2.1 Applied climate models and setup

We carry out modern-day equilibrium climate simulations with two independent climate models, ECHAM6.1 and NorESM1. ECHAM6.1 (Stevens et al., 2013) is the sixth generation of the ECHAM general circulation model developed in the Max Planck Institute with 47 sigma hybrid vertical levels, with the model top at 0.01 hPa and a horizontal resolution of $\mathrm{1.9}{}^{\circ }\times \mathrm{1.9}{}^{\circ }$. The original ECHAM model branched from an early version of the European Centre for Medium-Range Weather Forecasts (ECMWF) model for climate studies. NorESM1 is the Norwegian Earth system model with 26 sigma hybrid vertical levels (the highest model level at 2.9 hPa) and $\mathrm{1.9}{}^{\circ }\times \mathrm{2.5}{}^{\circ }$ horizontal resolution (Bentsen et al., 2013; Iversen et al., 2013; Kirkevåg et al., 2013). NorESM1 is based on the Community Climate System Model version 4 (CCSM4) operated at the National Center for Atmospheric Research (NCAR). Thus, the two models applied in our study do not share a common development history. Here, both models were run with identical fixed modern-day greenhouse gas concentrations. Oceans were simulated with the intrinsic slab ocean configurations of the models. This idealization removes the effect of natural and aerosol-induced variations in ocean circulation and restricts our study to the response in atmospheric circulation, oceanic heat exchange and sea ice dynamics only.

2.2 Standardized aerosol representation

MACv2-SP is a standardized representation of anthropogenic aerosol radiative effects, accounting for the direct radiative as well as the cloud albedo effect of anthropogenic aerosol (Stevens et al., 2017). However, the cloud lifetime effect is not taken into account. Anthropogenic aerosols are represented by nine 3-D time-varying Gaussian plumes defining the aerosol optical depth, single scattering albedo and asymmetry parameter. Four of these plumes represent aerosol emissions from biomass burning and the other five are associated with industrial emissions. The industrial plumes originate from Europe, North America, east Asia, south Asia and Australia, and the biomass plumes from north Africa, South America, south central Africa and the Maritime Continent (Fig. 1 and Table 1 in Stevens et al., 2017). The plumes differ in their annual cycle and optical properties, and have a realistic horizontal and vertical structure that represents the transports of aerosols with prevailing winds. The aerosol properties are based on aerosol climatology by Kinne et al. (2013), derived from ground-based Sun photometer networks (AERONET) merged onto background maps from global models participating in the Aerosol Model Intercomparison Project (AeroCom). The cloud albedo effect in MACv2-SP is parameterized by modifying the model-intrinsic natural cloud droplet number concentration (CDNC) via a relation based on the total change in aerosol optical depth (AOD). This parametrization is derived from Moderate Resolution Imaging Spectroradiometer (MODIS) data. MACv2-SP allows for a simple and observation-based representation of the changes in aerosol optical properties and cloud droplet number concentrations due to anthropogenic aerosols.

2.3 Model experiments and analysis

Sets of 100-year equilibrium climate runs for the year 2005 were conducted with both models, with the last 60 years used for the analysis. (1) The control run (CTRL) included only natural aerosols and was constructed from two runs for each model with small initial condition perturbations. (2) The MACSP run included both natural and anthropogenic aerosols for the year 2005. In addition, for NorESM1, a third run (EF) was carried out. This run employed the time-varying 3-D aerosol radiative forcing field computed from the ECHAM6's MACSP run. A more detailed description of the implementation is given in the Appendix A. A summary of the runs is given in Table 1.

Table 1Summary of the performed model runs.

Download Print Version | Download XLSX

Based on these runs, the following three experiments were defined to estimate the effect of anthropogenic aerosols: ECHAM6-MACSP (the difference between the MACSP and CTRL runs for ECHAM6), NorESM1-MACSP (MACSP minus CTRL for NorESM1) and NorESM1-EF (EF minus CTRL for NorESM1). The analysis of the results was based on monthly-mean values of data and focused on the effects of MACv2-SP aerosols on near-surface temperature, precipitation, surface albedo and total cloud cover. The statistical significance of the responses was evaluated using a Student's t test with an auto-correlation correction according to Zwiers and von Storch (1995). The response uncertainties in global-mean values were estimated by the standard error of means taking into account lag-1 auto-correlation according to Zwiers and von Storch (1995). The instantaneous radiative forcing was calculated using double radiation calls with and without MACv2-SP aerosols during the slab ocean runs.

3.1 Aerosol radiative forcing

The total radiative forcing from the MACv2-SP anthropogenic aerosol description was found to be very similar for the two models (see Fig. 1). For ECHAM6, the MACv2-SP aerosol scheme produces a $-\mathrm{0.64}{\mathrm{Wm}}^{-\mathrm{2}}$ global average total shortwave radiative forcing at the top of the atmosphere (TOA) for the year 2005, with $-\mathrm{0.35}{\mathrm{Wm}}^{-\mathrm{2}}$ arising from direct and $-\mathrm{0.29}{\mathrm{Wm}}^{-\mathrm{2}}$ from indirect radiative forcing. For NorESM1, the same aerosol scheme produces a slightly higher global radiative forcing of $-\mathrm{0.69}{\mathrm{Wm}}^{-\mathrm{2}}$ at TOA, with $-\mathrm{0.36}{\mathrm{Wm}}^{-\mathrm{2}}$ direct and $-\mathrm{0.33}{\mathrm{Wm}}^{-\mathrm{2}}$ indirect radiative forcing. Figure C1 shows the maps of aerosol direct and indirect radiative forcing in the two models as calculated here. The largest difference in the total forcing was found over southeast Asia (up to 3.20 Wm⁻²), where also the largest absolute forcing was found in both models. Fiedler et al. (2019) have calculated both the MACv2-SP effective radiative forcing and the instantaneous radiative forcing using double radiation calls with fixed sea surface temperature for the two climate models used here. They showed that with fixed sea surface temperature the MACv2-SP aerosols produce an instantaneous radiative forcing of −0.60 and $-\mathrm{0.68}{\mathrm{Wm}}^{-\mathrm{2}}$ in ECHAM6 and NorESM1, respectively. The correlation coefficient for the regional total forcing in the two models due to MACv2-SP is 0.97, and 0.90 for direct and 0.89 indirect forcing only. Thus, the regional differences in direct and indirect forcing somewhat compensate for each other.

Figure 1The total radiative forcing at top of the atmosphere produced by MACv2-SP aerosols. Panel (a) shows the forcing in ECHAM6-MACSP experiment and (b) in NorESM1-MACSP experiment. Panels (c) and (d) show the difference in forcing between the two models and difference between ECHAM-MACSP and NorESM1-EF runs. Small green circles mask the areas where results are not statistically significant at the p<0.05 level.

We used a Gaussian process emulation technique (O'Hagan, 2006) to assess the causes for the regional differences in aerosol radiative forcing (see Appendix B for details). Our analysis showed that differences in cloud cover and surface albedo can explain nearly all of the variance in the difference in total instantaneous shortwave radiative forcing between ECHAM6 and NorESM1. Our sensitivity analysis reveals that in the regions with the largest radiative forcing (close to the center of the MACv2-SP plumes) the difference in model cloud cover dominates the difference in model shortwave forcing. In contrast, in regions with low aerosol radiative forcing, the differences in surface albedo dominate the differences in forcing. We note that these results apply only to fixed aerosol fields produced by the MACv2-SP representation. Previous research shows that the aerosol radiative forcing can also depend on the meteorology (surface winds and precipitation) produced by the models, partly driven by the natural variability of the climate system (Fiedler et al., 2019).

3.2 Climate response to the addition of anthropogenic aerosols

3.2.1 Temperature

We obtain a robust global temperature response of −0.5 K due to the inclusion of MACv2-SP anthropogenic aerosols in both models. For the ECHAM6-MACSP experiment, the global-mean near-surface temperature response is −0.50 (±0.03) K, with regional values ranging from +0.30 to −2.10 K. For the NorESM1-MACSP experiment, the global-mean value is −0.48 (±0.02) K and the regional values range between +0.39 and −2.28 K.

Figure 2Near-surface temperature response to the addition of anthropogenic (MACv2-SP) aerosols. Panel (a) shows the response for ECHAM6-MACSP experiment and (b) for NorESM1-MACSP experiment. Panels (c) and (d) show the difference in the responses between the two models and difference between NorESM1-MACSP and NorESM1-EF. Small green circles mask the areas where results are not statistically significant at the p<0.05 level.

Figure 2 shows the regional temperature response to the inclusion of anthropogenic MACv2-SP aerosols. The spatial correlation between ECHAM6-MACSP and NorESM1-MACSP experiments is 0.81 for full experiments with 60+120 years of MACSP and CTRL runs in both models. The largest cooling in ECHAM6 is located in southeast Asia, whereas in NorESM1 the largest cooling is found near the Russian Far East and north of Japan, with a second minimum over the Greenland Sea. Small positive temperature responses are found close to the Antarctic coast in both models, but these temperature responses are not statistically significant and are related to natural variations in sea ice. We found some significant correlation between the regional aerosol forcing and regional temperature response in both models: 0.39 in ECHAM6 and 0.29 in NorESM1, respectively. Among the CMIP5 model considered in Shindell et al. (2015), the multi-model mean regional correlation between the combined effective aerosol and ozone forcing and temperature response was slightly negative (−0.1), varying between negative values in some models and positive values among others.

Figure 3Impact of MACSP anthropogenic aerosols on zonal-mean temperature (K) and precipitation (%) in ECHAM6-MACSP, NorESM1-MACSP and NorESM1-EF experiments. The shaded area shows the standard error of the mean as a function of latitude.

Download

Figure 3a shows the zonal-mean temperature responses obtained from ECHAM6-MACSP and NorESM1-MACSP experiments. These experiments show a moderate cooling due to anthropogenic aerosols across the Southern Hemisphere latitudes, whereas in the Northern Hemisphere the cooling response clearly strengthens towards the high latitudes. The modeled regional temperature responses between ECHAM6 and NorESM1 simulations disagree the most in mid- and high-latitude regions, as seen in Fig. 2c. In high-latitude regions, temperature differences are associated with surface albedo responses (snow/sea ice) between the models (see Fig. C2). Changes in surface albedo are known to amplify changes in Arctic temperatures (albedo feedback). Hence, differences in snow and sea ice responses may partly explain the difference in temperature responses in the high latitudes. This feedback, together with ocean circulation feedback, also dominates at high latitudes the regional differences in temperature responses to homogeneous greenhouse gas forcing among different climate models (Shindell et al., 2015).

Figure 4Panel (a) shows the ECHAM6-MACSP experiment precipitation response to adding MACv2-SP aerosols and (b) shows the same for the NorESM1-MACSP experiment. Panels (c) and (d) show the intermodel difference in precipitation response and difference between NorESM1-MACSP and NorESM1-EF. The green dots mark the regions where the MACv2-SP aerosols do not have a statistically significant impact at the p<0.05 level.

3.2.2 Precipitation

The inclusion of anthropogenic aerosols results in a similar global reduction of precipitation in all experiments, with ECHAM6-MACSP showing a change of $-\mathrm{1.79}\pm \mathrm{0.05}$ % and NorESM1-MACSP showing a change of $-\mathrm{1.69}\pm \mathrm{0.04}$ % in annual precipitation (Table 2). The regional changes of the precipitation patterns are shown in Fig. 4. The spatial correlation between the precipitation responses in the full ECHAM6-MACSP and NorESM1-MACSP experiments is 0.47, which is much lower than the corresponding correlation for temperature. In addition, while the temperature responses are negative almost globally, both positive and negative responses occur for precipitation, with relatively sharp edges between regions with different signs of changes. While similar large-scale features of precipitation changes can be seen in both models, their dislocation leads to a weaker regional correlation than for the temperature response. In both models, the relative changes in the convective precipitation are larger than the relative changes in large-scale precipitation. Also consistently across the two models, the seasonal response in the total precipitation is similar, with the largest changes in June–July–August (see Table C1). Both models consistently show an overall drying of the Northern Hemisphere, with some statistically significant regional increases in precipitation over northwest Africa.

Table 2Summary of modern-day global-mean change of temperature and precipitation. Standard errors of means are shown in brackets.

Download Print Version | Download XLSX

Both models show a maximum reduction in total precipitation around 15–20^∘ N and a maximum increase around 10–15^∘ S, associated with an asymmetric response in Hadley circulation across the Equator (see Figs. 3b and 4). Changes in precipitation in the tropics are also related to changes in vertical motion in the same region (see Fig. C4). This is suggestive of a southward shift of the Intertropical Convergence Zone (ITCZ) associated with a change in hemispheric temperature gradient (Broccoli et al., 2006). The inclusion of anthropogenic aerosols results in decreased precipitation in the South Asian monsoon region (defined here as the land region over 5–25^∘ N, 65–110^∘ E) (Fig. 3). In June–August, the monsoon precipitation decreased by 12.8 % in the ECHAM6-MACSP and 15.3 % in the NorESM1-MACSP experiments. Reduction of monsoon precipitation due to the anthropogenic aerosols has also been reported in several previous studies (Ganguly et al., 2012; Li et al., 2018b; Polson et al., 2014; Bollasina et al., 2011). In contrast with the seasonal cycle in temperature response, the largest precipitation response occurs in Northern Hemisphere summer during the Asian monsoon season. The two models show a different response over the West African monsoon region (5^∘ S–25^∘ N, 20^∘ W–20^∘ E), with the NorESM1-MACSP experiment showing a statistically significant reduction in precipitation of −5.3 %, while the ECHAM6-MACSP experiment does not show a significant change (−1.8 %). In the vicinity of the Australian continent, the ECHAM6-MACSP experiment shows an area of increased precipitation extending from the Indian Ocean to western Australia, while in the NorESM1-MACSP experiment, the increase is located entirely over the Indian Ocean.

There appear to be several causes for the differences in the precipitation response between the two models. For instance, there is a relationship between the difference of the regional precipitation response and the difference in vertical velocity response (correlation coefficient 0.44 between Figs. 4c and C4c). However, it cannot be concluded that change in precipitation is caused by the change in vertical velocity. Probably, both the changes in vertical velocity and precipitation are related to changes in circulation. Also the difference in the initial equilibrium state of precipitation patterns correlates weakly with the difference in the precipitation response (correlation coefficient 0.23). Furthermore, differences in the cloud cover responses (see Fig. C3) are also related to differences in precipitation responses (correlation coefficient 0.32). The vertical velocity correlates also with the total cloud cover response (correlation coefficient 0.41).

3.3 Comparison to models with interactive aerosols

Finally, we compare the obtained equilibrium temperature and precipitation responses with prescribed MACv2-SP aerosols in ECHAM6 and NorESM1 against those equilibrium climate responses from four fully coupled climate models (CESM1, GISS, HadGEMS2 and NorESM1) with intrinsic aerosol schemes but the same aerosol emissions, reported by Samset et al. (2018). In the four models considered by Samset et al. (2018), the global average temperature responses were −(0.5, 0.5, 1.1 and 0.6) K, and precipitation responses were −(1.5 %, 1.8 %, 2.6 % and 3.1 %), respectively. We obtain similar temperature responses of −(0.48–0.50) K and precipitation responses of −(1.69–1.82) % using the prescribed MACv2-SP aerosol description.

Table 3Intermodel correlations of regional temperature response for the Samset et al. (2018) models and our models. The average correlation coefficient between the Samset et al. (2018) models is 0.79 with a standard deviation of 0.05; the average correlation coefficient between the models used in this study and the Samset et al. (2018) models is 0.76. The correlations are calculated for 50 years with and 50 years without anthropogenic aerosols. Correlation for our whole dataset (60+120 years) is shown in brackets. The range is the standard deviation between results obtained for two different CTRL runs.

Download Print Version | Download XLSX

Table 4Intermodel correlations of regional precipitation response for the Samset et al. (2018) models and our models. The average correlation coefficient between the models is 0.34 with a standard deviation of 0.10; the average correlation coefficient between the models used in this study and the Samset et al. (2018) models is 0.38. The correlations are calculated for 50 years with and 50 years without anthropogenic aerosols. The range of the correlation coefficient shows the standard deviations between results obtained for two different CTRL runs. The correlation for our whole dataset (60+120 years) is shown in brackets.

Download Print Version | Download XLSX

Tables 3 and 4 show the correlation coefficients for regional climate responses between all experiments in our datasets and the Samset et al. (2018) datasets. The correlations are calculated for equilibrium climate runs with equal time averaging over 50 years with and without anthropogenic aerosols both for our datasets and the Samset et al. (2018) datasets. Note that these coefficients do not depend on the magnitude of the average responses in the models but only on the relative regional distributions of the responses. Perhaps surprisingly, the average correlation coefficient for regional temperature response between interactive aerosol models (i.e., the Samset et al., 2018 models), 0.79, is almost identical to the correlation between our prescribed aerosol models (0.78). Also, the average correlation coefficient between experiments using interactive aerosols and a fully coupled ocean model (Samset et al., 2018) and experiments using prescribed aerosols and a slab ocean model (our models) is 0.76, nearly the same as for the fully coupled interactive aerosol models only. The similar regional correlation between different experiments is remarkable considering large differences in the aerosol descriptions between the different models. It appears that the differences in aerosol descriptions do not dominate the differences in regional temperature response. The average correlation coefficient for regional precipitation changes within Samset et al. (2018) models with intrinsic aerosol descriptions is 0.34, while it is 0.41 within our models with prescribed aerosols. The average correlation coefficient for regional precipitation changes between the Samset et al. (2018) models with fully coupled ocean and our models with a slab ocean is 0.39, which is similar to the mean correlation within the Samset et al. (2018) models.

The correlation coefficient between NorESM1 experiments using different aerosol descriptions and ocean models is now only 0.33∕0.38. Thus, differences in aerosol descriptions, ocean models and atmospheric responses all contribute to differences in regional precipitation responses. The correlation coefficients for precipitation responses are, however, more uncertain than those for temperature responses, due to a stronger impact of natural variability.

Even long equilibrium climate runs cannot fully eliminate the natural climate variability on a regional level. With our full dataset (60 years of MACSP runs +120 years of the CTRL run), we obtain a spatial correlation of 0.47 between NorESM1-MACSP and ECHAM6-MACSP precipitation responses, a slight improvement over the correlation coefficient of 0.41 (±0.02) for 50+50-year datasets. The spatial correlation for temperature improves from 0.78 (±0.02) to 0.81. The fully coupled ocean models in the Samset et al. (2018) dataset also feature long-term internal variability in the ocean states that adds to the level of natural variation with respect to our models with simpler slab ocean representations used in this paper. Therefore, we would expect the Samset et al. (2018) data to include more noise than our results with slab ocean configurations. Furthermore, it is important to note that differences in the ocean descriptions are known to have a large impact in the regional climate responses between different models (Deser et al., 2016; Kay et al., 2016). Overall, we would expect that due to these differences the climate signals obtained from fully coupled models would intrinsically correlate less well with each other than those from models with slab ocean configurations. Somewhat surprisingly, this turns out not to be the case.

The dependence of the calculation of time-averaged correlation coefficients on the simulation length for our data is shown in Fig. 5. There, the blue and red shaded regions represent the level of expected variation in the regional correlation coefficients between two climate models obtained from equilibrium model experiments with and without anthropogenic aerosols. We obtained a correlation coefficient of 0.78 with a standard deviation of ±0.02 for temperature response and 0.41 (±0.02) for precipitation after 50 years of simulation, these periods being representative for the Samset experiments but neglecting the impact of long-term ocean variations. The corresponding correlation coefficients for full model runs (60+120 years of simulation) are 0.47 for precipitation and 0.81 for temperature.

Figure 5Correlation coefficient of temperature (precipitation) response as a function of the number of averaged years. Blue (red) is the correlation between the temperature responses to MACv2-SP aerosols in the two models. The shaded area shows the variation between different control runs. The same number of years is used for the CTRL run and MACSP run.

Download