Prediction of Asian Dust Days Over Northern China Using the KMA-ADAM2 Model

DECEMBER 2019 H O N G E T A L . 1777

Prediction of Asian Dust Days over Northern China Using the KMA-ADAM2 Model

SEUNGKYU K. HONG AND SANG-BOOM RYOO Environmental Meteorology Research Division, National Institute of Meteorological Sciences, Seogwipo, South Korea

JINWON KIM Climate Research Division, National Institute of Meteorological Sciences, Seogwipo, South Korea

SANG-SAM LEE Environmental Meteorology Research Division, National Institute of Meteorological Sciences, Seogwipo, South Korea

(Manuscript received 9 February 2019, in ﬁnal form 7 August 2019)

ABSTRACT

This study evaluates the Korea Meteorological Administration (KMA) Asian Dust Aerosol Model 2 (ADAM2) for Asian dust events over the dust source regions in northern China during the first half of 2017. Using the observed hourly particulate matter (PM) concentration from the China Ministry of Environmental Protection (MEP) and station weather reports, we find that a threshold value of 23 PM10–PM2.5 5 400 mgm works well in defining an Asian dust event for both the MEP-observed and the ADAM2-simulated data. In northwestern China, ADAM2 underestimates the observed dust days mainly due to underestimation of dust emissions; ADAM2 overestimates the observed Asian dust days over Manchuria due to overestimation of dust emissions. Performance of ADAM2 in estimating Asian dust emissions varies quite systematically according to dominant soil types within each region. The current formulation works well for the Gobi and sand soil types, but substantially overestimates dust emissions for the loess-type soils. This suggests that the ADAM2 model errors are likely to originate from the soil-type- dependent dust emissions formulation and that the formulation for the mixed and loess-type soils needs to be recalibrated. In addition, inability to account for the concentration of fine PMs from anthropogenic sources results

in large false-alarm rates over heavily industrialized regions. Direct calculation of PM2.5 in the upcoming ADAM3 model is expected to alleviate the problems related to anthropogenic PMs in identifying Asian dust events.

1. Introduction et al. 2014; Hong et al. 2010; Kashima et al. 2016; Kwon et al. 2002; Zhang et al. 2016). Asian dusts composed of dry sands, clay, and loam, In northeastern Asia regions, Asian dust events are exert great social and economic impacts on East Asia most frequent in spring, from March to May (Chun et al. (EA). Severe Asian dust events often lead to canceled 2001; Kim 2008; Shao et al. 2013). However, changes in flights and adversely affect the manufacturing of semi- the wind field and precipitation patterns due to desert- conductors and precision instruments (Shao and Dong ification and climate change in northern China may re- 2006). Increased optical thickness by Asian dusts inhibits duce or enlarge major Asian dust source regions to alter photosynthesis, causing plant growth disorders and eco- the intensity and frequency of Asian dust events (Kim system disturbances (Huang et al. 2006; Kim et al. 2005). 2008; Kurosaki and Mikami 2003; Kurosaki et al. 2011; Previous studies also showed that Asian dusts cause a Lee and Sohn 2011; Wang et al. 2006; Zhang et al. 2003). number of adverse effects on human health (Giannadaki Forecasting and early warning of severe dust events are critical for alleviating the adverse impacts of Asian dusts

Denotes content that is immediately available upon publica- tion as open access. Publisher’s Note: This article was revised on 3 January 2020 to correct a typographical error in the second paragraph in the Con- Corresponding author: Sang-Sam Lee, [email protected] clusions section.

DOI: 10.1175/WAF-D-19-0008.1 Ó 2019 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses). Unauthenticated | Downloaded 10/02/21 03:25 AM UTC 1778 WEATHER AND FORECASTING VOLUME 34 on the economy and environment in EA. Despite its simulating Asian dust events in the Asian dust source importance, current Asian dust forecasting suffers regions in northern China. This study further exam- from large uncertainties (Lee et al. 2017; Shao et al. ines the model performance according to regions and 2003; UNEP, WMO, and UNCCD 2016). soil types. The Korea Meteorological Administration (KMA) employs the Asian Dust Aerosol Model 2 (ADAM2) (In and Park 2003; Park et al. 2010; Park and In 2003)to 2. Data and analysis methods forecast Asian dust events. The model has been used in a. ADAM2 various Asian dust studies: simulation of dry deposition of Asian dusts in EA (Park et al. 2011; Lee The ADAM2 model was developed by incorporating et al. 2005), the effects of particulate matter (PM) as- the Asian dust emissions algorithm (In and Park 2003; similation on Asian dust forecasting (Lee et al. 2013a), Park and In 2003) into the U.S. Environmental Protec- and intercomparison of Asian dust simulations using tion Agency (EPA) Community Multiscale Air Quality ADAM2 and Lagrangian models (Kim and Lee 2013). (CMAQ) model, version 4.1.7 (Byun and Schere 2006). In ADAM2 is also a current Asian dust forecasting model the model, calculations for the atmospheric chemistry and at the Mongolia National Agency for Meteorology and transports of air pollutants are carried out by the CMAQ Environmental Monitoring (Lee et al. 2013b). model and the dust generation algorithm calculates the The Sand and Dust Storm Warning Advisory and amount of dust emissions for given meteorological con- Assessment System (SDS-WAS) Asian node of the ditions and soil types. Note that the dust emissions algo- World Meteorological Organization (WMO) in- rithm in ADAM2 is not a part of any EPA CMAQ tercompares and manages Asian dust predictions versions. The meteorological fields to drive ADAM2 for at KMA, China’s Ministry of Environmental Protec- the operational Asian dust forecasting are obtained from tion (MEP), the European Centre for Medium-Range the Unified Model (UM) of the Met Office (Davies et al. Weather Forecasts (ECMWF), and the National Cen- 2005) that is the current global forecast model at KMA. ters for Environmental Prediction (NCEP). The Dust To run the KMA Asian dust forecasts, the UM forecast Model Intercomparison Project (DMIP) has inter- data are interpolated onto the ADAM2 grid using the compared the performance of several Asian dust models UM Meteorology–Chemistry Interface Processor (MCIP) including ADAM (Park and In 2003), Chemical Weather conversion system of the Aeolus and the National Center Forecasting System (CFORS; Uno et al. 2004), Model of for Atmospheric Science (Taylor 2006). The current op- Aerosol Species in the Global Atmosphere (MASINGAR; erational forecast performs 84-h forecast cycles at 6-h Tanaka and Chiba 2005), and Chinese Unified Atmo- intervals over an EA domain covered horizontally by spheric Chemistry Environment for Dust (CUACE/Dust; a 25-km resolution 340 (east–west) 3 220 (north–south) Gong and Zhang 2008; Uno et al. 2006)insimu- grid using the Lambert conformal conic projection and lating Asian dust events. Previous studies compared vertically by irregularly spaced 47 layers. Asian dust simulations against those from the Global The Asian dust algorithm in ADAM2 calculates par- Telecommunications System (GTS) synoptic obser- ticulates in 11 particle size bins. The particle size dis- vatories or PM observations at limited observatories tribution is adjusted by a weighting function based on (Zhou et al. 2008). However, most of the early Asian the friction velocity using the logarithmic minimum and dust model evaluation was performed in the regions maximum variance distributions (Park and Lee 2004). A downstream of the source regions where dust storms af- basic framework for the Asian dust algorithm was de- fect large populations; evaluation of Asian dust forecast veloped by studying the critical friction of the Asian dust models against observations in the Asian dust source re- phenomenon in the EA region (In and Park 2002). The gions is not available in previously published studies. Asian dust algorithm in ADAM2 has been improved by This study aims to evaluate ADAM2, the current considering the reduction factor to account for the sea- KMA Asian dust forecast model, using the hourly sonal variations in the vegetation cover (Park et al. 2010).

PM10 and PM2.5 observations from MEP for the dust The Asian dust source regions are largely limited to source region within 358–508Nand758–1308E dur- northern China, Mongolia, northern India, and central ing the ﬁrst half of 2017. This study ﬁrst establishes Asia (Fig. 1). Details on the Asian dust source regions a threshold value for identifying an Asian dust day and the formulations for dust emission used in the for the PM values from the MEP observations and model are described in Park et al. (2010), and will not ADAM2 simulations. The threshold value is sub- be elaborated in this paper. The colored areas indicate sequently used to analyze the observed and simulated soil types used in ADAM2; Gobi (black), sand (yellow), dust days to examine the model performance in loess (red), mixed (blue), and Tibetan (green). Different

Unauthenticated | Downloaded 10/02/21 03:25 AM UTC DECEMBER 2019 H O N G E T A L . 1779

FIG. 1. Surface soil over the dust source region in ADAM2 categorized as Gobi (black), sand (yellow), loess (red), mixed (blue), and Tibetan (green). critical values for wind speeds, humidity, and surface codes 7 and 8 are selected as Asian dust events. Of the temperature are assigned according to soil types. total 19 958 weather events reported in terms of the codes, 651 are codes 7 and 8 (3.26%) with high values b. PM-based Asian dust threshold value occurring in and near the Gobi Desert. Using only the To determine the threshold for determining an codes 7 and 8 to define Asian dust days is likely to Asian dust event, we first examine the ratio of Asian underestimate the total Asian dust days because these dust events to the total weather events. Figure 2 shows two weather codes represent only the events related the percentages of the Asian dust events in the GTS to local sources due to updrafts. Figure 2b shows the synoptic observations over the region 358–508Nand percentage of the Asian dust days determined by con- 758–1308E during the first half of 2017. Figure 2a shows sidering the codes 6, 7, 8, 9, 30, 31, 32, 33, 34, 35, and 98 as the ratio of the Asian dust events when the weather Asian dust events. Unlike in the previous estimate based

FIG. 2. (a) Ratio of blowing dust phenomena (weather codes 5 7 and 8) and (b) ratio of all dust phenomena (weather codes 5 6, 7, 8, 9, 30, 31, 32, 33, 34, 35, and 98) of the GTS stations in northern China in the ﬁrst half of 2017.

Unauthenticated | Downloaded 10/02/21 03:25 AM UTC 1780 WEATHER AND FORECASTING VOLUME 34

FIG. 3. (a) Cumulative relative frequency of PM10–PM2.5 in the ADAM2 simulation depicted as a solid red line and that in the MEP observations depicted as a solid blue line in the ﬁrst half of 2017 over the dust source region. (b) Boxplots

of PM10 in the dust case and nondust case in the ADAM2 simulation (Model) and those in the MEP observations (OBS). Bars from the top represent 95%, 90%, 50%, 10%, and 5% of the total data, and the dot point is the average value. solely on the codes of 7 and 8 (Fig. 2a), these cases in- The threshold value obtained here is larger than PM10 5 2 clude not only the Asian dust events from local origins 200 mgm 3 used in Wang et al. (2008) for identifying 23 but also those due to dusts transported into the re- Asian dust days, but is in line with the PM10 5 400 mgm gion. These cases occupy 1665 out of the 19 958 events traditionally used to identify Asian dust days (Lee (8.34%)withhighratiosinandneartheGobiand et al. 2012). Figure 3b shows a box-and-whisker plot

Taklamakan Deserts. Based on these observation-based of the PM10 concentrations for the Asian dust days and estimates, we assume that Asian dust events represent 5% non-Asian dust days from the MEP data and the ADAM2 of the entire weather types in the dust emission region. simulations determined by applying PM10–PM2.5 5 2 To determine the threshold PM values for identify- 400 mgm 3 as the threshold value. Using the threshold ing the Asian dust events, the cumulative distribution value, PM10 concentrations for the Asian dust days are function (CDF) of the MEP-observed daily coarse par- clearly separated from those for the non-Asian days in ticle concentration values (PM10–PM2.5) are compared both observations and simulations. against the collocated ADAM2-simulated values within the dust source region, 358–508N and 758–1308E, in the 3. Analysis and discussion ﬁrst half of 2017 (Fig. 3a). Compared to using the PM10 values, using the PM10–PM2.5 value as the threshold The Asian dust day threshold value obtained in for identifying the Asian dust days helps to reduce the section 2 is used to analyze the Asian dust events in the possibility of mistakenly identifying haze-dominated observed and simulated data. There are 355 MEP data days as Asian dust days; haze is mainly affected by ﬁne points in the Asian dust source region located within

PMs of anthropogenic origins that contribute mostly to 358–508Nand758–1308E. When the maximum PM10–PM2.5 PM2.5 values. The PM10–PM2.5 values in Fig. 3a are se- value satisﬁes the Asian dust threshold value even lected for the days on which observations were taken for for a single hour, the day is considered as an Asian 23 more than 12 h or PM10 values exceeded 200 mgm for dust day. Figure 4a shows the MEP-observed Asian over 3 h. The 95th percentile values of the observed CDF dust days. The highest Asian dust frequencies are of PM10–PM2.5 (i.e., the upper 5% of the total weather observed in northwestern China near the border with events that are regarded as Asian dust days in the weather- Kyrgyzstan, Tajikistan, and Pakistan, as well as in Inner 2 code-based Asian dust days, is 369 mgm 3); the corre- Mongolia around the Badain Juran and Tengger 2 sponding value in ADAM2 simulations is 385 mgm 3. Deserts. Asian dust days are also frequent in the vi- 23 Based on this analysis, we adopt PM10–PM2.5 5 400 mgm cinity of the four provinces around the Huabei Plain. as the threshold for identifying Asian dust days for The PM2.5/PM10 value is known to decrease to below 0.4 both observed and simulated PM data for simplicity. because large size particles are dominant in Asian dust

Unauthenticated | Downloaded 10/02/21 03:25 AM UTC DECEMBER 2019 H O N G E T A L . 1781

FIG. 4. Dust days in the ﬁrst half of 2017 derived from (a) MEP observations and (b) the ADAM2 simulation. (c) Average PM2.5/PM10 ratio in the ﬁrst half of 2017 obtained from MEP observations during dust events. events (Claiborn et al. 2000). Figure 4c shows the ratio observation, the simulation shows (Fig. 4b) a large number

PM2.5/PM10 (%) when an Asian dust day was identi- of Asian dust days in the Badain Juran and Tengger ﬁed in the MEP data. The PM2.5/PM10 values are in the Deserts in Inner Mongolia; however, the number of four provinces near the Huabei Plain and northeastern Asian dust days for northwestern China is well below China are generally larger than in other regions due to the the MEP observations. The simulation overestimates the effects of anthropogenic aerosols from industrial sources. observed Asian dust days in northeastern China. Although anthropogenic aerosols are screened using the Figure 5 evaluates the accuracy index of the simu-

PM10–PM2.5 threshold, as shown in Plaza et al. (2011), lated Asian dust days against the observed data using a nitrates, one of the major anthropogenic aerosols, some- contingency table constructed as follows: When a day is times occur in the range 3.2–5.6 mm, leading to overesti- identiﬁed as an Asian dust day in both the observation mation of Asian dust days for the heavily industrialized and the simulation, it is counted toward the hit (H); if an four provinces in the Huabei Plain. Similar to the observed Asian dust day is not an Asian dust day in the

FIG. 5. Hit rate, threat score, probability of detection, and false-alarm ratio of the dust simulation in ADAM2 compared with those in the MEP observations in the ﬁrst half of 2017.

Unauthenticated | Downloaded 10/02/21 03:25 AM UTC 1782 WEATHER AND FORECASTING VOLUME 34

FIG. 6. Dust day differences (ADAM2 2 MEP) according to the average PM2.5/PM10 ratio of the observations at each site in dust day cases over each region from A to D. Bars from the top represent 95%, 90%, 50%, 10%, and 5% of the total data, and the dot point represents the average value. simulation, it is a miss (M); when a non-Asian dust day in The TS values are high in Badain Juran and Tengger the observations is an Asian dust day in the simulations, it Desert in Inner Mongolia (region B) and in Manchuria is a false alarm (FA); when both the observations and (region D). The TS values are relatively low in the simulations indicate a non-Asian dust day, it is a correct four provinces around the Huabei Plain (region C). rejection (CR). For a total number of forecasts n, n 5 H 1 Similar to HR, the TS values are low in northwestern China M 1 FA 1 CR, the hit rate (HR) is HR 5 (H 1 CR)/n,the (region A). The POD values are high in Inner Mongolia threat score (TS) is TS 5 H/(H 1 M 1 FA), the prob- (region B) and Manchuria (region D) and are relatively low ability of detection (POD) is POD 5 H/(H 1 M), and in the four provinces around the Huabei Plain (region C). the false-alarm rate (FAR) is FAR 5 FA/(H 1 FA). To POD values are low in northwestern China (region A). examine the ADAM2 performance in simulating Asian FAR values are large in region D. The POD values are also dust events according to regions, the dust source re- high in region D, implying that ADAM2 substantially gion in northern China is divided into four regions overestimates Asian dust days in the region. (Fig. 5): the northwestern China region (region A), the To examine the effect of anthropogenic aerosols in Inner Mongolia region (region B); the four provinces determining Asian dust days in the four regions, the near the Huabei Plain (region C), and the Manchuria variations of the model errors according to PM2.5/PM10, region (region D). Note that region C and region D an approximate measure of the portion of anthropo- include heavily industrialized areas and thus the aero- genic aerosols within PM10, on the observed Asian dust sol concentration in these regions is strongly affected days are summarized in Fig. 6. As the PM2.5/PM10 ratio by anthropogenic sources. The HR values are relatively increases, the simulation overestimates the Asian dust high except for northwestern China (region A); that is, days; this is common in all four regions. This may be ADAM2 substantially underperformed in detecting the due to the combined effects of the model overestimating Asian dust days in region A compared to other regions. the downstream transport of Asian dusts and the

Unauthenticated | Downloaded 10/02/21 03:25 AM UTC DECEMBER 2019 H O N G E T A L . 1783

TABLE 1. Accuracy indices at all MEP sites over regions A, B, C, and D in the ADAM2 simulation compared with those in the MEP observations in the ﬁrst half of 2017 (bold indicates the highest value among the indices). HR 5 (H 1 CR)/(H 1 M 1 FA 1 CR) 3 100; TS 5 H/(H 1 M 1 FA) 3 100; POD 5 H/(H 1 M) 3 100; and FAR 5 FA/(H 1 FA) 3 100.

Region HMFA CR HR (%) TS (%) POD (%) FAR (%) All 1014 1151 1782 54 012 94.3 23.3 39.5 63.7 A 30 519 12 3900 88.1 5.4 5.5 28.6 B 455 582 454 14 756 93.6 30.5 43.9 50.1 C 160 253 434 19 258 96.6 18.9 38.7 73.1 D 369 197 882 16 098 93.9 25.5 65.2 70.5 observation overestimating Asian dust days by in- anthropogenic aerosol emission in region C yields large accurately counting anthropogenic aerosols toward FAR (73.1%) and small POD. large mineral dust particles. This also implies that the To further examine the relationship between soil threshold for identifying Asian dust days does not types and the regional variations of model performance, work well when anthropogenic particulate concentration the ratio of the soil types in each region is examined is very high as often occurring in heavily industrialized re- (Fig. 7). In the study area, mixed soil and loess are the gions. Regionally, the model tends to underestimate most common soil types. In the Tibetan Plateau area, Asian dust days in northwestern China (region A), but there are no observation stations, so it is difﬁcult to de- the PM2.5/PM10 values in region A are small, indicating termine whether Asian dust was present. Thus, we exclude that most of the Asian dusts in the region are large the Tibetan soil type in discussion. In the northwestern re- particles of natural origins. The PM2.5/PM10 values are gion (region A), sand is most dominant, and mixed and Gobi small in Inner Mongolia (region B), and the model are most dominant in Inner Mongolia (region B). Mixed error in the number of Asian dust days for region B soil and loess are present in region C. Last, Manchuria is smaller than that in the Taklamakan Desert. Major (region D) shows a pattern similar to that in the Huabei Plain simulation errors occur in the four provinces near the (region C), but the sand type is also present in region D. Huabei Plain (region C). This indicates using the Asian- To determine the accuracy index of the model accord- dust detecting threshold value based on PM10–PM2.5 may ing to soil types, an analysis similar to that in Table 1 is overestimate Asian dust days in the regions of heavy performed for each soil type (Table 2) in all regions ex- anthropogenic aerosols. To examine the possibility of cept for northwestern China (region A) where the model erroneously overestimating Asian dust days due to heavy performance is problematic. There are only a few HR concentrations of anthropogenic particles, we examine the average PM2.5/PM10 values in each region. Compared with the entire distribution, high PM2.5/PM10 ratios of 0.4–0.45 are mostly found in the regions C and D that include heavily industrialized areas. As the PM2.5/PM10 value increases in region D, the number of Asian dust days in the simulation is slightly larger than in the observation; this also contributes to the tendency of the model to overestimate the Asian dust events for the region. The accuracy indices in each region are summarized in Table 1 with the corresponding formulations. The highest HR value of 96.6% appears in the four provinces near the Huabei Plain (region C), and the lowest value is 88.1% in northwestern China (region A). The TS value is highest in Inner Mongolia (region B) (30.5%), fol- lowed by Manchuria (region D) (25.5%). This yields a high POD value (65.2%) for region D, but FAR is also large (70.5%). FAR is 50.0% in Inner Mongolia (region B), comparable to that in region D, leading to the largest TS. In northwestern China (region A), the simulation substantially underestimates Asian dust days to result in FIG. 7. Surface soil over regions A, B, C, and D categorized as the smallest POD value (5.5%); the smallest FAR also Gobi (black), sand (yellow), loess (red), mixed (blue), and Tibetan occurs (28.6%) in the region. The large amount of (green) soil types.

Unauthenticated | Downloaded 10/02/21 03:25 AM UTC 1784 WEATHER AND FORECASTING VOLUME 34

TABLE 2. Accuracy indices at all MEP sites except for northwestern China with Gobi, sand, loess, and mixed surface soil types in the ADAM2 simulation compared with those in the MEP observations in the ﬁrst half of 2017 (bold indicates the highest value among the indexes).

Soil type HMFA CR HR (%) TS (%) POD (%) FAR (%) All 984 1032 1770 50 112 94.8 26 48.8 64.3 Gobi 71 120 54 1377 89.3 29 37.2 43.2 Sand 74 72 43 1601 93.6 39.2 50.7 36.8 Loess 398 314 1018 26 253 95.2 23 55.9 71.9 Mixed 441 526 655 20 881 94.8 27.2 45.6 59.8 values in the Gobi area because of a small number of types. The systematic model errors according to soil types samples, but the Gobi area shows the second largest appear to be the main cause of the regional variations of TS value (29.0%). The largest TS (39.2%) appears for model errors as the model performed poorly in the regions the sand type. In conjunction with very small FAR of the mixed and loess soil types.

(36.8%), this indicates that the model simulates Asian The observed and simulated PM10 concentration distri- dust events well for the sand type soil. Thus, ADAM2 butions are analyzed according to the H, M,FA,andCR is found to work well for the Gobi and sand types. For values in the three regions except for region A (Fig. 9). In the loess type, the FAR value is largest (71.9%), show- Inner Mongolia (region B), the simulated PM10 con- ing that the area in which Asian dusts occur excessively centrations agree closely for H and CR although the in the current soil classiﬁcation system is the loess area. high-concentration events are underestimated. For M Considering the overestimation of Asian dusts in region and FA, the model underestimates and overestimates

D, this result means that Asian dusts that originating in the observed PM10 concentrations, respectively. For the the regions of loess and mixed soil types (the ADAM2 Huabei Plain (region C), the model simulations notably soil types in region D) need to be reduced. underestimate the observed mean and high percentile of

Figure 8 compares the simulated number of Asian PM10 values for the H and CR cases. The exact cause for dust days to that in the MEP observations according this model bias is not clear, however, the lack of con- to soil types and regions. The number of the simulated sideration of the dusts generated in Mongolia and sub- Asian dust days agrees closely with that of the obser- sequently transported into the region as well as the dusts vation for the Gobi and sand types. Unlike for the mixed from agricultural activities in the simulation may have and loess soil type which show large spread, the model contributed to the underestimation. Because of high con- tends to overestimate the number of Asian dust days for centrations of anthropogenic PMs in the four provinces the loess type. As shown in Fig. 8b, the model over- around the Huabei Plain (region C), the M and FA cases estimates the observed number of Asian dust days in in the region may be due to the inability of ADAM2 in region D. This means that, in the model, Asian dust simulating anthropogenic PMs. Unlike for other regions, emission needs to be modiﬁed for the mixed and loess soil the simulation overestimates the observed PM10 values in

FIG. 8. Scatterplots of dust days in the ﬁrst half of 2017 in the ADAM2 simulation and MEP observations (a) as Gobi (black), sand (blue), loess (red), and mixed (green) surface soil types (except for region A) and (b) over region B (blue), region C (green), and region D (red).

Unauthenticated | Downloaded 10/02/21 03:25 AM UTC DECEMBER 2019 H O N G E T A L . 1785

FIG. 9. Boxplots of PM10 of MEP (black) and ADAM2 (red) in case of a hit, miss, false alarm (FA), or correct rejection (CR) over each region. Bars from the top represent 95%, 90%, 50%, 10%, and 5% of the total data, and the dot point represents the average value. the H cases for Manchuria (region D); the mean and the poor performance for the mixed and loess-type soil. The upper/lower limit of the PM10 concentration in the CR regional variations in ADAM2 performance appear to cases are underestimated for this region like for other be related to the dependence of the model performance regions. This may be the consequence of the soil-type- according to soil types. For example, the large FAR value dependent ADAM2 model errors characterized by over- in the Manchuria appears to result from overestimation estimation and underestimation of dust emission over of dust emissions in the source algorithm. Another source loess/mixed soil types and Gobi/sand types, respectively. of the ADAM2 model errors may be due to inadequate handling of anthropogenic PMs, most seriously for the

heavily industrialized Huabei Plain in which large PM2.5 4. Conclusions concentrations of anthropogenic PMs can yield large Using the hourly PM observations and ADAM2 PM10–PM2.5 values to meet the Asian dust day threshold simulations in the first half of 2017, we have developed a when the concentration of large dust particles is small. threshold value for identifying Asian dust days from the This problem will be alleviated in the upcoming ADAM3 observed as well as the ADAM2-simulated PM data. The model that simulates anthropogenic PMs. threshold value in turn is applied to evaluate the fidelity of As alterations in winds, vapor transports, and pre- ADAM2 in simulating Asian dust days in the dust source cipitation are expected due to future climate change, the regions in northern China. Based on the Asian dust events proposed threshold value in the current Asian dust al- reported at the GTS stations, we have estimated that 5% of gorithm may not work well in future Asian dust fore- all weather events correspond to Asian dust days in the casting. Thus, in addition to the model improvements source region. Based on the weather-code deduced Asian listed above, it is necessary to regularly verify the threshold 23 value for identifying an Asian dust forecast day. Contin- dust days, the daily maximum PM10–PM2.5 5 400 mgm is adopted as a threshold value for identifying an Asian ued evaluations and verifications of Asian dust simulations dust day in the ADAM2 simulations and the MEP obser- are important for improving Asian dust forecasts. vations. The model performance in identifying Asian dust days are evaluated based on the accuracy index according Acknowledgments. This work was funded by the Korea to regions and soil types based on the threshold value. Meteorological Administration Research and Development The ADAM2 model notably underestimates the Asian Program ‘‘Development of Asian Dust and Haze Monitor- dust days in northwestern China and overestimates in ing and Prediction Technology’’ under Grant 1365003013. Manchuria. Both the probability of detection and false- alarm rate are high in the Manchuria, resulting in over- REFERENCES estimation of Asian dust days. In terms of the threat score, the model fidelity for the number of Asian dust Byun, D., and K. L. Schere, 2006: Review of the governing days is highest in Inner Mongolia. The predictability in- equations, computational algorithms, and other components dex for the number of Asian dust days according to soil of the Models-3 Community Multiscale Air Quality (CMAQ) modeling system. Appl. Mech. Rev., 59, 51–77, https://doi.org/ types shows that the model performance in identifying 10.1115/1.2128636. Asian dust days varies quite systematically according to Chun, Y., K. O. Boo, J. Kim, S. U. Park, and M. Lee, 2001: Synopsis, soil types: good performance for the sand-type soil and transport, and physical characteristics of Asian dust in

Unauthenticated | Downloaded 10/02/21 03:25 AM UTC 1786 WEATHER AND FORECASTING VOLUME 34

Korea. J. Geophys. Res., 106, 18 461–18 469, https://doi.org/ its association with climate and surface condition change. 10.1029/2001JD900184. Atmos. Environ., 45, 4611–4616, https://doi.org/10.1016/ Claiborn, C. S., D. Finn, T. V. Larson, and J. Q. Koenig, 2000: j.atmosenv.2011.05.065. Windblown dust contributes to high PM25 concentrations. ——,S.Kim,J.C.Ha,andY.Chun,2012:Performance J. Air Waste Manag. Assoc., 50, 1440–1445, https://doi.org/ analysis of simulation of Asian dust observed in 2010 by 10.1080/10473289.2000.10464179. the all-season dust forecasting model, UM-ADAM2 (in Davies, T., M. J. Cullen, A. J. Malcolm, M. H. Mawson, Korean). Atmosphere, 22, 245–257, https://doi.org/10.14191/ A. Staniforth, A. A. White, and N. Wood, 2005: A new dy- Atmos.2012.22.2.245. namical core for the Met Office’s global and regional mod- ——, J. C. Ha, S. S. Lee, and Y. Chun, 2013a: PM 10 data as- elling of the atmosphere. Quart. J. Roy. Meteor. Soc., 131, similation over South Korea to Asian dust forecasting 1759–1782, https://doi.org/10.1256/qj.04.101. model with the optimal interpolation method. Asia-Pac. Giannadaki, D., A. Pozzer, and J. Lelieveld, 2014: Modeled global J. Atmos. Sci., 49, 73–85, https://doi.org/10.1007/s13143- effects of airborne desert dust on air quality and premature 013-0009-y. mortality. Atmos. Chem. Phys., 14, 957–968, https://doi.org/ ——, E. Munkhtsetseg, S. B. Kim, J. C. Ha, S. S. Lee, and 10.5194/acp-14-957-2014. Y. Chun, 2013b: Numerical simulation and evaluation of Gong, S. L., and X. Y. Zhang, 2008: CUACE/Dust-an integrated Asian dust events observed in Mongolia in spring 2011. system of observation and modeling systems for operational Asia-Pac.J.Atmos.Sci., 49, 111–120, https://doi.org/10.1007/ dust forecasting in Asia. Atmos. Chem. Phys., 8, 2333–2340, s13143-013-0012-3. https://doi.org/10.5194/acp-8-2333-2008. Lee, S. S., Y. Chun, J. C. Nam, S. U. Park, and E. H. Lee, 2005: Hong, Y. C., and Coauthors, 2010: Asian dust storm and pulmonary Estimation of dry deposition during Asian dust events in function of school children in Seoul. Sci. Total Environ., 408, spring of 2002. J. Meteor. Soc. Japan, 83A, 241–254, https:// 754–759, https://doi.org/10.1016/j.scitotenv.2009.11.015. doi.org/10.2151/jmsj.83A.241. Huang, J., P. Minnis, B. Lin, T. Wang, Y. Yi, Y. Hu, S. Sun-Mack, ——, E. H. Lee, B. J. Sohn, H. C. Lee, J. H. Cho, and S. B. Ryoo, and K. Ayers, 2006: Possible influences of Asian dust aero- 2017: Improved dust forecast by assimilating MODIS sols on cloud properties and radiative forcing observed from IR-based nighttime AOT in the ADAM2 Model. SOLA, MODIS and CERES. Geophys. Res. Lett., 33, L06824, https:// 13, 192–198, https://doi.org/10.2151/sola.2017-035. doi.org/10.1029/2005GL024724. Park, S. U., and H. J. In, 2003: Parameterization of dust emission In,H.J.,andS.U.Park,2002:Asimulationoflong-range for the simulation of the yellow sand (Asian dust) event ob- transport of Yellow Sand observed in April 1998 in Korea. served in March 2002 in Korea. J. Geophys. Res., 108, 4618, Atmos. Environ., 36, 4173–4187, https://doi.org/10.1016/ https://doi.org/10.1029/2003JD003484. S1352-2310(02)00361-8. ——, and E. H. Lee, 2004: Parameterization of Asian dust ——, and ——, 2003: The soil particle size dependent emission (Hwangsa) particle-size distributions for use in dust emission parameterization for an Asian dust (Yellow Sand) observed in models. Atmos. Environ., 38, 2155–2162, https://doi.org/10.1016/ Korea in April 2002. Atmos. Environ., 37, 4625–4636, https:// j.atmosenv.2004.01.024. doi.org/10.1016/j.atmosenv.2003.07.009. ——, A. Choe, E. H. Lee, M. S. Park, and X. Song, 2010: The Kashima, S., T. Yorifuji, S. Bae, Y. Honda, Y. H. Lim, and Y. C. Asian dust aerosol model 2 (ADAM2) with the use Hong, 2016: Asian dust effect on cause-specific mortality in of normalized difference vegetation index (NDVI) ob- five cities across South Korea and Japan. Atmos. Environ., tained from the Spot4/vegetation data. Theor. Appl. 128, 20–27, https://doi.org/10.1016/j.atmosenv.2015.12.063. Climatol., 101, 191–208, https://doi.org/10.1007/s00704- Kim, J., 2008: Transport routes and source regions of Asian dust 009-0244-4. observed in Korea during the past 40 years (1965;2004). ——, ——, and M. S. Park, 2011: Asian dust depositions over the Atmos. Environ., 42, 4778–4789, https://doi.org/10.1016/ Asian region during March 2010 estimated by ADAM2. Theor. j.atmosenv.2008.01.040. Appl. Climatol., 105, 129–142, https://doi.org/10.1007/s00704- Kim, C. H., and H. J. Lee, 2013: Numerical simulations of Asian 010-0380-x. dust events: A Lagrangian Dust Model and its applications. Plaza, J., M. Pujadas, F. J. Gomez-Moreno, M. Sanchez, and Asia-Pac. J. Atmos. Sci., 49, 571–586, https://doi.org/10.1007/ B. Artinano, 2011: Mass size distributions of soluble s13143-013-0051-9. sulfate, nitrate and ammonium in the Madrid urban aerosol. Kim, S. W., and Coauthors, 2005: Aerosol optical, chemical and Atmos. Environ., 45, 4966–4976, https://doi.org/10.1016/ physical properties at Gosan, Korea during Asian dust and j.atmosenv.2011.05.075. pollution episodes in 2001. Atmos. Environ., 39, 39–50, https:// Shao, Y., and C. H. Dong, 2006: A review on East Asian dust storm doi.org/10.1016/j.atmosenv.2004.09.056. climate, modelling and monitoring. Global Planet. Change, 52, Kurosaki, Y., and M. Mikami, 2003: Recent frequent dust events 1–22, https://doi.org/10.1016/j.gloplacha.2006.02.011. and their relation to surface wind in East Asia. Geophys. Res. ——, and Coauthors, 2003: Northeast Asian dust storms: Real-time Lett., 30, 1736, https://doi.org/10.1029/2003GL017261. numerical prediction and validation. J. Geophys. Res., 108, ——, M. Shinoda, and M. Mikami, 2011: What caused a recent 4691, https://doi.org/10.1029/2003JD003667. increase in dust outbreaks over East Asia? Geophys. Res. ——, M. Klose, and K. H. Wyrwoll, 2013: Recent global dust Lett., 38, L11702, https://doi.org/10.1029/2011GL047494. trend and connections to climate forcing. J. Geophys. Kwon, H. J., S. H. Cho, Y. Chun, F. Lagarde, and G. Pershagen, Res. Atmos., 118, 11 107–11 118, https://doi.org/10.1002/ 2002: Effects of the Asian dust events on daily mortality in jgrd.50836. Seoul, Korea. Environ. Res., 90, 1–5, https://doi.org/10.1006/ Tanaka, T. Y., and M. Chiba, 2005: Global simulation of dust enrs.2002.4377. aerosol with a chemical transport model, MASINGAR. Lee, E. H., and B. J. Sohn, 2011: Recent increasing trend in dust J. Meteor. Soc. Japan, 83, 255–278, https://doi.org/10.2151/ frequency over Mongolia and Inner Mongolia regions and jmsj.83a.255.

Unauthenticated | Downloaded 10/02/21 03:25 AM UTC DECEMBER 2019 H O N G E T A L . 1787

Taylor, G. J., 2006: Instruction manual for the UM-MCIP meteo- Wang, Y. Q., X. Y. Zhang, S. L. Gong, C. H. Zhou, X. Q. Hu, rological pre-processor. Hatfield, University of Hertfordshire, H. L. Liu, T. Niu, and Y. Q. Yang, 2008: Surface observation 123 pp. of sand and dust storm in East Asia and its application in UNEP, WMO, and UNCCD, 2016: Global Assessment of Sand CUACE/Dust. Atmos. Chem. Phys., 8, 545–553, https://doi.org/ and Dust Storms. United Nations Environment Programme, 10.5194/acp-8-545-2008. Nairobi, 16 pp. Zhang, X. Y., S. L. Gong, T. L. Zhao, R. Arimoto, Y. Q. Wang, and Uno, I., and Coauthors, 2004: Numerical study of Asian dust trans- Z. J. Zhou, 2003: Sources of Asian dust and role of climate port during the springtime of 2001 simulated with the Chemical change versus desertification in Asian dust emission. Geophys. Weather Forecasting System (CFORS) model. J. Geophys. Res. Lett., 30, 2272, https://doi.org/10.1029/2003GL018206. Res., 109,D19S24,https://doi.org/10.1029/2003JD004222. Zhang, X., L. Zhao, D. Tong, G. Wu, M. Dan, and B. Teng, 2016: A ——, and Coauthors, 2006: Dust model intercomparison (DMIP) systematic review of global desert dust and associated human study over Asia: Overview. J. Geophys. Res., 111, D12213, health effects. Atmosphere, 7, 158, https://doi.org/10.3390/ https://doi.org/10.1029/2005JD006575. atmos7120158. Wang, X., F. Chen, and Z. Dong, 2006: The relative role of cli- Zhou, C. H., and Coauthors, 2008: Development and evaluation matic and human factors in desertification in semiarid China. of an operational SDS forecasting system for East Asia: Global Environ. Change, 16, 48–57, https://doi.org/10.1016/ CUACE/Dust. Atmos. Chem. Phys., 8, 787–798, https://doi.org/ j.gloenvcha.2005.06.006. 10.5194/acp-8-787-2008.

Unauthenticated | Downloaded 10/02/21 03:25 AM UTC