Weaker cooling by aerosols due to dust-pollution interactions

The interactions between aeolian dust and anthropogenic air pollution, notably chemical ageing of mineral dust and coagulation of dust and pollution particles, modify the atmospheric aerosol composition and burden. Since the aerosol particles can act as cloud condensation nuclei, this not only affects the radiative transfer directly via aerosol-radiation interactions, but also indirectly through cloud adjustments. We study both radiative effects using the global ECHAM/MESSy atmospheric chemistry5 climate model (EMAC) which combines the Modular Earth Submodel System (MESSy) with the European Centre/Hamburg (ECHAM) climate model. Our simulations show that dust-pollution interactions reduce the cloud water path and hence the reflection of solar radiation. The associated climate warming outweighs the cooling which the dust-pollution interactions exert through the direct radiative effect. In total, this results in a net warming by dust-pollution interactions which moderates the negative global anthropogenic aerosol forcing at the top of the atmosphere by (0.2 ± 0.1) Wm−2. 10

Large-scale clouds are simulated by the submodel CLOUD (Jöckel et al., 2006) where different parametrisations of cloud droplet formation and ice nucleation are implemented. We use a two-moment stratiform cloud microphysics scheme (Lohmann et al., 1999(Lohmann et al., , 2007Lohmann and Kärcher, 2002) in combination with the UAF (Unified Activation Framework) cloud droplet activation parametrisation (Kumar et al., 2011;Karydis et al., 2011Karydis et al., , 2017. For the ice crystal formation we use the comprehensive parametrisation for cirrus and mixed-phase clouds implemented by Bacer et al. (2018) based on Barahona and Nenes 5 (2009). Convective clouds are calculated by the CONVECT submodel (Jöckel et al., 2006), where interactions with aerosols are not taken into account. CONVECT provides a choice of convection schemes (Tost et al., 2006b), and here we use the scheme of Tiedtke (1989) including modifications by Nordeng (1994). The optical properties of clouds which serve as input for the radiative transfer submodel RAD are computed by the submodel CLOUDOPT (Dietmüller et al., 2016). The model yields a global annual mean cloud liquid water path around 80 g m −2 (Table 1 and Table S3 in the supplement), which is well 10 within the range of other climate model results (32 to 125 g m −2 , Lebsock and Su, 2014) and observations (30 to 90 g m −2 , Lohmann and Neubauer, 2018). Likewise, the modelled annual mean global cloud ice water path of about 15 g m −2 (Table 1 and Table S3 in the supplement) is consistent with results from other models (e.g., 14.8 g m −2 , Lohmann and Neubauer, 2018) and close to observed values (e.g., (25 ± 7) g m −2 , Li et al., 2012).
A complete list of the MESSy submodels used in our simulations is provided in Table S2 in the supplement. Descriptions of 15 each submodel and further references can be found online in the MESSy submodel list (MESSy 2020).

Methodology
To estimate the global aerosol ERF, we study the radiative fluxes in simulations with prescribed sea surface temperature (SST) and sea ice climatologies. The ERF accounts for rapid adjustments by radiative and dynamical feedbacks, whereas it excludes long-term climate responses involving the much slower thermal equilibration of the oceans. Due to the limited constraints on 20 the atmospheric dynamics in SST simulations, the meteorological variability is large and hence a sufficient number of years has to be simulated to obtain statistically significant results. We perform SST simulations long enough to yield significant globally averaged results, however detailed regional analysis would require much longer SST simulations. In order to nevertheless gain insights from regional evaluation, we additionally use simulations where the model dynamics above the boundary layer is nudged to meteorological analyses of the European Centre for Medium-Range Weather Forecasts (ECMWF). Within the 25 boundary layer and to some extend above, nudged quantities like the temperature may still respond to other variables such as radiative fluxes (soft nudging). The nudging greatly reduces the influence of inter-annual variability on statistical analysis. The results from the nudged simulations turn out to be largely consistent with those of the SST simulations (Tabs. 1, 2 vs. Tabs. S3, S4 in the supplement), in particular the estimates for the total global radiative effect of the dust-pollution interactions agree within the error bounds, so that the use of nudged simulations for the regional analysis is reasonable and helpful.

30
With prescribed SST we run ensembles of 16 simulations, each covering one year. The ensemble members are obtained by perturbing temperature and humidity in the fourth year of a common spin-up simulation, followed by an additional spin-up of the individual ensemble members to attain a total of 5 spin-up years. The perturbation is implemented by adding a uniformly distributed random variable ranging from -0.1 ‰ to 0.1 ‰ of the perturbed quantity so that the perturbation is numerically but not meteorologically relevant. Emission data for 2010 is used for all simulations. The nudged simulations cover 10 years from 2006 to 2015, and two simulation years prior to that period were used for the model spin-up. To estimate the uncertainties of the 10-year mean values for the nudged simulations and the ensemble mean values for the SST simulations, we compute the standard error of the mean of the annual values. 5 We apply a similar analysis as Klingmüller et al. (2019) which is based on simulations with four different emission set-ups: a baseline simulation with neither dust nor anthropogenic emissions ("0"), a simulation with dust but without anthropogenic emissions ("dust"), a simulation with anthropogenic pollution but without dust emissions ("pol") and a full simulation considering all emissions. Thus, in total we have performed four nudged simulations and four times 16 SST simulations.
In the anthropogenic pollution free simulations ("0", "dust") we disable the EDGAR emissions including SO 2 , NH 3 , NO x , 10 black-and organic carbon emissions, but retain the greenhouse gases. We attribute 90 % of the GFED biomass burning emissions to human activities (Levine, 2014) and reduce them accordingly, whereas we do not consider anthropogenic factors on dust emissions such as land use and climate change (Klingmüller et al., 2016), assuming all dust emissions to be natural.
A result x from the full simulation (e.g., the annual global mean cloud liquid water content) is related to the corresponding result from the baseline simulation x 0 by which represents the effect of the dust-pollution interactions. In the absence of any interactions, the term ∆ int x vanishes. We apply Eq.
(2) to the annual mean cloud liquid-and ice-water paths and radiative fluxes, both globally averaged and, only for 20 the nudged simulations, per grid cell.
To compute the direct radiative effect of aerosols, the radiative transfer code is called twice for every model time step. The first call considers the aerosol-radiation interactions and is used to calculated the heating rates affecting the temperature, the second call ignores the aerosol-radiation interactions and computes the radiative fluxes and heating rates only for diagnostic output. The difference of the radiative fluxes from both calls yields the instantaneous forcing due to the direct radiative effect of 25 aerosols. Since both calls are performed with identical clouds, only little statistical noise is introduced by the strong variability of clouds. Nevertheless, in this way we obtain the direct radiative forcing in the presence of clouds, which is typically smaller than the clear sky forcing. The direct radiative forcing is subtracted from the total radiative forcing to extract the indirect radiative forcing of aerosols.

30
Hydrophilic particulate anthropogenic pollution enhances the cloud droplet formation and thus the liquid water content (Table 1). However, in the presence of mineral dust particles this effect is reduced because fine pollution particles coagulate with coarse dust particles decreasing the particle number and virtually cleaning the atmosphere from fine particulate pollution.
Moreover, the adsorption activation of mineral dust particles occurs early on in the cloud formation process (Kumar et al., 2011), reducing the maximum supersaturation and inhibiting the activation of small pollution particles. These effects reduce the number of cloud condensation nuclei (Karydis et al., 2011(Karydis et al., , 2017 and decrease the cloud liquid water path as shown in Fig. 1 (a). Especially over East and South Asia, where strong pollution emissions mix with aeolian dust from the Taklamakan,

5
Gobi and Thar deserts, the reduction is substantial and regionally exceeds -40 g m −2 . Even over polluted regions in Europe and the USA which are only occasionally exposed to dust intrusions, we obtain a small but significant reduction. This negative impact of the dust-pollution interactions over large parts of the northern hemisphere leads to a reduction of the global mean cloud liquid water path in Fig. 1 (a) by (-1.10 ± 0.03) g m −2 . A comparable reduction by (-1.5 ± 0.2) g m −2 is obtained in the SST simulations (Tab. 1). Relative to the mean liquid water path in the SST simulation considering all emissions (85.5 10 ± 0.1) g m −2 , these reductions appear to be rather moderate. The reason is that the transport time periods between most of the major dust sources, especially the Sahara and the Middle East, and major pollution sources like Northern America and Europe to a large degree exceed the dust aerosol lifetime. In Asia these sources are less distant while pollution emissions are generally larger. Thus, the strong effects over Asia might provide an outlook for regions with emerging pollution sources close to dust sources in Africa and the Middle East. But already today, due to the critical influence of clouds on radiative transfer, 15 the relatively small changes of the water paths cause substantial radiative forcings as will be discussed in the next section.
The dust-pollution interaction effect on the cloud ice water path, shown in Fig. 1 (b), is less distinct. A negative impact is obtained over the Sahel. The direct radiative effect of mineral dust over the Sahara warms the atmosphere by absorption of solar radiation (Fig. S1 (a) in the supplement). This increases the atmospheric capacity to hold moisture and the vertical water vapour transport (Fig. S1 (b) in the supplement). As a result, more moisture is available for ice cloud formation (Fig. S1 (c) 20 in the supplement). Since the net direct radiative effect of the dust-pollution interactions cools the atmosphere over the Sahara The reduction of the cloud water content by dust-pollution interactions has a significant impact on the transfer of solar radiation ("shortwave", SW), which is shown in Fig. 2 (a). With reduced liquid cloud water, less solar radiation is reflected back to space resulting in a net positive forcing at the top of the atmosphere (TOA). Comparing Fig. 2 (a) and Fig. 1 (a) reveals the one-to-one correspondence of the dust-pollution interaction effect on the liquid cloud water and solar radiation. Over the polluted regions 5 of the northern hemisphere, i.e., Asia, Europe and North America, and over the Atlantic Ocean along the North African coast in the Saharan dust outflow, the positive forcing can exceed 2 Wm −2 . Globally averaged, the net forcing in the solar spectrum shown in Fig. 2 (a) is (0.23 ± 0.01) Wm −2 , the SST simulations yield an ERF of (0.3 ± 0.1) Wm −2 (Tab. 2).
On the other hand, the dust-pollution interaction effect on the terrestrial spectrum ("longwave", LW), Fig. 2 (b), is directly related to the effects on ice clouds, Fig. 1 (b). This is most distinct over the Sahel, but also apparent over the East Asian deserts.

10
The reduced cloud ice water path over these regions traps less outgoing terrestrial radiation resulting in a net cooling from the dust-pollution interactions. Over the Sahel the terrestrial TOA forcing reaches -2 Wm −2 . With regard to the radiative energy budget, the regions with a significant dust-pollution interaction effect on cloud ice in Fig. 1 (b) are of different relevance.
The Sahel, where the dust-pollution interactions reduce cloud ice, is relatively close to the equator and accordingly stronger radiative fluxes are affected by the cloud ice changes than in the other regions, hence the global net radiative effect related 15 to cloud ice is more relevant than the global net effect on cloud ice itself. Globally averaged, the net forcing in the terrestrial spectrum shown in Fig. 2 (b) is (-0.05 ± 0.01) Wm −2 , and the SST simulations yield an ERF of (-0.08 ± 0.09) Wm −2 (Tab. 2).
Thus, a substantial positive forcing in the solar spectrum is partially compensated by a negative forcing in the terrestrial spectrum to yield a still considerable, positive net-forcing associated with the effect of dust-pollution interactions on clouds.
The global distribution of the total net-forcing at the TOA including the direct radiative effect is shown in Fig. 3. The regional 20 forcing ranges from below -2 Wm −2 over the Sahel to above 2 Wm −2 over Asia. Even though overall these contributions partially counterbalance, with (0.15 ± 0.02) Wm −2 the corresponding global mean forcing in Fig. 3 is significantly positive.
Consistently, the ERF in the SST simulations is (0.2 ± 0.1) Wm −2 (Tab. 2). The cloud water path is reduced by the dust-pollution interactions as they moderate the cloud water path increase caused by 5 anthropogenic pollution. The reason for this moderation is that mineral dust particles decrease the number of anthropogenic cloud condensation nuclei by coagulation and additionally limit the activation of the fine hydrophilic anthropogenic particles by lowering the maximum supersaturation through adsorption activation. Dust-pollution interaction effects on the cloud ice content are noticeable as well, but less relevant.
The atmospheric radiative transfer is very sensitive to the reduction of the cloud water path. Generally, dust-pollution in-