Dominant Role of Arctic Dust With High Ice Nucleating Ability in the Arctic Lower Troposphere

Recent observations show that dust emitted within the Arctic (Arctic dust) has a remarkably high ice nucleating ability, especially between −20°C and −5°C, but its impacts on the number concentrations of ice nucleating particles (INPs) and radiative balance in the Arctic are not well understood. Here we incorporate an observation‐based ice‐nucleation parameterization indicating the high ice nucleating ability of Arctic dust into a global aerosol‐climate model. A simulation using this parameterization better reproduces INP observations in the Arctic and estimates >100 times higher dust INP number concentrations with ∼100% contribution from Arctic dust in the Arctic lower troposphere (>60°N and >700 hPa) during summer and fall (June–November) than a simulation applying a standard ice‐nucleation parameterization suitable for desert dust to Arctic dust. Our results demonstrate the importance of considering an ice‐nucleation parameterization suitable for Arctic dust when simulating INPs and their effects on aerosol‐cloud interactions in the Arctic.

study also showed that Arctic dust is mostly distributed in the Arctic lower troposphere due to the stable atmosphere and limited convection in the cold climate of the Arctic, whereas dust transported over long distances from Africa and Asia is dominant in the Arctic middle and upper troposphere. These different height distributions between Arctic dust and long-range transported dust affect the vertical distribution and seasonal variation of dust INPs over the Arctic (Shi et al., 2022), which affects mixed-phase clouds, found in the Arctic lower troposphere throughout the year (Mioche et al., 2015;Shupe, 2011), and radiative balance in the Arctic (Prenni et al., 2007;Tan & Storelvmo, 2019;Zhang et al., 2019).
Recent studies have suggested the importance of Arctic dust as INPs under conditions relevant for the formation of mixed-phase clouds (Sanchez-Marroquin et al., 2020;Tobo et al., 2019;Xi et al., 2022;Yun et al., 2022). Tobo et al. (2019) showed that Arctic dust collected from the surface of a glacial outwash plain in the Svalbard Islands has a remarkably high ice nucleating ability, especially at high temperatures (>−20°C), and suggested that this high ice nucleating ability is caused by the presence of certain organic matter contained in the dust. Although Arctic dust exists mainly in the Arctic lower troposphere (Groot Zwaaftink et al., 2016), Arctic dust might substantially affect the number concentrations of dust INPs in this lower level of the Arctic because of its high ice nucleating ability. However, this impact is still unknown because there have been no modeling studies that consider the high ice nucleating ability of Arctic dust.
In this study, we incorporate the observed high ice nucleating ability of Arctic dust (Tobo et al., 2019) into a global aerosol-climate model for the first time. Then, we investigate its impacts on the vertical distribution and number concentrations of dust INPs and their effects on cloud radiative forcing (CRF) and surface downwelling radiative flux in the Arctic and reveal the importance of Arctic dust with high ice nucleating ability in simulations of INPs and their effects on clouds in the Arctic.
In CLM4, dust emission fluxes are calculated online from wind friction velocity and land-surface parameters such as snow cover, soil moisture, and vegetation based on the Dust Entrainment and Deposition (DEAD) model (Zender et al., 2003). We used a revised version of the DEAD model (Kok, Albani, et al., 2014;Kok, Mahowald, et al., 2014), which calculates dust emission fluxes based on vertical dust flux measurements over arid regions of the world. The revised DEAD model also considers dust emissions at high latitudes (north of 60°N). The size distribution of emitted dust particles in this study is based on that of Kok (2011).
In this study, Arctic averages of dust mass and INP number concentrations were calculated for north of 60°N. We define the source region of Arctic dust as north of 57°N to include dust emissions from southern Greenland ( Figure S1 in Supporting Information S1). Our model has different tracers for Arctic dust and other dust (emitted from south of 57°N) and simulates their emission, transport, deposition, and ice nucleation processes separately. Total concentrations were calculated as the sum of the concentrations of Arctic dust and other dust for both dust mass and INP number concentrations.
As with Kawai, Matsui, and Tobo (2021), we replaced the ice nucleation scheme for mixed-phase clouds used in the default CAM5 (Meyers et al., 1992), which estimates INP number concentrations from ice supersaturation only, with the physically based parameterization of DeMott et al. (2015) (hereafter D15), which focuses on immersion/condensation freezing of dust (>0.5 μm in diameter) and obtains INP number concentrations related to the variation of dust. The D15 parameterization is based on laboratory experiments for dust particles sampled from Asian and Saharan deserts and field measurements made within Asian and Saharan dust plumes. Using this parameterization, the model calculated INP number concentrations at every timestep (30 min) from ambient temperature (between −37°C and −5°C) and interstitial and cloud-phase dust number concentrations (>0.5 μm diameter) when liquid water is present (i.e., relative humidity exceeds 100% locally within a grid of 1.9° × 2.5°). This study shows grid-mean number concentrations of INPs that exist in clouds, calculated from grid-mean INP number concentrations multiplied by stratus cloud fractions, which is consistent with the treatment of INPs in the cloud microphysics scheme of CAM5 Morrison & Gettelman, 2008). Because immersion freezing is the dominant mechanism for heterogeneous ice nucleation in mixed-phase clouds (Hoose et al., 2010), this study focuses on INPs from immersion/condensation freezing of dust.
The INP number concentrations of Arctic dust were calculated based on the INP observations of Tobo et al. (2019) (hereafter T19), which obtained high ice nucleation active site density per unit mass n m (g −1 ) for Arctic dust as where T is air temperature (°C) ( Figure S2 in Supporting Information S1). In this study, this equation was used in the same temperature range as the D15 parameterization (between −37°C and −5°C). We calculated the INP number concentrations of Arctic dust N inp (m −3 ) from n m and the number concentrations of Arctic dust N dst,j (m −3 ) in particle size bin j as where m dst,j is dust mass per particle in particle size bin j (g) (in analogy to Equation 2 of Niemand et al., 2012).
We performed two simulations for 11 years (2009-2019) with a horizontal resolution of 1.9° × 2.5° and 30 vertical layers (0-40 km altitude) with monthly data of sea surface temperature and sea ice concentrations (Hurrell et al., 2008), which were used to reproduce more realistic conditions in the Arctic (e.g., sea surface albedo).
The first year of the simulations was treated as model spin-up, and the last 10 years (2010-2019) were used for analysis. In the BASE simulation, the INP number concentrations of both Arctic dust and other dust were calculated using the D15 parameterization (Table S1 in Supporting Information S1). In the HIGH simulation, the INP number concentrations of Arctic dust were calculated using the T19 parameterization, while the INP number concentrations of other dust were calculated using the D15 parameterization. Note that the spatial distributions of dust emissions, dust mass concentrations, and clouds are almost the same in both simulations. Temperature and horizontal wind fields (<800 hPa) were nudged to the Modern-Era Retrospective analysis for Research and Applications version 2 (MERRA2) data set. The Coupled Model Intercomparison Project Phase 6 (CMIP6) emission data set for the year 2010 (Hoesly et al., 2018;van Marle et al., 2017) was used for all aerosol and precursor species other than dust and sea salt emissions, which were calculated online.
We also performed sensitivity simulations in which the number concentrations of total dust INPs calculated in the ice nucleation scheme were turned off (INPx0) or increased by a factor of 10 (BASEx10 and HIGHx10) to investigate the effects of total dust INPs on CRF at the top of the atmosphere (TOA) and downwelling radiative flux at the surface in the Arctic (Table S1 in Supporting Information S1). These effects are calculated from the difference in CRF and downwelling radiative flux between the 10-fold and INPx0 simulations (BASEx10 − INPx0 and HIGHx10 − INPx0). The 10-fold sensitivity simulations were performed to increase the signal of CRF and surface downwelling radiative flux from dust INPs. CRF was calculated as the difference in radiative fluxes at TOA between all sky and clear sky conditions (Ramanathan et al., 1989).
For comparisons of model simulations with observations (Section 3.1), we used dust aerosol optical depth (AOD) from the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO) lidar level 3 tropospheric aerosol profile product version 4.20 (Kim et al., 2018) and cloud liquid water path from the Multisensor Advanced Climatology (MAC) data set (Elsaesser et al., 2017). We also used INP observations at various locations in the Arctic region (Conen et al., 2016;Creamean et al., 2018;Mason et al., 2016;

Comparisons With Observations and Other Simulations
The simulated annual-mean global and regional dust emission flux in our model is within the estimate of the AeroCom models (Huneeus et al., 2011), except for the Middle East (Table S2 in Supporting Information S1). The simulated annual-mean dust emission flux in the Arctic (>57°N) is 45 Tg yr −1 and contributes to 1.3% of the annual-mean global dust emission flux (3,573 Tg yr −1 ). These values are close to the observation-based estimate of Bullard (2017) ≥60°N). In our simulations, Arctic dust is mostly emitted between June and October, with high surface temperatures and low snow coverage in the Arctic ( Figure S3 in Supporting Information S1). The seasonality of the Arctic dust emissions is consistent with those simulated by Tobo et al. (2019) and Shi et al. (2022). Arctic dust is mainly emitted from northern Canada, Greenland, and Arctic islands, such as the Svalbard Islands and Severny Island ( Figure S1 in Supporting Information S1).
The simulated annual-mean globally averaged dust AOD is 0.024, which is close to the observation-based estimate (0.030 ± 0.005) of Ridley et al. (2016). Our model relatively well reproduces the tendency of the latitudinal distribution of annual-mean observed dust AOD obtained from the CALIPSO retrievals ( Figure S4 in Supporting Information S1), although the magnitude of observed dust AOD is overestimated at 10-30°N and underestimated at 50-80°N. The overestimation at 10-30°N may be because dust mixed with smoke or urban pollution is not included in the observed values. Matsui and Mahowald (2017) and Shi and Liu (2019) showed that simulated dust concentrations were overestimated near dust source regions and underestimated in remote regions, which is consistent with the comparison between our simulations and the CALIPSO retrievals.
We used cloud liquid water content as the threshold for the presence of liquid water when calculating INP number concentrations. Our model reasonably well reproduces the annual-mean cloud liquid water path in the Arctic obtained from the MAC satellite observation data, although these observation data are available only over the ice-free ocean ( Figure S5 in Supporting Information S1). The annual-mean cloud liquid water path averaged over the Arctic is 78 g m −2 for the MAC observation data and 76 g m −2 for our simulations (averaged when and where the observation data are available).
We compared the BASE and HIGH simulations with INP observations during summer and fall in northern Norway (Conen et al., 2016); Alert, Canada (Mason et al., 2016); Ny-Ålesund, Svalbard (Tobo et al., 2019); Utqiagvik, Alaska; and northern Greenland (Wex et al., 2019) in the Arctic region. The INP number concentrations in the HIGH simulation are in better agreement with those observed at the four locations other than Svalbard, compared with those in the BASE simulation ( Figure S6 and Table S3 in Supporting Information S1). In the HIGH simulation, the INP number concentrations observed in Svalbard are overestimated by about 1-3 orders of magnitude, but the good agreement with the INP observations at the other four locations suggests that our model may overestimate local dust emissions from Svalbard. During spring, the INP number concentrations observed in Alaska (Creamean et al., 2018) and Svalbard (Tobo et al., 2019) are underestimated by about 1-2 orders of magnitude in our simulations (Table S4 in Supporting Information S1). This discrepancy suggests that INPs other than dust, such as terrestrial and marine biogenic aerosols (Burrows et al., 2013;Creamean et al., 2020Creamean et al., , 2022Šantl-Temkiv et al., 2019;Wilson et al., 2015), are dominant near the surface in the Arctic during the period when Arctic dust is not emitted (winter-spring) and that such non-dust INPs need to be incorporated into models in future studies.

Contributions of Arctic Dust to Dust Mass and INPs in the Arctic
Arctic dust is mainly distributed over northern Canada, the coast of Greenland, and Arctic islands (Figure 1a). The mass concentrations of Arctic dust averaged over the Arctic (>60°N) are high near the surface in summer and fall (Figure 2a), with high contributions (>80%) to total dust (Figure 2c), because Arctic dust is emitted during these seasons ( Figure S3c in Supporting Information S1). Arctic dust is mostly distributed in the lower troposphere because of the stable atmosphere and limited convection in the cold climate of the Arctic (Groot Zwaaftink et al., 2016;Stohl, 2006). The vertically and horizontally integrated mass of Arctic dust in the Arctic lower troposphere (>700 hPa) averaged in summer and fall (June-November) is 64 Gg, corresponding to 83% of total dust (77 Gg).

10.1029/2022GL102470 5 of 10
Other dust transported over long distances from regions outside of the Arctic, such as North Africa, the Middle East, and Asia (Groot Zwaaftink et al., 2016;Shi et al., 2022), is mainly distributed in the middle and upper troposphere (200-700 hPa) during spring and in the lower troposphere from spring to fall (Figure 2b). The long-range transport of dust from mid-and low-latitudes to the Arctic middle and upper troposphere is consistent with isentropic transport of air pollution to the Arctic (Heidam et al., 2004;Massling et al., 2015;Stohl, 2006).
The spatial distribution and temporal variation of INP number concentrations differ significantly from those of dust mass concentrations because INP number concentrations are determined by dust concentrations, air temperature, and cloud properties. In the Arctic lower troposphere during summer and fall, the mass concentrations of Arctic dust are high (Figure 2a), but Arctic dust does not act efficiently as INPs in the BASE simulation (Figures 1b, 2d, and 2f) because the D15 parameterization used for both Arctic and other dust INPs indicates low ice nucleating ability of desert dust at temperatures warmer than about −15°C. However, when we consider the observed remarkably high ice nucleating ability of Arctic dust using the T19 parameterization, Arctic dust acts efficiently as INPs in the Arctic lower troposphere (>700 hPa) during summer and fall in the HIGH simulation ( Figure 2g), with substantial contributions (∼100%) to total dust INPs (Figure 2i). In the Arctic lower troposphere during summer and fall, the vertically integrated number concentration of Arctic dust INPs in the HIGH simulation (1,301 m −2 ) is 186 times higher than that in the BASE simulation (7.0 m −2 ) (Figure 3a and Figure S7 in Supporting Information S1), which substantially increases the vertically integrated number concentration of total dust INPs by a factor of 109 from the BASE simulation (12 m −2 ) to the HIGH simulation (1,305 m −2 ). As a result, Arctic dust INPs contribute to ∼100% of total dust INPs in the Arctic lower troposphere during summer and fall in the HIGH simulation. Arctic dust INPs are mainly distributed over northern Canada, Greenland, and Arctic islands in the HIGH simulation (Figure 1c).
The substantial increase in Arctic dust INPs in the Arctic lower troposphere during summer and fall in the HIGH simulation occurs at temperatures between −20°C and −5°C ( Figure S8 in Supporting Information S1). In the Arctic lower troposphere during summer and fall, the vertically integrated number concentrations of Arctic dust INPs in the HIGH simulation are 94 times higher at temperatures between −20°C and −10°C and 206 times higher at temperatures between −10°C and −5°C than those in the BASE simulation. The temperature range of −20°C to −5°C is typical of the Arctic lower troposphere during these seasons (Figure 2), which is also the temperature range in which the ice nucleating ability of Arctic dust is higher than that of desert dust (Tobo et al., 2019).
In the Arctic middle and upper troposphere (<700 hPa), Arctic dust INPs contribute to 3.7% of total dust INPs even in summer and fall in the HIGH simulation (Figure 3b), while other dust INPs dominate throughout the year in both the BASE and HIGH simulations (Figures 2e and 2h). Other dust also acts efficiently as INPs in the Arctic lower troposphere in winter and spring because of low temperatures (<−10°C) in these seasons.
Our results suggest that an ice-nucleation parameterization suitable for Arctic dust, such as the T19 parameterization, should be used when simulating INPs in the Arctic. In addition, because Arctic dust may contain various amounts and types of organic matter and have different ice nucleating ability depending on its sources (e.g., Alaska, Canada, Greenland, and Iceland) (Sanchez-Marroquin et al., 2020;Xi et al., 2022), such spatial variability of Arctic dust ice nucleating ability needs to be considered in future INP modeling. Furthermore, the

Impacts of the Increase in Arctic Dust INPs on CRF and Surface Radiative Flux
We further investigate the effects of total dust INPs on CRF at TOA and downwelling radiative flux at the surface in the Arctic and their changes due to the increase in Arctic dust INPs. The annual-mean effect of total dust INPs on CRF averaged over the Arctic (>60°N) for the BASE and HIGH simulations is estimated to be 0.81 ± 0.08 (average and standard deviation (1σ) of annual-mean values for the years 2010-2019) and 0.85 ± 0.08 W m −2 for shortwave (SW), −0.39 ± 0.07 and −0.39 ± 0.06 W m −2 for longwave (LW), and 0.42 ± 0.08 and 0.45 ± 0.07 W m −2 for net radiation, respectively (Table S5 in Supporting Information S1). The signs of these values are consistent with those of Shi and Liu (2019)

Conclusions
We investigated the impacts of the observed remarkably high ice nucleating ability of Arctic dust (  CRF at TOA and surface downwelling radiative flux in the Arctic by using the global aerosol-climate model CAM-ATRAS. Arctic dust is mostly emitted in summer and fall, when the Arctic surface temperatures are high and snow coverage is low, and distributed in the Arctic lower troposphere. If Arctic dust has the same ice nucleating ability as desert dust (BASE simulation), it does not act efficiently as INPs in the Arctic lower troposphere in summer and fall because the ice nucleating ability of desert dust is low at temperatures warmer than about −15°C. However, when considering the high ice nucleating ability of Arctic dust (HIGH simulation), INP observations at various locations in the Arctic region are generally better reproduced, the number concentrations of Arctic and total dust INPs in the Arctic lower troposphere during summer and fall increase by more than a factor of 100, and almost all of the total dust INPs are contributed from Arctic dust. This substantial increase is caused by the higher ice nucleating ability of Arctic dust than desert dust at temperatures between −20°C and −5°C, which is typical of the Arctic lower troposphere in summer and fall. The increase in Arctic dust INPs enhances the annual-mean effects of total dust INPs on net CRF at TOA and net downwelling radiative flux at the surface in the Arctic by 0.03 and 0.04 W m −2 , respectively. These values should be further investigated because these estimates are influenced by uncertainties in many factors (e.g., the spatial distributions and concentrations of dust, INPs, and cloud ice and water). Our results show that Arctic dust with the observed high ice nucleating ability plays a dominant role in the number concentrations of Arctic and total dust INPs in the Arctic lower troposphere during summer and fall. This study therefore demonstrates the importance of considering an ice-nucleation parameterization suitable for Arctic dust when simulating INPs in the Arctic, and their impacts on aerosol-cloud interactions in the Arctic need to be more accurately estimated.