The SCS is one of the world's largest marginal seas, characterized by complex air–sea interactions and a pronounced seasonal reversal of wind systems (Wang et al., 2009). This region is strongly influenced by the Asian monsoon circulation, exhibiting distinct southwest and northeast monsoon regimes that profoundly modulate its thermodynamic and dynamic environments (Zheng et al., 2025). These alternating monsoon circulations govern the regional aerosol loading (Ou et al., 2025), boundary-layer structure (Chen et al., 2025; Peng et al., 2016; Wang et al., 2024a), and cloud microphysical processes (Miller et al., 2023; Wang et al., 2024b), making the SCS an ideal natural laboratory for investigating ACI under contrasting meteorological conditions. In this study, the analysis domain (blue polygon in Fig. 1) encompasses the entire SCS, including both coastal and open-ocean areas, to capture the spatial variability of aerosol, cloud, and environmental parameters. The red box in Fig. 1 delineates the subregion (3–22° N, 110–120° E) used for defining the monsoon regimes following Wang et al. (2004).

Cloud retrievals used in this study are derived from the Moderate Resolution Imaging Spectroradiometer (MODIS) aboard the Aqua satellite, which has an equatorial crossing time of approximately 13:30 local time (LT). The MODIS cloud products analyzed are obtained from the Clouds and the Earth's Radiant Energy System (CERES) MODIS (CERES–MODIS hereafter) Edition 4 (Minnis et al., 2011a, 2021) Single Scanner Footprint (SSF) daily products (Level 3, 1° × 1° grid resolution), which provide physically consistent cloud and radiative properties based on synergistic MODIS and CERES observations. The CERES–MODIS cloud retrievals have been extensively validated (Minnis et al., 2011b; Yost et al., 2021) and have been widely employed in studies of ACI (Jia et al., 2021; Painemal, 2018). CERES–MODIS Edition 4 SSF cloud parameters utilized here include cloud optical depth (τ), cloud effective radius (CER), cloud-top temperature (CTT), cloud-top pressure (CTP), liquid cloud area fraction (LCAF), and liquid water path (LWP). CER is retrieved from the 3.7 µm channel, which is less affected by retrieval biases than its 2.1 and 1.6 µm counterparts (Grosvenor et al., 2018).

Nd used in this study is obtained from the community-standard gridded dataset of Gryspeerdt et al. (2022a), which provides 1° × 1° Level-3 Nd products derived from pixel-level MODIS Collection 6.1 retrievals. In this dataset, Nd is first estimated at the native MODIS pixel scale using retrieved cloud optical thickness and cloud effective radius, and then aggregated to a common grid using established sampling strategies to ensure robustness and consistency. The Nd retrievals are based on the adiabatic cloud assumption and are subject to strict quality control procedures, including screening for optically thin clouds, large solar zenith angle and viewing zenith angle conditions, sub-pixel heterogeneity, and potential retrieval contamination. Only single-layer liquid cloud scenes are retained. This dataset has been evaluated against observations and is widely used in aerosol–cloud interaction studies (e.g., Jia et al., 2024; Wall et al., 2023). In this study, we use the Nd_G18_37 product from this community-standard dataset.

Aerosol properties used in this study are obtained from the Modern-Era Retrospective Analysis for Research and Applications, Version 2 (MERRA-2; Gelaro et al., 2017), which assimilates a wide range of satellite observations to provide a physically consistent representation of global aerosol distributions. The MERRA-2 aerosol dataset used here includes the total aerosol extinction (AOD, 550 nm) and the total aerosol Ångström parameter (AE, 470–870 nm), with a spatial resolution of 0.5° × 0.625° and a temporal resolution of 1 h. To ensure spatiotemporal consistency with the CERES–MODIS cloud products, the MERRA-2 aerosol fields at 14:00 local solar time (LST), closest to the Aqua overpass (∼ 13:30 LST), were regridded to a 1° × 1° grid. AOD represents the column-integrated aerosol extinction, whereas AE characterizes the wavelength dependence of AOD and is commonly used as an indicator of aerosol particle size. A higher AE generally indicates dominance of fine-mode aerosols, while a lower AE suggests coarse-mode particles. The aerosol index (AI) is defined as the product of AOD and AE: 3AI=AOD×AE AI, in comparison to AOD, is considered a better parameter for representing aerosols in ACI studies, because it incorporates information on aerosol particle size, which is critical for cloud droplet activation and microphysical properties (Ma et al., 2018b; Nakajima et al., 2001). Therefore, AI is employed in this study as the aerosol proxy in the calculation of ACI.

Atmospheric fields were obtained from the fifth-generation ECMWF reanalysis (ERA5; Hersbach et al., 2020). ERA5 assimilates a comprehensive suite of ground-based and satellite observations through a state-of-the-art four-dimensional variational data assimilation system, offering physically consistent and dynamically balanced representations of the atmosphere. The dataset used here includes three-dimensional fields of specific humidity and horizontal wind components (U and V) at all standard pressure levels, temperature at 1000 and 700 hPa, and mean sea level pressure. The daily mean specific humidity and wind speed at 850 hPa were used to distinguish the three periods over the SCS (Wang et al., 2004). Wind and humidity fields at all pressure levels were analyzed to characterize the large-scale atmospheric circulation and vertical moisture structure during these periods. The 1000 hPa specific humidity was employed as a proxy for the ambient water vapor available to warm clouds (Dadashazar et al., 2020). Temperatures at 1000 and 700 hPa were used to compute the LTS, which quantifies the thermodynamic stability of the lower atmosphere (Klein and Hartmann, 1993), as follows: 4LTS=θ700hPa-θ1000hPaθ=TP00PRCp where θ700hPa and θ1000hPa denote the potential temperatures at 700 and 1000 hPa, respectively, P00 is the standard reference pressure (typically 1000 hPa), P is the pressure at a given level, R is the gas constant for dry air, and Cp is the specific heat capacity of dry air at constant pressure. To ensure spatiotemporal consistency with the aerosol and cloud parameters, both the 1000 hPa specific humidity and LTS at 14:00 LST were regridded to a 1° × 1° grid. Sea surface temperature (SST) was obtained from the National Oceanic and Atmospheric Administration (NOAA) Optimum Interpolation SST, version 2 (OI SST v2, Reynolds et al., 2002), which incorporates both in-situ and satellite observations and provides monthly fields at 1° × 1° resolution.

Precipitation data were obtained from the Integrated Multi-satellite Retrievals for GPM (IMERG) Version 07 Final Run. IMERG is the flagship precipitation product of the Global Precipitation Measurement (GPM) mission, a collaborative effort between NASA and JAXA (Huffman et al., 2023). The IMERG algorithm calibrates, merges, and interpolates precipitation estimates from the constellation of Low Earth Orbit (LEO) passive microwave (PMW) radiometers onboard GPM satellites. These estimates are subsequently integrated with geostationary satellite infrared observations, particularly in regions with sparse PMW coverage, to produce a global precipitation product with 0.1° spatial and 30 min temporal resolutions (Huffman et al., 2020; Watters et al., 2021). IMERG precipitation data, owing to its global coverage and high spatiotemporal resolution, have been widely adopted by the research community (Dezfuli et al., 2017; Durden, 2024; Hayden et al., 2023; Tan et al., 2019a; Watters et al., 2021; Watters and Battaglia, 2019; Zhang and Wang, 2024; Zhu et al., 2024). Consequently, IMERG V07 Final Run PrecipitationCal, the gauge-calibrated multi-satellite product, is used here to determine whether precipitation occurred within each CERES–MODIS grid cell.

To minimize the influence of satellite retrieval biases when investigating ACI, a rigorous quality-control procedure was applied to the CERES–MODIS cloud dataset following the method of Saponaro et al. (2017). The selection criteria were as follows:

Only liquid-phase warm clouds were retained, identified by a cloud-phase flag of “liquid” and CTT > 273 K.

ACI are strongly modulated by environmental factors such as humidity, vertical wind velocity, and atmospheric stability (Zhu et al., 2022). Therefore, considering the prevailing atmospheric conditions is crucial when examining their variability. The atmospheric environment over the SCS exhibits distinct characteristics under the influence of the southwest monsoon and the northeast monsoon. A distinctive feature of the SCS summer monsoon is its nearly simultaneous onset across a broad latitudinal range (3–22° N) (Wang et al., 2004). On top of the previous studies that defined monsoon periods over the SCS based on wind direction (Wang et al., 2004, 2009), we further incorporated precipitation and specific humidity within this 3–22° N band (the red box in Fig. 1) to classify the study periods into three regimes: the southwest monsoon wet period (SWMW), the northeast monsoon dry period (NEMD), and the northeast monsoon wet period (NEMW). As shown in Fig. 3, the SCS is dominated by the southwest monsoon from May to August, during which specific humidity reaches its maximum and precipitation is strongest. From October to December, the northeast monsoon prevails, accompanied by relatively high specific humidity and intense precipitation. During January to March, the SCS remains under the influence of the northeast monsoon, but specific humidity is at its lowest and precipitation is minimal, representing a dry period. Accordingly, we defined May–August as SWMW, October–December as NEMW, and January–March as NEMD. These three periods not only reflect significant differences in atmospheric circulation and moisture conditions, along with aerosol regimes over the SCS, but also provide distinct environmental backgrounds for ACI, which may influence their characteristics and intensity.

Twomey (1977) proposed that atmospheric aerosol particles can act as cloud condensation nuclei, such that an increase in aerosol loading leads to a higher cloud droplet number concentration. Under a nearly constant liquid water content, this results in smaller cloud droplet effective radius. To examine the Twomey effect during the three periods, the 25th and 75th percentiles of the AI were used to define clean (AI < 25th percentile) and polluted (AI > 75th percentile) conditions, respectively. For each condition, two-dimensional probability density distributions were calculated as a function of CER and LWP. The difference in the two-dimensional probability densities between polluted and clean conditions therefore illustrates how CER varies with aerosol loading under quasi-constant LWP. As shown in Fig. 7, CER is mostly smaller than 15 µm, justifying the effectiveness of the filtering of non-raining cases. Under clean conditions, the CER of warm clouds increases with LWP across all three periods. Under polluted conditions, however, CER initially increases with LWP and then tends to level off during the SWMW and NEMW periods. During the NEMD period, CER exhibits an increase followed by a decrease with increasing LWP, and subsequently remains nearly constant. The difference plots between polluted and clean conditions clearly demonstrate that polluted samples are more concentrated toward smaller CER values compared with clean samples in each quasi-constant LWP bin. This consistent shift toward smaller droplet sizes under higher aerosol loading demonstrates a clear manifestation of the Twomey effect during all three periods.

As shown in Fig. 6, the warm-cloud fraction during the Southwest Monsoon Wet period is approximately 45 %, while it increases to about 77 % during the Northeast Monsoon Wet period and further to about 80 % during the Northeast Monsoon Dry period. This substantial difference suggests that the warm-cloud populations sampled during different monsoon periods are fundamentally distinct and likely reflect different cloud dynamical regimes. To better distinguish cloud morphological types within each monsoon period and reduce potential regime-mixing effects, we further classified warm clouds over the SCS according to CTP. Specifically, warm clouds were separated into shallow stratocumulus (CTP: 800–950 hPa) and deeper cumulus (CTP: 650–800 hPa) clouds. The ACI index was then quantified separately for these two cloud regimes.

According to the assumption of the Twomey effect (Twomey, 1977), an essential prerequisite for investigating the aerosol indirect effect based on the CER–AI relationship is to keep the LWP constant. Since CER is a function of both LWP and AI, and generally increases with LWP, variations in LWP associated with changes in aerosols can in turn modulate the CER–AI relationship. Therefore, when applying the CER–AI relationship to analyze the first aerosol indirect effect, it is essential to constrain LWP to ensure its constancy. To satisfy this requirement while maintaining sufficient sample sizes, LWP was binned at 10 g m⁻² intervals, within which ACI_r was examined separately for shallow stratocumulus (Fig. 8a–b) and deeper cumulus clouds (Fig. 8e–f) during the three periods. The results show that for both shallow stratocumulus and deeper cumulus clouds, under both all warm-cloud conditions and non-raining warm-cloud conditions, ACI_r is generally weak when LWP < 50 g m⁻², and even exhibits the anti-Twomey effect (i.e., an increase in CER with increasing aerosol loading). Clouds in this LWP regime are typically very thin or broken, as well as post-precipitation remnants (McComiskey et al., 2009). A similar phenomenon is observed over the northern Indian Ocean, which may be attributed to the intense competition for available water vapor under high aerosol concentrations, combined with the entrainment of dry air at cloud tops (Jose et al., 2020). When LWP > 50 g m⁻², the ACI for all cloud regimes across the three periods are consistent with the Twomey effect. An exception occurs in Fig. 8b and d, where the ACI at LWP = 120 g m⁻² during the southwest monsoon is negative. However, this result is not statistically robust due to the limited sample size.

Precipitation formation efficiently reduces cloud droplet number concentration and scavenges aerosols from clouds (Gryspeerdt et al., 2015), introducing a sink of that does not reflect the Twomey effect (Jia et al., 2022). Therefore, when analyzing the aerosol first indirect effect in warm clouds, the influence of precipitation should be separated in order to accurately quantify the sensitivity of CER to aerosols. Figure 8 shows that, for both shallow stratocumulus and deeper cumulus clouds over the SCS, the ACI_r values for all warm clouds are consistently larger than those for non-raining warm clouds across all three periods when LWP exceeds approximately 80 g m⁻². This indicates that the inclusion of raining samples amplifies ACI_r, a phenomenon also identified in the AI/AOD–Nd relationship by Jia et al. (2022) and Painemal et al. (2020). But this amplification is just an artifact governed by the joint impacts of the suppression of precipitation by aerosols and the aerosol removal by precipitation (Jia et al., 2022). Therefore, after removing precipitating clouds from all warm-cloud samples, the ACI_r obtained from non-raining warm clouds provides a more realistic representation. Nevertheless, the possibility of undetected light drizzle under high-LWP conditions may introduce additional uncertainty in the derived ACI estimates, despite precipitation screening using IMERG.

Figure 8b shows that, under non-raining warm-cloud conditions with LWP > 50 g m⁻², the period-to-period variations in ACI_r differ between shallow stratocumulus and deeper cumulus clouds over the SCS. For shallow stratocumulus, ACI_r values are broadly comparable across the three periods, with no significant differences (Fig. 8b). In contrast, deeper cumulus clouds exhibit clear differences among the three periods (Fig. 8f), with ACI_r generally strongest during the NEMD period, while the NEMW and SWMW periods show comparatively weaker values, with the NEMW period exceeding the SWMW period in some LWP bins. For LWP > 140 g m⁻², the ACI_r during the southwest monsoon exhibits large fluctuations due to the limited number of samples, even exceeding that of the northeast monsoon. The analysis in Sect. 3.2 reveals substantial differences in atmospheric conditions over the SCS among the three periods. During the SWMW period, atmospheric moisture and sea surface temperatures reach their highest levels, and upward motion dominates over the region, while aerosol concentrations remain relatively low. During the NEMW period, moisture and sea surface temperatures are still relatively high, with upward motion primarily confined to the southern areas near the equator, and aerosol concentrations are elevated due to pollution transported from continental China. In contrast, during the NEMD period, atmospheric moisture and sea surface temperatures are at their lowest, subsidence dominates, and aerosol concentrations reach their maximum. These results suggest that although the large-scale environmental statistics differ substantially among the three monsoon periods, the local thermodynamic conditions favorable for shallow stratocumulus formation may be relatively similar across periods. Such cloud-favorable environments may not be fully resolved by the period-mean large-scale statistics presented here, which could explain the broadly comparable ACI_r values for shallow stratocumulus. In contrast, ACI_r for deeper cumulus clouds over the SCS generally tends to strengthen under drier and more stable monsoon environments, with the strongest signals occurring during the NEMD period, while no consistent ordering is observed between the NEMW and SWMW periods across different LWP bins. Accordingly, the subsequent analysis focuses on the period-to-period differences of deeper cumulus clouds ACI and its relationship with variations in the monsoon environmental background.

In addition to the radius-based ACI_r, we further examined the droplet-number susceptibility (ACI_Nd) separately for shallow stratocumulus (Fig. 8c–d) and deeper cumulus clouds (Fig. 8g–h). Consistent with the ACI_r results, ACI_Nd for shallow stratocumulus remains broadly comparable across the three monsoon periods, with no significant differences. Deeper cumulus clouds exhibit clear period-to-period variations, with the strongest signals occurring during the NEMD period, while comparatively weaker values are observed during the SWMW and NEMW periods. The consistency between ACI_r and ACI_Nd suggests that the inferred ACI variations are robust to the choice of metric. However, because the two metrics are not independent, their agreement should be interpreted primarily as a consistency check rather than as independent evidence for the underlying physical mechanisms. Nevertheless, the similar inter-period variations exhibited by both metrics motivate a further examination of the environmental factors associated with these ACI differences. Therefore, Sect. 3.5 explores the potential influences of moisture and LTS on deeper cumulus cloud ACI across the three periods.

The progressive enhancement of ACI_r and ACI_Nd for deeper cumulus clouds from the SWMW to the NEMD period (Fig. 8) may potentially be influenced by the aerosol hygroscopic swelling artifact. The MERRA-2 AOD used to construct AI is calculated under ambient RH, with the extinction coefficients of sulfate, hydrophilic carbonaceous aerosols, and sea salt explicitly parameterized as functions of RH (Randles et al., 2017). Under humid conditions, aerosol hygroscopic growth can increase AOD, and therefore AI, without a corresponding increase in CCN-active particle number. This may artificially flatten the CER–AI and Nd–AI regression slopes, leading to apparently weaker ACI under moister conditions. Given that the SWMW and NEMW periods are characterized by substantially higher moisture levels than the NEMD period, the weaker ACI observed during these moister periods could partly reflect this hygroscopic swelling artifact rather than intrinsic differences in cloud microphysical sensitivity.

To assess whether the observed period-to-period differences are substantially affected by systematic humidity differences, we further stratified the data from each monsoon period into three RH ranges (0 %–45 %, 45 %–80 %, and 80 %–100 %). Under constrained LWP (50–200 g m⁻²) and cloud-top pressure (650–800 hPa) conditions, ACI_Nd was recalculated separately for each subsample. The RH used in this study is from ERA5 at 950 hPa at 06:00 UTC, representing the ambient environmental humidity within the marine atmospheric boundary layer. Previous studies show that aerosols are predominantly confined below ∼ 3 km in the marine atmosphere over the SCS (Li et al., 2020; Su et al., 2022), with a peak in extinction occurring at approximately ∼ 480 m (Su et al., 2022). Therefore, the 950 hPa RH is adopted as a representative proxy for the ambient humidity governing aerosol hygroscopic growth.

As shown in Table 2, the enhancement from SWMW to NEMW and further to NEMD is evident in the moderate and high RH bins (45 %–80 % and 80 %–100 %), whereas the low RH bin (0 %–45 %) does not exhibit statistically robust differences among the three periods. These results suggest that although AI may be affected by hygroscopic swelling under humid environments, this effect is insufficient to explain the systematic ACI differences among the three monsoon periods. Therefore, aerosol hygroscopic swelling is unlikely to be the dominant cause of the observed period-to-period variability, motivating further examination of the thermodynamic and moisture controls on ACI.

To provide an integrated view of how the ACI of deeper cumulus clouds co-varies with the thermodynamic and moisture background across the three periods, Fig. 9 shows the ACI_Nd together with the corresponding q and LTS. Both ACI_Nd and the key environmental regulators display a coherent evolution across the three periods. From the SWMW to the NEMW and NEMD, ACI_Nd intensifies steadily, in parallel with declining moisture and increasing LTS. Quantitatively, the ACI_Nd increases progressively from 0.250 ± 0.027 (95 % confidence interval, 95 % CI) during the SWMW period to 0.286 ± 0.016 during NEMW and further to 0.399 ± 0.029 during NEMD. Meanwhile, q decreases from 12.111 ± 0.540 g kg⁻¹ during SWMW to 11.072 ± 0.931 g kg⁻¹ during NEMW and 9.540 ± 1.120 g kg⁻¹ during NEMD, while the LTS increases from 13.341 ± 0.358 to 14.565 ± 0.875 and 15.343 ± 0.977 K, respectively. (All q and LTS uncertainties represent one standard deviation, SD) These co-varying changes indicate that both q and LTS may regulate the strengthening of ACI across the three periods over the SCS. In the following subsections, we separately examine the roles of q (Sect. 3.5.1) and LTS (Sect. 3.5.2) in regulating ACI.