Volcanic aerosols lend causality to the indicated substantial susceptibility of clouds to aerosol over global oceans

Abstract

The large indicated associations between aerosols and cloud radiative effects imply large negative radiative forcing, i.e., cooling incurred by the aerosols’ effects on clouds, if their relationships are causal. The alternative explanation is aerosol-meteorology co-variability. Here, we examine whether aerosols are the primary driver of aerosol-cloud co-variability, i.e., constituting susceptibility of the cloud properties to aerosols. It is done by domains affected by volcanic aerosols, where the aerosol-meteorology co-variability is expected to be minimized. We hypothesize that volcanic aerosols would reduce aerosol-meteorology co-variability under similar meteorology, thus diminishing aerosol-cloud co-variability. However, our findings in both volcanic and non-volcanic regions across the global oceans indicate a consistent pattern of aerosol-cloud co-variability. This does not prove definitively a causal link between aerosols and cloud properties, but mininimizes the probability that meteorological co-variability is a major cause.

Introduction

The uncertainty of aerosol-cloud interactions (ACI) plays an important role in anthropogenic climate forcing^1,2. The inherent nonlinearity of cloud responses to aerosol perturbations complicates the attribution of causality in observed relationships of aerosol radiative forcing³, leading to persistent uncertainties in its regional and global climate effect⁴. A major challenge is separating the aerosol effects from meteorology on clouds. Therefore, cases where aerosol variability is well-defined with minimal meteorological influence, such as ship tracks and volcanic emissions, can be natural laboratories for investigating ACI⁵. However, research on ship tracks faces limitations. Ship tracks occur under special conditions of marine stratocumulus with high lower tropospheric stability (LTS), and even in such conditions, they manifest only about 25% of the time. Moreover, ship tracks generally do not form in most boundary layer clouds with scales of cloud self-organization which can extend over hundreds of kilometers⁶. Notably, the properties of clouds that are not conducive to ship tracks still show a very large susceptibility to aerosols⁷, which translates to aerosol indirect forcing of −1.1 ± 0.8 W m⁻² only due to the Twomey effect, in addition to the cloud cover effect⁸. Its absolute value exceeds substantially the absolute value of −0.84 (−1.45 to −0.25) W m⁻² as assessed by the 6^th assessment report⁹. This large value begs the question of whether the large indicated susceptibility of the general population of marine boundary layer clouds is from aerosol perturbations, or whether non-causal co-variability of aerosols with meteorology that determines cloud properties can serve as an alternative explanation. To answer this question, we use here volcanic plumes, which are produced by volcanic activity and found to be independent of meteorology^10,11, and have already been used for ACI studies at a regional scale^12,13. Substantial effects from volcanic eruptions on boundary layer clouds are found even if the volcanoes are weakly explosive¹⁰.

Numerous studies on the interaction between volcanic aerosols and clouds have been conducted in the past. Reduced size of cloud drop effective radius (r_e) was indicated in stratocumulus during large volcanic eruptions in Iceland^12,14, with the increases of albedo. Increased cloud cover was detected in trade cumulus during the activity of the Kilauea volcano in Hawaii^11,15. Nonetheless, the reliability of applying the observed relationships from a regional to global scale remains unclear³. In this study, we quantify the aerosol-cloud interaction (ACI) in clouds affected by volcanic activity, employing the methodology previously applied over the Southern Oceans between the equator and 40°S⁷ (hereafter R19), where R19 established correlations between variability in cloud fraction (Cf), liquid water path (LWP) and cloud radiative effect (CRE) with changes in cloud drop concentrations (N_d). Despite the significant susceptibility of Cf and CRE in R19, it did not provide a comprehensive cause-and-effect understanding of ACI. A major objective of this study is to replicate the analytical approach of R19, particularly in scenarios where the aerosols are known to be decoupled from meteorological variability.

When assessing ACI, a fundamental step is establishing the relationships between cloud condensation nuclei (CCN) and N_d. Retrieval algorithms are reliant on their underlying assumptions, parameters, and the limited constraints applied, which could introduce potential uncertainties¹⁶. One approach involves utilizing the pixels with the highest 10% cloud optical depth (COD) to retrieve N_d (N_d-core)¹⁷. This segment corresponds to the clouds’ cores closest to the adiabatic assumption and exhibits the strongest updrafts. By employing the highest 10% of COD for N_d retrieval, it best captures the effects of subcloud CCN¹⁸. The accuracy of this N_d retrieval method was assessed, revealing a significant correlation coefficient of 0.88 between N_d and CCN in the case of coupled clouds, relying on shipborne observations¹⁸. Compelling evidence of the pronounced reliance of the satellite-retrieved N_d on CCN for marine shallow clouds was presented¹⁹. It achieved an impressive ±30% accuracy in satellite-retrieving CCN under both clean and polluted environmental conditions. Additionally, the threshold for COD used in N_d retrieval has a minimal influence on the relationship between cloud properties and aerosols²⁰.

Here we chose the N_d-core retrieving method to investigate aerosols’ effect on cloud properties and radiative forcing downwind of large and persistent effusive volcanic eruptions. In this study, we have analyzed a collection of massive volcanic events documented across different geographical regions^{11,12,14,21,22}. These events include sequences of volcanic eruptions in the Southwest Pacific during the summer of 2018, including volcanoes in Vanuatu and Galapagos Islands, the Holuhraun fissure eruption in Iceland in 2014, and the eruption of Mt Kilauea in Hawaii in 2018. These events were chosen to ensure the inclusion of a large range of meteorological conditions and cloud regimes.

Our study focuses on clarifying the preponderance of volcanic aerosols as the primary driver of aerosol variability in regions influenced by volcanic activities, to non-volcanic regions where the co-varying influence of meteorology is likely to be more important. Therefore the observed associations of cloud properties and N_d within the volcanic aerosol-dominated regions can be compared to volcano-free areas under similar meteorological conditions elsewhere over global oceans. Further investigations validate the similar susceptibilities of predicted and observed cloud properties to aerosols across various regions with varying levels of volcanic and non-volcanic aerosol variability, reinforcing the importance of aerosol perturbations to cloud properties and associated aerosol cloud-mediated forcing.

Results

The domination of volcanic aerosol variability over meteorological influences

In volcanic aerosol-influenced areas, the variability of aerosols exhibits a large degree of meteorological independence or substantially reduced relationships with meteorological factors. Here we introduce sulfur compound mixing ratio (SC), which combines integrated low-level SO₂ and sulfate aerosols (expressed in units of kg/kg), as a “tracer” of the volcanic plume. The SC acts as a flag to areas where volcanic aerosols dominate the aerosol variability. The volcanic aerosol-influenced regions were selected when average SC exceeding a threshold of 10^−8.6 kg/kg, and these regions were further divided into southwest and east volcanic clusters for the Vanuatu and Galapagos Islands, respectively, as described in “Materials and Methods”. In contrast to the commonly used natural laboratories that rely on cloud properties^8,23 or on broader regular regions^7,12,24 to describe aerosol-influenced areas, our approach introduced aerosol properties (SC) to capture volcano-affected domains and demonstrated the domination of volcanic aerosols.

Figure 1 shows the relationships between N_d-core and SC in the volcanic aerosol-influenced domains within the Southwest Pacific, under different bins of meteorological factors. These parameters are CGT (cloud geometrical thickness), LTS (lower troposphere stability, the difference in potential temperature between 700 hPa and the surface), CTRC (cloud top radiative cooling rate), SST (sea surface temperature), RH (surface relative humidity) and ω850 (vertical velocity at 850 hPa). The data were subsequently classified into bins corresponding to SC intervals.

Fig. 1: N_d-core as a function of sulfur compounds stratified by various values of meteorological cloud-controlling factors within the volcanic-affected regions in the southwest and the east Pacific volcanoes.

Different bins of CGT (A), LTS (B), CTRC (C), SST (D), RH (E), and ω850 (F) are shown.

The presented data indicates the average N_d-core values corresponding to scenes within each SC bin. The relationships between N_d-core and SC remained relatively consistent under different meteorological factors. This suggests that volcanic aerosols, rather than meteorological factors, primarily drive changes in aerosol concentration.

The control of cloud properties by N_d and meteorological conditions

Volcanic aerosols have significant effects on cloud microphysical properties. Strong cloud responses to increased aerosols are indicated in Fig. 2, particularly in cleaner background conditions, consistent with previous research findings^8,23,25. For the southwest and the east Pacific volcanoes, Fig. 2 shows a positive correlation of N_d-core with Cf (Fig. 2A, E), and with albedo (Fig. 2C, G), which resulted in a strong negative correlation between N_d-core and CRE (Fig. 2D, H). No systematic and discernible connection was observed between the average LWP (LWP_average) (Fig. 2B, F) and N_d-core. The response of cloud water in polluted clouds can vary, influenced by meteorological conditions and cloud types²⁶. The co-variability between N_d and LWP arises from the combined effects of meteorological conditions and microphysical processes, which influence LWP and N_d in opposite directions²⁷. For instance, the suppression of rain formation due to smaller cloud droplets may lead to an increase in LWP, while enhanced evaporation can cause a decrease. LWP was found to be slightly increased in a clean background and somewhat decreased in a polluted background²⁶. However, the two volcanic clusters show different strengths of the dependence of cloud properties on N_d-core. This could potentially caused by differences in cloud regimes.

Fig. 2: The dependence of Cf, LWP, cloud albedo, and CRE on N_d-core for different intervals of CGT in the southwest and east Pacific volcanic regions.

Panels A–D correspond to the southwest Pacific volcanoes, while Panels E–H correspond to the east Pacific volcanoes. The data are for all scenes which have an average SC ≥ 10^−8.6 kg/kg box. The values are the mean values for each bin. Bins of N_d-core and CGT are shown only if they contain at least 50 scenes. “N” at the upper right corner represents the total number of scenes. The error bars represent the standard error.

Regional variation of cloud radiative effect could be related to the change in large-scale flow and local thermodynamic conditions²⁸. Remarkable differences in meteorological conditions are observed within these two volcanic clusters. As illustrated in Fig. 3, the east volcanic cluster has a larger modal LTS in comparison to the southwest volcanic cluster, although the average LTS values are similar for these two clusters which are 17.0 K and 15.7 K, respectively. A larger LTS favors the formation of marine stratocumulus and a larger Cf. Conversely, a smaller LTS is prevalent in the trade wind cumulus cluster, which dominates the southwest volcanic cluster. According to the probability distribution density of LTS, it is clarified that ~23% of the scenes within the southwest volcanic cluster have LTS < 14 K, indicative of the cumulus regime. While only 4% of the scenes have LTS < 14 K in the east volcanic cluster. In addition to higher LTS, the east volcanic cluster also has a smaller SST and marginally higher RH, collectively leading to higher cloud cover^29,30,31,32. While the southwest volcanic cluster’s CTRC is slightly larger, its overall contribution to the differences is much smaller in comparison to the other meteorological factors (Fig. 3C). The effects of volcanic aerosols on cloud properties within aerosol-perturbed and unperturbed scenarios are shown in Supplementary Section 1 of the Supplementary Information. Within the volcano-affected region, all scenes were further classified based on a threshold SC = 10^−8.6 kg/kg, distinguishing between aerosol-perturbed and unperturbed scenarios. This classification allowed for an initial control over meteorological influences, and served to assist in validating the dominant role of volcanic aerosols.

Fig. 3: The meteorological conditions and sea surface temperatures for the southwest (blue lines) and east (red lines) volcanic clusters.

The probability density functions (PDFs) and mean values of LTS (A), SST (B), CTRC (C) and RH (D) are shown.

The aerosol-perturbed scenario tends to have a large increase in N_d-core, albedo, Cf, and associated negative CRE, as well as a decrease in r_e, with LWP remaining relatively consistent. This analysis calculated the relative change of cloud properties due to aerosol perturbations (Supplementary Table S1), accounting for the background cloud conditions.

Applicability of the volcanic aerosols and cloud property relationships to global aerosol variability

Investigating how perturbed aerosols affect clouds demands controlled conditions with minimal meteorological variations^3,33. Volcanic activities reveal independent aerosol effects on clouds, which exhibit a large degree of meteorological independence. However, cloud-aerosol connections might differ in unperturbed aerosol conditions due to covarying meteorological conditions, potentially obscuring part of the aerosol impact on clouds. The co-variability of cloud properties with aerosols arises from two sources. One is the effects of ACI on cloud properties, and the other is the effects of meteorological cloud-controlling factors that co-vary with the aerosols.

To investigate the dominant factors driving cloud responses, we compared the susceptibilities of cloud properties to aerosol perturbations in regions influenced by volcanic emissions with that of volcano-free global areas. We ensured similar distribution patterns of meteorological conditions in both cases. This approach enables us to analyze different aerosol effects (both volcanic and non-volcanic) on clouds under similar meteorological conditions for both cumulus and stratocumulus clouds, respectively.

The correlation analysis between meteorological conditions and cloud properties is detailed in Supplementary Figs. S8–S11. The relatively low coefficients of determination (R²) observed can be attributed to factors such as the inherent co-variability of meteorological parameters and spatial resolution discrepancies between satellite data and reanalysis products³⁴. This study utilized cloud-controlling factors (CCFs, including CGT, LTS, CTRC, SST, and RH). Before applying the meteorological filtering approach, volcanic data from 2018 and global ocean data from 2014 were stratified into bins based on identical classifications for each parameter, utilizing six bins each for CGT, LTS, CTRC, SST, and RH. This binning process resulted in 7776 combinations (calculated as 6⁵), each representing a distinct environmental condition of CCFs. We retained only those combinations (referred to as valid bins) where the global data had more scenes than the volcanic data. We calculated the ratio between the number of volcanic and global data for each valid bin. The number of volcanic data and valid bins decreases as the ratio increases, while the number of filtered global data increases (Supplementary Fig. S12). The optimal ratio was determined when the number of valid bins decreased to 90% of the total, as indicated by the broken black line in Supplementary Fig. S12. This method preserved the majority of valid bins while ensuring proportional sample sizes. Finally, global data within each valid bin were randomly selected, with the number of scenes in each bin based on the optimal ratio and the corresponding number of volcanic data points, thereby ensuring comparability between volcanic and filtered global data sample size. This approach directly compares volcanic and filtered global ocean data under similar meteorological conditions. By applying this method, we aim to more comprehensively capture the nonlinear interactions between cloud properties, aerosol perturbations, and meteorological variability. More details on this method are described in Supplementary Section 5. After constraining for the similar distributions of meteorological conditions, 98,912 (Fig. 4A) and 291,816 (Fig. 4B) scenes of 2014 global ocean data were selected to compare with southwest and east volcanic clusters, respectively.

Fig. 4: Occurrence of the global distribution of filtered scenes.

Filtered global ocean data from 2014 which was under similar meteorological conditions with scenes of southwest (A) and east (B) volcanic clusters, Holuhraun fissure (C), and Kilauea volcano (D).

The rationale behind incorporating volcanic emissions into this study lies in the intention to disentangle the variability in aerosols from meteorology. Figure 5 and Supplementary Fig. S7 show a large similarity of the indicated susceptibility between N_d-core and cloud properties for the volcano-forced aerosol variability and for the freely varying aerosols over the global ocean. Such similarity is also apparent during the Holuhraun (Supplementary Fig. S3) and Kilauea (Supplementary Fig. S6) volcanic effusive eruptions.

Fig. 5: The dependence of cloud properties on N_d-core of filtered 2014 global data.

Panels A–D show the filtered data with similar meteorological conditions to the southwest volcanic cluster. Panels E–H show the same for the east volcanic cluster.

Meteorology-free susceptibilities should be able to predict the results of the ideal natural experiment. Cloud-controlling factors, including N_d-core, CGT, LTS, SST, RH and CTRC, were applied to predict cloud properties. Prior to regression process, binning was applied to partition each parameter based on its cumulative distribution function^7,35,36, ensuring distinct environmental conditions were captured. For each bin combination, corresponding parameter values were averaged, and these mean values served as inputs for training the regression models. We used a third-power polynomial regression to build prediction models for different cloud properties based on N_d-core and meteorological conditions, respectively. In this context, we employed regression models derived from the observed volcanic data and filtered global ocean data, respectively. Each regression model was applied to predict cloud properties of both volcanic and filtered global ocean data. The regression models showed a large predictive capacity for the cloud properties, with an overall R² > 0.83. Detailed information about the prediction process can be found in the Supplementary Information (Supplementary Section 6).

The susceptibilities of CRE, Cf, albedo and LWP to N_d-core were further calculated and presented in the form of horizontal bars in Fig. 6. For example, the susceptibilities of CRE to N_d-core (represent as red bars) were described here for panel (A): for the observations, two solid bars showed the susceptibilities of observed CRE to N_d-core (0.50 ± 0.07 and 0.51 ± 0.08, for the volcanic and global data, respectively). For the prediction, two faint bars represent the susceptibilities of predicted CRE to N_d-core based on regression models derived from volcanic data applied to both volcanic (0.47 ± 0.13) and filtered global cases (0.50 ± 0.13). The same was done based on the regression models derived by filtered global ocean data applied to volcanic (0.43 ± 0.08) and filtered global cases (0.47 ± 0.10), presented by the hollow bars. The increase in aerosols does not directly lead to an increase in Cf over the mid-to-high-latitudes of the Northern Hemisphere³⁷. This might be the case that the background aerosol is high or the added volcanic aerosol mainly influences the high cloud coverage area which does not have the potential to have high cloud coverage. Consequently, when employing the regression models derived from volcanic data, the susceptibilities of Cf and CRE to N_d-core are slightly lower (faint bar in panel (C)). The similarity of the relationships between the volcanic or non-volcanic aerosols and cloud properties under similar meteorological conditions, and predicted properties derived by quite different data sets, means that such relationships can also be attributed to the freely and naturally varying aerosol relationships on clouds, such as documented by R19.

Fig. 6: Observed and predicted susceptibilities of cloud properties to N_d-core.

It contains scenes of southwest (A) and east (B) volcanic clusters, Holuhraun fissure (C), and Kilauea volcano (D), and their corresponding filtered global ocean data. For the Holuhraun fissure eruption case, we excluded the volcanic data from the same year’s global ocean data before filtering. The overall susceptibilities of LWP, Albedo, Cf, and CRE to N_d-core, based on the slopes of the lines of cases of volcanic data and filtered global ocean data of 2014 for similar meteorological conditions (From Fig. 5, Supplementary Figs. S3, S6, and S7). Solid dark bars represent the susceptibilities calculated by observed cloud properties and radiative effect. Also presented are susceptibilities calculated based on predicted cloud properties by the volcanic data (faint bars) and based on global data filtered for meteorological conditions in the volcanic areas (hollow bars).

We also quantified the correlation between meteorological factors and N_d-core. We aim to verify the degree to which N_d-core can be effectively predicted only through meteorological parameters, as presented in Supplementary Table S5. Our findings reveal R² ranging from 0.42 to 0.90. The explained N_d variability by meteorological conditions for the volcanic aerosols is lower than that for the non-volcanic aerosols. This difference underscores the substantial contribution of volcanic aerosols to the overall aerosol variability. Given that most volcanic areas are more than several hours downwind from the volcanoes, the slow adjustment process has sufficient time to occur and induces similar relationships between the aerosol and cloud properties as in non-volcanic aerosols. The generally slower rate of aerosol scavenging affects the aerosols and N_d-core, but apparently not the relationships between N_d-core and the other cloud properties.

Discussion

This study addresses two persistent challenges to ACI research posed by the co-variability of aerosols and meteorology. The initial challenge revolves around determining whether the identified correlations between aerosols and cloud properties result from aerosol perturbations or if the aerosols simply indicate specific meteorological factors that influence cloud properties³⁸. Figure 1 and Supplementary Table S5 provide compelling evidence suggesting volcanic aerosols rather than meteorological factors are the primary drivers of aerosol variability. Therefore, the changes in cloud properties directly respond to aerosol perturbations. The second challenge lies in confirming whether the established connections between aerosols and cloud properties are limited to a regional scale or if they represent a cause-and-effect relationship that extends to the global variability of aerosol. Figure 6 provides compelling evidence and suggests that the susceptibility observed in naturally varying aerosols closely aligns with the susceptibility observed in volcanic plumes under similar meteorological conditions. This alignment, supported by both observation and prediction, suggests the predominant role of aerosols in aerosol-cloud co-variability on the global ocean scale. Any systematic biases that may affect the relationships, such as biases in the retrieved N_d-core or Cf values and the nonlinear relationship between aerosol and N_d³⁹, are similar both within and outside volcanic plumes. Therefore, the biases effectively cancel out when comparing susceptibilities within and outside the plumes.

Our findings, while not being a definite proof, indicate that aerosol plays a dominant role in the co-variability of aerosol and cloud properties within a fixed meteorological background, contradicting the alternative explanations attributing the correlations to simultaneous changes in aerosols and cloud properties driven by meteorology. The underlying logic operates as follows:

Suppose that a co-variability of meteorological with aerosols and cloud properties explains part of the indicated susceptibility, but the aerosols also affect cloud properties. Then, adding volcanic aerosols that do not correlate with meteorology but affect cloud properties will modify the indicated susceptibility under similar meteorological conditions.

Our finding that susceptibilities within and outside volcanic plumes are similar suggests two potential explanations

• 1. Aerosols do not affect cloud properties, or,

• 2. Aerosols do affect cloud properties, but in this case, meteorological co-variability did not contribute to the indicated susceptibility. This implies the effect of aerosols on the indicated susceptibility.

The lack of indication for indicated susceptibility is caused by meteorological co-variability and lack of other alternative explanations, we are left with the aerosols affect cloud properties as the likely explanation, although not a definite proof.

The assessments reveal a large magnitude of these observed and predicted relationships between aerosols and cloud properties, in contrast to other observations and model simulations⁴⁰, particularly concerning CRE and the associated aerosol-forcing mediated by clouds on a global ocean scale, as shown in Table 1. The susceptibility of CRE to aerosols could be considered as a forcing in cases where variations in aerosols are forced externally to the natural Earth radiative balance⁴¹. An example of such externally induced changes in aerosols is exemplified by the volcanic emissions utilized in this study. Previous studies showed similar susceptibilities of CRE to both N_d-core and N_d-average²⁰. Using N_d differentiates our study from others employing alternative aerosol proxies, such as aerosol optical depth (AOD) or aerosol index (AI). These alternatives may introduce biases, particularly in cases involving aerosol swelling, wet scavenging, and other aerosol processes⁴². The susceptibility of N_d to CCN proxies was greater for finer aerosol proxies, increasing from AOD to AI and reaching 0.66 for polarimetric measurements, with corresponding aerosol-cloud interactions radiative forcing (RF_aci) of −1.14 (−0.84 to 1.72) Wm^−2 42, which is much larger than the RF_aci of −0.84 (−0.25 to −1.45) of the 6^th IPCC Assessment Report. The strong radiative forcing occurs mainly in clean cloud backgrounds (N_d less than 60 cm⁻³)^8,23,25. This fact was further emphasized by a recent study on a larger scale⁸. In pristine environments, the susceptibility of CRE to aerosols is 0.78 and 0.70 for the southwest and east volcanic clusters, respectively. In contrast, these susceptibilities decrease under polluted conditions to 0.36 and 0.53 (Supplementary Table S6).

Table 1 Susceptibilities of CRE to aerosols in previous and this studies

In summary, our findings underscore the primary driver of aerosols in driving aerosol-cloud co-variability, revealing large susceptibilities of cloud properties to aerosols. Although the causal relationship between aerosols and cloud properties remains unproven, these results leave limited scope for alternative measurable interpretations. A causal relationship would imply a greater RF_aci than reported in the 6^th IPCC Assessment Report.

Materials and methods

The summer of 2018 was a remarkable period of effusive volcanic eruptions in the southwest Pacific^43,44. The Manaro Voui volcano on Ambae, located in Vanuatu, made a record for sulfur dioxide (SO₂) emission, injecting 400,000 tons of SO₂ into the atmosphere in July. During the same period, the Galapagos Islands’ Fernandina and Sierra Negra volcanoes erupted one after the other. The annual emission of this series of eruptions at Sierra Negra emitted about 160 kt/y SO₂, which was nearly ten times higher than that from recent years (SO₂ emissions by explosive volcanic eruptions, https://so2.gsfc.nasa.gov/measures.html). The study is performed over the Southwest Pacific between 40°S and 10°N, 160°E and 90°W from June to August 2018. The study area was divided into two volcanic clusters. The southwest volcanic cluster lies near Vanuatu, while the east volcanic cluster is near South America.

In this study, satellite observations of clouds were obtained from MODIS level-1 and level-2 Collection 6 data⁴⁵. Cloud properties such as N_d-core, r_e-core, LWP_average, Cf and CRE were retrieved by dividing each MODIS (Terra) granule into scenes with a size of 110 km × 110 km (~1° × 1° at the equator) as done by R19. In this study, N_d and r_e were obtained from cloud cores (N_d-core, r_e-core). CRE is the difference between the top of the atmosphere radiative flux under cloudy conditions and the corresponding flux under clear sky conditions, subsequently multiplied by the water Cf in the 1° × 1° grid box. This approach is widely applied to investigate the influence of clouds on the Earth’s radiation balance^46,47. In this study, we utilized the MODIS level-2 Cloud Mask to derive Cf. Specifically, the liquid water Cf for each pixel was computed using the “Confident Cloudy” flag, ensuring accurate estimation of cloud properties⁴⁸. This methodology enabled us to quantify the cloud response to corresponding aerosol perturbations. All derived cloud properties were averaged to support subsequent analyses.

The corresponding meteorological parameters (LTS and RH) were provided by the National Centers for Environmental Prediction-National Center for Atmospheric Research (NCEP-NCAR) reanalysis data. CTRC was calculated based on the relation between it and the updraft at the cloud base⁴⁹. Daily average SST was derived from the National Oceanic and Atmospheric Administration (NOAA) High-Resolution SST data.

The volcanic plume diffuses, and its border cannot be identified as it spreads over a large area far downwind. In this study, we integrated low-level SO₂ and sulfate aerosols from MERRA-2 reanalysis data (Modern-Era Retrospective analysis for Research and Applications, Version 2) between the surface and 850 hPa to account for the boundary-layer SC. A robust correlation between sulfate aerosols and N_d has been established, providing a more accurate representation of the sensitivity of N_d to SO₄⁵⁰. Given the inherent uncertainties in both N_d observations and aerosol reanalysis, this represents a significant portion of the explained variance. The SC mean value in this study area is about 6.42 × 10⁻⁹kg/kg, reaching a relative change of 208% concerning the same period and area in 2014. This dramatic rise of SC produces an enormous aerosol plume near the volcanoes and provides an ideal zone where clouds are most susceptible to aerosol perturbation¹².

Here we focused on single-layer marine low clouds with CGT up to 800 m. In our study, CGT is derived as the difference between cloud top height and cloud base height⁵¹. Cloud top height and cloud base height are both uniquely defined for the combination of LWP, cloud top temperature and SST⁷. A critical assumption is the dry adiabatic lapse rate from the sea surface to the cloud base and the moist adiabatic from the cloud base to the cloud top. In total, there were 155,308 available scenes within the study area. N_d-core starts to exceed 50 cm⁻³ when SC ≥ 10^−8.6 kg/kg according to the relationship between N_d-core and SC (Fig. 7). Here we selected the scenes which were located in the 1° × 1° box with average SC ≥ 10^−8.6 kg/kg as shown in Fig. 7. Here 38,469 scenes were selected and further analyzed (divided by the red line as shown in Fig. 8), which are ~25% of the total available cases. We also examined global active volcanoes in recent years and identified 2014 as a background year with relatively low volcanic activity. This approach allowed us to compare cloud responses to volcanic aerosols (non-meteorological aerosols) with those to non-volcanic aerosols (meteorological aerosols).

Fig. 7: N_d-core as a function of SC bins.

The red horizontal bars are the medians, and the blue dots are the means. The horizontal broken blue line represents N_d-core = 50 cm⁻³, which is the value of the selection SC = 10^−8.6 kg/kg.

Fig. 8: The study area and distribution of SC.

The map shows the spatial distribution of SC under the influence of volcanic eruptions in the Southwest Pacific in June, July and August of 2018. The colors over the ocean represent average concentrations of SC.