud morphology and the organization of the cloud field, such as the self-aggregation patterns found in trade cumulus and the precipitation-driven oscillation between open- and closed-cellular structures in marine stratocumulus . For marine warm clouds, distinct cloud albedo-fraction relationships have been observed for different mesoscale cloud morphologies . As discussed in Sect. 3.4, this relationship is further influenced by mixed-phase microphysical processes, given the strong dependence of shortwave reflectivity on hydrometeor phase (Fig. ). investigated the evolution of cloud morphological properties during two Arctic cold-air-outbreaks using a watershed approach applied to satellite observations, and found a convergence in the scaling between “cloud size” and “nearest-neighbor distance” (the distance between two adjacent cloud objects) as the cloud field transitions into cellular structures.

Here we examine cloud organizational evolution during the overcast-to-broken transition as described by a measure of deviation from randomness, following . Essentially, this approach assesses the departure of a given 2D cloud field from a predefined, “perfectly” random cloud field constructed using a Bernoulli random matrix . Specifically, a cloud field is first converted into a binary cloud mask based on a τc threshold of 5. The distributions of cloud-chord length (analogous to cloud size) and void-chord length (analogous to nearest-neighbor distance) in this binary mask are then compared against those from a random cloud mask, which is defined by a single parameter, the observed cloud fraction calculated from the original cloud field. This comparison yields a Goodness-of-Fit score based on the Kolmogorov–Smirnov (KS) test , defining the deviation from randomness, which ranges from 0 (perfectly random) to 1. Figure  depicts the evolution in cloud organization since the transition as a function of cloud fraction, with cases colored by minimum cloud top temperature. The degree of organization (or deviation from randomness) in cloud size (or cloud-chord length) scales linearly with CF (Fig. a), whereas organization in void size (or nearest-neighbor distance) remains relatively invariant throughout the transition (Fig. b). The scaling between cloud-size organization and CF is slightly separated by cloud-top temperature, such that the coldest (and highest) cloud tops are associated with the highest degree of organization and highest scene albedo during the transition (Figs. , , and ). Nevertheless, the degree of organization in both cloud size and void size converges to a common range of 0.3–0.4 (Fig. a and b), despite differences in boundary layer thermodynamic and dynamical structures, Nd, LWP, IWP, CRE, and even the visual pattern of the cloud field (Figs. , , and c–e). This convergence is consistent with the findings of . We note a relatively large spread in cloud organization under high cloud-fraction conditions (CF>0.9; Fig. a–c). This is likely due to the 2°×2° cloud-mask domain, which inevitably includes a mixture of overcast and broken conditions. In addition, the τc-based cloud mask is particularly sensitive to the domain ice fraction, which varies considerably across the five events (Figs.  and ).

Leveraging the high temporal coverage of GOES-16 satellite, we examine variations in cloud evolution along instantaneous trajectories between 14:00–20:00 UTC (09:00–15:00 LT) as indicators of the diurnal cycle. Figure  shows the traces in LWP-Nd (column A), cloud-top height vs. temperature (column B), buoyancy flux vs. M-index (column C), and 700 hPa subsidence vs. SST (column D) spaces, as a function of time (colors). Overall, traces in LWP-Nd space exhibit similar patterns of evolution throughout the day for each event (Fig.  column A). However, more noticeable differences emerge during the cloud-thickening stage, also evident in cloud-top characteristics (Fig.  column B). This diurnal variation is marked by a cyclic pattern centered around local noon, with a clear directionality in re: traces around noon (yellow-ish) tend to begin with larger re (i.e. higher LWP and lower Nd), compared to morning (green) and late afternoon (red). For a given cloud-top height, cloud tops become warmer later during sunlit hours (Fig.  column B). Buoyancy flux, M-index, and SST gradients along the evolution trajectories remain highly invariant throughout sunlit hours, whereas subsidence at 700 hPa varies the most across the large-scale conditions (Fig.  columns C and D). Together, these observations underscore the coupled roles of large-scale dynamics, cloud microphysics, and solar radiation in shaping process fingerprints during sunlit hours.

This daytime evolution in the developing stage of MCAO subsequently leads to a similar cyclic pattern in the albedo-CF scaling (not shown) and in the timing of cloud breakup (open circles in Fig. ), with cloud transitions occurring later (or farther downstream) in near-noon traces. One possible explanation for this delayed transition is the shortwave heating induced cloud thinning e.g.. Alternatively, this pattern may reflect a diurnally evolving aerosol size distribution in the MBL prior to cloud formation so that prescribing the observed afternoon size distribution in simulations leads to delayed precipitation .

As discussed in Sect. 2.1, SZA-dependent biases can potentially contribute to the observed diurnal variations in cloud evolution. By restricting our analysis to 09:00–15:00 LT, we largely constrain SZA-dependent biases in LWP , with the aim of attributing the observed variations to physical processes rather than retrieval artifacts. That said, we acknowledge that differences between traces in the early morning (09:00 LT) or later afternoon (15:00 LT) and those near midday may still be affected by residual SZA-related artifacts.