Strong impacts on aerosol indirect effects from historical oxidant changes

Uncertainties in effective radiative forcings through aerosol–cloud interactions (ERFaci, also called aerosol indirect effects) contribute strongly to the uncertainty in the total preindustrial-to-present-day anthropogenic forcing. Some forcing estimates of the total aerosol indirect effect are so negative that they even offset the greenhouse gas forcing. This study highlights the role of oxidants in modeling of preindustrial-to-present-day aerosol indirect effects. We argue that the aerosol precursor gases should be exposed to oxidants of its era to get a more correct representation of secondary aerosol formation. Our model simulations show that the total aerosol indirect effect changes from −1.32 to −1.07 W m−2 when the precursor gases in the preindustrial simulation are exposed to preindustrial instead of presentday oxidants. This happens because of a brightening of the clouds in the preindustrial simulation, mainly due to large changes in the nitrate radical (NO3). The weaker oxidative power of the preindustrial atmosphere extends the lifetime of the precursor gases, enabling them to be transported higher up in the atmosphere and towards more remote areas where the susceptibility of the cloud albedo to aerosol changes is high. The oxidation changes also shift the importance of different chemical reactions and produce more condensate, thus increasing the size of the aerosols and making it easier for them to activate as cloud condensation nuclei.

. NO 3 experiences a very large relative change between PI and PD (up to more than 1000 % in the northern hemisphere), which is also seen in other model studies (Khan et al., 2015). CAM5.3-Oslo also applies a daily cycle to OH and HO 2 , which is not included in CAM5.3. The prescribed oxidants in Fig. 1 are only applied in the chemistry of the model, and not for radiation calculations. O 3 is also important for radiation calculations, but here another prescribed O 3 -field is applied, which is the same in both PI and PD.

Configurations
The model was configured with a horizontal resolution of 0.9 • (latitude) by 1.25 • (longitude) and 30 hybrid levels between the surface and ∼3 hPa. The simulations were carried out using nudged meteorology produced by the model itself to constrain 5 the natural variability (Kooperman et al., 2012). The horizontal wind components (U, V) were nudged with a relaxation time scale of six hours, while the temperature was allowed to run free, enabling impacts by aerosol perturbations, which could be important when calculating indirect effects (Zhang et al., 2014). Prescribed climatological SSTs and sea-ice extent from the mean of 1982-2001 were used in all simulations, as well as greenhouse gases and land use information from year 2000.  Table 2. 15 Each of them was restarted from an already spun up case with emissions and oxidants from its era. The last three years of the simulations were analyzed. The first set of simulations used CAM5.3-Oslo as described above, without any other modifications to the code. We name these simulations ORG, and the impact of historical oxidant changes on the PD-PI indirect effects in CAM5.3-Oslo are quantified by the difference we get (relative to the PD simulation PDAER_PDOXI_ORG) when switching between the two PI-simulations PIAER_PDOXI_ORG and PIAER_PIOXI_ORG.

Decomposing the oxidant change
To estimate the importance of the different changes in the individual oxidants between PI and PD, four additional simulations with PI-aerosols were carried out. In these simulations, the oxidant of interest was changed to PI-concentrations, while all other oxidants were kept at PD-levels. Acknowledging the complexity of oxidant chemistry, one can not expect that separate oxidant changes in separate simulations will add up to the same result as changing them all simultaneously. To explore the importance 25 of this non-linearity, another four additional simulations were performed, keeping all oxidants from PI except for the one of interest, which was set to PD-levels.
4 Results and discussion

Original setup
The top panels of Fig. 3 show the PD-PI indirect effect for shortwave radiation (a), longwave radiation (b), and total radiation (c) when using the standard setup with PD-oxidants in both simulations. The bottom panels of Fig. 3 show the impact of historical oxidant changes on the PD-PI indirect effect. Figure 3(d) shows that letting the precursor gases in the PI-simulation 5 be exposed to oxidants from its era, instead of oxidants from PD, makes the shortwave indirect effect 0.39 Wm −2 less negative (changing from -1.48 Wm −2 to -1.09 Wm −2 ). This implies that the clouds in the PI-simulation with PI-oxidants are cooling the climate more through SW-effects than the clouds in the PI-simulation with PD-oxidants, reducing the difference in shortwave cloud forcing between PI and PD. Figure 3(e) shows that the change in longwave indirect effect is -0.14 Wm −2 (from 0.16 Wm −2 to 0.02 Wm −2 ), meaning that the clouds in the PI-simulation with PI-oxidants are warming the climate more through 10 increased absorption of longwave radiation, reducing the difference in longwave cloud forcing between PI and PD. Figure 3(f) shows a total (shortwave + longwave) change in the indirect effects of +0.25 Wm −2 (changing from -1.32 Wm −2 to -1.07 Wm −2 ), meaning that the PI-clouds with PI-oxidants are cooling the climate more than the PI-clouds with PD-oxidants, thus making the indirect effect less negative. The largest changes in the shortwave indirect effect occur over ocean, especially over the North Pacific, off the west coast of America, in remote areas between 30 • S and 60 • S and over the Indian Ocean. The 15 changes in the longwave indirect effect mainly take place in the polar regions and over the Indian Ocean.
Different cloud-and aerosol changes can help explain the resulting change in the indirect effect. Some of these are presented in Fig. 4. In the global mean, switching to PI-oxidants in the PI-simulation results in (a) more numerous aerosol particles (+9.2 %), (b) more numerous cloud droplets (CDNC) (+3.7 %), (c) smaller cloud droplets (-1.5 %), (d) larger cloud fraction (+0.26 %), which is mainly caused by changes in the low cloud fraction, and (e) larger total gridbox averaged liquid water path (LWP) 20 (+1.7 %). The size of the cloud droplets in Fig. 4(c) is taken from the cloud top layer of the stratiform clouds.
The sign of the changes in the global mean cloud and radiative properties seen in Figs. 3 and 4 is as expected for an increase in the global mean aerosol number concentration. We will now further investigate why the oxidant changes enhance the aerosol number concentration. Figures 3 and 4 show that the distribution of the changes in aerosol number concentration does not always correspond directly to the distribution of the changes in the cloud and radiative properties. This indicates that it is not 25 only the change in aerosol number concentration that is important for the result, but also changes in the composition of the aerosols and in the atmospheric conditions where the aerosol changes take place.

The increase in aerosol number concentration
Since the formation of new aerosols depends on the availability of low volatile gases, and the PI-atmosphere consisted of relatively small amounts of oxidants to produce secondary gases with reduced volatility, one could expect a reduction in the 30 aerosol number concentration when switching from PD-to PI-oxidants. This is the opposite of what Fig. 4(a) shows. The increased lifetime of the precursor gases and the aerosols seen in Table 3 partly explain this. When the oxidizing power of the atmosphere is reduced, the precursor gases with high volatility are transported higher up in the atmosphere before they are oxidized. This is seen in Fig. 5, where the relative change in chemical loss of (a) DMS, (b) SO 2 , (c) isoprene, and (d) monoterpene through oxidation is negative close to the surface, but positive higher up in the atmosphere when switching from PD-to PI-oxidants in the PI-simulation. This results in the change in the vertical profile of the aerosol number concentration seen in Fig. 8(a), with lower values close to the ground, but larger values above ∼900 hPa. Aerosols formed from gases higher up in the atmosphere are not removed by dry-and wet-deposition as easily as aerosols formed closer to the ground, explaining 5 the longer aerosol lifetime seen in Table 3 (Jaenicke, 1980;Williams et al., 2002).
It is not only the vertical transport of the gases that changes. The reduced oxidation capacity also increases the horizontal transport of the primary precursors away from the source regions. This is seen in Fig. 6, focusing on DMS, the main precursor gas over ocean, where most of the aerosol-, cloud-and radiation changes occur. Figure 6(a) shows the distribution of DMSemissions, which is equal in all PI-simulations, while Fig. 6(b) shows the change in the net chemical loss of DMS through 10 oxidation when switching from PD-to PI-oxidants. Increased horizontal transport happens from areas with negative values to areas with positive values, since chemical loss through oxidation is the only way DMS can be lost in the model. The increase is especially pronounced in the North Pacific, with increased transport further south and towards the Arctic, but is also found in the southern oceans with increased transport from the large emission sources close to the coast and towards the remote ocean.
Figure 6(c) shows that this transport results in increased aerosol formation close to the ground in areas that receive more DMS 15 with PI-oxidants. Since the precursor gases are spread more in space with PI-oxidants, towards more remote areas where the background concentration of aerosols are low, the coagulation sink during the nucleation process is reduced, contributing to an increase in the formation rate. In CAM5.3-Oslo, "formation rate" describes the formation of aerosol particles of 12 nm, which is the size limit a particle must achieve to be accounted for in the aerosol number concentration (Figs. 4(a) and 6(e)). "Nucleation rate" describe the formation of aerosol particles of 2 nm. As for all other aerosols, the particles between 2 and 12 nm can 20 also be lost through coagulation with background aerosols. Figure 6(d) shows how the coagulation sink of these particles changes when switching from PD-to PI-oxidants in the PI-simulation. The reduction in the coagulation sink is especially large close to the strong DMS-emissions sources ( Fig. 6(d)). The areas over ocean with increased formation rate close to the ground corresponds well with the areas in Fig. 6(e) with increased aerosol number concentrations, indicating that the horizontal transport of DMS due to its longer lifetime in an atmosphere with PI-oxidants is important for the increase in aerosol number 25 concentration. Higher up in the atmosphere (above ∼850 hPa), the formation rate of aerosols also increase over the emission sources and at higher latitudes (not shown). The change in the total vertically integrated coagulation sink decreases by 17.7 % when switcing from PD-to PI-oxidants in the PI-simulation, favouring enhanced formation of new aerosols. Increased lifetime of the precursor gases also results in an increased deposition rate of SO 2 of 12.8 % (DMS, isoprene and monoterpene are only lost through atmospheric chemistry), favouring a decrease in the formation of new aerosols. As a result of these two competing 30 effects, the total vertically integrated formation of new aerosols increases by 5.4 %. which in turn allows for more DMS transport to and subsequently increased aerosol formation in remote regions like the South Pacific (SP) and the Arctic Ocean (AO), as defined in Fig. 7. The region named North Pacific (NP) in Fig. 7 experiences a local minimum in the change in the aerosol number concentration. Figure 6 shows that this is caused by less aerosol formation in this region. Nevertheless, NP also experiences a relatively large increase in CDNC. The vertical profiles in Fig. 8 show that the regions which receive more precursor gases with PI-oxidants (AO and SP) experience an increase in both aerosol number 5 concentration and CDNC for all altitudes, while the NP region experiences a decrease close to the ground, but an increase higher up. The latter can be explained by the vertical shift in the oxidation (Fig. 5). In NP, the height above which the change in CDNC is positive is located lower down in the atmosphere than the height at which the aerosol number concentration starts to increase ( Fig. 8(i) and 8(l)). This can be explained by the change in the size of the aerosols ( Fig. 8(j)), caused by the increased aerosol condensate relative to the aerosol number concentration ( Fig. 8(k)). The mean size of the aerosols is calculated as a 10 mean of the number mean radius of all mixtures in the model, weighted by the number of aerosols in each mixture. The relative amount of condensate increases in the global mean (Fig. 8(c)) and in the northern hemisphere ( Fig. 8(g) and 8(k)) because of the strong shift in the importance of the different oxidation reactions (Fig. 9). Both DMS, isoprene, monoterpene and SO 2 have the potential of being oxidized in three different ways. Figure 9 shows how many percent of the oxidant reactions of a specie happening through the different reactions. The largest change in the oxidant level when switching from PD-to PI-oxidants is 15 found for NO 3 in the northern hemisphere ( Fig. 1(c)). When switching to PI-oxidants, the relative fraction of DMS oxidized by NO 3 is reduced ( Fig. 9(a,c,d), red curves), while the the oxidation fraction involving the other oxidants become more important. For the dominant precursor gas over the remote oceans, DMS, this means that instead of mostly getting 1·SO 2 and no SOA out off a DMS-oxidation through the reaction (R4), the PI-atmosphere will to a larger extent produce 0.75·SO 2 and some SOA (R3). SO 2 nucleates easier than SOA, and 80 % of the SOA from (R3) comes as SOA SV , which is only allowed to 20 condense. The change in aerosol size in SP ( Fig. 8(n)) deviates from the other regions. This is due to the increase in OH in SP when switching to PI-oxidants (blue colors in Fig. 1(a)), giving rise to enhanced nucleation of small SO 4 -aerosols followed by an enhanced H 2 SO 4 -production through (R1). This also happens in AO, where the OH-level also is larger in PI, but here this effect is small relative to the effect of the increased SOA SV -production due to the large NO 3 -change in the northern hemisphere ( Fig. 1(c)).

The change in aerosol indirect effect
The SW radiative effect of a change in CDNC varies depending on where these changes take place. Twomey (1991) showed that dA/d(CDNC), where A is the cloud albedo, is largest in clean regions with low CDNC and where the cloud albedo is approximately 0.5. The SW radiative effect will also be larger in areas with low surface albedo, in areas close to the equator due to more incoming solar radiation, and in areas where the cloud fraction is high. The last two factors, in addition to the ocean being relatively more susceptible than others to cloud albedo changes caused by changes in CDNC. The cloud-weighted susceptibility function is normalized by its maximum value. Applying this function to three years of daily output from the PIAER_PDOXI_ORG-simulation in this study results in Fig. 10(a). Areas with high cloud-weighted susceptibility are found off the west coast of the continents and in the remote southern ocean storm tracks. The large increase in CDNC ( Fig. 4(b)) in the North and South Pacific regions efficiently increase the albedo of the clouds, thus resulting in the large change in the SW 5 indirect effect seen in Fig. 3(d). Due to less insolation in the Arctic, the cloud-weighted susceptibility in this region is low, resulting in a negligible effect on the SW indirect effect, even though this is the region that experiences the relatively largest increase in both CDNC ( Fig. 4(b)), cloud fraction ( Fig. 4(d)) and LWP (Fig. 4(e)) due to the oxidant changes. The LW indirect effect is not dependent on the incoming solar radiation, so the large changes in cloud properties seen in the Arctic affect the LW indirect effect. The thicker and longer-lived clouds in the simulation with PI-oxidants act to reduce the difference in LW heat-10 ing between the PD-and PI-simulations (Fig. 3(e)). Figure 10(b) shows the vertical profile of the global mean cloud-weighted susceptibility. It shows that the decrease in CDNC close to the ground (Fig. 8(d)) does not affect the cloud albedo as much as the increase in CDNC between 900 and 800 hPa.

Decomposing the oxidant change
To get a better understanding of the results in the original experiment, results from the sensitivity tests where only one oxidant 15 at a time was changed are analyzed. Figure 11 shows differences in the global mean shortwave and longwave indirect effect between the setups with modified PI-simulations (PIOXI, PIOH, PIO3, PINO3 and PIHO2) and the original setup with only PD-oxidants in both simulations. Figure 12 shows the same for the horizontal distribution. Changing only NO 3 (PINO3) gives almost the same result as changing all of the oxidants (PIOXI), indicating that the historical change in NO 3 is the most important oxidant change for indirect effect calculations. This corresponds well with Fig. 1 showing that NO 3 is the oxidant that 20 has experienced the largest relative change since PI, and Fig. 9 showing that the importance of the oxidation reactions involving NO 3 drop the most when switching from PD-to PI-oxidants in the PI-simulation. The negative pattern over land in the tropics in PINO3 that is missing in PIOXI (Fig. 12) Table 5.

NOSOALVDMS and NOSOALVBVOC
When moving from a high NO 3 -regime (PD-oxidants) to a low NO 3 -regime (PI-oxidants), the oxidation reactions giving SOA LV as a product ((R3) and (R6)) become more important. This is seen from the large change in the global mean column burden of SOA LV (+49.6 %). Since SOA LV can take part in nucleation and give rise to the increased aerosol concentration seen in Fig. 4(a), the additional SOA LV that is produced when using PI-oxidants may explain the change in the indirect effects seen 15 in Fig. 3. When replacing all of the originally produced SOA LV from the DMS-oxidation in (R3) with SOA SV , the change in the total aerosol indirect effects is almost the same as for the original setup (∆AIE tot : +0.25 Wm −2 ), and the geographical pattern looks largely the same (not shown here). This also holds when doing the same for the oxidation of monoterpene (R6) (∆AIE tot : +0.26 Wm −2 ). The pattern of the resulting AIE from the oxidant changes in the NOSOALVBVOC-simulations looks almost the same as for the original simulations, except over the Amazon where the signal from the O 3 -changes explained in the last 20 section is gone. This does not change the global mean AIE by more than 0.01 Wm −2 , however. These sensitivity tests indicate that even though the global mean burden of SOA LV changes a lot when using PI-oxidants, this plays a minor role for the change in the indirect effects seen in Fig. 3.

NOSOA
The increased production of total SOA(g) (SOA SV and SOA LV ) when switching from PD to PI-oxidants has potential to cause 25 changes in the indirect effects even though the nucleation effect is negligible. All SOA(g) can condense onto already nucleated aerosols and make it easier for them to grow to the critical size for cloud droplet activation, except for cases where the reduction in hygroscopicity is more important than the increase in size. The impact of the hygroscopicity changes due to the changes in the oxidant levels has been tested and found to be negligible (not shown here). The change in total global mean column burden of SOA(g) due to changes in the oxidant levels with the original setup was +40.7 %. To find out if this increase is causing the 30 change in the indirect effects seen in Fig. 3, the model was run with the NOSOA-setup described in Table 5. This resulted in a change in the total aerosol indirect effects (∆AIE tot ) of +0.14 Wm −2 , deviating by more than 0.10 Wm −2 from the original setup. Removing products from the reaction makes the atmosphere cleaner, thus creating a different regime for both aerosol growth through reduced competition for condensable gases and for activation of aerosols through reduced competition for water vapor. This means that one cannot conclude that 0.11 Wm −2 of the 0.25 Wm −2 is caused by an increase in condensable SOA(g) when switching from PD-to PI-oxidants, but this sensitivity test indicates that it may have contributed to the overall result seen in Fig. 3.

NACTOFF
This test is performed in order to see how important the change in the droplet activation on the smallest aerosols are. When modifying the oxidant level, the smallest aerosols are affected by the change in formation rate, while all aerosols are affected by the change in condensation. The results from this test give an indication of how important the changes associated with the smallest aerosols are. When not allowing the smallest aerosols in mixture number 1 (corresponding to the nucleation 10 mode in modal aerosol schemes) to activate, the change in the total aerosol indirect effects found when switching from PDog PI-oxidants in the PI-simulation is small (∆AIE tot : -0.03 Wm −2 ). This confirms that it is the difference in the number concentration of the smallest SO 4 -and SOA-aerosols that gives the large difference in the indirect effect seen in Fig. 3.

Summary and conclusions
Here we have used the global atmospheric model CAM5.3-Oslo to study the effect of historical oxidant changes on the PD-PI 15 aerosol indirect effect. The precursor gases in the PI-simulation were exposed to PI-oxidants instead of PD-oxidants. Our main findings are: -The total aerosol indirect effect is reduced from -1.32 Wm −2 to -1.07 Wm −2 , mainly due to a cloud brightening in the modified PI-simulation.
-NO 3 is the oxidant that contributes the most to the changes.

20
-When the precursor gases are exposed to an atmosphere with relatively lower oxidative power (PI-oxidants vs. PDoxidants), their lifetime increases and they are transported higher up in the atmosphere and horizontally towards more remote areas before they are oxidized and can contribute to new aerosol formation.
-A large portion of the new aerosol formation and the increase in aerosol number concentration occurs where the cloudweighted susceptibility is high, giving a large impact on the radiative effects.

25
-The change from PD-to PI-oxidants in the PI-simulation yields a shift in the chemical reactions towards increased production of condensate relative to the amount of gases that can nucleate, which increases the size of the aerosols, making it easier for them to activate.