Table 4 shows the budgets for the different species in the model simulations. For each term in the table, results from both present day (PD) and preindustrial (PI) conditions are listed, together with the respective values found in K13. Unless otherwise stated, the discussed model values are from the NUDGE_PD simulation.

The result of the change in DMS emission parameterization described in Sect. 2.4 is an almost doubled DMS emission (34–35 Tg S yr-1) compared to the 18.1 Tg S yr-1 found in K13, accompanied by a similar increase in the SO2 source term from the oxidation of DMS. The main reason for the increase is that the DMS emissions in Dentener et al. (2006) (applied in K13) were based on the DMS climatology of Kettle and Andreae (2000), with generally lower DMS concentrations in seawater than in the updated version of Lana et al. (2011). An experiment with wind-driven DMS emissions in a research version of CAM4-Oslo using the same transfer function gave 22.0 Tg S yr-1 with the Kettle and Andreae (2000) data and 34.2 Tg S yr-1 with the Lana et al. (2011) data. The shorter lifetime of DMS (1.5 days) compared to K13 (2.4 days) is likely caused by the use of different oxidant fields. Liu et al. (2012) obtain a lifetime of 1.3 days using nearly the same chemical mechanism (see Sect. 2) and the same oxidant fields as in the present work, but with emissions from Dentener et al. (2006). An additional test simulation with CAM5.3-Oslo with the AMIP PD setup and 2∘ resolution shows that the effect of increased resolution (to 1∘) on DMS emissions and lifetime alone is only about 5 % and 0.2 %, respectively (not shown). Note also that the increase in column burden from CAM4-Oslo to CAM5.3-Oslo is much smaller than the increase in emissions (see Table 1), going from 0.12 to 0.14 Tg S. These both lie well within the range of estimates (0.015–0.17 Tg S) from other model studies reported by Liu et al. (2007); see their Table 1.

The chemical source for SO4 is divided into clear-air sources through the SO2+OH reaction and production in cloud water. The chemical sources of OM (via SOA) are mainly from monoterpene and isoprene. This gives a total of 78 Tg(OM) yr-1 of SOA produced from terpenes, which lies within the range of AeroCom models published by Tsigaridis et al. (2014). For comparison, the total amounts of BVOC emitted as isoprene and monoterpene are 438 and 119 Tg yr-1, respectively. There is also a source from the oxidation of DMS to MSA assumed to form organics (ca. 9 Tg(OM) yr-1), adding up to a total of 87 Tg(OM) yr-1. As mentioned in Sect. 2.2, for the MSA contribution to SOA, 20 % and 80 % of the MSA mass is added to the SOAGLV and SOAGSV tracers, respectively. The exact LV-to-SV ratio is unknown, but we find a quite low sensitivity of the anthropogenic change in cloud effective radiative forcing (i.e., the indirect effect, which is the most important in a climate change perspective) to the assumed apportionment of MSA: test simulations indicate that the total shortwave and longwave indirect effect only changes by about -0.03 W m-2 if all MSA goes into SOAGSV (no nucleation) and by 0.00 W m-2 if we instead feed all MSA into the SOAGLV tracer. The effect of neglecting the MSA contribution to SOA altogether is similarly estimated to give a -0.10 W m-2 change.

The zonal mean mass mixing ratios and their variation with height for SO2, BC, OM, sulfate, mineral dust, and sea salt (SS) are shown in Fig. 4, both for NUDGE_PD and AMIP_PD. The figure shows that some BC is transported to the stratosphere where the lifetime is longer. OM and sulfate also have this secondary maximum in the stratosphere, but the concentrations aloft are smaller in CAM5.3-Oslo than in CAM4-Oslo (not shown). Dust and sea salt do not exhibit the same clear secondary maxima in the stratosphere, since these particles are generally larger and more readily removed by sedimentation. The additional 2∘ test simulation with CAM5.3-Oslo reveals that the effect of increased resolution on vertical profiles is very small compared to the differences between the two model versions for all species (not shown).

For BC we can compare the model with profiles from the HIAPER (High-Performance Instrumented Airborne Platform for Environmental Research) Pole-to-Pole Observations (HIPPO) of carbon and greenhouse gases study over the (mainly remote) Pacific Ocean in January and November 2009, March–April 2010, and June–July and August–September 2011 (Wofsy et al., 2011; Schwarz et al., 2013); see Fig. 5. It is clear that the new model version does produce less excessive BC concentrations in the upper troposphere and in the stratosphere globally (Fig. 5a) and that it now compares better with the HIPPO observations in the Pacific (Fig. 5b), although the concentrations are still too high in the upper troposphere and lower stratosphere for this region, similar to the findings for CAM5.3-MAM4 in Liu et al. (2016). This is probably related to the way aerosols are transported and scavenged in deep convective clouds in the model (see, e.g., Kipling et al., 2013, 2016). There are currently ongoing tests with alternative treatments of convective transport and mixing (see Sect. 2.1.5 of K13 for a sensitivity test on this in CAM4-Oslo); these are improved treatments which will possibly be included in the upcoming CAM6-Oslo version for CMIP6. The additional 2∘ test simulation reveals that the effect of increased resolution on the lifetime of BC is only about 0.3 % (not shown). Note that NUDGE_PD and AMIP_PD yield almost identical results in the troposphere. This indicates that the nudging, as long as we are not nudging the temperature, only has modest effects on the convective transport and mixing of BC in the model (see also Fig. 4).

Some of the changes in aerosol concentration fields are connected to changes in cloud microphysics in the host model. Two major factors that affect both aerosols and aerosol precursors are the amount of liquid cloud water and the cloud fraction. Globally averaged, CAM5.3-Oslo has only about one-third as high cloud liquid water path (LWP) as CAM4-Oslo, while the precipitation rate is slightly (7 %) larger. Since the loss rate of aerosol activated to cloud droplets in the model is assumed to be proportional to the precipitation-to-LWP ratio, an increased scavenging efficiency and a subsequent reduction in aerosols away from source regions as a result of the reduced LWP is to be expected. A reduction in aerosol transport to remote regions is indeed found for all aerosol components and is particularly pronounced in the Arctic and Antarctic regions. At the same time, the total (low) cloud cover has increased from 53 % (34 %) in CAM4-Oslo to 66 % (43 %) in CAM5.3-Oslo, with the largest changes at high latitudes. This increase in cloud cover likely also gives an increase in the frequency of precipitation events, which tends to reduce aerosol lifetimes. The additional 2∘ test simulation (note that this by default setup has a slightly different cloud tuning) reveals that the effect of increased resolution on LWP and on total (low) cloud cover is small compared to the differences between the two model versions, only about 1 % and -1 % (-3 %), respectively (not shown).

Even sea salt burdens have been reduced away from the source regions, despite an almost 4 times increase in global lifetime, which is now 1.07 days. This is to a large degree due to the shift towards more long-lived (i.e., accumulation mode) particle sizes (compare Table 2 with Table 1 in K13). While the lifetime is now longer, the emissions have decreased even more so that the overall sea salt burdens are about 35 % smaller than in CAM4-Oslo. In Liu et al. (2012), the sea salt lifetime lies between CAM4-Oslo and CAM5.3-Oslo, but is quite dependent on the aerosol microphysics (0.76 days in MAM3 and 0.55 days in MAM7). The effect of increased resolution from 2 to 1∘ is found here to be 11 % for the emissions (due to stronger winds), 9 % for the burden, and only -2 % for the lifetime (not shown).

As for BC, the concentrations of OM, sulfate, and mineral dust have also dropped in the upper troposphere and lower stratosphere when going from CAM4-Oslo to CAM5.3-Oslo. This reduction is more substantial for carbonaceous aerosols than for the other species, however. In addition to the increased overall scavenging efficiency, BC and primary OM now experience a more rapid transition from external to internal mixtures; see Sect. 2.3. The lifetimes of BC and OM of approximately 5 days in CAM5.3-Oslo are now more comparable to the MAM7 values in Liu et al. (2012), which are 4.4, 4.9, and 4.1 days for BC, OM, and SOA, respectively. The additional 2∘ test simulation reveals that the effect of increased resolution on the OM lifetime is only about 1 % (not shown).

The situation for sulfur is more complex. While the scavenging efficiency of SO2 is increased on the one hand, as for aerosols, the lower liquid water content in CAM5.3-Oslo, on the other hand, acts to reduce the aqueous-phase reaction rates. The net effect of all changes is a ca. 20 % increase in lifetime. Furthermore, while SO2 (and thus the potential for the formation of sulfate) is now transported higher into the atmosphere, the increase in aerosol activation scavenging tends to counteract the effect of this enhanced transport. The combined effect of a longer lifetime for SO2 and increased aerosol loss rates in the lower troposphere is just a 3 % overall reduction in the atmospheric residence time of SO4. The estimate at 3.70 days is very close to the sulfate lifetimes in Liu et al. (2012): 3.72 days for MAM7 and 3.77 days for MAM3. The additional 2∘ test simulation reveals that the effect of increased resolution on the SO2 lifetime is only about -0.4 % (not shown).

As for carbonaceous aerosols, the lifetime of mineral dust is also reduced. The main reason for this reduction is most likely the general increase in activation scavenging. Below-cloud collection efficiencies are still as in Seland et al. (2008), so any changes in below-cloud scavenging are due to changes in precipitation and aerosol life cycling. The relative amount of dust emitted in the accumulation mode (DST_A2) in the new emission parameterization (13 %) is larger than for the prescribed emissions in CAM4-Oslo (11 %), which should rather contribute to a longer dust lifetime in CAM5.3-Oslo due to reduced gravitational settling. A test simulation performed with an earlier model version showed that a tuning of the relative amount of emissions taking place through the accumulation mode from 13 % to 20 % led to a 20 % increase in lifetime globally. The inherent assumption of OsloAero that there is a constant size background aerosol – the particles cannot shrink to smaller sizes than that of the background as the largest particles are deposited – may result in a shorter lifetime of the coarse mode compared to the modal aerosol schemes (MAM3 and MAM7) in Liu et al. (2012). Since Liu et al. (2012) calculate number and mass independently, the size of the coarse-mode particles may decrease with time, thereby increasing the lifetime of that mode. The estimated dust lifetime of 1.9 days is shorter than in both MAM3 (2.6 days) and MAM7 (3.1 days) in Liu et al. (2012). The additional 2∘ test simulation reveals that the effect of increased resolution on the mineral dust lifetime is only about 2 % (not shown).

According to Kok et al. (2017), mineral dust in global models is probably often too fine based on constrained atmospheric dust properties and abundance. AeroCom emission rates and loadings (Textor et al., 2006) are below the central estimates in Kok et al. (2017) of 1000–2700 Tg yr-1 and 13–29 Tg, respectively. We get a slightly higher global emission rate of 3100 Tg yr-1 in NUDGE_PD, but 2500 Tg yr-1 in the AMIP_PD simulation. The estimated global dust burden of 13 (NUDGE_PD) or 16 Tg (AMIP_PD) that follows, however, falls within the central estimates of Kok et al. (2017). The global emission rate may be adjusted by a tuning factor (a constant in the emission flux term) in CAM5.3, but in the present version we have retained the value used in the original CAM5.3 code.

Some of the aerosol burden changes from CAM4-Oslo to CAM5.3-Oslo are due to differences in meteorology. To roughly estimate the magnitude of such an effect, we compare the NUDGE_PD and AMIP_PD results in Table 4.

The globally averaged burdens of DMS and SO2 differ by less than 2 %, and SO4 and sea salt by less than 1 %, while BC and OM and mineral dust are ca. 5 % and 7 % lower in the free-running AMIP simulations, respectively. So for these species the differences between NUDGE and AMIP are quite small. We would probably have obtained even smaller changes if the model was self-nudged, i.e., being nudged to a meteorology produced by the model itself (e.g., the AMIP_PD simulation) instead of the ERA meteorology. In a similar comparison by Liu et al. (2016), they obtain as much as ca. 20 % lower BC and OM burdens with nudged (towards 1-year recurrent meteorology) vs. a free simulation (10 years). They partly attribute this to interannual variability, but mainly to (climatological) differences in the meteorological conditions between the free and nudged model simulations, which affect aerosol transport and cloud processing. Unlike the nudging procedure applied here, Liu et al. (2016) also nudged the model meteorology to reanalyzed temperatures (Tilmes et al., 2015), which may explain the larger difference in simulated aerosol burdens between their nudged and free AMIP simulations. A similar effect was found in an older version of CAM5.3-Oslo as we went from nudging temperatures, specific humidity, and U, V, PS, and some surface fields to only nudging U, V, and PS: the difference in globally averaged LWP between the nudged and free simulations was reduced by an order of magnitude. An important effect of nudging is that it constrains the model's natural variability (Kooperman et al., 2012), which is useful in calculations of the indirect effect of aerosols since it reduces the simulation length required to obtain sufficiently high signal-to-noise ratios. When nudging to an atmospheric circulation produced by the model itself (self-nudging) instead of using data from reanalysis (such as the ERA data), the circulation mean and variability characteristics are less affected, resulting in ERF estimates more consistent with the model's own innate behavior. However, since the circulation variability is not “synchronized” here with the observed variability of a specific time period, self-nudging does not facilitate a comparison of modeled aerosol properties with observations for that time period.

The 3 % increase in sea salt emissions in going from NUDGE_PD to AMIP_PD, which is consistent with larger simulated 10 m wind speeds in the extratropical storm track regions, is almost offset by a reduction in lifetime (more wet scavenging), giving only a 1 % net increase in column burden. There is one exception for which the difference between NUDGE_PD and AMIP_PD seems to be important, namely for mineral dust. This is most readily seen from the global dust emissions, varying with wind speed and soil humidity, which are 19 % lower in AMIP_PD than in NUDGE_PD, very close to the 18 % difference in atmospheric burden.

The contribution by interannual variations in the NUDGE_PD simulation to global aerosol or aerosol precursor burdens, given here as normalized standard deviations, is found to be about 3.6 % for DMS, 0.8 % for SO2, 1.2 % for SO4, only 0.1 % for BC, 1.0 % for OM, 2.6 % for sea salt, and 2.5 % for mineral dust. Hence, the above-estimated changes in burdens from NUDGE_PD to AMIP_PD are actually smaller than 1 standard deviation of the interannual variation (in NUDGE_PD) for DMS, SO4, and sea salt so that only SO2, BC, OM, and mineral dust can be said with some confidence to be different (smaller) in the AMIP_PD than the NUDGE_PD simulation.

Column burdens cannot be measured and observed surface concentrations are used here for validating the aerosol masses in the model. Figure 6 and Table 5 show surface mass concentrations of BC, SO2, OA (modeled OM vs. observed OC*1.4; see explanation in the figure caption and below), SS (sea salt), SO4 (sulfate), and DUST (mineral dust) in NUDGE_PD compared with various observations as available via the AeroCom intercomparison project (http://aerocom.met.no, last access: 24 September 2018). Note that the amount of data and spatiotemporal coverage available for the different parameters is inhomogeneous because of data network fluctuations and incomplete storage in the databases used (EBAS: Tørseth et al., 2012; see also http://ebas.nilu.no, last access: 25 September 2018; AEROCE: Arimoto et al., 1995; Huneeus et al., 2011). Tables 6–8 give an overview of statistical evaluation for the NUDGE_PD and AMIP_PD simulations as well as a range of AeroCom Phase II (AP2) and AeroCom Phase III (AP3) models. These are compared for different years, both for individual years (meteorology of 2006 for AP2 and 2010 for AP3) and our model climatology against a climatology from the observational data.

We find that the model mainly overestimates SO2 concentrations. One possible explanation for the large positive bias is the low vertical and horizontal resolution in the model. With such low resolution the model does not capture the dispersion of primary emissions of SO2 well from large point sources or shipping routes. A part of this bias probably comes from the fact that we are comparing concentrations at the midpoint of the lowermost model layer (∼50 m) with ground-based observations (see discussion in Simpson et al., 2012). For the climatologically averaged SO2 data, the Pearson correlation coefficient R (hereafter often just referred to as the correlation) is slightly better for the nudged than for the un-nudged AMIP simulation, in which instead the normalized mean bias (NMB; hereafter often just referred to as the bias) is slightly better. The bias and correlation for each of the continents are 216 % and 0.52 for Europe, 134 % and 0.94 for North America, and -53 % and -0.02 for Asia. None of the AP3 models have available SO2 statistics, while four of the five AP2 models that do exhibit higher biases than ours. The correlations are also lower than ours in all the AP2 models, while three of them have a higher percentage of monthly model values within a factor of 2 of the observed values (Fact2).

Sulfate is also somewhat overestimated, with a positive bias of 22 % and a correlation as high as 0.72 for the monthly climatological data, slightly above that of the free AMIP simulation. CAM4-Oslo exhibits a smaller, slightly negative bias, but is less correlated with the observations. The new model version still yields a lower Fact2 value, all in all performing slightly worse than the predecessor. Biases and correlations for each of the continents are 15 % and 0.54 for Europe, 38 % and 0.92 for North America, and -9 % and 0.59 for Asia. The bias for sulfate is better than in four of the eight AP3 models with available concentration data (for year 2010), while the correlation falls just below the AP3 range. Comparing against the 23 AP2 models (for year 2006), however, CAM5.3-Oslo has a lower bias than only 6 of the AP2 models, while outperforming or matching 14 models with respect to correlation.

We see that the model mainly underestimates BC, especially the highest concentrations. The bias is -28 % and the correlation 0.38, which is also higher here than for the AMIP simulation. CAM4-Oslo has an almost twice as large bias and a much lower correlation coefficient, so apparently there has been an improvement in modeled BC surface concentrations for the very limited number and geographical coverage of stations available (in Europe only). As much as 75 % of the model values lie within a factor of 2 of the observed values, compared to 68 % for CAM4-Oslo. The BC bias is also better than in six of the eight AP3 models. Although the correlations for BC are quite low for all the AP3 models, only one has a lower correlation than CAM5.3-Oslo. Similarly, comparing against the 23 AP2 models, CAM5.3-Oslo outperforms only 7 of the models bias-wise and 6 with respect to correlation.

For the calculation of mass concentrations of OM from OC the model does not distinguish between tracers from different source types, since they are lumped together for each of the background and size-modifying tracers. This has been done in order to limit the CPU requirements as much as possible, as the model (when fully coupled with the ocean and sea ice modules) is built for use in long climate simulations. We compare modeled OM with observed OC values that have been multiplied by 1.4 (defined as OA for the observations in Fig. 6, while OA simply means OM for the model values) to account for the conversion factor in going from fossil fuel OC to OM in the model (K13). For OM from biomass burning, defined as agricultural waste burning, grass fires, and forest fires in the model, the respective conversion factor is assumed to be 2.6 (K13; see also Formenti et al., 2003), i.e., 1.86 that of the fossil fuel emissions. If all OM originated from biomass burning, the bias would therefore be 19 % instead of 122 %. The latter value is simply based on the assumption of zero OC contribution from biomass burning. The truth concerning the validation probably lies somewhere between these two estimates, even though OM / OC ratios exceeding 2.6 might be more representative for some sources, such as MSA (see Sect. 4.2.1 in K13). For comparison, the respective bias values in CAM4-Oslo are 108 % and 12 %. The correlation coefficient for OM in CAM5.3-Oslo's NUDGE_PD is substantially lower than for both BC and SO4, but very close to that for OM in both AMIP_PD and CAM4-Oslo. Regional bias and correlation values are 143 % and 0.44 for North America, where most of the observation sites are located, and -26 % and 0.01 for Europe.

Assuming that OA is representative for the modeled OM, in North America the concentrations are most overestimated in the months JJA, while being underestimated in DJF. In Europe OM is overestimated only in JJA. This may indicate that OC is overestimated in summer or that sources with OM/OC ratios exceeding 1.4 dominate during summer, as should be expected since relative contributions to OM from SOA (e.g., Gelencsér et al., 2007) and forest fires are generally larger in this season. As discussed in K13, in addition to the various OM/OC ratios in the model, as in nature, a further complicating factor comes from the use of different standards and methods for measuring OC mass concentrations. While being integrated over all particle sizes in the model, the measured quantities may be based on PM2.5 or PM10 values in different observation networks, as is the case for North America (PM2.5 in IMPROVE) vs. Europe (PM10 in EMEP). This hampers reliable validation of OM in the model in its present form. Ideally the model should carry separate tracers for OC from SOA (preferably speciated), fossil fuel, and biomass burning sources and also have separate mass diagnostics for the different size intervals, which would better facilitate a more comprehensive evaluation of organic matter in the model.

Compared to the eight AP3 models, the bias (i.e., modeled OM–measured OA) is found to be smaller than in all but one model. The correlation, however, is just below the range for the AP3 models. It is slightly negative for the whole year of 2010 (Europe only), as for the months MAM that year, while being 0.15 or higher in the other seasons. Comparing against the 22 AP2 models (1 model is missing surface concentration data), CAM5.3-Oslo has a smaller bias than only 1 of the models, while it performs better than 6 models with respect to correlations for the year 2006. The Pearson correlation in our model varies between 0.16 and 0.25 for years 2004–2006, when both North American and European stations are included, while being closer to zero or negative in 2007–2010 based only on European station data, in which it varies between -0.12 and 0.12. It is surprising that there has been practically no change in correlation for the all-year climatology since K13 (CAM4-Oslo in Table 7), for which the SOA treatment was very simplistic. This should be investigated in future studies.

For the sea salt surface concentrations we obtain a bias of 22 % and a correlation of 0.54, and 31 % of the model values are within a factor of 2 of the observations. Compared to CAM4-Oslo this is much better bias-wise, but with nearly the same Fact2 value. The bias is also about half of that in the free-running AMIP_PD simulation. Regional biases and correlations are 59 % and 0.76 for Europe, 19 % and 0.72 for North America, and 31 % and -0.04 for Asia. A considerable number of the observation stations for sea salt are coastal and inland, however, and are perhaps therefore not very representative for sea salt aerosol as such in the model. CAM5.3-Oslo performs better than all the AP3 models bias-wise, and only one of the AP3 models has a higher Pearson correlation for sea salt. Our model is also less biased than 20 of the 23 AP2 models, and with higher correlation than 21 models.

For mineral dust we only have climatological observations to compare with. The bias for all stations and months is found to be -39 %, with a correlation of 0.52, which is slightly lower here than in the free AMIP_PD simulation. The observation stations for mineral dust surface concentrations are all quite distant from the largest dust source regions. Hence, the negative bias found in CAM5.3-Oslo may very well be a result of underestimated long-range transport rather than too-small emissions. This is corroborated by the fact that aerosol optical depths in the largest source regions (see Sect. 4.2) are biased high compared to the remotely retrieved values. Although the correlation coefficient is slightly better than in CAM4-Oslo, in which mineral dust emissions are simply prescribed, CAM5.3-Oslo is more biased and has a lower Fact2 value. We note, however, that even for the nudged simulation, the year-to-year variation for mineral dust is large enough to affect these validation results. Comparing monthly data from each individual model year with the observed climatological dust concentrations, the bias here varies between -46 % and -23 % and the correlation between 0.29 and 0.71. Part of this variability may be due to a varying number of stations for which there are enough data to be included in the multiyear climatology. Compared to the eight AP3 models, our model performs better than only three models bias-wise, but lies above the middle of the AP3 range with respect to correlations. It is also less biased than 14 of the 23 AP2 models and has a higher correlation than 7 of the models.

Table 9 gives the modeled mass extinction coefficients (MEC) for each of the aerosol components, calculated as the component's aerosol optical depth at 550 nm divided by its atmospheric burden. What determines MEC for a mono-disperse aerosol consisting of spherical (which we assume) and homogeneous particles is the particle size (divided by the radiative wavelength of interest), its mass density, and refractive index. For an internally mixed component of an aerosol size distribution, the size-integrated and atmospheric column-averaged MEC depends on a range of factors in the model. In addition to the refractive index of the components in a given mixture and the mixture's lognormal modal parameters (median radii and standard deviations) at the point of emission or nucleation, the growth by added process tracers and by hygroscopic swelling also play important roles. Aerosol lifetimes and aerosol life cycling in general, including transport and deposition, can further affect the results by shifting the “center of mass” of the aerosol components in question to areas and altitudes with different relative humidity, which consequently also affects the globally averaged MEC value.

Since neither the assumed mass density nor the initial lognormal modal parameters of the sulfate background modes in mixture nos. 1 and 5 have changed relative to the treatment in K13, i.e., in CAM4-Oslo, the ca 14 % reduction in MEC globally must be due to changes in growth, including the effects of life cycling on growth. As outlined in Sect. 2.3, the hygroscopicity of sulfuric acid has been reduced by about 17 % for relative humidities close to RHmax, while for ammonium sulfate there has been an equally large increase for these highest RH values but a larger reduction in large parts of the hysteresis domain (RH≈50–80 %), up to a 50 % reduction at RH≈80 %. The net effect of this when introduced into the model at the time (in an older model version), however, was small compared to the change from CAM4-Oslo to the present model version, which points to changes in meteorology and life cycling as the main cause. Although the atmospheric residence times and burdens of sulfate are quite similar globally (Table 4), in CAM5.3-Oslo they are both considerably smaller at middle to high latitudes and somewhat larger in the subtropics. At these low latitudes the relative humidity (and cloud cover) in the lower troposphere is also somewhat lower in CAM5.3-Oslo. Hence the sulfate “center of mass” is in effect shifted towards typically less humid regions, which is consistent with less hygroscopic growth and the smaller MEC values found in CAM5.3-Oslo. Some of the reduction may in addition be a result of having relatively larger amounts of (less hygroscopic) OM internally mixed with sulfate in the present model version, due to the co-nucleation of sulfate and SOA (mixture no. 1) and to the condensation of sulfuric acid and SOAGSV/SOAGLV onto larger particles (mixture nos. 1–10). The sulfate MEC estimates lie within the inter-model variability of the AeroCom Phase I models (Kinne et al., 2006) for both configurations of CAM5.3-Oslo, as for CAM4-Oslo.

MEC for OM aerosol has decreased by about 30 % compared to CAM4-Oslo, also still within the range of the AeroCom I models, but now closer to the AeroCom I median value. Looking back on results from earlier model versions of CAM5.3-Oslo, we find that the larger part of this change is most likely due to a shift in OM burdens to less humid areas (mainly at lower latitudes), just as for sulfate. An additional change that might be of importance is that SOA now comes as nucleation- or Aitken-mode particles (mixture no. 1) and is distributed onto larger particles by condensation instead of in the internally mixed primary OM/BC(a) mode (mixtures 4 and 14), which generally has a higher specific extinction. For instance, MEC is about 0.4 (0.6) m2 g-1 for mixture 1 if only consisting of nucleated OM at RH=0 % (80 %) compared to 3.0 (4.5) m2 g-1 for mixtures 4 and 14 when only consisting of OM (and condensed water).

Despite a shift in burdens towards lower latitudes also for BC, the mass specific extinction for BC (7.6 m2 g-1) has increased by about 17 % from CAM4-Oslo to CAM5.3-Oslo. This is also closer to the AeroCom I median value (8.9 m2 g-1). Regionally the increase is largest in areas downwind of relatively large sulfate and SOA or biomass burning sources in northern South America (where MEC is now at its largest at about 20 m2 g-1) and Indonesia (∼15 m2 g-1), as well as over and downwind of eastern North America to eastern Europe (∼10–15 m2 g-1). As mentioned, there is more growth by condensation in CAM5.3-Oslo since SOA is no longer treated as primary particles as in CAM4-Oslo. It is reasonable to assume that this extra aerosol growth may also be linked to the increase in MEC. Most importantly, however, the changes in BC emissions size, mass density, and refractive index (see Sect. 2.3) did change MEC for the pure and dry (RH=0 %) background particles of mixture nos. 2 (when containing only BC) and 12 from about 7.0 to 8.5 m2 g-1, i.e., a 20 % increase from the background tracer with the largest mass-wise contribution (90 %) to fossil fuel BC emissions. For mixture no. 0, the fractal fossil fuel BC particles, the net change in MEC from altered size, density, and refractive index is just a 0.3 % increase to 8.2 m2 g-1, due to compensating effects. The increase in MEC is also very small (∼3 %) for fresh BC particles from biomass burning in mixture nos. 4 and 14 if we assume that only BC is present in the OM/BC(a) mode.

MEC for mineral dust has increased by about 19 % globally and with a regional pattern quite similar to that of BC. Mass densities and particle sizes at the point of emission are the same here as in CAM4-Oslo for both tracers (DST_A2 and DST_A3). The effect of the change in refractive index (see Sect. 2.3) only yields a 0.4 % increase in MEC at 550 nm for pure dust in both mixture 6 and 7. Dust hygroscopicity has increased somewhat (see Sect. 2.3), which together with the extra growth potential from SOA is consistent with an increase in MEC. We note, however, that MEC is now higher even in the most arid source regions (e.g., Sahara) due to a slightly larger fraction of accumulation-mode (DST_A2) to total dust mass in the new emission parameterization (0.13) compared to CAM4-Oslo (0.11). With MEC=2.44 m2 g-1 and 0.335 m2 g-1 for DST_A1 and DST_A2 (assuming no growth), this shift towards smaller sizes alone (i.e., before further growth and deposition) can account for a 7 % larger MEC for dust in CAM5.3-Oslo. Mineral dust MEC is still within the range of the AeroCom I models, although it is now closer to the highest model estimates. Note that the (common) assumption that dust particles are spherical leads to a substantial underestimate in MEC for coarse particles, while the error is much smaller for particles with geometric diameters below about 0.6 µm (Kok et al., 2017). The bias towards smaller emission sizes, however (see the discussion above), should lead to an opposite-directed bias in MEC, since coarse mineral dust has much lower MEC than submicron dust (e.g., 86 % lower for DST_A3 than for DST_A2).

The hygroscopicity of sea salt has increased by about 4 % for high ambient relative humidities, now being smaller throughout much of the hysteresis domain compared to CAM4-Oslo (see Sect. 2.3). Together with changes in particle growth by the process tracers, such as by the condensation of SOA (missing in K13 and M14), this might explain some of the changes in sea salt MEC in moving to CAM5.3-Oslo. The main cause of the about 63 % increase, however, is the shift in particle effective radii towards sizes with higher specific extinction: globally averaged MEC for sea salt in CAM5.3-Oslo (5.04 m2 g-1) is just 1 % lower than in the CAM4-Oslo development version of Salter et al. (2015), which used the same model parameters for sea salt as in Table 2 while otherwise being the same as in K13.

Note that these MEC estimates are based on the common assumption that an internally mixed component's contribution to the total extinction increases linearly with its volume fraction, which in our model (in AeroTab) is allowed to vary with size. The same goes for the absorption or scattering when we focus on either of their contributions to the extinction separately. In this way nonabsorbing aerosols, such as sulfate and sea salt, contribute to the total aerosol absorption wherever internally mixed with absorptive aerosols, such as BC. Although the total extinction, scattering, and absorption is thus correctly found by summing up the contributions from each of the aerosol components, the method is expected to give biased extinction estimates, especially for the absorption part, compared to in situ measurements for each aerosol component separately (or for less aged and/or internally mixed particles close to the sources). Furthermore, the refractive indices of mixtures consisting of absorbing and nonabsorbing components are calculated by using the semi-empirical Maxwell–Garnett mixing rule, which gives less absorption (in better agreement with measurements) than the volume mixing rule for homogeneous mixtures (Chýlek et al., 1998), but more absorption than for purely external mixtures (Chýlek et al., 1998; see also Fig. 6 in Kirkevåg et al., 2005).

To obtain a first rough estimate of the magnitude of at least parts of the uncertainty in connection with the choice of methodology for calculating BC MEC and the corresponding mass specific absorption, MAC (defined as absorption aerosol optical depth (AOD) divided by aerosol burden), we have also calculated the corresponding coefficients for the anthropogenic part (i.e., using PD–PI AODs and burdens). This means a shift towards sizes and specific extinctions more representative of fossil fuel sources. The anthropogenic MEC is found to be about 8 % larger than for PD BC, 8.18 m2 g-1 for NUDGE_PD and 8.28 m2 g-1 for AMIP_PD, and 10 % higher (7.14 m2 g-1) than for PD BC in CAM4-Oslo. Similarly, the anthropogenic MAC value is as much as 30 % higher than for PD BC, 3.15 m2 g-1 for NUDGE_PD and 3.27 m2 g-1 for AMIP_PD, and 31 % higher for anthropogenic BC (3.15 m2 g-1) than for PD BC in CAM4-Oslo. We note that this is still low compared to measured values and the recommended range of 7.5±1.2 m2 g-1 for fresh, uncoated BC in Bond and Bergström (2006). According to that review paper, MAC can drop to about 5 m2 g-1 for collapsed BC aggregates, while coating by negligibly absorbing aerosol typically enhances MAC by 50 % (to ca. 11 m2 g-1).

One may also calculate alternative MAC values from the PD simulations by assuming that nonabsorptive or less absorptive components do not contribute to the light absorption of the mixture containing BC. First, leaving out only sulfate and sea salt and letting MAC = ABS(BC+SO4+SS)/BBC, we find that MAC=4.82 and 4.95 m2 g-1 in NUDGE_PD and AMIP_PD, respectively, compared to 5.07 m2 g-1 in CAM4-Oslo. MAC here exceeds 7 m2 g-1 over large areas (for all the above simulations) somewhat downstream of major BC emissions in North and South America and over several smaller areas in Southeast Asia. Similarly, assuming that mineral dust and OM do not contribute to the absorption either (as in Stjern et al., 2017), which is a much less realistic assumption in many regions, we obtain global MAC values of 23.2 and 21.4 m2 g-1 in NUDGE_PD and AMIP_PD, respectively, and 13.6 m2 g-1 in CAM4-Oslo. Assuming that the truth lies somewhere between the two last assumptions we could even obtain globally averaged BC MAC values within the recommended range of Bond and Bergström (2006). The problem with this line of reasoning is, of course, that BC is not the only absorbing aerosol component and that nonabsorptive components also add to, and even enhance (e.g., Chen et al., 2017), the total absorption for internal mixtures. Finally, although both mineral dust and OM individually have small MAC values, they have much larger atmospheric burdens than BC and thus also contribute considerably to the total absorption, even dominating regionally. In a test simulation with less absorptive mineral dust at most wavelengths – the imaginary refractive index at 550 nm is reduced from 0.0055 to 0.0024 – otherwise being identical to NUDGE_PD, the latter BC MAC estimate is reduced by 25 % globally, from 23.2 to 17.5 m2 g-1. Assuming linearity in MAC with respect to the imaginary part of the refractive index, MAC for BC partially internally mixed with nonabsorptive dust can be estimated from this at ca. 10.1 m2 g-1. The absorption by OM is still included in this estimate, however.

Figure 7 shows aerosol optical depth and absorptive optical depth at 550 nm as well as the Ångström parameter for wavelengths 440 to 870 nm in the NUDGE_PD simulation compared with AERONET (Holben et al., 1998). The results discussed and referred to below are shown in Fig. 7 and Tables 6–8.

We first look at modeled clear-sky aerosol optical depth at 550 nm (OD550CS). This is in the model calculated as the all-sky optical depth weighted (at each time step in the simulation) with the clear-sky fraction and with hygroscopic swelling calculated from average grid cell RH. This is the method adapted in K13, while a more common method for simulating the cloud-screened remote sensing assumes hygroscopic swelling based on the clear-sky RH fraction, but for all-sky conditions (no sampling or weighting). Due to the relatively large coverage (which we somewhat loosely call global here; see Fig. 7 and http://aerocom.met.no (last access: 25 September 2018) for the actual coverage) we find an apparently wide spread for modeled vs. observed (monthly) values, but with a relatively low NMB of -16 %, R=0.59, and Fact2=42 %. The all-sky values (OD550) look slightly better, with a positive NMB of 15 %, R=0.64, and Fact2=68 %. In comparison, CAM4-Oslo has a slightly stronger negative bias for OD550CS and a slightly smaller, but negative, bias for OD550.

Across the various available observation years 2004–2010, MNB for OD550CS varies between -27 % and -6 %. Regionally, OD550CS is most underestimated in East Asia (NMB=-59 %), followed by North America (-56 %), Europe (-38 %), South and Central America (-21 %), and India (-19 %). Europe is also defined here to include sites at the northern coast of Africa. For northern Africa, which again is defined to include sites at the Mediterranean coast in Europe, the bias is positive (12 %). The positive bias is even larger in Australia (71 %), where mineral dust is also estimated to dominate as the most optically thick aerosol. In spite of the apparent underestimation of near-surface dust mass concentrations discussed in Sect. 4.1.2, we may add here that the global all-sky optical depth contribution to mineral dust is biased by 65 % (not shown; see results at http://aerocom.met.no, last access: 25 September 2018), i.e., much more than the 15 % bias for total OD550. It is furthermore clear from Fig. 7 that OD550CS is underestimated at high latitudes. What is not known, however, is how much of this negative bias is caused by missing or inaccurate emissions (see, e.g., Stohl et al., 2013) and how much of it is a result of other systematic biases such as deficiencies in the modeling of transport, aerosol chemistry, microphysics, and subsequent scavenging or dry deposition.

Comparing OD550CS with the simulated aerosol optical depth from the same eight AP3 models as in Sect. 4.1.2, we find that four of the models have approximately the same (one model) or larger (in absolute value) NMB values than ours. The correlations are higher in six of the AP3 models, and five models also exhibit higher Fact2 values. Comparing with the 20 models with available data among the 23 AP2 models, we find that only 7 of these are less biased, but the correlations are smaller than ours in only 3 of the models. Although CAM5.3-Oslo performs well in terms of NMB, the spread is so large that the Fact2 value is lower than in all of these AP2 models.

Moving on to the clear-sky absorption aerosol optical depth at 550 nm (ABS550CS), we similarly find that NMB=-25 %, R=0.47, and Fact2=47 %. The all-sky values also look slightly better here, with NMB=-20 %, R=0.47, and Fact2=50 %. In comparison, CAM4-Oslo has a slightly stronger negative bias for both ABS550CS and ABS550.

Across the individual years 2004–2010, MNB for ABS550CS only varies very little between -41 % and -36 %. Regionally it is most underestimated in India (-52 %), followed by East Asia (-44 %), North America (-39 %), Europe (defined as above, -29 %), South and Central America (-20 %), and Australia (-13 %). For northern Africa (defined as above), the bias is slightly positive (4 %). Regionally the biases in ABS550CS and OD550CS thus mainly have the same sign, which is consistent with too-low or too-high (depending on the sign of the bias) modeled aerosol burdens. Some exceptions are found, however, such as for Australia as a whole and in (e.g.) some mineral-dust-dominated areas over and downwind of the Sahara Desert where the absorption optical depth is underestimated, while the optical depth (at 550 nm) is overestimated. This may indicate that the assumed imaginary part of the refractive index at 550 nm is too small or that the effective size of the mineral dust particles is underestimated, which has been identified as a problem in many AeroCom models (Kok et al., 2017).

A few AP2 models and one AP3 model also have absorption data available (ABS550). Comparing with the AP3 model (MNB=-56 %) CAM5.3-Oslo is less biased (-38 %). Comparing with the 16 models with available data among the 23 AP2 models, we find that only 5 of these are less biased than our model. The correlations are higher than ours in all these AP2 models, however, and nine models have larger Fact2 values.

Finally, we look at the statistics for the clear-sky Ångström parameter, defined here through the clear-sky aerosol optical depths at the wavelengths 440 and 870 nm in OD440CS and OD870CS, respectively: ANG4487CS=-ln(OD870CS/OD440CS)ln(870/440).

Globally, for ANG4487CS we obtain NMB=-17 %, R=0.75, and Fact2=83 %, a quite decent result indicating that the aerosol size for the clear-sky atmospheric column is fairly well modeled in terms of its relative abundance of large vs. small particles. The all-sky equivalent ANG4487, however, yields a much poorer match with AERONET, having NMB=-44 %, R=0.46, and Fact2=49 %. In comparison, CAM4-Oslo has a smaller negative bias for ANG4487 (no clear-sky value is available from that model version), indicating that the effective particle sizes are indeed smaller there than in CAM5.3-Oslo. For all-sky conditions the aerosol sizes are biased much more towards large particles (small ANG values), which is consistent with higher relative humidities and thus more extensive hygroscopic swelling.

Across the individual years 2004–2010, the bias varies as little as between -15 % and -16 %. Regionally, ANG4487CS is most underestimated in northern Africa (defined as above, i.e., extended to include sites along the European coast of the Mediterranean, -35 %), followed by Europe (defined as above, -32 %), Australia (-31 %), India (-20 %), East Asia (-10 %), and North America (-7 %). The pattern (see also Fig. 7) seems to point towards dust as a source of large negative biases, which is consistent with an excessive mineral dust contribution to the total aerosol (as also indicated by the regional OD550CS biases) or, alternatively, overestimated dust particles sizes (opposite of what we found as a potential cause of the positive bias in ABS550CS). For South and Central America NMB=15 %, i.e., an overestimate indicating slightly too-fine particles. This positive bias is smallest (8 %) for the SON months, i.e., late in the biomass burning season for the region, while it is largest (22 %) for DJF. Since the negative biases for OD550CS and ABS550CS here are smallest (-5 % and -6 %, respectively) in JJA and largest (-32 % and -35 %) in DJF, there is still a theoretical possibility that the biomass burning aerosol contribution is exaggerated. This could be the case if contributions from other sources are generally underestimated, e.g., due to missing emissions or exaggerated scavenging. Just based on these results, however, we cannot conclude whether this is the case or not nor whether the assumed OM / OC ratio of 2.6 for biomass burning aerosols is too high or not.

None of the AP3 models but 13 of the AP2 models also have ANG4487 information available at http://aerocom.met.no (last access: 25 September 2018). Comparing with the 13 AP2 models, we find that 5 have larger biases than in ours. The correlations are smaller in six models, and the Fact2 values are also smaller than ours in six models.

The particle sizes globally seem to be well represented. Based on modeled ANG4487CS, the consistent low bias in OD550CS and ABS550CS, and the assumption that the intrinsic optical properties and other factors that might affect the result are fairly well represented, the modeled aerosol column burdens may be underestimated. For the surface concentrations, only BC and mineral dust are underestimated compared to in situ observations, as discussed in Sect. 4.1.2. However, considering that the available in situ measurements are very sparsely distributed globally and that we know little about the model performance in terms of the vertical distribution of mass concentrations (except for BC), we cannot expect these very different measures of model performance to fully agree.

Further aerosol model validation is taking place through ongoing multi-model studies that include results from the present model version. These studies are the AeroCom Control EXPERIMENT 2016, the remote sensing evaluation for AeroCom Control 2016, the AeroCom in situ measurement comparison (for optical properties) (https://wiki.met.no/aerocom/phase3-experiments, last access: 25 September), and the BACCHUS CCN global model intercomparison exercise (http://lists.met.no/pipermail/aerocom-modeller/2017-January/000109.html, last access: 25 September 2018).