Thin Clouds Control the Cloud Radiative Effect Along the Sc‐Cu Transition

In situ and spaceborne studies reveal the prevalence of thin clouds in the major Stratocumulus‐to‐Cumulus Transition (SCT) regions. Using instantaneous satellite and reanalysis data, this study investigates the properties of thin clouds in the Southeast Pacific Ocean and their impact on the cloud radiative effect (CRE). Our findings demonstrate that thin clouds are intrinsic to the SCT. The overcast stratocumulus‐dominated regime exhibits a minimal presence of thin clouds, which become notably prominent after the clouds breakup into the cumulus‐dominated regime. The regime dependence of the occurrence of thin clouds is also observed in terms of the marine cold‐air outbreak parameter and the sea surface temperature. Thin clouds at a given cloud cover significantly modulate the shortwave (SW) and longwave (LW) components of CRE. SW CRE decreases by 46 %–65 % with increasing thin cloud cover. They account for a larger variance in cloud albedo than the combined influence of the liquid water path and effective radius. Furthermore, LW CRE decreases by about 12 %–52 % with thin cloud cover. An increase in the fraction of thin clouds also leads to a larger fraction of negative SW CRE offset by positive LW CRE at a given cloud cover. This LW compensation ranges from approximately 8 % at overcast cloud cover to as much as 19 % at about 50 % cloud cover. These findings elucidate the crucial role of thin clouds, and thus cloud morphology, in modulating CRE and underscore the necessity of their accurate representation in climate models.


Introduction
Shallow marine clouds are amongst the most crucial components of Earth's radiative budget.Particularly significant are the bright marine stratocumulus (Sc) cloud decks that typically occur over the colder eastern sectors of the Pacific and Atlantic Oceans (Muhlbauer et al., 2014;Wood, 2012), covering approximately 20 % of the Earth's dark ocean on an annual scale (Hahn & Warren, 2007;Warren et al., 1988).Due to their high albedo and extensive global coverage, specifically in regions of high solar influx, Sc clouds exert a substantial impact on the global cloud radiative effect (CRE; Hartmann et al., 1992;Hartmann & Short, 1980).With a top-of-theatmosphere net negative annual CRE of 8.3 W m 2 (L' Ecuyer et al., 2019), they contribute significantly to cooling the climate.Understanding how the radiative properties of these clouds will evolve in a warmer future climate is critical to reducing uncertainties in climate predictions (Sherwood et al., 2020;Zelinka et al., 2020).
Cloud fraction ( f c ), representing the horizontal extent of clouds, constitutes a fundamental component of CRE (Y. C. Chen et al., 2014;Y. Chen et al., 2022;Goren & Rosenfeld, 2014;Rosenfeld et al., 2019;Christensen et al., 2020;Wall et al., 2023).Its variations within a warming climate have been investigated to better understand and constrain the shortwave (SW) and longwave (LW) cloud feedback (Klein et al., 2017;Qu et al., 2015;Zelinka et al., 2012Zelinka et al., , 2016)).However, the CRE of shallow clouds exhibit substantial variations that cannot be solely explained by the binary cloud fraction (i.e., either cloudy or clear).Variations in SW CRE have been mainly linked to differences in cloud properties, such as cloud water content and droplet concentration (Bender et al., 2016;Christensen et al., 2020;Engström et al., 2015;Zelinka et al., 2016;Y. Chen et al., 2022).The occurrence of diverse cloud morphology patterns exhibiting distinct horizontal and vertical cloud-water distributions may also complicate the CRE-f c relationship (Alinaghi et al., 2023;Dauhut et al., 2023;Denby, 2023;McCoy et al., 2017McCoy et al., , 2023;;Mohrmann et al., 2021;Muhlbauer et al., 2014;Stevens et al., 2020;Zhou et al., 2021).For instance, McCoy et al. (2017) highlight significant albedo (and thus SW CRE) variations among closed, open, and disorganised mesoscale cloud morphology patterns, even at fixed cloud cover.This difference in CRE has been linked to the occurrence of optically thin clouds associated with these morphology patterns (McCoy et al., 2023).
LW CRE is predominantly a function of cloud-top temperature (CTT) and f c (Zelinka et al., 2012(Zelinka et al., , 2016(Zelinka et al., , 2020)).Although shallow marine clouds have a relatively small temperature difference between their tops and the ocean surface underneath, optically thin clouds, due to their semi-transparency, can allow a portion of the surface thermal emission to penetrate through them and thus affect the LW CRE (Arouf et al., 2022;Guzman et al., 2017;Vaillant de Guélis, Chepfer, Noel, Guzman, Dubuisson, et al., 2017;Vaillant de Guélis, Chepfer, Noel, Guzman, Winker, et al., 2017).This suggests that optically thin clouds influence the entire spectrum of radiative effects associated with shallow clouds.
A prominent feature observed in Sc-dominant regions across the globe is the transition of overcast Sc cloud decks to deeper and broken cumulus (Cu) clouds (the so-called Sc-Cu transition, hereby abbreviated as SCT) (Bretherton & Wyant, 1997;Wood, 2012;Wood & Bretherton, 2004).Along the transition, the overcast Sc clouds gradually thin out as they become decoupled from the surface moisture supply and are eventually replaced by the Cu clouds that develop underneath (Wood, 2012).These Cu clouds are frequently associated with the presence of geometrically and optically thin clouds that exhibit a veil-like appearance (Wood et al., 2018).Satellite observations indicate a significant presence of thin clouds within the SCT regions across the globe, accounting for approximately 30 %-50 % of the annual cloud cover (Leahy et al., 2012;Guzman et al., 2017;O et al., 2018).
In-situ observations demonstrate that thin veil clouds form within the detrained flow of Cu clouds as they spread horizontally at the top of the boundary layer (Wood et al., 2018;O et al., 2018).These veil clouds are usually associated with a low cloud droplet number concentration (N d ; less than 10 cm 3 ) and a large droplet radius (>15 μm) (Wood et al., 2018;O et al., 2018).Such a low N d is also a key reason for their low optical thickness (usually less than 3) (Wood et al., 2018).Simulations from a Lagrangian adiabatic parcel model suggest that the collision-coalescence scavenging in the active Cu updraft is primarily responsible for such low droplet concentrations in the veil clouds (O et al., 2018).This hypothesis is further supported by a satellite-derived negative correlation between the fraction of optically thin clouds and N d (O et al., 2018).
The SCT displays a diverse array of cloud morphology across a wide range of cloud cover, starting with overcast Sc in the shallow boundary layer, transitioning to Cu in the deeper boundary layer, and eventually dissipating in later stages.Thin cloud cover may also vary at such stages and therefore impact the SW (McCoy et al., 2017(McCoy et al., , 2023) ) and LW components of CRE differently (Guzman et al., 2017;Vaillant de Guélis, Chepfer, Noel, Guzman, Dubuisson, et al., 2017).Given the prevalence of thin clouds in the SCT region (Leahy et al., 2012;Guzman et al., 2017;O et al., 2018), quantifying their radiative effects is critical to furthering our understanding of the role of clouds in the global radiative budget.In this paper, we use simultaneous satellite and reanalysis data to investigate the properties of thin clouds and untangle their contribution to the radiative effects of shallow clouds.We further discuss the key role of cloud morphology in cloud-radiation and aerosol-cloud-radiation interactions.The manuscript is organized as follows: Section 2 describes the data and methods used in the study; Section 3 presents the results, which are discussed in Section 4 and summarized in Section 5.

Data and Methods
Spaceborne lidar measurements over global oceans indicate a dominant presence of optically thin clouds to the west of Sc-dominant regions, with their highest occurrence over the Southeast Pacific Ocean (SEP) (Leahy et al., 2012;O et al., 2018).Therefore, we select a domain over the SEP bounded by latitudes ranging from 10°S to 30°S and longitudes from 80°W to 110°W, as shown in Figure 1a.The data sets considered in this work include cloud observations from the Moderate Resolution Imaging Spectroradiometer (MODIS) instrument (Platnick et al., 2015), radiation measurements from the Clouds and the Earth's Radiant Energy System (CERES) sensor (Doelling et al., 2013), and meteorological properties from the European Centre for Mid-range Weather Forecasting ERA5 reanalysis data ("The ERA5 global reanalysis", 2020).The MODIS, CERES, and ERA5 data are considered over a span of 6 years, from 2015 to 2020.The data products, along with the post-processing methods, are described in the subsequent sections.

MODIS Cloud Observations
For inferring instantaneous cloud properties, we use the MODIS Aqua level-2 collection 6.1 cloud product (Platnick et al., 2015), which includes cloud optical and microphysical properties at a nadir resolution of 1 km by 1 km.Parameters used in this study include the corrected reflectance at 0.64 μm (R), cloud fraction ( f c ), cloud optical thickness (τ), liquid water path (L), CTT and cloud effective radius (R e ).We filter the MODIS scenes to only include single-layered liquid-phase clouds.Single-layered clouds are identified using the MODIS cloud multi-layer flag, and liquid clouds are filtered using the MODIS cloud-phase metric.We exclude pixels with sensor and solar zenith angles greater than 41°, as they may have retrieval-related issues (Grosvenor & Wood, 2014).The filtered cloud properties are then gridded into a uniform latitude and longitude grid of 1°by 1°.As thin clouds mostly occur at the edges of cloud cores (Eytan et al., 2020;Koren et al., 2007;Wood et al., 2018;O et al., 2018;Mieslinger et al., 2022), their CTT may not be reliable due to partial pixel filling.Therefore, assuming a uniform cloud top height within a scene, we use the core CTT as a representative of the CTT of the entire cloud within the scene.Cores are defined as the pixels corresponding to the top 10 percentile of R (indicating the brightest pixels) within a scene (Goren & Rosenfeld, 2014).Additionally, we classify scenes with a mean R e larger than 14 μm as precipitating (Fan et al., 2020;Goren & Rosenfeld, 2014;Rosenfeld et al., 2012).To quantify the heterogeneity in the distribution of cloud water within a scene, we use the dispersion ratio (D) of L, defined as the ratio between the standard deviation and the mean of L (Feingold et al., 2022;Wood, 2006).Studies based on spaceborne lidar suggest that thin veil clouds, which do not completely attenuate the lidar signal, usually have a τ ≤ 3 (Leahy et al., 2012;O et al., 2018).Following this, McCoy et al. (2023) used the MODIS cloud product and defined thin clouds as those pixels that have a τ ≤ 3.However, this way of defining thin cloud cover ( f c,thin ) using τ alone excludes those cloudy pixels that do not have a valid τ retrieval (King et al., 1997;Platnick et al., 2015) and may underestimate f c,thin (see Figures 1b and 1c).To overcome this issue, we use a combination of MODIS-derived R and τ to identify thin clouds.We define f c,thin for a scene as the fraction of pixels with R less than or equal to the R matched to a τ of 3 (R τ = 3 ).R τ = 3 is estimated for each scene by fitting a second-degree polynomial regression between τ and R. As shown for an example case in Figures 1d and 1e, combining τ and R yields a better estimate of f c,thin , which would be otherwise underestimated by about 55 % if only τ was used.The fraction of thin cloud cover ( f thin ) is defined as the ratio between f c,thin and f c (f thin = f c,thin f c ) .Note that the spatial variation of f thin over the considered study domain (see Fig. S1a in Supporting Information S2) is found to be consistent with previous spaceborne lidar-based results (Leahy et al., 2012;O et al., 2018).A limitation of defining veil clouds in terms of cloud optical properties is that it may also include optically thin shallow Cu and stratus clouds.Because our focus is on thin detrained clouds located at the uppermost part of the decoupled boundary layer, typically associated with an active cloud core (Wood et al., 2018;O et al., 2018), we restrict our analysis to cloudy scenes with a core L greater than 100 g m 2 .This criterion allows to exclude scenes that are primarily dominated by thin Cu and stratus clouds.

CERES Radiation Measurements
For inferring information on cloud radiative properties, we use the CERES SYN Edition 4A product (Doelling et al., 2013(Doelling et al., , 2016)), which includes radiative fluxes and albedo for cloudy and clear-sky conditions at a uniform latitude-longitude resolution of 1°by 1°and a temporal resolution of 1 hr.Radiative fluxes and albedo are interpolated temporally to MODIS observations.The LW and SW CRE at the top of the atmosphere (TOA) are estimated by subtracting the corresponding fluxes for clear-sky and all-sky conditions.Similarly, cloud albedo (A c ) is estimated by subtracting the albedo corresponding to all-sky and clear-sky conditions.

ERA5 Reanalysis Data
SST, temperature profile, and pressure profile are derived from ERA5 hourly data, available at a uniform latitudelongitude resolution of 0.25°by 0.25° (Hersbach et al., 2023a(Hersbach et al., , 2023b)).We use these properties to estimate the marine cold-air outbreak parameter (M) (Fletcher et al., 2016;Kolstad et al., 2009), a measure of lower tropospheric stability defined as where θ s and θ 800 represents the potential temperatures at the surface and 800 hPa, respectively.Note that the surface here refers to the sea surface and not the near-surface layer in the atmosphere.M is strongly correlated with the boundary layer height (BLH;McCoy et al., 2023;Naud et al., 2018Naud et al., , 2020)), which has previously been found to have a strong correlation with the thin cloud fraction (O et al., 2018).The M index and SST are then re-gridded to a uniform latitude-longitude grid of 1°by 1°and interpolated temporally to match the MODIS observations.

Non-Monotonic Relationship Between f c,thin and f c
In SCT regions, the climatological progression of Sc development involves the formation of overcast clouds ( f c = 1) over colder SSTs and shallow boundary layers near the coastal region, which breakup ( f c < 1) as they move towards warmer SSTs and deeper boundary layers (Sandu & Stevens, 2011;Wood, 2012;Wood & Bretherton, 2004).This cloud breaking is also associated with the formation of precipitation and the development of Cu clouds (Diamond et al., 2022;Feingold et al., 2010;Rosenfeld et al., 2006;Wang & Feingold, 2009;Yamaguchi et al., 2017).The variation in f c,thin with f c for the study region is shown in Figure 2. It can be seen that overcast clouds are associated with the least coverage of thin clouds.This is consistent with the in-situ study of Wood et al. (2018), which report a lower likelihood of the occurrence of thin clouds in the overcast Sc regime.
The variation in f c,thin with f c is non-monotonic, featuring three distinct linear phases (I, II, and III).In Phase I (0.85 < f c ≤ 1), there is a steep increase in f c,thin along with a decrease in f c , reaching a maximum at an f c of about 0.85.Additionally, this increase is accompanied by a rise in the fraction of precipitating scenes (defined as those with R e > 14 μm).A similar negative correlation between f c and f c,thin at high cloud cover is also observed in trade wind Cu clouds, albeit to a lesser degree (Mieslinger et al., 2022).Below an f c of 0.85 (Phase II), we observe a correlation reversal, with decreasing f c,thin along with f c .However, f c,thin decreases slower than f c , implying that thin clouds become more prominent with decreasing f c in this phase.The fraction of precipitating scenes remains consistently high, at about 85 %.In Phase III ( f c < 0.44), f c,thin declines steeply, approaching the identity line at low f c .The fraction of precipitating scenes also decreases as one approaches low f c .Upon analysing random cloudy scenes in Phase-III, we observe that scenes with low f c often consist of small scattered Cu clouds that are optically thin (e.g., the top right panel of Figure 3a).These small clouds are most likely in their dissipating stage, which may account for the decrease in the fraction of precipitating scenes.A similar pattern of correlation reversal is also observed in the geographical variation of the correlation coefficient between f c,thin and f c (see Fig. S1b in Supporting Information S2), where a negative correlation close to the coast gradually shifts to a positive correlation as one moves offshore.
It is important to note that this analysis focuses on the relationship between f c, thin and f c rather than the previously explored relationship between f thin (= f c,thin f c ) and f c (Leahy et al., 2012;McCoy et al., 2023).The latter fails to identify the three phases shown in Figure 2 and only reports a strong negative correlation.A negative correlation is expected, as f thin , by definition

f thin and Cloud Heterogeneity
In order to explore the relationship between the fraction of thin clouds at a given cloud cover and cloud water heterogeneity, we use the relative dispersion of L (D L ) .The relative dispersion was shown to be an effective measure of cloud heterogeneity (Feingold et al., 2022;Wood, 2006).
Figure 3b illustrates the variation in D L with f thin at different cloud cover.Quantitatively, Figure 3b shows a strong positive correlation between f thin and D L , irrespective of the f c .It can be seen that the extent of cloud heterogeneity at a given f c is constrained by the degree of variability in f thin .For example, at f c = 1, we observe an f thin range of 0-0.3, which limits D L variations to between 45 % and 95 %.As f c decreases, the f thin range shifts towards higher values, initially undergoing a significant shift (from 0 to 0.3 under overcast conditions to 0.2-0.6 at an f c of 0.85) followed by a more gradual increase.The degree of variability in D L also exhibits a similar increasing pattern as f thin with decreasing f c .A plausible reason for the initial significant increase in cloud heterogeneity after cloud breakup (i.e., f c < 1 in Figure 3b) could be the shift in cloud regime from Sc-dominated overcast scenes, which are characterized by a homogeneous cloud distribution (Wood, 2006), to a relatively heterogeneous Cu-dominated regime (Goren et al., 2023;Zheng et al., 2018).This is also evident in the MODIS true-color example snapshots depicted in Figure 3a, which shows that scenes with a lower f thin appear more clustered and uniform than those with a higher f thin at all f c .The results are consistent with the observations of McCoy et al. (2023), which report a higher fraction of thin clouds in disorganised mesoscale convective cells compared to relatively more organized open and closed cells.

Cloud Properties in F c -f thin Space
In this section, we investigate the variations in cloud properties in f c -f thin space.The analysis is limited to cases with f c > 0.45 because of three primary reasons: (a) there are potential retrieval issues related to cloud properties in scenes with low f c and high f thin , arising from high sub-pixel heterogeneity, an increased occurrence of partially filled cloudy pixels, and an increased likelihood of pixels corresponding to failed retrievals (Cho et al., 2015; Figures 4a-4d) depicts the mean cloud properties in f c -f thin space.We observe an increase in R e with f thin at a given cloud cover across the entire cloud cover range (Figure 4a).R e values are typically larger than 14 μm for broken clouds and can increase up to 22 μm at high f thin , implying that most cloudy scenes associated with high f thin are precipitating.These results are consistent with the in-situ findings of Wood et al. (2018), which report a volume median radii of thin veil clouds ranging between 15 and 30 μm. τ, on the other hand (Figure 4b), decreases with increasing f thin at a given cloud cover.Such an inverse correlation is expected, given the definition of optically thin clouds as those that have a τ ≤ 3.
The relationship between L and f thin at a fixed f c (Figure 4c) exhibits more complexity.For f c < 0.8, L first increases with f thin , attains a maximum, and then decreases with a further increase in f thin .This non-monotonic variation is contrary to what one would expect since thin clouds typically have low L (approximately 25 g m 2 ) (Wood et al., 2018), and their addition at fixed f c should reduce the mean L. A possible explanation of the non-linear relationship between L and f thin for f c < 0.8 relies on the results of Goren et al. (2023), who demonstrated that the deeper the cloud cores, the thinner their anvils.These observations also aligns with our results of increase in cloud heterogeneity with f thin (as discussed previously in Section 3.2).Following these results, the contribution of deepening cloud cores overtakes the accompanying increase in f thin dominating the scene mean L at low f thin values, explaining the initial increase in the scene mean L with f thin .At high f thin (ranging from 0.4 to 0.65 depending on f c ; see Fig. S3 in Supporting Information S2), the presence of thin anvils likely becomes predominant, leading to a decrease in the scene-mean L with f thin .At higher cloud covers ( f c > 0.8), L is more spatially homogenous (Figure 3b), and the increase of f thin has a dominant effect leading to a decrease in L.
The CTT is a crucial parameter that governs the upward thermal emissions of the cloud.Figure 4d depicts the core

M and SST in F c -f thin Space
Using long-term monthly mean cloud and meteorological data, O, Wood and Tseng (2018) identified a statistically significant positive correlation between BLH and f thin within most SCT regions, except for the SEP region (our study's focus).Here, we use the M index as a proxy for BLH, given its strong correlation with BLH (McCoy et al., 2023;Naud et al., 2018Naud et al., , 2020)), to study its relationship with f thin values.Figures 4e and 4f illustrate the variations in M index in the f c -f thin space.The parameter ΔM in Figure 4f is defined within each f thin bin at a given f c as the difference from the mean M at that f c .We observe a statistically significant positive correlation between f thin and M, with a Pearson correlation coefficient (ρ) of 0.46 (refer to Fig. S4a in in Supporting Information S2), which was not identified in O, Wood and Tseng (2018) for the SEP region.Additionally, this positive correlation remains consistent even at a constant cloud cover (Figure 4f), with ρ values varying between 0.26 and 0.4, depending on the cloud cover (see Fig. S4c in Supporting Information S2).
In a Cu-dominated regime, the variability of Cu cloud thickness is primarily governed by the BLH since the Cu base height tends to exhibit less variations compared to the BLH (Daub & Lareau, 2022;Lareau et al., 2018;Lenhardt et al., 2024;Wood, 2006).Consequently, the positive correlation between f thin and M (and thus BLH) supports the earlier-discussed positive association between deeper Cu clouds and a high fraction of thin clouds at a given cloud cover (see Sect. 3.3).
The variation of SST in the f c -f thin space is shown in Figures 4g and 4h.Overall, f thin shows a statistically significant positive correlation with SST with a ρ of 0.28 (Fig. S4b in Supporting Information S2), which is less pronounced compared to M. Also, the variation in SST with f thin at a given f c is minimal (see Figure 4h), with the correlation coefficient being somewhat significant only at overcast f c (ρ of 0.21; see Fig. S4c in Supporting Information S2).
From the collective results of Figures 4e and 4g, we can infer that overcast f c with the lowest f thin are associated with the lowest SSTs and M values.As the M parameter is correlated to BLH (Naud et al., 2020), this suggests that shallow boundary layers and colder SSTs, which are usually favorable for the development of overcast Sc decks, are not favorable for the formation of thin clouds.Consequently, warmer SSTs and higher M values, conducive to the development of deeper Cu clouds, are associated with lower f c and higher f thin .These results are consistent with the in-situ findings of Wood et al. (2018) and the hypothesis of O, Wood and Bretherton (2018).

Effects of f thin on CRE
Optically thin clouds, due to their semi-transparent nature, may interact differently with SW and LW radiation compared to an equivalent amount of optically thick clouds.Results from Figures 2 and 3 highlight that the occurrence of thin clouds is strongly dependent on the cloud cover.Overcast Sc-dominated cloud covers have the least occurrence of thin clouds, which increases drastically as the cloud cover decreases below unity, where heterogeneous Cu clouds are dominant (see Sections 3.2 and 3.4).These variations in the prevalence of thin clouds across different cloud covers may lead to a corresponding cloud cover dependence of thin cloud impacts on CRE.This may have significant implications for a future warmer climate, in which studies indicate a faster Sc-Cu transition due cleaner boundary layer (Goren et al., 2022), or changes in cloud morphologies to those associated with greater amounts of thin clouds throughout the cloud cover range due to environmental cloud controlling factors (Chemke & Polvani, 2021;McCoy et al., 2023).In the following section, we quantify this cloud cover dependence on f thin -CRE relationship by separating the analysis into changes in the SW and LW components of the CRE.

Effects of f thin on SW CRE
SW CRE is primarily governed by the changes in total cloud cover and cloud optical thickness (King, 1987;Platnick & Twomey, 1994;Twomey, 1977;Zelinka et al., 2012Zelinka et al., , 2016)).Since the presence of thin clouds at a given cloud cover reduces the scene cloud optical thickness (see Figure 4b), they can significantly reduce the SW albedo and therefore the magnitude of SW CRE (Goren et al., 2023;McCoy et al., 2017McCoy et al., , 2023)).Figure 5a shows the CRE at the TOA for SW radiation (CRE SW,TOA ) in f c -f thin space.It can be seen that the CRE SW,TOA reduces significantly with a decrease in f c , from an average of about 321.2 W m 2 at overcast conditions to about 69.4 W m 2 at f c = 0.5.The figure also demonstrates that the cooling extent of CRE SW,TOA at a given cloud cover shows a substantial reduction with an increase in f thin .ΔCRE SW,TOA in Figure 5b within each f thin bin at a given f c represents the percentage change in CRE SW,TOA from the mean CRE SW,TOA at that f c .For overcast scenes, the reduction in SW cooling due to an increase in f thin is approximately 46 % over the entire range of f thin .This can also be seen in the example case presented in Figure 3a (bottom row), where the increase in f thin dims the overcast scene.Furthermore, the relative extent of cloud dimming increases with decreasing f c , reaching about 65 % at an f c of 0.55.This increase in the percentage of cloud dimming due to thin clouds with a decrease in f c may be attributed to the high amounts and extent of variation in f thin , which are minimum for Sc-dominated overcast cloud regime and increases coherently with scene heterogeneity as f c decreases in the Cu-dominated regime.Such a wide range of possible variations in CRE SW,TOA , even at fixed cloud cover, highlights the dominant role of thin clouds in controlling the SW radiative effect (Goren et al., 2023;McCoy et al., 2017McCoy et al., , 2023)).

Effects of f thin on LW CRE
For optically thick clouds, the CRE LW,TOA is primarily controlled by CTT, assuming that their emissivity is approximately 1 (Vaillant de Guélis, Chepfer, Noel, Guzman, Dubuisson, et al., 2017).However, for partiallytransparent optically thin clouds, the CRE LW,TOA depends on the combination of thin-cloud radiative emission and the fraction of surface thermal emission that passes through the thin clouds (Arouf et al., 2022;Vaillant de Guélis, Chepfer, Noel, Guzman, Dubuisson, et al., 2017;Vaillant de Guélis, Chepfer, Noel, Guzman, Winker, et al., 2017).The thermal emission of thin clouds depends on the CTT and emissivity, and is relatively lower compared to that of optically thick clouds with the same CTT.The amount of surface thermal emission transmitted through the thin clouds depends on the cloud transmissivity, SST, and surface emissivity.Optically thick clouds are opaque to surface emission, and do not easily transmit the surface emission (Vaillant de Guélis, Chepfer, Noel, Guzman, Dubuisson, et al., 2017).More details regarding the outgoing LW flux of a cloudy scene composed of optically thin and thick clouds are provided in Text S1 in Supporting Information S1.CRE LW,TOA variations in f c -f thin space is depicted in Figure 5c.Similar to SW, the CRE LW,TOA reduces significantly with a decrease in cloud cover, from an average of about 14.1 W m 2 under overcast conditions to about 7.2 W m 2 at f c = 0.5.Figure 5d illustrates the percentage change in the warming extent of CRE LW,TOA from the mean for various f c bins due to a change in f thin .The figure shows a slight initial increase in CRE LW,TOA with f thin for overcast scenes.At such high f c that are associated with a low fraction of thin clouds, optically-thick opaque clouds are likely the primary contributors to CRE LW,TOA .As shown in Figure 4d, the CTT at a given cloud cover decreases with increasing f thin , implying that less LW radiation is emitted out into space, which may explain the slight increase in CRE LW,TOA .
For f c < 0.9, LW CRE decreases with f thin at a given f c (Figure 5d).Since these f c are associated with relatively higher f thin amounts (between 0.2 and 0.8; see Figure 3b), thin clouds likely exert significant control over CRE LW, TOA via LW transmission through the clouds.Increasing f thin at a given f c leads to higher transmissivity of the scene, allowing more thermal emission from the ocean surface to escape through the thin clouds into space.As a result, CRE LW,TOA decreases with f thin at a given f c .The variability in LW warming at a given f c due to changes in f thin can range from approximately 12 % for nearly overcast scenes to as high as 52 % at an f c of about 0.6 (Figure 5d).Note that these percentage variations are estimated separately for each f c bin, and the absolute magnitude of CRE LW,TOA variations is relatively lower at low f c compared to high f c .These results suggest a notable role of thin clouds in modulating the CRE LW,TOA of shallow marine clouds in the SCT region, particularly in the Cu-dominated regime.

f thin and the Balance Between LW and SW CRE
We also assess the balance between the instantaneous positive LW CRE (warming) and negative SW CRE (cooling).We find an average CRE LW,TOA of 9.8 W m 2 , which is approximately 7.1 % of the average CRE SW, TOA (mean of 138.2 W m 2 ).We also investigate how f thin modulates the balance between the LW warming and SW cooling.Figure 5e shows the percentage of SW cooling compensated by LW warming calculated as 100 × CRE LW,TOA |CRE SW,TOA | (hereafter termed LW compensation) in f c -f thin space.At a given f c , the LW compensation increases with f thin at all f c .At high f thin , this compensation can range from 8 % for overcast scenes to 19 % at a cloud cover of 0.5.For the considered range of cloud cover in Figure 5e, about 13 % of the cloudy scenes correspond to a LW compensation of 15 % or more.This percentage of scenes increases to 21 % for a minimum LW compensation of 12 %.While not as prominent as in tropical anvil clouds (Hartmann & Berry, 2017), the instantaneous LW compensation due to thin clouds can nevertheless be non-negligible, depending on their lifetime.Therefore, future research to quantify the longevity of thin veil clouds is required to accurately assess their overall contribution to the radiative effects of low marine clouds.

f thin , Cloud Morphology, and Cloud Albedo Predictability
To identify the relative influence of key cloud parameters on A c , we have constructed multiple linear regression models to predict A c using three sets of predictors: (a) only f c ; (b) f c , R e , and L; and (c) f c , R e , L, and f thin .Note that the model assumes linearity between each of the cloud properties and A c .The advantages of this assumption lie in identifying the key factors that control albedo variability, due to which it has been employed in multiple studies on cloud-radiation interaction (for instance, McCoy et al. (2023); Wall et al. (2023)).Figure 6a illustrates and compares the performance of these models in predicting A c .We find that f c alone explains 69 % of the variability in A c , as indicated by the R-squared value.Including R e and L in the regression improves the explained variance to 81 %.Adding f thin to the regression model further improves the explained variance to approximately 89 %.
We further repeat a similar regression analysis for cloud cover intervals of 2.5 % ranging from 45 % to 100 %. Figure 6b illustrates the amount of variance in A c explained by f thin and combined R e and L at those f c intervals.The figure shows that R e and L together explain about 64 % of the variance in A c for overcast scenes, which reduces to about 20 % at a cloud cover of 0.45.In contrast, f thin explains about 56 % of the A c variability for overcast scenes, which comes down to about 39 at low f c .For overcast scenes that are dominated by homogeneous cloud morphology (see Figures 3b and 3c) and are usually associated with less variations in f thin , R e and L explain greater amounts of variance in A c than f thin .However, as the cloud cover decreases and cloud heterogeneity increases, the variance in A c explained by f thin becomes larger than that of R e and L.
f thin is associated with the distribution of L within a scene (see Figure 3b).An increase in f thin at a given cloud cover indicates a relative reduction in the horizontal extent of cloud water corresponding to optically thick clouds, which are comparatively more efficient at reflecting the incoming solar radiation back into space.Therefore, f thin at a given cloud cover can be considered a geometrical constraint that governs the interaction between L and insolation.This impact is shown in Figure 7, which shows the relationship between A c and L for various f thin intervals for cloud cover ranging from 0.7 to 0.8.The figure demonstrates that f thin regulates the sensitivity of A c to variations in L, with higher sensitivity (indicated by a steeper linear slope) observed at low f thin amounts.An increase in f thin leads to a decrease in L-A c sensitivity.Since thin clouds restrict the extent of optically thick cloud water at a given cloud cover, high amounts of f thin within a scene limit the increase in L to a smaller horizontal extent, potentially explaining the low L-A c sensitivity.
The morphological constraint imposed by f thin is expected to be more pronounced at low cloud cover, given that the variability in f thin increases as cloud cover decreases below 1.This could account for the relatively higher proportion of variance in A c explained by f thin at lower cloud cover (Figure 6b), thereby contributing to an improved R-squared value when f thin is incorporated into the multiple linear regression model alongside L and R e (Figure 6a).According to Bender et al. (2016), the primary source of variability in A c in subtropical marine clouds, at a given cloud cover, originates from variations in L. However, as shown in Figure 7, it is evident that even with constant cloud cover and L, a change in f thin can modulate A c by 63 %-69 %, depending on L, highlighting the significant influence of f thin (and thus cloud morphology) on A c .

Discussion
Climate models indicate a weakening of subtropical subsidence and an increase in SST in response to global warming (Chemke & Polvani, 2021;Webb et al., 2013).These changes are likely to result in a deeper boundary layer, which, based on our findings and previously established correlations (O et al., 2018), may lead to a higher occurrence of optically thin clouds throughout the entire range of cloud cover in the SCT regions (McCoy et al., 2023).Model evaluation studies have highlighted the underestimation of cloud cover and the overestimation of cloud brightness in the tropics, often referred to as the "too few, too bright" bias (Konsta et al., 2022;Nam et al., 2012), which may occur due to an inaccurate representation of thin clouds in simulations (Konsta et al., 2022).Our findings, which elucidate the significant impact of thin clouds on the SW and LW components of CRE, further embolden the necessity for their accurate representation in climate models.
Recently, McCoy et al. (2023) reported a global positive SW cloud feedback between 0.04 and 0.07 W m 2 K 1 due to a shift in mesoscale low cloud morphology, driven by changes in optically thin clouds (McCoy et al., 2023).Our results further show that a change in thin cloud fraction also impacts the LW component of CRE.This LW CRE can subdue about 8 %-19 % of the SW CRE in the SCT region.It is essential to note that the LW contributions reported in our study are based on instantaneous CERES measurements and are likely to compensate for larger amounts of SW CRE when viewed at a daily resolution, considering the negligible SW CRE at night (Allan, 2011).Also, we do not explicitly consider the LW contribution from the cloud-twilight zone present at the vicinity of cloud boundaries (Eytan et al., 2020), that can be considered as an extension to the veil clouds.The LW contribution from this zone is expected to further add to the total LW CRE.These factors highlight the importance of quantifying the LW component in cloud morphology feedback.
Although our study does not consider the influence of aerosols, they can influence the occurrence of thin clouds.Studies have shown that an increase in aerosol concentration can postpone cloud breakup in SCT regions by delaying the onset of precipitation (Christensen et al., 2020;Goren et al., 2019Goren et al., , 2022)).Our results show a prevalence of thin clouds after cloud breakup.Therefore, in a cleaner climate, where the overcast clouds may precipitate and breakup sooner, the occurrence of thin clouds may increase substantially.Thin clouds can also modulate the aerosol-cloud interactions.Studies have found that, for a given scene-mean cloud albedo, homogeneous cloudy scenes display a larger albedo enhancement due to aerosols (Feingold et al., 2022;Goren et al., 2023).Given the intrinsic positive association between thin clouds and cloud heterogeneity (Figure 3b), high occurrence of thin clouds may dampen the susceptibility of the albedo to anthropogenic aerosols, and thus, the negative radiative forcing due to aerosol-cloud interactions.

Summary
In-situ and spaceborne studies (Leahy et al., 2012;Wood et al., 2018) have highlighted the significant presence of thin clouds in major SCT regions.In this study, we use simultaneous MODIS, CERES, and re-analysis data to study thin clouds over the Southeast Pacific Ocean.We find a non-monotonic relationship between f c,thin and f c (Figure 2).Sc-dominated overcast scenes are associated with the least f c,thin .As f c decreases, f c,thin first increases steeply, reaches a maxima at f c = 0.85, and thereafter decreases.An increase in f thin (= f c,thin f c ) is accompanied by an increase in the scene cloud-water heterogeneity (Figure 3b) and the occurrence of precipitating scenes (Figures 2  and 4a), indicating a transition to Cu-dominated regime.The cloud regime dependence of f thin is also observed in terms of SST and M. Both M and SST show positive correlations with f thin .The minimum of f thin is associated with minimum values of M and cold SST, conditions favorable for overcast clouds near the coastal region (McCoy et al., 2023).The maximum of f thin is associated with high values of M and warm SSTs.The positive correlation is more significant in the case of M (ρ = 0.46) than SST (ρ = 0.29).Furthermore, the correlation between M and f thin remains consistent even with constant cloud cover, throughout the cloud cover range, suggesting that the depth of the boundary layer exerts significant control over f thin during all stages of SCT.
f thin significantly influences SW CRE at all f c .At a given f c , increase in f thin leads to a substantial reduction in SW CRE by approximately 46 % for overcast scenes and about 65 % at 50 % f c , across the entire f thin range.Although Bender et al. (2016) showed that albedo variations at a given f c are mainly controlled by L, our findings indicate that variations of approximately 63%-69 % in albedo can occur even with constant f c and L, owing to changes in f thin .We further use multiple linear regressions to identify key parameters that influence cloud albedo.We show that integrating f thin in the regression that already incorporates L, R e , and f c as predictors, improves the explained variance in cloud albedo from 81 % to 89 %.Furthermore, at a given non-overcast cloud cover, f thin explains greater variability in cloud albedo than the combined effect of L and R e .These findings underscore the crucial role of thin clouds in regulating the SW cooling of clouds in SCT region.
The LW CRE is found to be non-negligible for low clouds in the SCT region.We find that at a given cloud cover, the LW warming increases with f thin for overcast scenes, while for other f c , it decreases with f thin .This variation likely results from the balance between the thermal emission by opaque clouds near the cloud top (which depends on CTT) and the portion of transmitted surface LW emission through the thin clouds.The degree of variation in LW CRE due to f thin is noteworthy, with an approximately 12 % increase observed for near-overcast scenes and a substantial 52 % decrease at 50 % cloud cover.
The positive LW CRE is also shown to offset a portion of SW CRE.This LW compensation is shown to increase with f thin .For overcast scenes, the relative increase in LW compensation with f thin is minimal, at about 8 % due to low occurrence of thin clouds.However, it increases to approximately 19 % at a cloud cover of 50 %.Approximately 13 % of the total cloudy scenes within the study region correspond to a LW compensation of 15 % or more.These results imply a non-negligible contribution of the LW component to the total CRE, specifically for scenarios with high f thin .
This study aims to enhance our understanding of thin clouds and their influence on cloud radiative effects.Our results demonstrate that thin clouds are intrinsic characteristics of the Sc-Cu transition and play a significant role in modulating the cloud-radiation interaction.Consequently, this research emphasizes the need for a more comprehensive representation of cloud morphology in climate models (Konsta et al., 2022;Nam et al., 2012).

Figure 1 .
Figure 1.Defining thin clouds using optical thickness (τ) and reflectance (R).Panel (a) depicts a MODIS true-color image, highlighting the study domain and a zoomed-in 1°by 1°scene, whose properties are shown in the remaining panels.Panels (b) and (c) show R at 0.64 μm and τ, respectively.Black pixels in Panel (d) represent the thin clouds identified using R and τ.Panel (e) shows the thin cloud pixels identified using only τ.The location and time of the observation are provided at the bottom of the figure.(Panel (a) credits: https://worldview.earthdata.nasa.gov/).

Figure 2 .
Figure 2. Joint histogram of cloud cover ( f c ) and thin cloud cover ( f c,thin ) depicting the conditional probabilities of the occurrence of f c,thin for the range of f c .A three-segmented linear regression following Bogartz (1968) is fitted to the f c,thin values corresponding to a maximum probability in each f c bin to highlight the non-monotonic relationship between f c and f c,thin .m represents the slope of the linear fit and ρ is the correlation coefficient.The filled circles at the top indicate the fraction of scenes with a cloud effective radius (R e ) greater than 14 μm.Subpanels at the top and right represent the distribution of the data with respect to f c and f c,thin , respectively.

Figure 3 .
Figure 3. (a) Example snapshots of MODIS true-color 1°by 1°scenes illustrating the variation in the fraction of thin cloud cover ( f thin ) at different cloud cover ( f c ). Panel (b) depicts the variation in the relative dispersion of liquid water path (D L ) with f thin at different f c .

Figure 4 .
Figure 4. Scene mean cloud and meteorological properties in f c -f thin space depicting cloud effective radius R e (a), cloud optical thickness τ (b), liquid water path L (c), core cloud top temperature CTT (d), marine cold-air outbreak parameter M (e), and sea surface temperature SST (g).Panels (f) and (h) represent the deviation of M (ΔM) and SST (ΔSST) from their corresponding means in each f c bin, respectively.

Figure 5 .
Figure 5. Scene mean CRE (W m 2 ) at the top of the atmosphere in f c -f thin space for SW (a) and LW (c) radiation.Panels (b) and (d) represent their percentage deviation from the mean in each f c bin, respectively.Panel (e) depicts the percentage of SW cooling compensated by LW warming.

Figure 6 .
Figure 6.(a) Predicting A c using multiple linear regression (MLR) with the predictors: f c (magenta); f c , R e , and L (blue); f c , R e , L, and f thin (red).The legend provides the R-squared and p-value of the regressions.(b) Variations in A c explained by different cloud properties at a given cloud cover ( f c ).The median and 90th percentile ranges of A c are indicated by the black dashed line and the grey polygon, respectively.At a given f c , the width of the yellow polygon represents the R-squared value for the linear regression between f thin and A c scaled to the 90th percentile range of A c , indicating the fraction of variance in A c explained by f thin .Similarly, the magenta polygon indicates the fraction of variance in A c explained by combined cloud effective radius (R e ) and liquid water path (L).

Figure 7 .
Figure 7. Variation of cloud albedo (A c ) with liquid water path (L) for different intervals of f thin for cloud cover ( f c ) between 0.7 and 0.8.m is 1,000 times the slope of the linear fit.Each f thin interval consists of 17,763 cloudy scenes.