Concentrations and radiative forcing of anthropogenic aerosols from 1750-2014 1 simulated with the OsloCTM 3 and CEDS emission inventory 2 3

Abstract. We document the ability of the new generation Oslo chemistry-transport model, OsloCTM3, to accurately simulate present-day aerosol distributions. The model is then used with the new Community Emission Data System (CEDS) historical emission inventory to provide updated time series of anthropogenic aerosol concentrations and consequent direct radiative forcing (RFari) from 1750 to 2014. Overall, the OsloCTM3 performs well compared with measurements of surface concentrations and remotely sensed aerosol optical depth. Concentrations are underestimated in Asia, but the higher emissions in CEDS than previous inventories result in improvements compared to observations. The black carbon (BC) treatment in OsloCTM3 gives better agreement with observed vertical BC profiles relative to the predecessor OsloCTM2. However, Arctic wintertime BC concentrations remain underestimated, and a range of sensitivity tests indicate that better physical understanding of processes associated with atmospheric BC processing is required to simultaneously reproduce both the observed features. Uncertainties in model input data, resolution and scavenging affects the distribution of all aerosols species, especially at high latitudes and altitudes. However, we find no evidence of consistently better model performance across all observables and regions in the sensitivity tests than in the baseline configuration. Using CEDS, we estimate a total net RFari in 2014 relative to 1750 of −0.17 W m−2, significantly weaker than the IPCC AR5 2010–1750 estimate. Differences are attributable to several factors, including stronger absorption by organic aerosol, updated parameterization of BC absorption, and reduced sulfate cooling. The trend towards a weaker RFari over recent years is more pronounced than in the IPCC AR5, illustrating the importance of capturing recent regional emission changes.



138
The OsloCTM3 is a global 3-dimensional chemistry-transport model driven by 3-hourly 139 meteorological forecast data [Søvde et al., 2012]. The OsloCTM3 has evolved from its predecessor  methylbenzene and other aromatics). The gas/aerosol partitioning of semi-volatile inorganic 176 aerosols is treated with a thermodynamic model . The chemical equilibrium 177 between inorganic species (ammonium, sodium, sulfate, nitrate and chlorine) is simulated with the 178 Equilibrium Simplified Aerosol model (EQSAM) [Metzger et al., 2002a;Metzger et al., 2002b]. 179 The aerosols are assumed to be metastable, internally mixed and obey thermodynamic gas/aerosol 180 equilibrium. Nitrate and ammonium aerosols are represented by a fine mode, associated with sulfur, 181 and a coarse mode associated with sea salt, and it is assumed that sulfate and sea salt do not interact 182 through chemical equilibrium ]. The sulfur cycle chemistry scheme and 183 aqueous-phase oxidation is described by Berglen et al. [2004].

185
The sea salt module originally introduced by Grini et al. [2002] has been updated with a new     Here we use the CEDS version released in 2016 (hereafter CEDSv16). In May 2017, after 220 completion of our historical simulations, an updated version of the CEDS emission inventory was 221 released after users reported year-to-year inconsistencies in the country/sector level gridded data.

222
The emission totals were not affected, but there were occasional shifts in the distribution within 223 countries (http://www.globalchange.umd.edu/ceds/ceds-cmip6-data/). The potential implications 224 for our simulations are discussed below.   We calculate the radiative forcing of anthropogenic aerosols due to aerosol-radiation interactions    The global mean aerosol burdens in the baseline simulation are summarized in Table 3 (top row), 356 with spatial distribution shown in Fig. S2. Table S3 also shows the split of OA between secondary 357 and primary sources. Compared to results from the AeroCom III experiment, the OsloCTM3 358 sulfate burden of 5.4 mg m -2 estimated here is about 50% higher than the multi-model mean of 3.5 359 mg m -2 and 35% higher than OsloCTM2 [Bian et al., 2017]. The nitrate burden is nearly a factor 360 three higher than both the AeroCom multi-model mean and OsloCTM2 burden, and higher than 361 all nine models contributing in AeroCom III [Bian et al., 2017]. This is mainly due to a higher and nitrate (-23%) and the highest for sulfate (-52%). There are, however, notable differences in 380 model performance between data sets in different regions, as seen from Table S2. For all species, 381 the NMB and RMSE are highest for measurements in China. For instance, excluding the 382 CAWNET measurements, reduces the NMB for sulfate in Fig. 1 from -52% to -31% (not shown).

383
In contrast, the correlation is generally similar to, or higher than, other regions. In the case of BC which could partly explain the NMB of opposite sign in these two networks in Table S2. For BC, 390 we also include measurements from across India compiled by Kumar et al. [2015]. This is a region  is one of the regions affected by the emission distribution bug (Fig. S4a). We limit the analysis to 424 BC, using a model data from one year, but note that emissions of other species are also affected.

425
The comparison against measurements from the IMPROVE network ( Fig. S4b-d) shows an   Excluding the coarse mode, the fraction of total mass attributable to nitrate decreases from 43% to 446 28%, which is much closer to the observed 30% contribution. However, this affects the comparison 447 in Figure 1, resulting in a negative NMB of -34%, compared to -23% when including both coarse 448 and fine mode. This suggest that part, but not all, of the nitrate represented as a coarse mode in the observations. Modeled global, annual mean AOD and AAOD is 0.13 (Fig. 3a) and 0.0051 (Fig.   455 3b), respectively. The overall spatial pattern of modeled AOD agrees well with MODIS (Fig. 3c), with measured AOD from the AERONET network, with an overall correlation of 0.82 and RMSE 463 of 0.11, when using monthly mean data from 266 stations (Fig. 3d). Note that the modeled global 464 mean AOD is 0.13, but the model mean at the AERONET stations is 0.175 (Fig 3d) and has only 465 a NMB of -11.8%. Many of the AERONET stations tend not to be regional background sites, but  The best agreement is found for Europe and North America, with NMB of -0.4% and -13%,    in the ECLv5 and CMIP5 inventories give a global BC burden that is 9 and 22% lower, respectively.

507
For sulfate, ammonium and OA, we also find the largest burden changes in the LSIDEC case, 508 followed by SOLDEC. The change in the LSIDEC is particularly large for OA and is driven by 509 changes in SOA. For SOA, the changes are determined not only by modifying the scavenging, but 510 also by changes in POA concentrations, which gas-phase secondary organics can partition onto.

511
Increasing the horizontal resolution results in a slightly higher burden for all species, except sea 512 salt.

513
While sensitivity tests may give similar changes in the total global burdens, the spatial distribution 514 of changes can differ substantially. Figure 4 shows the ratio of AOD and total burden by species 515 and altitude in each sensitivity simulation to the baseline. As expected, varying the emission 516 inventories results in changes that are largely confined to the main source regions (Figs.4a,b).

517
Using the CMIP5 inventory results in considerably lower concentrations over Asia, the Middle 518 East and North Africa, reflecting the higher emissions in the more recent inventory. Over Europe 519 and most of North America there is an increase, particularly for sulfate, nitrate and ammonium. A 520 similar pattern is found when using ECLv5, but the differences are smaller. Reducing the large-

548
For BC, the sensitivity tests have little or no impact on correlation, but there is a markedly better 549 agreement in terms of standard deviation (i.e., model becomes closer to observations) for 550 CEDSv16/CMIP6 compared to RCP/CMIP5, reflecting the higher emissions in the former. Similar, 551 but smaller, effects are also found for the other species. The improvement from RCP/CMIP5 to 552 CEDSv16/CMIP6 is especially seen for measurements in Asia. A higher resolution is also found 553 to reduce the bias, in particular for BC. Figure 5b shows the comparison against AERONET AOD 554 in each sensitivity simulation. Again, there is a higher correlation and lower bias in the 1x1RES 555 run than in the baseline, while the opposite is found in the RCP/CMIP5 and ECLv5 cases. The 556 most pronounced changes results from using meteorological data from year 2000, in which case 557 the correlation is reduced from around 0.8 to 0.7.

558
For both observables, the difference in model performance between the baseline and scavenging 559 sensitivity tests is small. This may partly be an effect of the geographical coverage of stations; the 560 majority of measurements are from stations in more urban regions, whereas simulated burden 561 changes occur in remote regions, particularly at high latitudes and altitudes (Fig. 4). We therefore 562 also perform evaluations against AOD from regional sub-sets of AERONET stations. Ten of the 563 AERONET stations used in the present analysis are located north of 65°N (Fig. S1) with a considerably higher bias when scavenging parameters are modified (Fig. S7a). This is 567 particularly the case in the LSIDEC run, where concentrations of all species increase at high 568 latitudes compared to the baseline (Fig. 4). In contrast, the reduced concentrations in LSIINC,  (Fig. S7b), where the dust influence is strong. Changing the meteorological year and reducing 579 scavenging results in higher dust burdens (Table 3). Again, the agreement is better in the baseline 580 run than in these sensitivity runs. In particular, the METDATA run result in a higher bias and a 581 lower correlation, which is not surprising as dust production depends also on meteorological 582 conditions. The changes compared to the baseline CEDSv16/CMIP6 simulation cannot be entirely

587
[2016] found significant reductions in NMB of BC AAOD relative to AERONET when using a 588 high resolution (10 km) emission data and model output. In our analysis, moving from 2°x2° to 589 1°x1° horizontal resolution also results in a slightly higher correlation and reduced bias and errors 590 when compared to all AERONET stations (Fig. 5b). The impact is largest for AOD in China and 591 India, the NMB is reduced (from -34% and -24% (Fig. S6) to -20% and -10%, respectively).  With confidence in the model ability to reasonably represent current aerosol distributions 620 established, we next present an updated historical evolution of anthropogenic aerosols from pre-621 industrial to present-day, and the consequent direct radiative effect (RFari). Figure 7 shows the net 622 change in total aerosol load from 1750 to 2014. Full times series by species are given in Table S4.  The most notable difference compared to the AeroCom II results is seen for biomass aerosols. OsloCTM3fast is well below the AeroCom multi-model mean for nitrate. The OsloCTM2 was 638 found to be in the low range, but the multi-model was also influenced by some models giving high  12.5 m 2 g -1 at 550 nm. This is 26% higher than the 9.94 m 2 g -1 using the approach from Bond and

734
This reflects the higher emissions in the CEDSv16 emission inventory than in Lamarque et al. 735 [2010], as well as a higher MAC.

736
In general, we find lower surface sulfate concentrations in the model compared with measurements.

737
This could contribute to an underestimation of the sulfate RFari, which is weaker in the present 738 study than in IPCC AR5. We also note that the global mean sulfate burden is higher in the 739 OsloCTM3 than in most of the global models participating in the AeroCom III experiment (Sect.  For instance, an improvemet in the baseline compared to using the CMIP5 emission inventory was 756 seen for BC surface concentrations, in particular in Asia, while the NMB of AOD compared to 757 AERONET stations in the same region was reduced in the simulation with higher spatial resolution.

758
The importance of using the correct meteorological year was also seen. Such uncertainties will 759 translate to the RFari estimates, along with uncertainties in optical properties such as absorption    The total AOD from the OsloCTM3 is in good agreement with observations from the AERONET Asia. The corresponding NMB range from -23% for BC and nitrate to -46% and -52% for OC and 806 sulfate, respectively. The OsloCTM3 performs notably better than its predecessor OsloCTM2 in 807 terms of high-altitude BC distribution as compared with observed BC concentration profiles over 808 the Pacific Ocean from the HIPPO3 campaign. In constrast, the model continues to underestimate 809 observed surface levels of BC during winter and spring. Compared with other recent estimates of 810 aerosol burdens, the OsloCTM3 generally lies close to or above the mean of other global models.