Interannual changes of land surface conditions in Asian dust source regions since 2000

The Taklimakan and Gobi deserts and the Loess Plateau in China and Mongolia are generally recognized source areas for Asian dust. However, dust emissions depend on meteorological factors such as air pressure and land surface conditions, and precise information on land surface conditions in the dust source areas and the frequency of dust events in Japan is lacking. In this study, interannual changes of land surface conditions in the springtime since 2000 were examined in a target source region ( 35 ° N - 50 ° N, 100 ° E - 120 ° E ) . Back trajectory analysis results showed that most dust trajectories of the past 10 years mainly followed three routes passing over this target region. Both the number of Asian dust events observed in Japan and the area with a threshold wind speed U t of <10 m s -1 in the target region significantly decreased after 2000. Further, the area of U t < 10 m s -1 and the number of events were significantly correlated. These results may reflect a decrease in the bare land surface area, which is associated with dust outbreaks.


Introduction
Asian dust is mineral dust that is emitted from arid regions in China and Mongolia and transported toward and often deposited in remote leeward regions Tian et al., 2007;Onishi et al., 2012;Kawai et al., 2021 . In Japan, Asian dust degrades visibility and contaminates and soils household items, including laundry and cars. Moreover, dust events can aggravate human health problems, including asthma, allergic symptoms, and contact dermatitis Watanabe et al., 2011;Otani et al., 2011;Higashi et al., 2014 . For these reasons, dust monitoring and prediction are conducted in Japan by numerical simulation, remote sensing, and weather forecasting to mitigate damage caused by Asian dust MOE, 2005 . Increased desertification land degradation in drylands and droughts in arid regions of northeast Asia are important factors contributing to increases in the number of Asian dust events observed in Japan ADE Igarashi et al., 2011;Kurosaki et al., 2011 . Thus, to improve the accuracy of numerical dust emission models it is important to clarify the effect of land surface conditions on dust occurrence Kawai et al., 2021 . For example, the dust outbreak frequency tends to decrease as the vegetation cover, as indicated by the normalized difference vegetation index NDVI , increases Kimura et al., 2009;Kimura and Shinoda, 2010 , and dust emission modeling results are improved when vegetation is taken into account Kang et al., 2014 . A method for assessing land surface conditions in the dust source area is needed to obtain precise information on dust outbreaks on the Asian continent.
In this study, a method for assessing land surface conditions is proposed based on the distribution of threshold wind speeds determined by using satellite data Kimura, 2016 . Then, based on the relationship between the number of ADE in springtime and land surface conditions since 2000, a simple equation to predict the number of ADE from the area where the threshold wind speed is < 10 m s -1 in the dust source region is introduced.

Target region
Kimura 2012a, b defined a target region for assessing land surface conditions for Asian dust emission as the region within 35 N -45 N and 100 E -115 E, which includes the Gobi Desert and the Loess Plateau, major dust source regions on the Asian continent Sun et al., 2001;Lim and Chun, 2006;Shao andDong, 2006 . Kimura 2012a, b showed that the increasing trend in vegetation coverage decreasing trend in the bare surface area from 2000 to 2011 in this target region corresponded to a decreasing trend in dust storm frequency and ADE. In addition, Tian et al. 2007 , who examined the relationship between ADE and dust storm frequency in northern China, reported that dust storm frequency in Inner Mongolia and the number of ADE varied mostly in parallel from 1967 to 2006. Therefore, in this study, the target region was expanded to the region within 35 N -50 N and 100 E -120 E 2,731,268 km 2 , which includes northeastern Mongolia, Inner Mongolia, and northeast China as well as the Gobi Desert and the Loess Plateau Fig. 1 . This target region approximately coincides with the CN region northeastern Asian continent defined by the Japan Meteorological Agency JMA for back trajectory analyses https://ds.data.jma.go.jp/ghg/kanshi/trajectory/info_trajectory.html and the major dust outbreak region 40 N -50 N, 100 E -120 E defined by the Japanese Ministry of the Environment MOE, 2018 .
To verify that the air masses transporting Asian dust passed over this target region in these past ten years 2012 -2021 , the NOAA HYSPLIT model Stein et al., 2015;Rolph et al., 2017 https://www.ready.noaa.gov/HYSPLIT_traj.php was used to calculate back trajectories for 96 hours 4 days from 500 m altitude over Fukuoka City 33.59 N, 130.40 E MOE, 2018 on dates when ADE were observed. Fukuoka was selected as the end point because on average 60 of ADE observed annually since 2000 affected Fukuoka.

Assessing land surface conditions in the target region
and ADE According to an analysis using weather data, strong wind significantly affects dust events in northeast Asia Lim and Chun, 2006;Shao and Dong, 2006 . However, the driving force of the strong wind varies among dust source regions because of differences in the land surface characteristics. Thus, consideration of surface conditions, such as vegetation and snow cover, which have an important effect on the dust events is also required. Firstly, Kimura 2012a, b examined simply the relationship between ADE and land surface conditions as indicated by the NDVI-based vegetation cover. Later, Kimura 2016 developed a satellite-based method for mapping dust erodibility in the area within 35 N -50 N and 75 E -120 E. In this method, based on the threshold wind speed for dust emission U t at 10 m height the region is categorized into severe bare land , moderate sparse vegetation , and nearly no dust hazard vegetation reducing dust outbreaks, snow cover, wet surface, and frozen soil areas Table 1 .
In this study, the areal coverage of each land surface type was mapped and annual changes in their coverages were compared to changes in the number of springtime ADE since 2000. The mapping system, which uses Moderate Resolution Imaging Spectroradiometer MODIS satellite data on snow cover, area of frozen soil, land surface wetness, and vegetation cover can be summarized as follows see Kimura, 2016, for details of the calculation process : First, if snow cover and frozen soil are absent, dust events are assumed to be possible. Second, if the land surface wetness determined from the Satellite-based Aridity Index SbAI Kimura and Moriyama, 2014 exceeds a threshold value = 0.03 , dust events are assumed to be possible. Finally, U t at 10 m height is calculated from the NDVI following the method of Kimura and Shinoda 2010 . Determinations of cloud and snow cover are derived from cloud and snow flag information in the satellite data. ''Frozen soil'' status is assumed when the satellite-determined land surface temperature is below 273.15 K.
For the calculations and analysis, MODIS data products MOD09A1, MOD09CMG, and MOD11C1 for spring March to May from 2000 to 2021 were used. Daily satellite data were downloaded from the Land Processes Distributed Active Archive Center https://earthdata.nasa.gov/eosdis/daacs/lpdaac , and the dust erodibility map was updated in semi-real time. To minimize the effect of clouds during analysis periods, which is large in 1-day data products, 16-day composite products were  prepared. The reason why 16-day composite was used is that the 16-day composite SbAI was a better indicator to avoid the effect of cloud cover than 8-day composite SbAI. The final spatial resolution was 0.05 . The 16-day data in each month was finally averaged as monthly. According to the above-described method, the assumed dust hazard level was based on U t and corresponded to the categories A to F in the land surface type, that is, bare land A , sparse vegetation B , vegetation reducing dust outbreaks C , snow cover D , wet surface E , and frozen soil F Table 1 . The approximate range of NDVI corresponding to each U t range is also shown Kimura and Shinoda, 2010 . Fig. 2 shows the distribution map of each land surface type categorized in the Table. Kurosaki and Mikami 2005 reported that U t was less than 10 m s -1 at 50 77 out of 65 SYNOP meteorological observatories within the target region, and at 15 23 observatories, U t was either more than 10 m s -1 or less than 15 m s -1 . When U t is 15 m s -1 , dust events are effectively eliminated because instances of wind speeds exceeding 15 m s -1 are rare in northeast Asia Kurosaki and Mikami, 2005 .
The number of ADE from March through May during the 22 years from 2000 was obtained from the JMA https://www. data.jma.go.jp/gmd/env/kosahp/kosa_Table_1.html . Note that this dataset contains the monthly "number of days" on which an ADE was observed at one or more of the 61 JMA observatories. Therefore, even if many observatories observed an event on a given day, only one ADE would be counted on that day.

Interannual changes in the number of ADE from 2000
The total number of ADE from March through May significantly decreased from 2000 to 2021 Fig. 3 . From 2012 to 2020, an ADE was observed on fewer than 10 days each year,  but in 2021 ADE were observed on 13 days. Although more ADE were observed in 2021 than had been observed during the previous decade, however, the normal value average values for the last 30 years observed annually from 1991 to 2020 as determined by JMA was 13 days; thus, the number of ADE in 2021 did not exceed the normal value.

Transport routes of Asian dust and dust source areas
Asian dust was observed in Fukuoka on 43 days in these past ten years 2012-2021 Fig. 4 ; on 40 days the trajectories passed through the defined target region, and on the other three days, they passed through the region between 120 E and 130 E 10 May 2021, 4 April 2020, and 7 April 2018 . Many other back trajectory analyses of ADE have also shown that the dust source was in the defined target area e.g., Onishi et al., 2012;MOE, 2014;Uno et al., 2017;Tsedendamba et al.,  Thus, in recent decades not only Mongolia, the Gobi Desert, and the Loess Plateau but also Inner Mongolia and northeastern China have been Asian dust source regions Tian et al., 2007;Minamoto et al., 2018. Moreover, Tian et al. 2007and Kurosaki et al. 2016 have reported that dust storm frequencies at SYNOP stations in Inner Mongolia and ADE varied nearly in parallel from the middle of the 1970s to the end of 1990s. In the region within 35 N -50 N and 105 E -120 E, high northerly or northwesterly wind speeds with cyclonic activity are remarkable Tian et al., 2007 , which favors dust outbreaks Tian et al., 2007;Yamamoto and Hayakawa, 2008 . Although in this study, the dust outbreak frequency in the target region was not examined, high dust outbreak frequencies were observed there during the 2000s Kurosaki and Mikami, 2005;Wu and Kai, 2016;Kawai et al., 2021;Bao et al., 2021 . Table 2 shows the correlation between the coverage of each land surface category see Table 1 in each month, which was calculated every 16-day interval of MODIS data, and ADE from March through May using the area of 35 N -50 N and 100 E -120 E. The number of ADE from 2000 to 2021 showed a significant positive correlation with land surface category A coverage bare land; U t < 10 m s -1 during the springtime and a significant negative correlation with category C coverage vegetation reducing dust outbreaks; U t ≥ 15 m s -1 except in May. A negative but nonsignificant correlation was found between category D coverage snow cover and the number of ADE. However, category H coverage = C + D showed a significant negative correlation with the number of ADE in April and during March -May considered together. In general, category E wet surfaces and F frozen soil surfaces are thought to inhibit dust occurrence, but a significant negative correlation was observed between category E and the number of ADE only in May. The effects of category A and B sparse vegetation; 10 ≤ U t < 15 m s -1 coverages observed here are similar to the findings of Kurosaki and Mikami 2005 , who reported that U t was less than 10 m s -1 at 77 of 65 SYNOP meteorological observatories within the target region, and between 10 and 15 m s -1 at the remaining 23 of the observatories. The significant correlations between category A and C coverages and the number of ADE may reflect less bare land or more vegetation. Both the number of ADE and category A coverage have been declining since 2000 Fig. 3 . The Chinese government has initiated national projects e.g., Grain for Green Program in 1999, and Phase 2 of the Three-North Shelter Forest Program in 2001 and invested substantial resources to combat desertification and promote greening in arid regions Cao et al., 2009;Li et al., 2012 . As a result of these efforts, vegetation has gradually increased, as shown by some satellite data IPCC, 2019; Chen et al., 2019;Kimura, 2017 . The spatial distribution of each land surface type in the target region during March 2001 the year with the largest number of ADE during the study period shows that the coverage of bare land was larger in 2001 than in 2020 the year with small number of ADE , especially in Inner Mongolia and on the Loess Plateau Fig. 2 . Although the reduction in the number of ADE cannot be definitely attributed to these Chinese projects, at least with regard to dust sources in China, statistical analyses suggest that the greening has had some effect on dust occurrences CAS, 2018; Long et al., 2018;Wu and Kai, 2016;Tan and Li, 2015 . The number of ADE was calculated using the following simple regression equation, where the explanatory variable was selected by considering multicollinearity. ADE = 9.23 10 -5 X 25.08 R = 0.74, p < 0.001 , 1

Effect of land surface conditions in the target region on ADE
where X is the category A coverage km 2 . Then, the relationship between the total observed and calculated numbers of ADE from March through May Table 2 was examined Fig. 5 . The root mean square error RMSE was 7 days, which is within the normal number of ADE = 13 days , but in 2002, 2008, and 2010, the estimated error was more than 10 days. Simulation results obtained with the Regional Atmospheric Modeling System and a synoptic weather analysis showed that slight changes in the air pressure distribution, the transport route of Asian dust, and the amount of dust transportation in the atmospheric boundary layer led to large differences in the number of ADE in those years Hara et al., 2004;MOE, 2014 . Therefore, the accuracy of ADE estimation with Eq. 1 depends on prevailing meteorological conditions. In addition, the instantaneous dust events would be suppressed by coverages of categories C, D, E, and F. However, Eq. 1 , which was obtained by using 16-day composite MODIS products, can be used to determine whether the number of ADE during March -May is likely to be more or less than the normal Fig. 4. continued value at the 77 probability level. The threshold category A coverage to obtain the normal number of ADE was 412,567 km 2 .

Conclusions
The spatial distribution of threshold wind speed during the springtime March to May in northeast Asia was mapped to assess land surface conditions in dust source regions since 2000. The region within 35 N -50 N and 100 E -120 E was targeted in this study, and most dust trajectories during the past ten years passed through this target region. Trajectories passing through Mongolia, the Gobi Desert, the Loess Plateau and Inner Mongolia, were roughly grouped into three routes.
The annual number of ADE significantly decreased after 2000. Therefore, the relationship between the coverage area of various threshold wind speed U t ranges and the number of ADE was analyzed. ADE showed a significant positive correlation with the coverage of U t < 10 m s -1 , and a significant negative correlation with the coverage of U t ≥ 15 m s -1 . These significant correlations  may reflect enhancements or reductions of dust outbreaks due to changes in the bare land surface and vegetation coverages. The RMSE between the observed number of ADE and the number calculated using the coverage of U t < 10 m s -1 from March through May was 7 days. Accuracy of this method is limited if the dust source region includes areas outside of the target region or if the transport route is different. However, the results of this study and previous studies indicate that most ADE from March to May are affected mainly by dust events in the target region. Thus, this method, if combined with consideration of the atmospheric transport route, is potentially a useful tool for predicting whether the number of ADE in a given year will be larger or smaller than the number in a normal year. It is my hope that the usefulness of the method presented here will be confirmed generally, and that it will become useful in the near future for assessing land surface conditions in dust source areas.