High Potential of Asian Dust to Act as Ice Nucleating Particles in Mixed‐Phase Clouds Simulated With a Global Aerosol‐Climate Model

Mineral dust affects the microphysical and radiative properties of mixed‐phase clouds and hence the radiative balance of the Earth by acting as ice nucleating particles (INPs). However, the importance of Asian dust as INPs is not well understood. In this study, we examined the contribution of Asian dust to global dust INPs and its effect on cloud radiative forcing (CRF) using a global aerosol‐climate model with an ice nucleation parameterization that links INP number concentrations to ambient temperature and dust number concentrations. Our model well reproduces INP number concentrations measured over the Tokyo Metropolitan area in Japan during May 2017, when Asian dust was transported to Japan. Our simulation for the years 2013–2017 shows that Asian dust extends from its source regions (e.g., the Gobi and Taklimakan Deserts) to the North Pacific, North America, and the Arctic. Notably, Asian dust is transported to higher altitudes (i.e., to temperature regimes relevant for the formation of mixed‐phase clouds) more efficiently than dust from other regions. The annual‐mean simulated contribution of Asian dust to global dust INPs is 15%, which is 4.4 times higher than its contribution to global atmospheric dust loading (3.4%). These characteristics of Asian dust show its high potential to act as INPs in mixed‐phase clouds. Sensitivity simulations show that Asian dust INPs have a positive net CRF of 0.054–0.19 W m−2 in East Asia and the North Pacific during 2013–2017 (cf. 0.092–1.0 W m−2 for dust from other regions).

to the North Pacific by the westerlies (Husar et al., 2001;Iwasaka et al., 1983;Kai et al., 1988;Yumimoto et al., 2009). A case has been reported of Asian dust being transported one full circuit of the globe . Therefore, the various dust sources and transport processes should induce different potentials of dust to be transported into low-temperature regions and to act as INPs in mixed-phase clouds.
The major source regions of Asian dust are the Taklimakan and Gobi Deserts (Kurosaki & Mikami, 2005;Sun et al., 2001;Wu et al., 2016), which are at high altitudes of 800-1,500 m. The Taklimakan Desert is located in the Tarim Basin and surrounded to the north, west, and south by high mountain ranges (>5,000 m altitude). Here, dust is often raised from the surface by strong easterly winds from the eastern corridor and then lifted to the upper troposphere along the slope of the high mountains (Kai et al., 2008;Tsunematsu, 2005;Uno et al., 2005;Yumimoto et al., 2009). In the Gobi Desert, dust events are commonly linked to the passage of a cold front (Hayasaki et al., 2006;Takemi & Seino, 2005), which can lead to the transport of dust to higher altitudes by ascending warm air (Hara et al., 2009;Kawai et al., 2015Kawai et al., , 2018. This transport of Asian dust to higher altitudes implies its high potential to be transported into low-temperature regions and to act there as INPs. Some studies have used Asian dust particles in cloud chamber experiments to develop ice nucleation parameterizations (e.g., Boose et al., 2016;Connolly et al., 2009;DeMott et al., 2015;Niemand et al., 2012;Ullrich et al., 2017). Overall, these studies showed that various desert dust, including Asian and Saharan dust, has a relatively similar ice nucleating ability for immersion freezing (ice nucleation by particles immersed within supercooled water droplets). In contrast, few studies have focused on the atmospheric distribution of INPs from Asian dust. Wiacek et al. (2010) suggested that Asian dust is more often transported into regions of mixed-phase clouds compared to Saharan dust based on forward trajectory analyses from their source regions, although they did not consider dust emission and deposition processes. Fung (2018a, 2018b) incorporated dust INPs into the Weather Research and Forecasting model coupled with Chemistry (WRF-Chem), thus improving simulations of cloud ice water content, shortwave and longwave radiation, surface temperature, and precipitation over East Asia during spring. However, the contributions of Asian dust to global INPs and cloud radiative forcing (CRF) are still not well understood.
In this study, we investigate the spatial distribution of Asian dust and its contributions to INPs and CRF by using a global aerosol-climate model and reveal the potential of Asian dust to act as INPs in mixed-phase clouds. In Section 2, we describe the methodology of our model simulations and the observational data used for model evaluations. Section 3 shows the simulation results of global dust emissions, model evaluations, the spatial distribution of Asian dust, its contribution to global dust INPs, and its effect on CRF. A summary and our conclusions are presented in Section 4.
In this study, we replaced the ice nucleation scheme for mixed-phase clouds used in the default CAM5 (Meyers et al., 1992) with the physically based parameterization of DeMott et al. (2015) (Table S1), which focuses on immersion/condensation freezing of dust particles larger than 0.5 μm in diameter. The parameterization of DeMott et al. (2015) is based on laboratory experiments for dust particles sampled from Asian and Saharan deserts and field measurements made within Asian and Saharan dust plumes. DeMott et al. (2015) suggested that the immersion freezing activity of dust particles derived from the field measurements might already be influenced by aging processes in the atmosphere. Some laboratory experiments showed that coatings of nitric acid on dust particles have little influence on immersion freezing (Sullivan, Miñambres, et al., 2010), while coatings of sulfuric acid on dust particles can decrease the immersion freezing ability of dust particles (Sullivan, Petters, et al., 2010;Tobo et al., 2012;Wex et al., 2014).
Our model calculates INP number concentrations online from ambient temperature (between −37°C and −5°C) and dust number concentrations (>0.5 μm diameter) ( Figure S1) when liquid water is present (i.e., relative humidity has exceeded 100% locally within a grid). The INP number concentrations averaged in a cloud portion (grid-mean INP number concentration × cloud fraction) are used in the two-moment stratiform cloud microphysics scheme in CAM5 Morrison & Gettelman, 2008) and change the number concentrations of cloud liquid and ice, liquid and ice water path, and cloud radiative balance. For comparisons with INP observations, we also calculated INP number concentrations offline (from simulation outputs) for temperatures between −25°C and −10°C (every 0.5°C) by using the parameterization of DeMott et al. (2015).
In our model, contact freezing for mixed-phase clouds is calculated by the parameterization of Young (1974), while deposition nucleation is not considered because of its small contribution to INPs in mixed-phase clouds (Ansmann et al., 2008) (Table S1). Because immersion freezing of mineral dust is the dominant mechanism for heterogeneous ice nucleation in mixed-phase clouds (Hoose et al., 2010), this study shows INPs from immersion/condensation freezing of mineral dust.
In CLM4, dust emission fluxes are calculated from wind friction velocity and land-surface parameters such as soil moisture and vegetation cover based on the Dust Entrainment and Deposition (DEAD) model (Zender et al., 2003). In this study, we used a revised version of the DEAD model (Kok, Albani, et al., 2014;, which calculates dust emission fluxes based on vertical dust flux measurements over arid regions of the world. We assumed the size distribution of dust emissions based on Kok (2011).
In this study, dust emitted from East Asia (the region bounded by 30-55°N and 75-130°E, but excluding the area bounded by 45-55°N and 75-85°E; Figure 1) is considered to be "Asian dust." Note that we use the term "Asian dust" for convenience, but do not include central Asian dust (e.g., Hofer et al., 2017Hofer et al., , 2020. Our model has two tracers for Asian dust and dust from other regions ("other dust" hereafter) in each size bin to simulate the emission, transport, and deposition processes of Asian and other dust separately. INP number concentrations are also calculated separately for Asian and other dust using the dust tracers. Total dust is the sum of Asian and other dust. The contribution of Asian dust to total dust can be calculated using the dust tracers for all three-dimensional grids and timesteps.
The base simulation was performed for the years 2012-2017 with a horizontal resolution of 1.9° × 2.5° and 30 vertical layers from the surface to ∼40 km. The monthly and daily mean output data for the last 5 years (2013-2017) were used for our analysis. The daily mean output data were used to determine the relationship between atmospheric dust loading and ambient temperature. Unless noted otherwise, the results KAWAI ET AL.  given in this study are 5-year averages (2013-2017). Temperature and horizontal wind fields were nudged to the Modern-Era Retrospective Analysis for Research and Applications version 2 (MERRA2). The Coupled Model Intercomparison Project Phase 6 (CMIP6) anthropogenic and biomass burning emissions (Hoesly et al., 2018;van Marle et al., 2017) for the year 2010 were used for aerosols and their precursor species. Sea salt emissions were calculated online by the parameterizations of Mårtensson et al. (2003) and Monahan et al. (1986).
We performed five sensitivity simulations to investigate the effects of INPs from Asian and other dust on cloud water path and CRF at the top of the atmosphere (ΔCRF) ( Table S2). The TOTx1 simulation is the base simulation described above. The ASAx1 (OTHx1) simulation considers INPs only from Asian dust (other dust). Because our model tends to underestimate dust concentrations in dust outflow regions by amounts ranging from tens of percent to two orders of magnitude (Matsui & Mahowald, 2017), we conducted additional simulations with increased INP number concentrations to estimate the possible ranges of ΔCRF (see Section 3.5). In the TOTx10 simulation, INP number concentrations are increased by a factor of 10 for total dust (Asian dust + other dust). In the ASAx10 (OTHx10) simulation, INP number concentrations from Asian dust (other dust) are increased by a factor of 10, while INPs from other dust (Asian dust) are not considered. Note that only INP number concentrations are modified in the sensitivity simulations; dust concentrations are not modified.
We calculated the effects of Asian dust INPs on cloud water path and CRF from the differences between TOTx1 and OTHx1 and between TOTx10 and OTHx10, and those of other dust INPs from the differences between TOTx1 and ASAx1 and between TOTx10 and ASAx10. This method was used to avoid noise produced by removing all INPs and obtain clear signals of the effects of Asian and other dust INPs (see Text S1 and Figure S2). CRF is the difference in net downward radiative fluxes between all sky and clear sky and is calculated for shortwave (SW) and longwave (LW) radiation (Ramanathan et al., 1989). Positive and negative CRF values indicate warming and cooling, respectively, of the underlying atmosphere and surface. In CAM5, radiative transfer for SW and LW radiation is calculated from cloud fraction and the mass and radius of cloud water droplets and ice crystals (Park et al., 2014) by using the Rapid Radiative Transfer Model for General Circulation Models (RRTMG) (Iacono et al., 2008;Mlawer et al., 1997).
We used observational data from a satellite, a ground observation network, and a tower to evaluate our model simulations. We obtained global distributions of aerosol optical depth (AOD) at a wavelength of 550 nm retrieved from the Moderate Resolution Imaging Spectroradiometer (MODIS) on board the Aqua satellite during 2013-2017. Surface PM 10 concentrations measured in Japan and Korea were obtained from the Acid Deposition Monitoring Network in East Asia (EANET) at 14 different sites for the years 2013-2017 ( Figure S3). We also used aerosol and INP concentration data observed at the 458-m level of Tokyo Skytree (35.71°N, 139.81°E) in Japan during May 2017, when dust concentrations were high (Tobo et al., 2020). Our simulation results were compared with size-resolved aerosol number concentrations measured with an optical particle counter (OPC) (KA-03, RION Co., Ltd.) with 5 channels (≥0.3, ≥0.5, ≥1, ≥2, and ≥5 μm). INP number concentrations for temperatures between −25°C and −10°C were observed every 3 days in freezing experiments using the National Institute of Polar Research Cryogenic Refrigerator Applied to Freezing Test (CRAFT) (Tobo, 2016). This result was used to evaluate the temperature dependence of simulated INP number concentrations.

Global Distribution of Dust Emission Fluxes
According to our base simulation, most of the world's major dust sources are in North Africa, the Middle East, central Asia, and East Asia (Figure 1), encompassing an area previously recognized as the "dust belt" (Prospero et al., 2002). Other minor dust source regions are in South and North America, South Africa, Australia, and the Arctic. The simulated annual mean global dust emission flux is 3,999 Tg yr −1 (Table 1), which is 3.6 times larger than the median flux of the AeroCom Phase I global dust model intercomparison (Huneeus et al., 2011), but it is within the range of the AeroCom models. Our simulated annual mean global averaged dust optical depth is 0.025, which is slightly lower but within the range estimated by Ridley et al. (2016) (0.030 ± 0.005). The slightly lower dust optical depth may be related to the tendency of our model to underestimate dust concentrations in dust outflow regions (Matsui & Mahowald, 2017). In our simulation, the maximum regional dust emission is in North Africa (52.5% of global dust emissions), followed by the Middle East (22.5%) and Asia (11.3%). Although our simulated regional dust emission fluxes are larger than the AeroCom medians, they are within the range of the AeroCom models, with the exception of the Middle East.
Our base simulation shows that the main source regions of Asian dust are the Gobi and Taklimakan Deserts ( Figure 1); this is consistent with the results of previous studies (Kurosaki & Mikami, 2005;Sun et al., 2001;Wu et al., 2016). The simulated annual mean emission flux of Asian dust (emitted from East Asia) is 283 Tg yr −1 , representing 7.1% of global dust emission flux (Table 1). This result is similar to the estimate of Tanaka and Chiba (2006) (214 Tg yr −1 and 11%). The simulated monthly mean emission flux of Asian dust reaches a maximum in May (67 Tg month −1 ), which accounts for 24% of the annual mean emission flux of Asian dust (Figures 1b and S4). Station-based observations show similar seasonality of Asian dust outbreaks (Kurosaki & Mikami, 2005).

Model Evaluations
The spatial distribution of AOD observed by MODIS during the years 2013-2017 is reasonably well reproduced in our base simulation, although AOD around the dust belt is overestimated by a factor of about 2 ( Figure S5). The distribution and magnitude of simulated AOD along dust transport routes, such as over the Atlantic and the North Pacific, are similar to the observed data. The simulated AOD averaged over the North Pacific (150°E-120°W and 20°N-60°N) is 74% of the observed value.
Our model reasonably well reproduces surface PM 10 concentrations observed at 14 different EANET sites in Japan and Korea during the years 2013-2017 (Figures 2 and S3). Kanghwa, Imsil, and Ijira are classified as rural sites, Banryu as an urban site, and the others as remote sites. The differences between observed and simulated PM 10 concentrations are within a factor of about 2 (except at Happo; Figure S3). We attribute higher observed and simulated PM 10 concentrations in Korea (Kanghwa, Cheju, and Imsil) than in Japan to the closer proximity of Korea to source regions of both Asian dust and anthropogenic aerosols. The timing of high PM 10 concentrations associated with high dust contributions is relatively well simulated ( Figure S6). Note that the differences between the observed and simulated PM 10 concentrations are caused in part by spatial representation errors of the observations for the larger model grids ( Figure S3) (Schutgens et al., 2016(Schutgens et al., , 2017. Such errors are also included in the comparisons at Tokyo Skytree shown below KAWAI ET AL.    (Figure 4). The observation suggests a stronger ice nucleating ability between −19°C and −10°C, which may be attributed to the presence of other aerosol particles such as biological aerosols (Tobo et al., 2013(Tobo et al., , 2020. During May 2017, the temporal variability of simulated dust INP number concentrations is greater than that of observed INP number concentrations ( Figure S8). This result suggests that aerosols other than Asian dust may contribute to the INPs observed for the period lacking a dust event (Tobo et al., 2020), but further investigation of this possibility is needed. However, the model simulation well reproduces the INP number concentrations measured during the dust event ( Figures S8b and S8c).
We also calculated INP number concentrations at Tokyo Skytree dur-    Figure 5 shows the simulated annual mean spatial distributions of total dust and Asian dust. As shown in Figure 5a, dust aerosol is distributed over and downwind of dust source regions (Figure 1). The annual mean atmospheric loading of total dust is 35 Tg (Table 2), which is higher than previously reported global dust loadings of 22.4 Tg (Liu et al., 2012) and 28.5 Tg (Matsui, 2017) because of the larger dust emissions in our simulation (Table 1). The higher dust loading and larger dust emissions may be related to our use of the revised version of the default dust emission scheme (see Section 2). Dust is transported to higher altitudes in the Northern Hemisphere (up to about 200 hPa) compared to the Southern Hemisphere (mainly up to 400 hPa) (Figure 5d).

Spatial Distribution of Asian Dust
Asian dust extends from its source regions to the North Pacific, North America, and the Arctic (Figure 5b). The transport range of Asian dust is largest in May ( Figure S10b), which is consistent with the timing of the maximum emission flux of Asian dust ( Figure S4). The annual mean atmospheric loading of Asian dust is 1.2 Tg (Table 2), and the mean for May is 3.2 Tg (Table S3). The simulated spatial distributions of Asian dust are consistent with previously reported eastward transport of Asian dust to the North Pacific by the westerlies (Husar et al., 2001;Kai et al., 1988;Yumimoto et al., 2009).
The contribution of Asian dust to the annual mean total dust loading is 3.4% (Table 2), which is about half of its contribution to global dust emissions (Table 1). This result implies that Asian dust has a shorter atmospheric lifetime than other dust, likely because Asian dust is efficiently removed from the atmosphere by precipitation associated with mid-latitude cyclones. Luo et al. (2003) and Tanaka and Chiba (2006) 20°N-60°N), the contribution of Asian dust decreases eastward from 80% to 30% (Figure 5c), with a regional average of 52% (Table 2).
Asian dust is distributed from the surface to the upper troposphere at middle latitudes in the Northern Hemisphere (Figure 5e). Asian dust is partly transported to temperature regimes relevant for the formation of mixed-phase clouds (between −38°C and 0°C). In these low-temperature regions, the contribution of Asian dust to annual mean total dust loading is 13%, which is 3.8 times greater than that for all temperature ranges ( Table 2). The fraction of dust loading in the low-temperature range (between −38°C and 0°C) is 35% for Asian dust and 9.1% for total dust. Asian and other dust show peak fractions between 0°C and 5°C and between 15°C and 20°C, respectively ( Figure 6). These differences show that Asian dust is efficiently transported into low-temperature regions and has high potential to act as INPs in mixed-phase clouds. This phenomenon is probably related to the high altitudes of Asian dust source regions (Wiacek et al., 2010). The upward transport of dust by cold frontal systems (Hara et al., 2009;Kawai et al., 2015Kawai et al., , 2018 that frequently pass over the Gobi Desert in spring (Hayasaki et al., 2006;Takemi & Seino, 2005) would also transport Asian dust to higher altitudes.
Our simulation shows that Asian dust is transported to the middle and upper troposphere in the Arctic (Figure 5e), which is consistent with the findings of Groot Zwaaftink et al. (2016). The contribution of Asian dust to total dust loading reaches about 30% in these levels (Figure 5f), with an annual mean contribution of 9.0% (north of 60°N; Table 2).
In Figure 7, we focus on the vertical distribution of Asian dust over its source regions. Asian dust is dominant in the lower and middle troposphere over the Gobi Desert (Figure 7b), but the contribution of other dust increases with height in the middle and upper troposphere (Figure 7c). Tanaka and Chiba (2006) also noted that the contribution of dust originating from regions other than East Asia cannot be ignored over East Asia and the North Pacific. Other dust accounts for 10%-80% of total dust at temperatures between −35°C and −20°C (Figure 7c). This result suggests that dust from regions other than East Asia (e.g., central Asia) may affect the radiation budget and hydrological cycle in the Gobi Desert by providing INPs and thus modifying the amount of mixed-phase clouds.
KAWAI ET AL.

Contribution of Asian Dust to Global Dust INPs
In Section 3.3, we discussed the contribution of Asian dust to total dust loading in the low-temperature range (between −38°C and 0°C). In contrast, the temperature dependence of the ice nucleating ability of dust particles plays an important role in simulating INP number concentrations, which vary substantially according to the choice of ice nucleation parameterizations (Shi & Liu, 2019;Xie et al., 2013). In this section, we discuss  (Figure 8a), although a large amount of Saharan dust is transported from North Africa (Figure 5a). This is attributed to the low-altitude, high-temperature transport of Saharan dust in the Saharan Air Layer (Karyampudi et al., 1999;Wiacek et al., 2010).
INPs are distributed mainly between 600 and 300 hPa with a maximum at middle latitudes in the Northern Hemisphere (Figure 8d). INP number concentrations are larger in the Northern Hemisphere than in the Southern Hemisphere due to larger dust source regions and higher dust emissions in the Northern Hemisphere (Figures 1, 5a and 5d). INPs are also distributed between 900 and 600 hPa at middle latitudes in the Northern Hemisphere (Figure 8d), which is attributed to the combination of decreasing dust concentrations and lower temperatures from south to north in the Northern Hemisphere (Figure 5d). These characteristics of global INP distributions are consistent with results presented by Xie et al. (2013).
Asian dust INPs extend from Asian dust source regions to the North Pacific, North America, and the Arctic (Figure 8b). They are distributed between 900 and 200 hPa at middle latitudes in the Northern Hemisphere and in the middle troposphere of the Arctic (Figure 8e). Asian dust INP number concentrations are high in the region from eastern Mongolia to northeastern China (Figures 8b and S11b), which is east of the highest concentration of Asian dust over the Gobi Desert (Figures 5b and S10b). This eastward shift is probably related to simultaneous eastward and upward transport of dust by cold frontal systems (Hara et al., 2009;Kawai et al., 2018). In this region, the contribution of Asian dust to total dust INPs exceeds 70% (Figure 8c) and is highest between 900 and 500 hPa (Figure 8f).

KAWAI ET AL.
10.1029/2020JD034263 9 of 16 The global contribution of Asian dust to annual mean total dust INPs is 15% (Table 3), which is 2.1 times higher than its contribution to global dust emissions (Table 1) and 4.4 times higher than its contribution to the global dust loading ( Table 2). The contribution of Asian dust to total dust INPs is similar to its contribution to the dust loading at temperatures between −38°C and 0°C (Table 2). Over the North Pacific (150°E-120°W and 20°N-60°N), the contribution of Asian dust to INPs is 27% (Table 3), which is about half of its contribution to the total dust loading (Table 2), probably because Asian dust is dominant in the lower troposphere (at higher temperatures) (Figure 7b). The contribution of Asian dust to INPs in the Arctic is 15%, similar to the global contribution (Table 3). In all regions described above, the contribution of Asian dust to INPs in May (27% globally, 39% over the North Pacific, and 19% in the Arctic in Table S4) is higher than the annual mean values.

Effect of Asian Dust INPs on Cloud Radiative Forcing
Sensitivity simulations were performed to investigate the effects of Asian and other dust INPs on cloud water path and CRF at the top of the atmosphere (ΔCRF) (see Section 2 and Table S2). We analyzed both the ×1 simulations (TOTx1, ASAx1, and OTHx1) and the ×10 simulations (TOTx10, ASAx10, and OTHx10) to estimate the possible ranges of ΔCRF (see Section 2). First, the ×10 simulations are used to clearly show the effects of Asian and other dust INPs. Then, the possible ranges of ΔCRF are shown by using both the ×1 and ×10 simulations. Abbreviation: INP, ice nucleating particle.  Table S2). The rectangles in the lower panels enclose the region of East Asia and the North Pacific used to generate the data given in Table 4. The global averages and the averages over the region of East Asia and the North Pacific are shown at the top right of the lower panels.

Table 3 Simulated Annual Mean Column-Integrated Number Concentrations of Total and Asian Dust INPs and Their Ratio
LWP at middle and high latitudes in the Northern Hemisphere ( Figure S12c). Asian and other dust INPs increase the mass, number, and effective radius of cloud ice in the middle troposphere at middle latitudes in the Northern Hemisphere (Figures S14a-S14c and S15a-S15c), increase high-cloud fraction at high latitudes in the Northern Hemisphere (Figures S16 and S17), and decrease the mass, number, and effective radius of cloud liquid in the lower troposphere at middle and high latitudes in the Northern Hemisphere (Figures S14d-S14f and S15d-S15f).  Figure 9f) and in the Northern Hemisphere by other dust INPs ( Figure S12f). These SW, LW, and net ΔCRFs are consistent with the simulation results for total dust INPs by Shi and Liu (2019).
In the base simulation (TOTx1), the annual mean SW, LW, and net CRFs are −53, 16, and −37 W m −2 globally and −62, 20, and −42 W m −2 in East Asia and the North Pacific (70°E-120°W and 20°N-60°N), respectively. The magnitude of the global mean CRFs is stronger (more negative) for SW CRF and weaker (less positive) for LW CRF compared to the standard CAM5 (−48.8 and 22.4 W m −2 , respectively) (Park et al., 2014). These differences are likely due to low biases of SW and LW CRFs at low latitudes (not shown).  Because simulated ΔCRF values are strongly dependent on the choice of ice nucleation parameterizations (Shi & Liu, 2019;Xie et al., 2013), we provided the possible ranges of ΔCRF by conducting the ×1 and ×10 simulations (Tables 4 and S5). The ΔCRFs of Asian and other dust INPs estimated in this study (Tables 4  and S5) differ from those simulated with and without INPs (Text S1 and Figure S2), especially at low latitudes, which suggests that ΔCRF values are also dependent on the choice of estimation methods. The large differences between the ×1 and ×10 simulation results (Tables 4 and S5) show that the response of ΔCRF to INP number concentrations may be highly nonlinear. Despite these uncertainties, our simulations show that Asian dust INPs produce an annual mean positive net ΔCRF of 0.054-0.19 W m −2 (cf. 0.092-1.0 W m −2 for dust from other regions) in East Asia and the North Pacific (Table 4). The default CAM5 underestimates LW CRF due to the horizontal heterogeneity assumption of water vapor within each grid layer in the radiation scheme (Park et al., 2014). This underestimation, as well as the biases of the SW and LW CRFs in our simulations, may lead to the low biases of ΔCRFs of Asian and other dust INPs in this study. Possible contributions of other aerosol species (e.g., biological aerosols [Hoose et al., 2010;Tobo et al., 2013] and marine organic aerosols [Burrows et al., 2013;Wilson et al., 2015]) to ice nucleation in mixed-phase clouds are not considered in this study. These aerosol species and their roles as INPs need to be considered in future modeling to improve our understanding of the effects of total INPs and dust INPs on clouds and CRF.

Summary and Conclusions
To examine the potential of Asian dust to act as INPs in mixed-phase clouds, we investigated the spatial distribution of Asian dust and its contributions to INPs and CRF during 2013-2017 by using the global aerosol-climate model, CAM5-chem/ATRAS2. We replaced the default ice nucleation scheme for mixed-phase clouds in the model with the parameterization of DeMott et al. (2015), which links INP number concentrations to ambient temperature and dust number concentrations. The model well reproduces INP number concentrations measured at Tokyo Skytree in Japan during May 2017, when Asian dust was transported to Japan, especially at temperatures between −25°C and −19°C.
Our base simulation shows that Asian dust is emitted from the Gobi and Taklimakan Deserts in East Asia and is transported to the North Pacific, North America, and the Arctic. Both the emission flux and transport range of Asian dust are largest in May. Asian dust is distributed from the surface to the upper troposphere at middle latitudes in the Northern Hemisphere. Asian dust is partly transported to low-temperature regions (between −38°C and 0°C) where INPs play an important role in the formation of mixed-phase clouds. The annual mean contribution of Asian dust to the global dust loading in this low-temperature range (13%) is higher than that in the all temperature ranges (3.4%). Asian and other dust show peak fractions between 0°C and 5°C and between 15°C and 20°C, respectively. These differences show that Asian dust is efficiently transported into low-temperature regions and therefore has high potential to act as INPs in mixed-phase clouds.
Over the Gobi Desert, Asian dust is dominant in the lower and middle troposphere, while other dust accounts for 10%-80% of total dust at temperatures between −35°C and −20°C. This result suggests that INPs from other dust (e.g., central Asian dust) may affect the radiation budget and hydrological cycle in the Gobi Desert, thus resulting in changes in Asian dust emissions (e.g., flux and frequency).
Asian dust INPs extend from Asian dust source regions to the North Pacific, North America, and the Arctic, and they are distributed between 900 and 200 hPa at middle latitudes in the Northern Hemisphere and in the middle troposphere of the Arctic. The annual mean contribution of Asian dust to total dust INPs is 15% globally, 27% in the North Pacific, and 15% in the Arctic. The global contribution of Asian dust to total dust INPs is 2.1 times higher than its contribution to global dust emissions (7.1%) and 4.4 times higher than its contribution to the global dust loading (3.4%), which highlights the importance of Asian dust to the provision of INPs in mixed-phase clouds. As this study shows that Asian dust has high potential to act as INPs, more attention should be paid to Asian dust in future modeling and observational studies. Since Asian dust emissions are expected to increase in the future (Zhang et al., 2016), the atmospheric loading of Asian dust INPs and their effects on cloud water path and CRF may increase. In contrast, dust may have to reach higher altitudes to act as INPs in a future warmer climate. Global warming may also change the spatial distribution (including cloud top height) and microphysical properties of mixed-phase clouds (Boucher et al., 2013). Further studies are necessary to evaluate the interactions of these complex effects because these interactions are important for prediction of the future influence of Asian dust on aerosol-cloud interactions.

Data Availability Statement
The MODIS AOD data are available from https://neo.sci.gsfc.nasa.gov/view.php?datasetId=MYDAL2_M_ AER_OD. The EANET data are available from https://www.eanet.asia/. The INP data at Tokyo Skytree are available from https://ads.nipr.ac.jp/data/meta/A20200727-004. Model simulation data used in this study are available from https://doi.org/10.5281/zenodo.4633727. (JPMEERF20202003) of the Environmental Restoration and Conservation Agency. We thank Natalie Mahowald (Cornell University) and Jasper Kok (University of California, Los Angeles) for providing us their dust emission data and model code, and Ryohei Misumi (National Research Institute for Earth Science and Disaster Resilience) for providing us the OPC observational data at Tokyo Skytree. The model simulations were conducted using supercomputer systems at Nagoya University and the University of Tokyo.