Multidecadal variability of dust activity in Gobi desert and its connection with the pacific decadal oscillation

The multidecadal changes of dust column mass density (DCMD) in Gobi desert (GD) in spring are investigated based on the Modern-Era Retrospective analysis for Research and Applications version 2 dataset. In addition, the possible effects of the atmospheric circulation and sea surface temperature (SST) forcing on the multidecadal changes are analyzed. Results show that the dust aerosol over GD experienced a decadal change in 1999 with about 30% higher dust loading during 2000–2013 in comparison to that during 1987–1999. Further analysis indicates that the decadal change of dust aerosol over GD is attributed to the more strengthened northwesterly wind anomaly extending from lower to middle troposphere and the anticyclonic anomaly in middle troposphere over GD during the latter epoch, which is favorable to the increase of local dust activities. Furthermore, the decadal change of DCMD in GD is associated with the switch of Pacific Decadal Oscillation (PDO) phase. From 2000 to 2013, the PDO was in the negative phase, which induced to a positive potential height anomaly and northwesterly wind anomalies in the middle troposphere over GD. The dry and cold air brought by the anomalous northwesterly wind associated with the negative PDO phase reduces the relative humidity in the lower troposphere further amplify the effect of strengthened wind speed, being favorable for the increase of local dust loading and the resultant increase of DCMD there.


Introduction
Dust particles, also called dust aerosols, are suspended in the atmosphere and are the main component of atmospheric aerosols, the aerodynamic diameter of which is usually less than 100 μm.The dust aerosols can modulate the cloud microphysical processes indirectly by acting as cloud condensation nuclei and ice nucleating particles (Mahowald et al 2007, Huang et al 2009, Mahowald et al 2010, Rodríguez et al 2012, Huang et al 2014, Nabat et al 2014, Chen et al 2014a, Kok et al 2023).The dust aerosols especially in East Asia have substantial impacts on human health, environment, ecosystems and climate (Abuduwaili et al 2010, Lin et al 2012, Huang et al 2013, Miao et al 2015, Schepanski 2018, Guo et al 2017, 2019).Hence, increasing attentions from the public and scientists have been paid to the variations of dust aerosols in East Asia.
East Asia is a major dust source in the world, which can eject large amount of dust particles into the atmosphere every year (Zhang et al 2003, Wang et al 2012, Wu et al 2016) and the dust activities in East Asia can be affected by a number of factors such as the changes of surface wind speed, precipitation, natural vegetation, and anthropogenic land use and so on (Liu 2004, Xu et al 2006, Ward et al 2014).The Taklimakan desert (TD) and Gobi Desert (GD) are the two of the most significant dust sources in East Asia (Ginoux et al 2004, Zhang et al 2008, Chen et al 2014b).Many previous studies have investigated some differences in dust emissions, transport and influences between the TD and the GD (Chen et al 2017a, 2017b, Liu et al 2019).For example, Chen et al (2017b) found that the dust aerosols originated from GD are more likely transported to the downstream region than that from TD. Liu et al (2019) indicated that comparing to TD, the dust aerosols generated in GD has greater effect on the western North America (WNA).However, compared to the dust studies in TD, there are limited attentions paid on the dust activities in GD.
The Gobi Desert (GD), characterized as a high-altitude plateau desert ranging in elevation from approximately 910 to 1520 meters, encompasses a substantially larger geographical expanse compared to the Taklimakan Desert (TD).Its territory extends from southern Mongolia to the northwestern and northern regions of China.Shao et al (2003) have conducted investigations that underscore the marked intensity of dust emissions within the GD.Their findings reveal that dust emissions in this region reach notable levels, with maximum dust emission rates reaching 5000 μg m −2 s −1 and net dust emissions amounting to 16 t km −2 d −1 in March and April of 2002.Moreover, Shao and Dong (2006) have drawn attention to the frequent occurrence of dust storms in the expansive Gobi and desert-steppe zones of Mongolia, which have been responsible for severe environmental and societal consequences.
In the GD, a noteworthy decadal variability characterizes the behavior of dust aerosols (Ding et al 2005, Liu et al 2019, Liu et al 2022).For instance, using a dust aerosol optical thickness (DAOT) index, Liu et al (2019) have discerned a substantial decadal shift in GD dust aerosols, with a preceding negative phase prior to 1999 and a subsequent positive phase spanning from 1999 to 2013.While these studies have successfully identified the decadal alterations in GD dust behavior commencing in 1999, the underlying factors contributing to the emergence of this decadal change remain obscured.Studying the causes of decadal variation of dust activity in the Gobi region will not only help to deepen our physical understanding of the multi-time-scale variation mechanism of dust activity in East Asia, but also lay a theoretical foundation for decadal scale dust prediction.Therefore, a comprehensive investigation is deemed both necessary and imperative to unravel the potential causes of this phase transition in dust aerosols within the GD, occurring in the pivotal year of 1999.
The remainder of the paper is organized as follows.Section 2 describes the data and methods.Section 3 presents the results of the decadal change of dust aerosols in GD in 1999 and its possible causes.Finally, the summary and discussion are provided in section 4.

Data and methods
In this study, the dust column mass density (DCMD) data are used to represent the dust aerosol.Many previous studies have reported that the quality of dust aerosol in MERRA2 data is generally reliable in East Asia (e.g., Buchard et al 2015, Guo et al 2019, Liu et al 2019, Sun et al 2019, Yao et al 2020), thus the long-term DCMD data are employed from the Modern-Era Retrospective analysis for Research and Applications version 2 (MERRA-2) Aerosol Reanalysis data of the National Aeronautics and Space Administration Global Modeling and Assimilation Office (Gelaro et al 2017), which has a horizontal resolution of 0.5°× 0.625°and covers the period from January 1980 to the present.We also use the independent satellite observed Multi-Sensor Absorbing Aerosol Index data that combine TOMS, GOME-1, SCIAMACHY, Ozone Monitoring Instrument, GOME-2A, and GOME-2B to further verify the reliability of MERRA-2 dust data.The results are basically consistent between two data sets, especially for the decadal changes of dust loading in East Asia (figure S1 in the supporting information).The monthly atmospheric circulation data including the zonal and meridional winds and the geopotential height (HGT) are obtained from the European Centre from Medium-Range Weather Forecast (ECMWF) Reanalysis v5 (ERA5) that covers the period from January 1959 to the present with a horizontal resolution of 0.25°× 0.25° (Hersbach et al 2020).The monthly sea surface temperature (SST) data are from the NOAA Extended Reconstructed SST version 5 (ERSST v5) dataset (Smith et al 2008).The ERSST v5 dataset has a horizontal resolution of 2°×2°and covers the period from January 1854 to the present.In order to make better comparisons, all the data are bilinearly interpolated to a resolution of 2.5°× 2.5°grid before conducting further analysis.Our analysis focuses on spring because the dust activity over GD is usually the most intense in spring.In this study, a two-tailed Student's t-test is used to evaluate the confidence level of correlation and regression.
In this study, the moving t-test method was used to detect the possible abrupt change points.The primarily principle of this method is to test a hypothesis based on a difference between two sub-sample means.The formula for calculation is as follows: where x1 and x2, s1 and s2, n1 and n2 are the mean, variance and length of the two sub-samples, respectively, and the statistic t follows t-distribution with degrees of freedom (n1+n2-2).The time series detects the abrupt change points if |ti| > tα at a particular significance level α.

Results
Figure 1 illustrates the spatial distributions of climatological spring DCMD and the corresponding 800 hPa wind fields during 1980-2017.It is observed that the dust loading over GD is very high, with a magnitude exceeding 3 × 10 −4 kg m −2 .The concentration of dust decreased significantly from the GD eastward to Japan and Korea, extending to the Pacific Ocean and even western North America, which is consistent with previous studies (Liu et al 2019, Liu et al 2022).Additionally, it is evident that the lower troposphere over the Gobi Desert is predominantly characterized by a climatological northwesterly wind flow, which shifts to a strong prevailing westerly in the downstream region during spring.To further explore the temporal changes of ECMD in GD, we define a DCMD index (DCMD_GD) by the method of area-averaged DCMD in GD (35°N-45°N, 95°E-110°E) region.Figure 2 It is worth noting that the maximum of t-values appears in 1999, which exceeds 99.9% significance level, confirming that the DCMD_GD experiences a decadal increase around 1999.
Since the variation of dust loading is very dependent on the near-surface wind speed and wind direction, we investigated the variation of near-surface wind fields before and after the decadal change.Figure 3(a) displays the spatial distributions of climatological spring 10m winds (UV10m) in spring during 1980-2017.It is clear that GD is dominated by northwesterly winds, which is conducive to the transport of dust aerosols from the upstream areas (high-concentration dust source regions) to GD.Here a WSD_10m_GD index during 1980-2017 is defined as the area-averaged 10m wind speed in GD region.It is noted that the 10m wind speed in GD also exhibits an evident decadal change, mainly presenting a negative phase from 1987 to 1999 and a positive phase during 2000-2013.The correlation coefficient between DCMD_GD index and WSD_10m_GD index from 1980 to 2017 on decadal time scale reaches 0.91.This result implies that the decadal fluctuation of dust activities in GD may be closely tied to the decadal changes of near-surface wind speed there.To further investigate the physical mechanism responsible for the decadal change in the dust activities in GD, two periods .The increased northwesterly wind speed facilitates the enhancement of intensity and frequency of dust activities in GD, contributing to the increase of dust concentrations over GD after 1999, thus leading to a decadal increase of DCMD_GD (figure 2(b)).The opposite atmospheric anomalies happen during the 1987-1999 (not shown).Hence, it is the enhanced northwesterly wind speed and wind direction over GD that is the direct cause for the decadal increase of DCMD_GD after 1999.
To find out why the enhanced northwesterly flows emerge in GD, figures 4(a) and (b) show the composited spring 500hpa geopotential height and the corresponding wind fields between these two periods.During 2000-2013, GD is located at the edge of the anomalous anticyclone in the 500 hPa accompanied by the downward motion anomaly.Corresponding to the anticyclonic anomalies, the significantly northwesterly anomalies appeared in the middle troposphere over the GD region (figure 4(b)).The configuration of these atmospheric circulations is favorable to the dust transport from the upstream high-loading dust regions to the GD region, increasing the local dust loading, thus leading to the significant decadal increase of DCMD_GD after 1999.These results suggest that there are direct influences of changes in wind speed associated with the anticyclonic anomalies on DCMD in GD area.
To further investigate the formation mechanism of the anticyclonic anomalies and accompanied wind changes in the middle troposphere, we focus on the possible roles of ocean signals on DCMD_GD.Here, a composite analysis of the spatial distribution of Pacific SST anomaly during the spring is shown in figure 5(a).The SST anomaly is negative in the eastern Pacific Ocean and positive in the western Pacific Ocean during 2000-2013, indicating a negative Pacific Decadal Oscillation (PDO) phase.This implies that the abrupt change of DCMD_GD in 1999 may be related to the transition of the PDO phase.Therefore, in order to study the relationship between DCMD_GD and PDO more directly, a PDO index is defined as the leading principal component of the monthly spring SST anomalies in the region of (20°N-70°N, 110°E-100°W) (Zhang et al 1997).Figure 5(b) exhibits the time series of PDO, DCMD_GD and their decadal components using a 9-year running mean during the period of 1980-2017.It is clear that there is a significant inverse relationship between the PDO index and DCMD_GD.The correlation coefficient is −0.56 between PDO index and DCMD_GD, while the correlation coefficient reaches −0.95 between their decadal components, both of which exceed the 95% confidence levels based on the two-tailed student's t-test.The results indicate the opposite decadal variations between PDO and DCMD in GD, and imply that the decadal increase of DCMD_GD in 1999 can partially attributed to the transition of the PDO phase.To further confirm the effect of the change of PDO phase on the decadal change of DCMD_GD, the regression patterns of anomalous HGT500, UV800, and DCMD onto the reversed PDO index are given (figures 6(a)-(c)).There is a positive potential height anomaly in the middle troposphere in GD area accompanied by the anomalous descending motion, which are favorable for the increase of local dust loading in GD and are consistent with the results in figure 4(a).It is clear that the anomalous northwesterly winds over the GD area are associated with negative PDO (figure 6(b)), which is conducive to the increase of wind speed (figure 6(c)) by enhancing the local climatological northwesterly wind (figure 3(a)), thus facilitates the increase of DCMD_GD.In addition, we find that the dry and cold air brought by the anomalous northwesterly wind in the GD area associated with the negative PDO phase reduces the relative humidity in the lower troposphere (figures 7(a), (b)).When the relative humidity is low, the air can hold very little moisture.As a result, the dry air is less effective at dampening or settling dust particles.This makes it easier for winds to lift and suspend fine soil particles into the atmosphere and induces increase of DCMD_GD.It is noted that the changes of soil moisture between these two periods, are small (figures 7(c), (d)), indicating that soil moisture is less correlated with PDO and it is not the main factor affecting the dust activity in GD area, supporting the previous finding (Guo et al 2019).In conclusion, the variation of DCMD_GD is strongly related to PDO, which can affect the decadal change of DCMD_GD by modulating the atmospheric circulations surrounding the GD region.

Conclusion
In this study, we investigate the decadal changes of the dust column mass density (DCMD) in Gobi desert (GD) based on Modern-Era Retrospective analysis for Research and Applications version 2 dataset.In particular, the possible roles of the atmospheric circulations and the SST forcing on the decadal changes are studied.Results show that the spring DCMD_GD experiences a decadal change in 1999, with a negative phase before 1999, and a positive phase during 2000-2013.
The strengthened anomalous northwesterly wind over GD during 2000-2013 compared to 1987-1999 play an important role in the decadal change of DCMD_GD.The near-surface anomalous northwesterly wind over GD region can strengthen the local wind speed by enhancing the climatological northwesterly wind, and is favorable to the dust transport from the upstream area to the GD region, thus leading to the increase of DCMD_GD.In addition, there is a significant anticyclonic anomaly in the middle troposphere over the GD region with a downward motion, which is conducive to intensify of the dust activities in the local regions.
Further analysis revealed that the decadal variations of DCMD_GD are attributed to the transition of the PDO phase.A negative phase of PDO can result in middle troposphere positive potential height and northwesterly wind anomalies in GD.Positive potential height with subsiding motion creates favorable conditions for local dust emission, and northwesterly wind anomalies can increase the local wind speed and the transportation of dust from the upstream high dust loading regions.Furthermore, the dry and cold air brought by the anomalous northwesterly wind associated with the negative PDO phase reduces the relative humidity in the lower troposphere further amplify the effect of strengthened wind speed, being favorable for the increase of local dust loading and the resultant increase of DCMD there.It should be noted that there are significant lowfrequency variations in the observed SST anomalies associated with PDO.Therefore, it may provide a possibility to predict near-future changes of PDO as well as the dust activities in GD.The decadal prediction of dust activities in GD need to be further investigated in the future study.

Figure 1 .
Figure 1.The climatological mean dust column mass density (DCMD, shading) and corresponding winds (UV800, vectors) during spring (March-April-May, MAM) during 1980-2017.The black box indicates the region of the Gobi desert.
Figure1illustrates the spatial distributions of climatological spring DCMD and the corresponding 800 hPa wind fields during 1980-2017.It is observed that the dust loading over GD is very high, with a magnitude exceeding 3 × 10 −4 kg m −2 .The concentration of dust decreased significantly from the GD eastward to Japan and Korea, extending to the Pacific Ocean and even western North America, which is consistent with previous studies(Liu et al 2019, Liu et al 2022).Additionally, it is evident that the lower troposphere over the Gobi Desert is predominantly characterized by a climatological northwesterly wind flow, which shifts to a strong prevailing westerly in the downstream region during spring.To further explore the temporal changes of ECMD in GD, we define a DCMD index (DCMD_GD) by the method of area-averaged DCMD in GD (35°N-45°N, 95°E-110°E) region.Figure2(a)shows the normalized time series of the spring DCMD_GD index.It is noted that the dust aerosols in GD are featured by a significant decadal change.The DCMD_GD index is dominated by a negative phase from 1987 to 1999, and is mainly positive phase during 2000-2013 with a decadal change in 1999.To verify whether the sudden change of 1999 occurred, figure 2(b) illustrates the results of a 11-years sliding t-test of spring DCMD in GD.It is worth noting that the maximum of t-values appears in 1999, which exceeds 99.9% significance level, confirming that the DCMD_GD experiences a decadal increase around 1999.Since the variation of dust loading is very dependent on the near-surface wind speed and wind direction, we investigated the variation of near-surface wind fields before and after the decadal change.Figure3(a) displays the spatial distributions of climatological spring 10m winds (UV10m) in spring during 1980-2017.It is clear that GD is dominated by northwesterly winds, which is conducive to the transport of dust aerosols from the upstream areas (high-concentration dust source regions) to GD.Here a WSD_10m_GD index during 1980-2017 is defined as the area-averaged 10m wind speed in GD region.It is noted that the 10m wind speed in GD also exhibits an evident decadal change, mainly presenting a negative phase from 1987 to 1999 and a positive phase during 2000-2013.The correlation coefficient between DCMD_GD index and WSD_10m_GD index from 1980 to 2017 on decadal time scale reaches 0.91.This result implies that the decadal fluctuation of dust activities in GD may be closely tied to the decadal changes of near-surface wind speed there.To further investigate the physical mechanism responsible for the decadal change in the dust activities in GD, two periods

Figure 3 .
Figure 3. (a) Geographical distributions of climatological spring 10m winds (UV10m, vectors) during 1980-2017.(b) Normalized time series of spring WSD_10m_GD index during 1980-2017.(c) As in (a), but for the period of 2000-2013 compared to 1987-1999.Arrows marked in red indicate winds significant at the 90% confidence level.(d) Spatial distribution of 10m wind speed (WSD_10m, shading) in spring for the period of 2000-2013 compared to 1987-1999.Stippling regions indicate values significant at the 90% confidence level.The black boxes in (a), (c), and (d) indicate the regions of the Gobi desert.

Figure 4 .
Figure 4. (a) The spring geopotential height at 500 hPa (HGT500, shading) in 2000-2013 compared to 1987-1999.Stippling regions indicate values significant at the 90% confidence level.(b) As in (a), but for the corresponding winds (UV500, vectors).Arrows marked in red indicate winds significant at the 90% confidence level.

Figure 5 .
Figure 5. (a) Spatial distributions of the spring sea surface temperature (SST_MAM, shading) of the Pacific for the period of 2000-2013 compared to 1987-1999.Stippling regions denote SST significant at the 90% level.(b) The normalized spring PDO index (red dashed line) and normalized DCMD_GD index (blue dashed line) from 1980 to 2017.The decadal components are also shown with the corresponding solid lines by using a 9-year running mean.

Figure 6 .
Figure 6.Regression maps of spring (a) geopotential height at 500 hPa (HGT500, shading), (b) the corresponding 800-hPa winds (UV800, vectors), and (c) dust column mass density (DCMD, shading) onto the reversed spring PDO index.The black box in (a-c) denotes the GD region.Stippling regions in (a) and (c) denote geopotential height and DCMD significant at the 90% level.Arrows marked in red indicate winds significant at the 90% level.

Figure 7 .
Figure 7. (a) The spring relative humidity at 800 hPa (RHUM800, shading) in 2000-2013 compared to 1987-1999.(b) Regression maps of spring relative humidity at 800 hPa onto the reversed spring PDO index.(c) As in (a), but for soil moisture.(d) As in (b), but for soil moisture.The black box in (a)-(d) denotes the GD region.Stippling regions in (a)-(d) denote relative humidity and soil moisture significant at the 90% level.