Impacts of New Particle Formation on Short-term Meteorology and Air Quality as Determined by the NPF-explicit WRF-Chem in the Midwestern United States

New particle formation (NPF) from nucleation and subsequent nuclei growth, which is frequently observed in the troposphere, is critical to aerosol-cloud interactions yet difficult to simulate. In this work, regional simulations with the fully coupled NPF-explicit WRF-Chem model link NPF to cloud properties and to changes in both meteorology and air quality in the U.S. Midwest during summer 2008. Simulations that include NPF have higher concentrations of condensation nuclei, as anticipated from the particle production associated with nucleation, leading to enhanced concentrations of cloud condensation nuclei (CCN) at high supersaturations. However, the online-coupled model develops a number of unexpected features that can be traced to a feedback loop involving aqueous (in-cloud) oxidation of sulfur combined with boundary layer NPF. Simulations with NPF (relative to simulations without) exhibit reduced PM2.5 sulfate mass, cloud dimming (reductions in the cloud frequency, CCN concentration at a low supersaturation, cloud optical depth, and cloud droplet number concentration), and enhanced surface-reaching shortwave radiation. This effect of NPF on the PM2.5 mass is mostly absent for other constituents of PM2.5. The implications of this feedback loop, which is not considered in most climate and air quality modeling, are discussed.

Model and emissions configurations are listed in Table 1 Table 2 lists the primary numerical experiments performed in this study. In case N11A1, 154 both PBL and FT nucleation are included, and aqueous chemistry is considered as well, and 155 N11A1 is the base simulation. In case N01A1, only FT nucleation is included, while in the 156 N00A1 case, both PBL and FT nucleation are turned off. For these three cases, the simulations 8 Accepted for publication in AAQR (Sept 2018) 8 are run continuously for the entire modeling period without reinitialization. N11A0 and N01A0 158 cases are one-week simulations (excluding 3-day spin-up) evaluating sensitivity to aqueous 159 chemistry from 20 June to 30 June. Sensitivity tests are performed in two additional scenarios to 160 explore the impact of model resolution and maximum supersaturation (Smax). Initial and 161 boundary conditions are the same for all simulations, and nudging is not used.  (Table S1). Soundings of temperature (T) and relative humidity (RH) at Davenport, Iowa  Table S1). AERONET AOD from Bondville, IL,   Table 3. Variables listed in Table 3 are also evaluated for N01A1 and N00A1. 175 Results for those cases, and statistical metric definitions, can be found in supplemental 176 information. Fractional bias (FB), fractional error (FE), mean bias (MB), mean error (ME), root 177 mean square error (RMSE) and correlation coefficient (R) are reported, with percentage bias and 178 error relative to measurement means. The performance statistics from different cases for both 179 meteorology and species are generally similar. Noticeable sensitivity occurs for SO4 2and SO2. 9 Accepted for publication in AAQR (Sept 2018) For SO4 2-, N11A1 shows the best performance for all statistical metrics; for SO2, the 181 performance of N11A1 is either better or between that of N01A1 and N00A1.  As the purpose of the study is sensitivity to NPF, bias and error in simulating observed 195 concentrations of air pollutants are most relevant to their influence on the sensitivity results.
196 Table 3 shows that the monthly average SO4 2concentration is overpredicted by 0.44 µg/m 3 197 (20.1%), while the black carbon (BC) concentration is underpredicted by 0.07 µg/m 3 (25.9%). A 198 possible source of positive bias of SO4 2could be SO2 overestimation. As shown in Table 3, SO2   199 is predicted with a positive bias of 0.92 ppb (29.8%). Underestimation of BC may be the result of 200 errors in anthropogenic emissions, the absence of wildfire emissions, and biases in the 201 meteorological fields (for example, the overprediction of temperature indicates PBL height might 202 be over-predicted as well). PM2.5 is underpredicted, which is expected, as this simulation did not Accepted for publication in AAQR (Sept 2018) include SOA formation except in sensitivity tests. Performance for PM2.5, SO4 2-, and BC exceed 204 the 75 th percentile for r 2 , and are consistent with the middle 50% on bias and error metrics 205 compared to prior U.S. photochemical grid modeling studies compiled by Simon et al. (2012).

206
Using FB and FE results criteria from Morris et al (2005), all chemical species fall into the 207 "good" or "average" classifications. Comparing to more stringent and recent performance goals 208 (Emery et al., 2017), black carbon meets suggested criteria for normalized mean bias (NMB), 209 normalized mean error (NME), and r. PM2.5 meets two of three criteria (NMB exceeds 30%).

210
SO4 2meets the correlation criteria (it exceeds r of 0.7) while slightly exceeding the NMB and 211 NME recommended criteria.  examined the global variation in particle size distribution using 12 global aerosol microphysics 230 models. They found that the best estimate of the annual mean particle concentrations (for CN3, 231 CN10 or CN14) were within a factor 2 of the observations at all 13 sites, including Bondville IL 232 (Midwestern U.S.).

233
The N01A1 scenario predicted a lower mean CN10-63 (4735/cm 3 ) that was closer to the   Table 4 lists the monthly mean CN and CCN concentrations from different simulations.

249
As expected, simulations that include nucleation in the FT (N01A1) or the FT and PBL (N11A1) 250 have higher number concentrations of CN20 relative to N00A1 (CNx is used throughout for the 251 number concentration larger than x nm). CN50 and CN100 behave similarly. FT nucleation has a 252 strong impact on surface CN concentrations, consistent with other studies (Merikanto et al., 253 2009;Spracklen et al., 2010); however, study of the FT nucleation impacts on boundary layer air 254 quality, meteorology, and cloud properties is beyond the scope of the current work.

255
Results for CCN at high supersaturations are similar to those for CN. At 0.5% and 1% 256 supersaturation, higher particle number concentrations in simulations with nucleation lead to 257 higher CCN concentrations (Table 4)  simulation. LWP and COD are also high at the same location (Fig. S5). Noticeable CDNC reductions, resulting from turning PBL nucleation on, can be seen in these high CDNC regions.

271
Changes in LWP and COD are similar in direction to the CDNC changes. PBL nucleation also 272 leads to lower cloud occurrence frequency (see Section 3.2.5). LWP and COD changes linked to 273 CDNC changes have been found in previous studies as well, but in those instances, CDNC 274 changes were not linked to NPF (Yang et al., 2011;Makar et al., 2015).   Midwest is shown in Figure 2(a). Midwest mean PM2.5 concentrations from the N11A1 and 289 N01A1 simulations are 6.04 and 6.56 µg/m 3 , respectively. By including boundary layer 290 nucleation, PM2.5 is reduced by 8.0% on average. This decrease in PM2.5 is due to reduction in 291 secondary inorganic species, especially SO4 2-. Compared to the N01A1 simulation, Midwest surface SO4 2in the N11A1 simulation decreases by 0.37 µg/m 3 on average, accounting for 293 71.2% of the PM2.5 reduction. Concentrations of other species are much less sensitive to NPF.

294
Ground level CO and O3 also decrease slightly in most areas in the N11A1 simulation, likely due 295 to increases in the average PBL height in the N11A1 simulation, as shown in Figure 4(d).

296
Changes in SO2 concentrations are significant in magnitude (up to 20%) but spatially 297 inhomogeneous. We are aware of no other studies reporting a reduction of PM2.5 sulfate due to 298 the inclusion of NPF in a model, and we discuss the causes and implications in section 3.2.5.

299
The spatial pattern to the PM2.5 decrease (Fig. 2a) is similar to that of the decrease in 300 CCN concentration at low supersaturations, as is expected since larger particles that can activate 301 at low supersaturations are a substantial portion of the mass distribution (Fig. 8).

302
In contrast to surface PM2.5, surface SO4 2-, and CCN concentrations at low 303 supersaturations, AOD is left mainly unchanged from the inclusion of NPF in the simulation 304 ( Fig. 3c), and the changes are spatially similar to changes in sulfate column loading .

305
Surface sulfate changes are slightly more pronounced than column sulfate changes because of 306 changes in the sulfate vertical profile and higher PBL height in simulations with NPF. The AOD 307 increase in the western part of the Midwest (Fig. 3i) is likely due to aerosol water.

316
Averaging across the entire domain, and considering grid cells with clouds in both N11A1 and 317 N01A1, turning NPF on increases CN25 by 497 cm -3 (9.4% increase relative to N01A1 for these 318 grid cells), but decreases cloud droplet number concentration (CDNC) from 66.2 cm -3 (average 319 for N01A1 in these paired grid cells) to 58.4 cm -3 , a 11.8% decrease in CDNC. Since the 320 Twomey effect establishes that the sensitivities of CDNC to increases in particle number are

326
The magnitude of changes reaches up to +66.4 W/m 2 for monthly averaged shortwave 327 radiation. T2 is influenced by a variety of factors, such as shortwave radiation, soil moisture and 328 soil temperature. In the southwest of the area analyzed, T2 decreases by more than 1 K, and Q2 329 increases by more than 1.6 g/kg, consistent with precipitation enhancement (Fig. S7). PBL 330 heights increases by up to 90.5 m as a result of surface warming in most areas of the Midwest.

331
Reduced PBL heights are simulated in locations where the N11A1 simulation has lower 332 shortwave radiation and lower surface temperature relative to N01A1. The PM2.5 reduction simulated with PBL nucleation on (mainly due to PM2.5 sulfate 335 reduction) is potentially important for a number of applications, and the modeling system was exploited to establish its cause. In general, a sulfate mass concentration decrease must be from 337 either weakening of a SO4 2source, or strengthening of a sink. SO4 2sources in WRF-Chem are 338 primary emissions, and gas phase and in-cloud oxidation of SO2. SO4 2is removed from the 339 model atmosphere by wet and dry deposition. A number of model-model comparisons and 340 sensitivity runs were conducted to investigate how SO4 2sources and sinks were changing. Key 341 results from these tests are discussed below; a full list of tests and associated inferences is also 342 reported (Table S4). Dry deposition is ruled out as a main cause of the modeled SO4 2decrease by comparison 358 of ∆SO4 2at the surface to ∆SO4 2at model layer 7 (~850 m above ground). Since dry deposition 359 acts at the surface, its effect would be most pronounced at the lowest model layer. However, 360 ∆SO4 2- (Fig. 6a) is similar at both elevations.

361
To elucidate the relative impacts of gas phase and aqueous chemistry in the weakening of 362 sulfur oxidation, model runs were performed with aqueous chemistry turned off, but with PBL 363 nucleation active (N11A0) or inactive (N01A0). The results of the sensitivity tests (for Midwest 364 averaged surface PM2.5 and SO4 2-) are shown in Table 6. These additional runs both show a 365 decrease in SO4 2from NPF, so the direction of the NPF effect is the same as in the longer base 366 and N01A1 runs. But the magnitude of the effect is much larger when both NPF and aqueous 367 chemistry are active. Specifically, a 1.1% sulfate decrease occurs without aqueous chemistry, 368 compared to a 13.0% sulfate decrease with aqueous chemistry. We therefore conclude that NPF 369 and aqueous chemistry act together as the primary cause of the nucleation effect on PM2.5 mass.  Table 6 discussed above. Without aqueous chemistry, the influence of NPF on SO4 2-372 concentration is small (Fig. 7b) and centered in the eastern portion of the Midwest and eastern 373 US. In other words, the combined effects of aqueous chemistry and NPF are needed to explain 374 the bulk of the reduced SO4 2concentrations and reduced SO2→ SO4 2conversion. 375 We also attribute a portion of the SO4 2reduction to decrease in the gas phase OH radical.

376
The OH reduction and its potential causes are explored in supplemental material (Fig. S8). In 377 summary, biogenic emissions (~10% increase due to higher temperatures and SW radiation) are enhanced and water vapor decreases in the NPF on simulations. Increased VOCs shortens the Accepted for publication in AAQR (Sept 2018) 18 OH lifetime (Archer-Nicholls et al., 2014;Yahya et al., 2014). Dilution of SO4 2through PBL 380 height increase was investigated and found to be a minor contribution, based on the insensitivity 381 of primary aerosols to PBL nucleation. With PBL nucleation on, cloud water in the lower troposphere (0-2 km) decreases by 385 over 30% in most parts of the region with high SO4 2- (Fig. S9), reducing the extent of aqueous 386 sulfate production. Cloud water content increases in some areas, but these are generally in 387 locations with low SO2 concentrations. In the Midwest, the largest decreases in absolute and 388 percentage SO4 2concentration are simulated in the most polluted region as shown in the red 389 square in Figure 7(a). In the following discussion of the influence of cloud chemistry, we limit 390 our analysis to this area.

391
The decrease in time-averaged mean cloud liquid water is from less frequent cloud 392 occurrence, and from decrease in LWC of simulated clouds. For frequency calculation, we use 393 hourly outputs for layer 1 -layer 10. Cloudy grid cells decreased from 3.08% (N01A1) to 2.44% 394 (N11A1) with PBL nucleation enabled. 395 Figure 8 shows the influence of NPF on the mean aerosol number distribution of the 396 boxed area (Fig. 8a), and mean aerosol volume distribution (Fig. 8b-d). Panels a-c are from 397 simulations without aqueous chemistry (N11A0 and N01A0) in order to isolate the influence of 398 NPF from the additional effect of aqueous chemistry. The increase in particle number as sizes 399 below 20 nm is clearly visible (Fig. 8a), as is the increase in particle volume in the same size  With aqueous chemistry active (Fig. 8d, ratios with the A1 suffix), the decrease in 408 particle volume at large sizes is much more pronounced than the cases without aqueous 409 chemistry, and shifts to larger sizes. respectively. The cause of the changes in Smax has not been quantified in this work; however, we 420 hypothesize that the decrease of Smax contributes to the feedback between NPF and reduction in 421 cloud water.
Further showing the importance of Smax is a sensitivity test with repetition of runs N11A1 423 and N01A1 with Smax fixed at 1%. In the two sensitivity tests (ten days including spin up) Smax 424 was coerced to 1.0% whenever a cloud is formed in a grid cell. This allowed more particles to be 425 activated with nucleation, and eliminated the large region of SO4 2decrease with NPF on (Fig. 426 S10), replacing it with a mix of small SO4 2increases and decreases.  In addition to liquid-water content, the conversion rate of SO2 to SO4 2depends on cloud 440 pH and oxidant concentrations (primarily dissolved H2O2 and O3). The SO2-H2O2 reaction is 441 roughly pH independent while SO2-O3 reaction rates decreases under acidic conditions. Under 442 both N11A1 and N01A1, over 96% of clouds had pH < 5, consistent with H2O2 oxidation of SO2 443 (Hung and Hoffmann, 2015). Figure 6(b) shows the gas phase H2O2 at a height of 850 m above ground. The N11A1 case has higher levels of H2O2 in most regions of the Midwest, suggesting 445 cloud water is a more important factor than H2O2 availability in decreasing SO4 2-.        Figure S1, ± represent the spatial standard 843 deviation of the monthly mean 844 Table 5 Midwest average precipitation for N11A1 and N01A1 runs 845 Table 6 Midwest mean surface PM2.5 and SO4 2in the sensitivity test period