Aerosol interactions with clouds and radiation constitute a major source of uncertainty in estimates of total radiative forcing. Aerosol particles can scatter and absorb solar radiation, causing a local cooling or heating. The altered temperature profile may in turn induce changes in cloud cover, whereby the so-called semi-direct effect describing dissipation of clouds in response to local heating is one of several possible adjustments . The resulting radiative forcing, including the cloud adjustments to the altered temperature profile, is referred to as effective radiative forcing from aerosol–radiation interactions. Aerosols can further modify the cloud albedo since an increase in the number of aerosol particles leads to more numerous and smaller cloud droplets for a cloud with a given liquid water content. This enhancement in cloud reflectivity is known as the cloud albedo effect . An increase in cloud droplet number concentration can further lead to suppression of precipitation since the formation of rain droplets is less efficient for a higher number concentration of smaller cloud droplets, and this rapid adjustment is referred to as the cloud lifetime effect . The cloud albedo and cloud lifetime effects are part of the effective radiative forcing from aerosol–cloud interactions. The overall effect of aerosol–radiation interactions, aerosol–cloud interactions, and the related rapid adjustments is estimated to be negative but with a substantial uncertainty (-0.9 (-1.9 to 0.1) Wm-2) . The vertical distribution of aerosols is one important factor for determining the aerosol effect on the radiative budget for both aerosol interaction with clouds, which requires vertical colocation, and aerosol interaction with radiation.

Model intercomparisons and comparisons with observations have shown large disagreement in the vertical distribution of aerosols in general and absorbing aerosols in particular, with large regional variation . Model diversity and uncertainty in radiative forcing from aerosol–radiation interactions have been found to be largely attributable to the vertical distribution of black carbon (BC), the main absorbing aerosol type . found that models overestimate BC concentrations over the remote Pacific compared to aircraft observations, whereas the quantity of biomass burning aerosols above clouds has been found to be underestimated in models over the southeast Atlantic and often prescribed as too reflective . This is in agreement with , who found that aerosols above the cloud layer occur in CMIP5 (Coupled Model Intercomparison Project phase 5) models without reducing the scene albedo.

Highlighting the diversity among climate models, compared vertical profiles of aerosol extinction of AeroCom (Aerosol Comparisons between Observations and Models) phase I models with satellite observations, and further investigated whether models from AeroCom phase II have improved compared to phase I models, focusing on regional and seasonal variability. Although the models were found able to reproduce the general features of the observed aerosol distribution, with a decrease in aerosol extinction from the surface up to 5 km, many models fail to capture the shape of the aerosol distribution in more detail.

The large model diversity and poor agreement with observations motivate further investigation of which processes are important for determining the vertical distribution of aerosols in global models. accordingly investigated various factors affecting the vertical aerosol distribution in two models (HadGEM3-UKCA and ECHAM5-HAM2), pointing at the importance of removal processes, which is also supported by the findings from , who found a large sensitivity of BC lifetime to wet scavenging in a chemical transport model. Studying biomass burning aerosols in particular, rather point at injection height and vertical transport as the main reasons for discrepancies between their chemical transport model and satellite observations. In the present study, we add to the generalizability of these previous results by testing the sensitivity to several processes that can control the vertical distribution of aerosol in another climate model, NorESM1-M. The sensitivity experiments performed are classified into five categories, following : emissions, transport, microphysics, deposition, and aerosol optical properties. Although some of the sensitivity experiments target specific aerosol types, we focus the evaluation on total aerosol extinction and number concentration, without discriminating between absorbing and reflecting aerosols, to give a full description of the vertical aerosol distribution in the model and to facilitate a comparison with observational estimates of total extinction.

While the analysis by is on global scale, we focus here on the regional scale and investigate five subtropical marine stratocumulus regions defined by . The radiative properties of the clouds in these regions, and their potential alteration by aerosol influence, remain a key challenge in climate models . Further, both absorbing and reflecting aerosols (BC, organics, and dust) located above the cloud layer have been identified in observations of these regions, which display a variety of aerosol signatures in terms of types and column burdens. To evaluate the model performance against observations, we use the 5 km aerosol profile product of CALIOP (Cloud–Aerosol Lidar with Orthogonal Polarization) version 4.10. A description of the satellite data retrievals can be found in Sect. , while a description of the climate model NorESM1-M and the model simulations performed is provided in Sect. . The results and further discussion are presented in Sects.  and , respectively. We summarize the most important processes that control the vertical aerosol distribution in the climate model NorESM1-M in the given regions in Sect.  and thereby give guidance for evaluating and improving this and other state-of-the-art climate models.

CALIOP is onboard the CALIPSO (Cloud–Aerosol Lidar and Infrared Pathfinder Satellite Observations) satellite as part of the A-train constellation. The satellite was launched in the year 2006, and we used data for the time period 2007 to 2016. We used the Level 2, 5 km aerosol profile product, version 4.10, of CALIOP lidar data, which has shown better agreement with the aerosol optical depth (AOD) from observations compared to the previous CALIOP version .

CALIOP measures backscattered radiation at two wavelengths and derives the aerosol extinction with an algorithm including iterative adjustment of the lidar ratio, i.e., the ratio between the extinction cross section and 180∘ backscatter cross section. We use the 5 km aerosol profile product , which provides profiles of the total aerosol extinction coefficient. A detailed product and data processing algorithm description can be found in and . Only wavelength 532 nm is considered here, as these measurements have a better signal-to-noise ratio than those at wavelength 1064 nm . Due to the higher detection sensitivity for aerosols in the night , we use only nighttime data, following and . We apply several additional data screening criteria, following . The cloud–aerosol discrimination (CAD) score distinguishes between clouds and aerosols, with a negative CAD score representing aerosol and a positive value representing cloud. We use a CAD score greater than -80 for a higher confidence in identifying aerosol . We also examine the quality of the extinction retrieval, represented by the extinction quality control (QC) filter, which stores information about the initial and final state of the lidar ratio at each layer. We use only cases in which the initial lidar ratio remains unchanged during the iterative solution process, referred to as an unconstrained retrieval (QC = 0), or use constrained retrievals (QC = 1) for which the initial lidar ratio was adjusted during the retrieval process by using measurements of a layer two-way transmittance, both with a higher confidence in the algorithm solution. Furthermore, we reject retrievals with a high extinction uncertainty of 99.9 km-1, thereby avoiding high biases in aerosol extinction.

Our analysis focuses on five regions of low marine stratocumulus clouds, following : Australian (25–35∘ S, 95–105∘ E), Californian (20–30∘ N, 120–130∘ W), Canarian (15–25∘ N, 25–35∘ W), Namibian (10–20∘ S, 0–10∘ E), and Peruvian (10–20∘ S, 80–90∘ W). The CALIPSO satellite overpasses the Equator twice per day. The temporal resolution of the lidar is 5 s, and the snapshots for each given satellite overpass are aggregated to a uniform 2∘ latitude by 5∘ longitude grid with a vertical resolution of 60 m so that each provided data file within each region contains multiple aerosol profiles. We average all profiles in the latitude and longitude range of each region to obtain a daily mean profile; a minimum number of 10 profiles at each vertical layer is thereby required to avoid high biases in aerosol extinction in the upper troposphere. In addition, to allow for a better comparison with the coarser model resolution of 26 vertical layers, we linearly interpolate the daily mean lidar profiles to the altitudes corresponding to the model levels. By choosing linear interpolation rather than averaging the vertical CALIOP levels, the original shape of the profile is still preserved. The daily mean profiles are further averaged over the whole 10-year period to obtain a climatological annual mean.

Discrepancies between the control simulation and CALIOP satellite data were found in all focus regions with regard to the total aerosol extinction and shape of the vertical distribution. In particular, the model underestimates the absolute values of aerosol extinction, showing a steady decrease from the surface, while observations indicate a maximum in the boundary layer. An adaptation of the CALIOP vertical resolution to the equivalent model resolution gives better agreement. The maximum in the boundary layer is not captured with a coarser, model-like vertical resolution for CALIOP. This emphasizes the importance of the vertical resolution to resolve mixing and transport processes in the lower troposphere. Also, increased model diffusion at lower model resolution might play a role. However, the model also underestimates aerosol extinction of elevated aerosol layers seen in two regions in the observations even if compared to the adapted CALIOP resolution.

It is also worth noting that while the observations are taken from the period 2007–2016, the emissions used in the model simulations (except in the Aero2010 experiment) are for the year 2000, and year-to-year variability in aerosol emissions may contribute to discrepancies between observed and modeled vertical profiles.

The sensitivity experiments performed suggest that the alterations that have the largest impact on the aerosol vertical profiles are found in the categories emissions, deposition, and vertical transport, whereas changes in the categories microphysics and aerosol optical properties have less of an effect. However, none of the chosen alterations of parameters and processes affecting the vertical distribution of aerosol extinction in the model are sufficient to reproduce the observed distribution. For instance, the emission height of biomass burning aerosols directly influences the aerosol vertical profile. This is despite the alterations in many cases going beyond what might be considered a realistic range, i.e., by turning processes off completely (e.g., in the case of wet deposition and autoconversion) or choosing extreme parameter values (e.g., in the case of vertical velocity). One example of a modification that does affect the vertical profile towards better agreement with observations in the Namibian region is the emission height of biomass burning. Inserting these absorbing aerosols above or within the boundary layer leads to increased aerosol extinction above the boundary layer, as expected. Biomass burning aerosol injection at the surface only, or uniformly in height, has less of an effect on the vertical profile, in agreement with .

The choice of the aerosol emission inventory was also found to be important for determining the magnitude of total vertically integrated aerosol extinction, in agreement with the findings of . By choosing aerosol emissions for the year 2010 a higher extinction and subsequently a higher AOD was produced, especially in biomass-burning-dominated areas. Considering the small interannual variability in biomass burning aerosol emissions from the main burning regions found by , the differences between the two emission data sets are more likely related to differences in resolution and data collection than to interannual variability. As discussed in and , emissions in GFED3 have increased compared to GFED2 due to an improved mapping approach of burned areas using MODIS and a higher resolution of 0.5∘ compared to GFED2 with 1∘ resolution. Previous studies have also pointed at the importance of the spatial and temporal resolution of aerosol emissions.

In terms of the vertical aerosol distribution, the updated emission data set leads to only a small change within the uncertainty range of the control simulation. similarly showed that using GFED3 instead of GFED2 biomass burning emissions leads to only a moderate improvement of the vertical BC distribution compared to observations, without statistical significance.

Another important factor that can control the vertical distribution of aerosol is the size of emitted aerosol particles. The sensitivity experiments performed are mass-conservative, except the experiment with an alternative emission data set, meaning that changes in emission particle sizes lead to a shift in the entire size and number distribution. Here we find that the shape of the vertical distribution in the model is highly sensitive to the size of emitted particles. Decreasing the size results in more numerous smaller particles and produces a maximum in aerosol extinction in the boundary layer in the Canarian, Namibian, and Peruvian regions. This is not only an effect of changes in aerosol number concentration and size distribution, but also of the resulting shift in aerosol composition produced by the model in response to the change in size distribution.

Large responses were also seen in the sensitivity experiments focusing on removal processes, particularly for the cases of altered wet deposition. Dry deposition mainly affects larger particles, and cutting this sink off leads to a small decrease in extinction throughout the vertical column, except in the Peruvian region. An additional reason for the small effect of reducing dry deposition is that this shifts the aerosol removal to wet deposition, which increases accordingly. Hence, the small sensitivity of aerosol extinction and number to turned-off dry deposition is not necessarily an indication that this process is not relevant, but rather that changes are compensated for by other processes. Wet deposition, on the other hand, affects all particles and is the major removal process for aerosol particles in the model. Cutting off this removal pathway leads to a large increase in extinction and a modified shape of the vertical distribution. In-cloud scavenging contributes more than below-cloud scavenging to the total wet deposition, and hence turning off in-cloud scavenging has similar effects as turning off wet deposition completely, while turning off below-cloud scavenging has little effect, in agreement with and . Hence, the representation of wet deposition is important for the vertical aerosol distribution in the model, in agreement with the findings of , , and . Changes in the removal processes also affect the aerosol composition in the model. Inhibited wet deposition increases the amount of sulfate, BC, and OM, as this is the main removal process for these aerosol types, but decreases the relative amount of dust, which is less affected by this removal process. The smaller portion of wet deposition that is due to below-cloud scavenging also affects composition but is less efficient for Aitken- and accumulation-mode particles, a size range in which, e.g., BC is found.

discussed the coupling between wet scavenging and convective transport as well as its importance for the representation of the vertical aerosol distribution by comparing HadGEM-UKCA with ECAHM5-HAM2 and with observations. The in-plume approach, with wet scavenging directly linked to the convective scheme, implemented in NorESM1-M is in line with the recommendations in .

Turning off either of the convective schemes, shallow or deep convection, does not switch off convective transport of aerosols completely; i.e., switching off shallow convection still allows deep convection and vice versa. However, the complete inhibition of the shallow convective scheme largely affects the aerosol distribution. Without the shallow convection scheme, i.e., allowing only deep convection, the shape of the vertical distribution changes, with a more pronounced increase close to the surface. Particles remain closer to the surface as they cannot be lifted higher, leading to an increase in aerosol number concentration and extinction, especially in the boundary layer. Hence, shallow convection in the model is essential for transporting aerosols to the middle troposphere in the focus regions, consistent with , who showed that vertical transport of aerosol on the global scale is dominated by convective processes on unresolved scales. further highlighted the importance of the parameterization of convective processes for tracers with a short lifetime. Another important transport process for aerosols is entrainment, and cutting off this mixing for convective clouds results in a decrease in extinction in the boundary layer and an increase in the upper troposphere in the biomass burning regions. However, the entrainment particularly controls the amount of aerosol above the boundary layer and is crucial for the formation of cloud droplets via provision of CCN. Entrainment can have a strong effect on the characteristics of parameterized convective clouds (see, e.g., ).

Microphysical processes, though linked to wet removal processes, have less of an impact on the vertical aerosol distribution. Altering the process of autoconversion results only in small changes in aerosol number and extinction, and only the extreme scenario of switching off autoconversion completely in warm clouds leads to a significant increase in aerosol number and extinction in the boundary layer. However, autoconversion and the subgrid vertical velocity are important processes in the model regarding cloud properties. Previous studies pointed at the importance of the autoconversion parameterization for aerosol indirect effects e.g., and the representation of the cloud lifetime effect in models . further showed that the difference between microphysics schemes (and their autoconversion in particular) can be greater than the non-albedo aerosol indirect effects. Also, the importance of the subgrid variability of the vertical velocity when estimating aerosol indirect effects was highlighted , and demonstrated the importance of subgrid vertical velocity variability in another model.

Finally, turning to optical properties, our results indicate that they have little impact on the vertical aerosol profile. point at aerosol above clouds in climate models as being underestimated in amount but too reflective. They found an improved representation of model output compared to satellite observations for climate models with an imaginary part of the refractive index of 0.71 compared to models with a lower value of 0.44. The refractive index was defined therein at a wavelength of 0.55 µm. NorESM1-M has a high default refractive index for pure BC with a value of 1.0 so that BC is prescribed as fully absorbing for the entire visible spectrum.

In contrast to other models, however, BC can be internally mixed and coated, thereby becoming more reflective. A decrease in the refractive index causes almost no change in the extinction coefficient. The SSA, on the other hand, is increased as expected. Hence, while found that climate models underestimate absorption by aerosol above clouds, primarily due to the properties of BC, our results indicate that for NorESM1-M it is the aerosol amount rather than the optical properties of pure BC that determines the aerosol extinction above clouds.

In this study the sensitivity of the climate model NorESM1-M to changes in processes affecting the vertical aerosol distribution was studied, focusing on five regions of subtropical marine stratocumulus clouds.

To evaluate the model performance, a control simulation was compared with satellite-borne lidar observations from CALIOP. The magnitude of aerosol extinction is underestimated in the model and displays a differently shaped vertical distribution. Discrepancies are of similar magnitude as those found for other models see, and the main difference in shape is the lack of a local maximum in aerosol extinction in the boundary layer, which is also a common feature among many previously investigated models. The model also underestimates the aerosol extinction of elevated aerosol layers above the boundary layer, as seen in observations in two of the studied regions.

None of the alterations made here were sufficient for reproducing the observed aerosol extinction, but better agreement between observations and the model in terms of the shape of distribution in the boundary layer was found by interpolating the vertical resolution of observations to the model levels. This highlights the importance of the vertical model resolution to capture aerosol processes, especially in the boundary layer. Observed local extinction maxima above the boundary layer appear in observations with both original and reduced vertical resolution, indicating that the model resolution does not restrict the representation of aerosol layers above clouds.

Among the categories in which sensitivity experiments are performed, the largest impact on the vertical distribution of aerosol extinction is found to result from alterations to emissions, deposition, and vertical transport and less from microphysics and aerosol optical properties. In this sense, the presented results show general agreement with , who conducted similar sensitivity experiments using a different model and focusing on the global mean. In particular, for our model the parameters and processes found to have the greatest effect on the shape of the vertical aerosol distribution in the dynamical regime studied are the altitude of emissions and size of emitted particles, as well as the representation of shallow convection, entrainment, and wet scavenging.

By emitting all biomass burning aerosol at the highest injection level or within the boundary layer in the model, an increase in aerosol extinction above the boundary layer can be produced but is still underestimated compared to the local maxima seen in observations in two regions. Emitting aerosol at higher altitude or within the boundary layer is the most efficient way of increasing extinction above cloud level, which highlights the importance of mixing processes in the boundary layer.

The shallow convection scheme is also important for transporting aerosols up from the boundary layer, and by switching off shallow convection, the aerosol extinction increases in the boundary layer. However, the resulting profile has an increase in aerosol extinction towards the surface that is much too strong compared to observations and does not indicate improved agreement with the observed shape compared to the control experiment.

Turning off in-cloud scavenging leads to a maximum in aerosol extinction in the boundary layer, in qualitative agreement with observations. Similar changes in vertical aerosol distribution are seen when the size of emitted particles is reduced. This qualitative improvement of the modeled aerosol profile suggests that wet scavenging might be too efficient in the model and that the emission size distribution may be shifted towards particles that are too large.

With a focus on a specific dynamic regime, our sensitivity experiments indicate which processes have the greatest potential to influence the vertical distribution of aerosol in NorESM1-M, finding general agreement with previous studies based on other models. Our results support and give guidance for further improvement of the representation of aerosol distribution and thereby aerosol–cloud interactions in this and other state-of-the-art climate models.