Volcanic eruptions influence the climate by emitting large quantities of solid particles (ash) and gaseous compounds into the atmosphere . Ash particles block sunlight and, therefore, decrease solar radiation reaching the surface. This leads to a cooling, even if the ash settles down relatively fast due to gravity .

The gas emissions mainly include water vapor, carbon dioxide, sulfur components (mainly sulfur dioxide, SO2), and nitrogen . Chemical processes convert SO2 to sulfuric acid (H2SO4; sulfate aerosol) in the troposphere at relatively short time spans of few days, while in the stratosphere, the conversion can take weeks to months .

Sulfate aerosols, injected from a large volcanic eruption, modify the Earth's radiative budget directly by scattering sunlight and indirectly via interaction with clouds . The latter is the focus of the present paper. A large volcanic eruption as a natural laboratory may help to better understand and quantify how cloud properties are modified in response to anthropogenic aerosol emissions .

Imposed effective radiative forcing by aerosol–cloud interactions in warm clouds can be separated into the Twomey effect and cloud adjustments to radiative forcing . An enhancement in cloud condensation nuclei (CCN) concentrations lead to an increase in cloud droplet number concentration (Nd), resulting in a smaller effective radius (re) if the cloud liquid water path (LWP) is constant. Consequently, scattering cross section and the cloud albedo are enhanced, causing clouds to reflect more sunlight back to space, which is known as the Twomey effect . Anthropogenic aerosols modify cloud particle size distributions, which reduces the efficiency of collision–coalescence processes, leading to a delay in precipitation onset consequently enhancing cloud lifetime . This infers on average an enhancement in cloud fraction and LWP . These longer-lived clouds reflect more sunlight back to space and cool the atmosphere and surface even more, which is known as the lifetime effect .

Along with the aforementioned effects, there is a large variety of processes that partially offset these effects on clouds, such as a reduced maximum supersaturation if more droplets compete for the available water vapor , a larger evaporation rate of smaller droplets , increased droplet spectrum dispersion , or enhanced evaporation due to cloud-top mixing . Because the different effects oppose each other, the overall changes in the effective radiative forcing could be minor on larger scales . In this study, the responses of clouds to aerosols emitted in the Holuhraun volcano eruption were examined. The Holuhraun eruption was the strongest in Europe since the 18th century and emitted substantial amounts of sulfate aerosol . This natural phenomenon triggered a large effort to investigate the impact of this large aerosol perturbation on cloud properties. .

found a significant reduction in re in satellite data but only insignificant alterations of LWP. They further concluded that several general circulation models overemphasize LWP increase in response to the additional aerosol. However, did find an increase in LWP when carefully conditioning on moisture convergence. In addition, ambiguous results, with LWP responses of either sign, were obtained by when analyzing multiple volcanic eruptions. Several cloud-resolving modeling studies on the sensitivity of LWP to variation in cloud droplet number concentration have been conducted. examined the effect of entertainment and sedimentation rate on LWP in stratocumulus cloud regimes using large-eddy simulation (LES). Their results explained the process details of the conclusions by , namely that sedimentation leads to a decrease in entrainment rates and an increase in LWP. conducted a set of LES simulations over fair weather cumulus cloud regimes over the subtropical ocean. They concluded that in this cloud regime, the response of LWP to enhancing Nd was almost negligible in equilibrium and slight reduction in cloud cover was obtained, leading to a negative cloud lifetime effect, compensating for the positive radiative forcing of the Twomey effect. performed LES simulations of two different cloud regimes of marine stratocumulus and trade wind cumulus clouds. They showed different relationships between the relative LWP response to the relative change in aerosol concentration Na, a term they called λ, and the precipitation probability susceptibility (SPOP). For the trade wind cumulus clouds regime, λ decreases with enhancement of SPOP because of the entrainment and evaporation rate effect in cumulus clouds. In stratocumulus clouds, λ and SPOP, in contrast, were positively related. In this case, aerosol-induced evaporation–entrainment and/or sedimentation–entrainment effects further restricted the increase in LWP in their simulations. conducted a 1-year global cloud-resolving simulation to examine the sensitivity of liquid water content (LWC) to aerosol loading. They demonstrated that in their model, the condensation process in the lower part of clouds is associated with a positive LWC response and the evaporation process in the upper part of clouds is responsible for a negative response to additional aerosol loading. Following these previous studies, we chose the Holuhraun eruption to investigate the response of LWP, cloud fraction, and its corresponding radiative effect in response to additional CCN in the emission plume of the volcano, employing simulations at cloud-resolving resolution and comparing them to satellite observations.

The present study focuses on a detection and attribution approach, using cloud-resolving simulations kilometer-scale resolution, in combination with the analysis of satellite data. A pair of simulations over the North Atlantic Ocean around the Holuhraun volcano in Iceland was employed (Fig. ). The model used is the ICOsahedral Non-hydrostatic model ICON,. The ICON model is developed by a collaboration between the German Meteorological Service and the Max Plank Institute for Meteorology . It can be used as a global simulation on the climate scale to high-resolution large-eddy simulations . Here, the physics package of the numerical weather prediction (NWP) variant is used (ICON-NWP). The resolution corresponds to approximately 2.5 km in the horizontal (R2B10 triangular grid). Vertically, 75 layers with a top height at 30 km were chosen. The vertical resolution increases towards the model top with a model layer thickness of 20 m in the boundary layer and a maximum layer thickens of 400 m near the model top.

The physics package of ICON-NWP includes a comprehensive double-moment cloud liquid and ice microphysical scheme . Because of using a rather fine resolution, deep convection is considered to be resolved, whereas, for shallow convection, the parameterization scheme by with modifications by was used. A grid-scale cloud cover scheme was employed in simulations. In this scheme, if the sum of specific cloud water content and specific cloud ice content is larger than a certain threshold, the cloud fraction is set to 1; otherwise, it is set to 0, and the shallow-convection scheme contributes to the computation of specific cloud water and ice content. To achieve a more realistic representation of the Twomey effect, we furthermore coupled the hydrometeor number concentrations from the double-moment microphysical scheme to the radiation scheme as proposed in .

Initial and boundary conditions were derived from the European Centre for Medium-Range Weather Forecast (ECMWF) Integrated Forecasting System (IFS) operational analysis. Variables such as temperature, wind, geopotential, humidity and hydrometeors were used in initial and boundary conditions. The 2014 Holuhraun eruption was a fissure eruption that started on 20 August 2014 and ended on 25 February 2015. By 7 September 2014, the lava field had extended more than 11 km to the north . We choose the period from 1 to 7 September 2014 for our analysis because the lava field had developed sufficiently in this period and substantial amounts of SO2 had been emitted into the atmosphere, while, at the same time, a well-defined plume was observable. The first 9 h of the simulations were excluded from analyses so that the spin-up effect is sufficiently small in our simulations. An additional feature in simulations that must be mentioned is the implementation of a satellite simulator into the model. Satellites are essential tools to assess the character of clouds due to their global coverage and availability . Differences between model simulations and satellite retrievals stem in part from a different definition of the respective quantities that are compared. Therefore, one approach to reduce inconstancy between model simulations and satellite retrievals is to use satellite simulators in models to mimic the observational processes . The COSP satellite simulator is an open-source work package developed by CFMIP (Cloud Feedback Model Intercomparison Project) to replicate active and passive satellite data using variables from the model as input . In this study, only the satellite simulator for MODIS (Moderate Resolution Imaging Spectroradiometer) observations was used. The COSP simulator uses several model variables as input such as temperature, pressure, cloud fraction, and cloud water content to generate what the MODIS retrievals would capture given the simulated clouds fields . In addition to the abovementioned variables, the effective radius of liquid cloud droplets and ice crystals is regarded as the MODIS simulator's input. The effective radius of cloud droplets and ice crystals was calculated from parameters derived from the size distribution function of the hydrometeor in the two-moment microphysic scheme. The satellite simulator has previously been implemented and used in ICON-NWP by .

In the cloud-resolving simulation (each grid box is either fully cloudy or clear), the use of sub-grid variability, one of the features of COSP for application in general circulation models, was not necessary. In order to evaluate COSP related variables in our simulations, the collection 6.1 Level-2 MODIS-Aqua optical and physical cloud data product was used ; therefore, swaths with 1 km spatial resolution for re, cloud optical thickness (τc), and LWP along with a cloud fraction with a 5 km spatial resolution were used and remapped to the model resolution to have an accurate comparison. To remap the MODIS granule to a latitude–longitude grid, for each specific point the mean value of each variable in each swath is computed. It should be mentioned that even when using a satellite simulator there is an uncertainty between the definition of LWP in the simulation and MODIS observations because the bulk microphysic scheme has a gap in the size distribution between cloud droplets and rain that is not necessarily the same as in the visible/near-infrared retrievals by MODIS. To a lesser extent, this issue may also affect the computation of Nd. Furthermore, the planetary albedo at the top of the atmosphere is analyzed as retrieved by the Clouds and the Earth's Radiant Energy System (CERES) instrument on board the Aqua satellite .

The present study aims at a detection and attribution approach, assessing the differences in cloud properties within and outside the volcanic plume by comparing simulations with satellite observations. It has been shown that meteorological conditions and cloud regimes are important to determine the effect of additional aerosol loading on cloud microphysical properties. Figure  indicates the visible image obtained from MODIS-AQUA satellite retrievals. A synoptic frontal system is located over the North Atlantic Ocean and contains large-scale, mostly stratiform ice and liquid phase clouds. These conditions remain similar during the simulation period. In order to select liquid phase clouds in the MODIS data, the Cloud Phase Optical Properties flag was used. For simulations, the COSP simulator produces the microphysical properties for the liquid and ice phase clouds separately, and we used only the outputs dedicated to liquid phase clouds in our analyses. This study aims to evaluate how cloud microphysical properties (Nd and LWP) behave differently in and outside the volcano plume. To address this scientific question, grid cells that are located inside and outside the volcano plume are analyzed and compared to each other in the volcano and no-volcano simulations along with MODIS satellite retrievals for the 7 d starting on 1 September 2014. In the no-volcano simulation, there is no CCN enhancement due to the volcanic emissions (Fig. a, c). Nevertheless, the grid points that are located inside the volcano plume are compared to the ones outside the plume to assess differences due to different meteorological conditions.

Nd is the first microphysical variable we assess. Nd is not directly retrieved by the operational MODIS satellite retrievals. Instead, re and τc are retrieved using the method described by . On the basis of such retrievals, assuming clouds that behave like adiabatic ones, Nd can be computed as follows : 1Nd=γτc12re-52.

In this relation, γ depends mainly on the adiabatic condensation rate and can be approximated as 1.37 × 10-5 m-12 . In order to obtain Nd by Eq. () in our analyses both in simulations and MODIS, re less than 4 µm and τc less than four were excluded from the data set because they are less reliable . For consistency, Nd is derived from the COSP diagnostics of τc and re (see Sect. ) in the same way as done in the MODIS retrievals. The model output is sampled at the time of the MODIS-Aqua overpass of approximately 13:30 LST (local sidereal time).

In the subsequent figures, in each panel the blue line is for the no-volcano run, the red line is for the volcano run, and the black line is for the MODIS observations. Figure  shows the relative frequency distribution of Nd. In order to define the plume, marine pixels that correspond to the SO2 concentrations in the lower troposphere exceeding 1 DU in Fig.  were chosen. These are assumed to be located inside the volcano plume, and the rest of the pixels are regarded as outside the volcano plume. Panel b (outside the plume) indicates that the Nd distribution outside the volcano plume for both simulations is, as expected, very similar because the meteorology is the same and there is no additional aerosol. Comparing both simulations to MODIS retrievals demonstrates that the simulated Nd distribution is close to what is obtained from the satellite retrievals. In contrast, for the grid points inside the plume, it can be seen that Nd is substantially enhanced in the volcano run compared to the no-volcano run as was expected due to the larger concentration of activated CCN inside the volcano plume. The Nd distribution for MODIS shows that these observations are considerably closer to the volcano run with respect to the higher probability of large Nd even if at lower concentrations there is a systematic discrepancy between MODIS data and both simulations. For such low concentrations, there is the possibility that the satellite data are biased . For broken clouds, MODIS shows overly large re, which implies overly low Nd (Eq. ). Nevertheless, the results for the large Nd concentrations and the overall good agreement between the simulations and satellite retrievals (also outside the plume) allow for clear detection of the enhancement of Nd inside the volcanic plume and its attribution to the volcanic aerosol.

The mean values for Nd are listed in Table . The mean Nd in the plume, compared to the mean of the distribution outside the plume, is enhanced by 77 % in the volcano run compared to no (0 %) change in the no-volcano run. The enhancement value in MODIS is 78 %, which is almost exactly the same as in the volcano run. The mean Nd outside the plume is 134, 128, and 135 cm-3 for no-volcano and volcano simulations and MODIS, respectively, showing that outside the plume Nd did not change considerably between simulations because the meteorology is the same and there is no additional activated CCN and showing good consistency between both model runs and the satellite retrievals. Table  further lists the mean values and changes for re and τc. The effective radius decreased inside the plume by 7 % compare to outside the plume in the volcano simulation. In the no-volcano simulation, there is no difference in re inside vs. outside the plume. In the MODIS retrievals, re decreased by 8 % inside the plume compared to outside the plume, very similar to the change in the simulation. This is consistent with the agreement in plume enhancement for Nd. Also the cloud optical thickness shows a consistent increase in MODIS as in the volcano simulation, whereas the no-volcano simulation shows a (very slight) decrease in τc inside the plume.

Figure  shows the same analyses as Fig.  but for LWP. The distribution of LWP for the region outside the volcano plume is not significantly different between the two simulations, as expected. The mean values for LWP (Table ) in both simulations are the same at 151 g m-2; furthermore, the MODIS mean value of 149 g m-2 is close to the simulations which demonstrate the accuracy of cloud simulations. This is also true for the entire distribution (Fig. ). Considering the simulated profiles, in the simulation with volcano emissions included, there is a decrease in the probability of clouds with lower LWP and an increase in the probability of clouds with higher LWP compared to the no-volcano simulation. The MODIS distribution for LWP inside the plume indicates that the probability for clouds with lower LWP is less than what the simulations show, but the probability for clouds with higher LWP is more than in the no-volcano run, albeit also less than in the volcano run. In terms of the mean values for LWP (Table ) for inside the plume, the simulations indicate a slight enhancement (+6 %) attributable to the different weather conditions (plume enhancement in the no-volcano run) and a strong enhancement (+30 %) in the volcano run. The difference suggests that the model shows an LWP enhanced by 24 % due to additional CCN inside the volcano plume. MODIS, however, is very close to the result of the no-volcano run for the average values. This almost zero enhancement on average, however, seems to come about by a decrease in LWP for the clouds with low LWP and an enhancement of LWP for large LWP values (Fig. ). This is qualitatively consistent with the results of the ICON-NWP model. The model, however, exaggerates the increase in large LWP values, leading to the exaggerated mean increase.

The question is now what is the underlying process leading to an increase in LWP in the volcano simulation? One reason is the suppression of precipitation e.g.,. Therefore, the distribution of rain water path (RWP) was analyzed to investigate the alteration of precipitation inside and outside the volcano plume in both the volcano and the no-volcano simulations. The comparison is shown in Fig. . Since the precipitation information is not available from MODIS or other satellite retrievals, RWP is only depicted for the simulations. Inside the volcano plume, there is a decrease in light rain and an increase in heavy rain for the volcano simulation, compared to the no-volcano simulation. In terms of mean values for RWP (Table ), there is a decrease in the volcano run by 15 % on average, while the precipitation profile for outside the plume is quite similar which is in the agreement of the fact that LWP for outside the plume did not alter significantly. In the volcano simulation compared to the no-volcano simulation, for the region inside the plume, an RWP of less than about 180 g m-2 occurs less frequently. This is because at larger Nd and thus smaller droplets, the occurrence of light rain is suppressed. Cloud droplets must grow larger, leading to deeper clouds in order to reach the size that they can start to precipitate. This leads to a shift in the LWP distribution to the higher values inside the plume. In turn, these deeper clouds, in which drops have more water available for growth, produce heavier precipitation. Therefore there is an enhancement in the probability of heavy rain (RWP larger than about 180 g m-2), compared to the no-volcano simulation inside the volcano plume, even if the average RWP (Table ) does not change significantly. Moreover, suppression of precipitation can also lead to enhancement in cloud horizontal extent (cloud fraction). Therefore, the modification in cloud fraction was examined in simulations and MODIS. The analyses for mean values of total cloud fraction in Table  demonstrate that, in the volcano simulation, cloud fraction is enhanced in the plume compared to outside the plume by 40 %, while the enhancement is only 32 % in the no-volcano simulation. However, even in the no-volcano simulation, cloud fraction inside the plume is higher than outside the plume by 32 % due to the different weather conditions, and this is consistent with what MODIS shows (29 %).

Finally, the effect on radiation (indicative of the effective radiative forcing due to the modification of cloud properties by the volcanic aerosol) is examined. Therefore the top of atmosphere (TOA) albedo was analyzed inside and outside the plume in simulations and CERES level-2 footprint data . For the comparison, the simulation output was remapped by distances-weighted average remapping of the four nearest-neighbor values method to a 20 km horizontal resolution to be consistent with the resolution of the CERES footprint. In Fig.  TOA albedo for the cloudy sky is depicted for inside and outside the volcano plume for both simulations and CERES data. Grid points with SO2 concentrations in the lower troposphere exceeding 1 DU are considered to constitute the plume, and SO2 concentration was obtained from OMPS satellite retrievals which are in 50 km footprint data in level-2. We remapped the level-2 data into the 50 km resolution, and due to the fact that CERES products in 20 km resolution, it has sufficient resolution to identify the plume. Clear sky was excluded because, in the model, no aerosol–radiation interactions are considered, but in the CERES this effect is in the data and would bias the analysis for clear sky. An additional important aspect that should be considered is that the TOA albedo distribution is considered here for liquid clouds with τc more than four because in obtaining Nd the data with τc less than four were excluded as well. Considering the TOA albedo distribution inside the plume, it is seen that in the volcano simulation, there is a higher probability for TOA albedo larger than 0.6 compared to the no-volcano simulation. In the CERES data, there is a peak at TOA albedo between 0.4 and 0.6 that is not as pronounced in either simulation. In turn, the probability for TOA albedo larger than 0.7 is smaller in the data than in both simulations. This bias, however, is clear outside the plume but much less so inside the plume – possibly indicative of the albedo enhancement due to the volcanic aerosol.

For the mean values (Table ), in turn, clear-sky data were taken into account to be able to see the influence of cloud fraction changes on modifying TOA albedo. The difference in mean values between inside and outside the plume in the volcano simulation is 15 % larger compared to the no-volcano simulation. In CERES data there is an 18 % enhancement inside the volcano plume compared to outside the plume. When compared to the difference between inside and outside the plume in the no-volcano simulation (27 %), it is difficult to conclude that there is a signal of alteration in TOA albedo in CERES data. We also analyzed cloudy-sky TOA albedo mean values in simulations and CERES. The values in Table 1 demonstrate an enhancement of 9 % in CERES and 7 % in the volcano simulation, while no changes were obtained in no-volcano simulation. The daily mean incoming solar radiation was obtained 260 W m-2; therefore, effective radiative forcing except cloud cover effect can be estimated as 10 W m-2 in CERES and 8 W m-2 in the volcano simulation.

In this study, the impact of aerosols emitted by the Holuhraun volcanic eruption on liquid clouds was assessed. For this, we used a pair of cloud-system-resolving simulations with and without the enhancement in CCN due to the volcanic emission as well as MODIS and CERES satellite retrievals. The COSP simulator was implemented in the model to allow for an apples-to-apples comparison between the simulations and satellite data. To identify the impact of the additional aerosol on cloud microphysical properties, areas located inside and outside the volcano plume were compared in terms of their statistical distributions. In the no-volcano simulation, only the differences in weather conditions are sampled. In the volcano simulation, in addition, there is the effect of the CCN enhancement on the clouds. To the extent the inside vs. outside-plume difference is consistent between the satellite retrievals and the volcano simulation, but not between the satellite retrievals and the no-volcano simulation, detection and attribution of the effect of the aerosol on the clouds are achieved. Our analyses indicated that Nd concentration is clearly enhanced inside the volcano plume. This enhancement by almost 80 % is attributable to the additional CCN inside the volcano plume. Our scientific goal in this study was to examine how LWP and cloud fraction respond to the enhancement of the Nd in the volcanic plume. The analysis reveals that in the simulations and MODIS, the LWP is increased inside the plume compared to outside the plume. However, the mean increase in MODIS is very close to the result of the no-volcano run. This almost zero enhancement in MODIS on average is because of decrease in LWP for the clouds with low LWP and an enhancement of LWP for large LWP values, which is consistent with the results of the ICON-NWP model; nevertheless, the model, exaggerates the increase in large LWP values. In the model the reason for the enhancement of LWP in the volcano simulation was the decrease in precipitation compared to the no-volcano simulation by 15 % on average, due to a decrease in light rain in the volcano simulation compared to the no-volcano simulation. When light rain is depressed, clouds droplets must grow deeper in order to reach the droplet size at which precipitation is initiated. This leads to a shift in the LWP distribution to the higher values. Since cloud droplets grow deeper, they precipitate more heavily because they have to fall through more clouds droplets. Examining cloud fraction (only possible for the mean value) demonstrates that the cloud fraction also increased inside the plume in the volcano simulation compared to the no-volcano simulation. Similar to the result for LWP, this mean increase cannot be attributed confidently to the volcanic aerosol. It is unclear for the MODIS data how much change in cloud fraction between inside and outside the plume is due to the enhancement of cloud lifetime due to the additional CCN and how much is simply because of different weather. To learn about the climate implications, it is essential to identify how the planetary albedo differs inside and outside the volcano plume. In this study, the difference in increase in TOA albedo between inside and outside the volcano plume in the volcano and no-volcano simulations was quantified at 42 % when considering the volcanic aerosol vs. only 27 % without it; however it is, again, not possible to attribute the enhancement in TOA albedo in the CERES observations.

Overall, the results from this detailed analysis using level-2 satellite observations and cloud-system-resolving simulations confirm the key result of that there is a clear, detectable, and attributable impact of the volcanic aerosol on Nd, but there is on average only a very small, not attributable, effect on both LWP and cloud fraction. This net result for the case of the Holuhraun volcano for LWP comes about by a slight enhancement of LWP for large-LWP clouds compensated for by a decrease in LWP in low-LWP clouds.