the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 28 Jul 2026
Mongolian Cyclone Intensification Reshapes Dust Activity in Northwest China
Abstract. Dust storms are a common hazardous weather phenomenon in arid regions. In recent years, dust storm intensity in northwest China has increased markedly under changing atmospheric circulation conditions, yet the underlying mechanisms remain debated. Previous studies have largely attributed dust variability to land-surface changes and broad-scale circulation anomalies, but the spatially divergent responses of different dust source regions to circulation changes remain poorly understood. Here, we investigate Mongolian cyclone variability during 2001–2025 and its impacts on dust activity in the Taklimakan Desert, Alxa Plateau, and Mu Us Sandy Land using multi-source observational data and a random forest regression model. Results showed that (1) Mongolian cyclone intensity increased from 94.84 × 10-5 s-1 to 97.76 × 10-5 s-1 and shifted southward by about 2.08° after 2019; (2) Strengthened Mongolian cyclone activity enhanced dust emissions in the Alxa Plateau (from 0.23 to 0.28 g·m-2 since 2019), whereas dust activity declined in the Taklimakan Desert, which is less affected by Mongolian cyclone forcing (from 0.63 to 0.57 g·m-2 since 2019); (3) Wind was the dominant driver of dust activity (contribution >0.60), while precipitation and the Pressure Gradient Index (PGI) were important secondary contributors in the Alxa and Mu Us regions, and PGI partially modulated wind effects in surrounding cities; and (4) Regional spring mean PM10 concentrations are projected to increase substantially during 2026–2100 (229.98–824.31 μg·m-3) and show weak sensitivity to emission scenarios. These findings highlight the importance of explicitly considering cyclone-wind regime-topography interactions in future dust projections, as circulation-driven variability may play a larger role than previously recognized in regulating East Asian dust activity.
- Preprint
(2382 KB) - Metadata XML
-
Supplement
(1678 KB) - BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2026-3940', Anonymous Referee #1, 14 Aug 2026
-
AC2: 'Reply on RC1', Wenhui Zhang, 15 Oct 2026
Response to RC1-1:
We thank the reviewer for this insightful comment. We agree that our original comparison between the Taklimakan Desert and the Alxa Plateau did not sufficiently isolate the contribution of Mongolian cyclone variability, because these regions differ substantially in topography, local circulation, and background surface conditions. We also acknowledge that the Taklimakan Desert cannot be characterized by a single prevailing circulation regime, as mesoscale and thermally driven circulations, as well as terrain effects, may substantially modify the regional wind field.
In the revised manuscript, we have therefore modified the regional framework and no longer interpret the contrasting dust trends between the Taklimakan Desert and the Alxa Plateau as direct evidence of a one-to-one cyclone effect. Instead, we consider three regions representing different degrees of exposure to Mongolian cyclone-related circulation. the Mongolia-Gobi region, which is closely associated with cyclone development and direct influence; the Alxa Plateau, which is located downstream of cyclone-related northwesterly circulation; and the Taklimakan Desert, where complex terrain and basin-scale circulation can substantially weaken the influence of synoptic-scale disturbances.
Based on this framework, we further added a cyclone-active versus cyclone-inactive year composite analysis to examine whether dust responses differ along this gradient of cyclone influence. Over the Mongolia-Gobi region (cyclone genesis and direct impact zone), DUCMASS is significantly higher in active years than in inactive years (+20.7%, P ≤ 0.05; Fig. 2d). Over the Alxa Plateau (downstream of cyclone tracks), DUCMASS is also significantly higher in active years (+12.1%, P ≤ 0.05; Fig. 2e). In contrast, over the topographically sheltered Taklimakan Desert, DUCMASS is slightly lower in active years (-6.1%, P > 0.05; Fig. 2f), though this change is not statistically significant. These results indicate that cyclone activity is associated with enhanced atmospheric dust loading primarily in exposed regions, whereas the Taklimakan exhibits a weaker response likely due to the combined effects of topographic constraints, basin-scale circulation, and local mesoscale processes (Relevant text: pp. 11, lines 314–320).
The methods are as follows: Cyclone-active vs inactive year classification. To test whether cyclone activity modulates dust emissions in a spatially divergent manner, we classified years based on the mean intensity of tracked cyclones (defined as the mean minimum central pressure of all cyclone tracks in each year). Years with the strongest 40 % of cyclone intensity (lowest central pressure) were designated as "cyclone-active years" and years with the weakest 40 % were designated as "cyclone-inactive years". The remaining 20% of years were excluded from the comparison. The 40 % threshold was selected as a compromise between maximizing cyclone intensity contrast and retaining sufficient sample size for statistical comparison. This yielded 10 active years (2001, 2003–2007, 2009, 2012, 2021, 2023) and 10 inactive years (2008, 2010, 2011, 2015–2017, 2019, 2020, 2022, 2024). We defined three regions representing a gradient of cyclone influence: the Mongolia-Gobi region (cyclone genesis and direct impact zone), the Alxa Plateau (downstream of cyclone tracks), and the Taklimakan Desert (topographically sheltered, weak cyclone influence). For each region, we extracted spring DUCMASS from MERRA-2 and calculated regional means. We then compared regional mean DUCMASS between active and inactive years using a two-sample t-test and the non-parametric Mann-Whitney U test (Relevant text: pp. 7, lines 194–205).
We have revised the Introduction, Results, and Discussion sections accordingly to clarify that our study investigates spatially divergent dust responses to Mongolian cyclone variability under contrasting environmental conditions, rather than attributing regional dust changes solely to cyclone forcing.
Response to RC1-2:
Response: We thank the reviewer for this important comment. We fully agree that relative vorticity anomalies alone are not sufficient to identify objectively defined cyclone centers, and that an explicit cyclone tracking algorithm is needed.In the revised manuscript, we have implemented an objective cyclone tracking algorithm following the approach of Mu and Fiedler (2025).
The specific cyclone identification and tracking methods have been supplemented in the Methods section, as follows: We identified and tracked Mongolian cyclones using 6-hourly ERA5 mean sea-level pressure (MSLP) fields during 2001–2025. Cyclone centers were detected as local minima in the MSLP field that were at least 4.0 hPa lower than the surrounding environment (computed as a 15×15 grid-point background) and with central pressure below 1006 hPa. Cyclones were tracked using a nearest-neighbor algorithm: at each time step, each active track was matched to the nearest detected center within a search radius of 600 km, with a maximum allowed displacement of 700 km between consecutive time steps. Tracks with a lifetime shorter than 18 hours, a total displacement less than 2.5°, or a predominantly westward movement were excluded. This procedure yielded 27–63 cyclone tracks per spring season, with a mean of approximately 42 tracks per year, consistent with previous cyclone-tracking studies over this region (Mu and Fiedler, 2025). Cyclone intensity was defined as the minimum central pressure along each track, and the annual mean intensity was calculated as the mean of all track minimum pressures in each year. Cyclone candidates were restricted to the Mongolian Plateau and adjacent regions (90–130 °E, 40–58 °N), where cyclones affecting northern China dust events are typically generated or intensified (Relevant text: pp. 6–7, lines 182–193).
We now report the interannual variation of cyclone frequency and mean central pressure (Fig. S1e, S1f). Objective cyclone tracking further shows that cyclone count increased from +0.42 yr-1 before 2019 to +2.43 yr-1 after 2019 (Fig. S1e), and mean central pressure shifted from a weakening trend (+0.09 hPa·yr-1, i.e., increasing central pressure) before 2019 to a strengthening trend (-0.08 hPa·yr-1, i.e., decreasing central pressure) after 2019 (Fig. S1f). However, the differences between the two segment slopes are not statistically significant (F-test P > 0.05), likely due to the short post-2019 period. Collectively, the strengthening of cyclone intensity and the increase in cyclone frequency after 2019 are conducive to the development of extreme near-surface gales and long-range dust transport towards Northern China in the past five years (Relevant text: pp. 11, lines 305–311).
Response to RC1-3:
Response: We agree that the simultaneous changes in Mongolian cyclone characteristics, near-surface wind, and dust activity after 2019 do not by themselves demonstrate a causal relationship, particularly given the relatively short post-break period. We acknowledge that our original wording may have overstated the causal interpretation of the observed temporal co-variations. In the revised manuscript, we have therefore modified our interpretation throughout the text. We no longer attribute the post-2019 dust changes solely to Mongolian cyclone changes. Instead, we interpret them as a spatially heterogeneous response associated with variations in cyclone-related circulation under different regional environmental conditions.
Firstly, we have added a cyclone-active vs inactive year composite analysis to move beyond simple co-occurrence, as detailed below:
To test whether cyclone activity modulates dust emissions in a spatially divergent manner, we compared spring DUCMASS between cyclone-active and inactive years over three regions representing a gradient of cyclone influence (Fig. 2d-f). Over the Mongolia-Gobi region (cyclone genesis and direct impact zone), DUCMASS is significantly higher in active years than in inactive years (+20.7%, P ≤ 0.05; Fig. 2d). Over the Alxa Plateau (downstream of cyclone tracks), DUCMASS is also significantly higher in active years (+12.1%, P ≤ 0.05; Fig. 2e). In contrast, over the topographically sheltered Taklimakan Desert, DUCMASS is slightly lower in active years (-6.1%, P > 0.05; Fig. 2f), though this change is not statistically significant. These results indicate that cyclone activity is associated with enhanced atmospheric dust loading primarily in exposed regions, whereas the Taklimakan exhibits a weaker response likely due to the combined effects of topographic constraints, basin-scale circulation, and local mesoscale processes (Relevant text: pp. 11, lines 312–320).
The methods are as follows: Cyclone-active vs inactive year classification. To test whether cyclone activity modulates dust emissions in a spatially divergent manner, we classified years based on the mean intensity of tracked cyclones (defined as the mean minimum central pressure of all cyclone tracks in each year). Years with the strongest 40 % of cyclone intensity (lowest central pressure) were designated as "cyclone-active years" and years with the weakest 40 % were designated as "cyclone-inactive years". The remaining 20% of years were excluded from the comparison. The 40 % threshold was selected as a compromise between maximizing cyclone intensity contrast and retaining sufficient sample size for statistical comparison. This yielded 10 active years (2001, 2003–2007, 2009, 2012, 2021, 2023) and 10 inactive years (2008, 2010, 2011, 2015–2017, 2019, 2020, 2022, 2024). We defined three regions representing a gradient of cyclone influence: the Mongolia-Gobi region (cyclone genesis and direct impact zone), the Alxa Plateau (downstream of cyclone tracks), and the Taklimakan Desert (topographically sheltered, weak cyclone influence). For each region, we extracted spring DUCMASS from MERRA-2 and calculated regional means. We then compared regional mean DUCMASS between active and inactive years using a two-sample t-test and the non-parametric Mann-Whitney U test (Relevant text: pp. 7, lines 194–205).
Secondly, we performed a segmented linear regression with a breakpoint at 2019 to examine whether the observed changes represent statistically significant trend reversals. The results show that cyclone frequency increased from +0.42 yr-1 before 2019 to +2.43 yr-1 after 2019, while cyclone intensity shifted from a weakening tendency (+0.09 hPa yr-1) to a strengthening tendency (-0.08 hPa yr-1). However, the differences between the two segments were not statistically significant, and no significant change was detected in cyclone latitude. These results have been reported transparently, and we now emphasize the uncertainty associated with the short post-2019 period (Relevant text: pp. 11, lines 305–309).
Lastly, we have added a limitation statement acknowledging that causal attribution requires numerical simulations. We acknowledge that our analyses are primarily observational and statistical, and that a full causal attribution would require numerical simulations with and without cyclone forcing. This limitation is now stated in the Conclusions: Our analyses are primarily observational and statistical. While the spatially divergent responses to cyclone activity provide stronger evidence than simple co-occurrence, a full causal attribution would require numerical simulations with and without cyclone forcing. This remains a direction for future work (Relevant text: pp. 22, lines 605–607).
Response to RC1-4:
Response: We sincerely thank the reviewer for this important suggestion. We fully agree that ERA5-derived surface wind trends require independent validation, particularly given that ERA5 is known to have biases in complex terrain regions such as northwest China. We have addressed this concern by validating the ERA5 wind trends against independent ground-based meteorological station observations from the NOAA Integrated Surface Database (ISD).
We have added the NOAA validation results to the Results section and presented the corresponding figure as Fig. S5, as detailed below:
To validate the ERA5-derived surface wind trends, we compared them with independent NOAA station observations. At four representative stations (ALAR, HALIUT, YUMENZHEN, and XIFENGZHEN) covering the four subregions, although ERA5 data underestimates wind speed, its trend of change is consistent with the station data (r = 0.42–0.82, P ≤ 0.05, Fig. S5a–d). For March, the observed declining trend before 2019 generally agreed with ERA5, whereas the post-2019 recovery was weaker in the station data and not statistically significant at most stations. To further assess the robustness of the wind speed trends, we examined spring and winter-spring mean wind speeds, which better represent the dust season and reduce the influence of sub-monthly variability. After 2019, spring wind speeds increased by 0.02 m·s-1·yr-1 over the Alxa Plateau and 0.18 m·s-1·yr-1 (P ≤ 0.05) over the Hexi Corridor (Fig. S5e). In winter-spring, wind speeds increased by 0.15 m·s-1·yr-1 over the Alxa Plateau, 0.18 m·s-1·yr-1 over the Hexi Corridor (P ≤ 0.05), and 0.02 m·s-1·yr-1 over the downwind cities, while they decreased over the Taklimakan (-0.11 m·s-1·yr-1), consistent with its topographic sheltering and weaker cyclonic influence (Fig. S5f). These results support the spatial heterogeneity of wind speed trends revealed by ERA5 (Relevant text: pp. 15, lines 414–424).
Also, we have added a detailed description of the NOAA station data in the Methods section. as detailed below:
To independently validate the ERA5-derived surface wind trends, we used hourly surface wind speed observations from the NOAA Integrated Surface Database (ISD; https://www.ncei.noaa.gov/data/global-hourly/). The ISD provides quality-controlled surface meteorological observations from more than 14,000 stations worldwide. First, we selected four typical stations (ALAR, HALIUT, YUMENZHEN, and XIFENGZHEN) and verified the correlation between the ERA5 data and NOAA data for these four stations in March. Second, we selected 32 stations within the study domain that had continuous records for at least five years during 2001–2025. These stations cover the four subregions: Taklimakan Desert (9 stations), Alxa Plateau (10 stations), Hexi Corridor (6 stations), and downwind cities (7 stations). Wind speed observations (variable WDSP, in m·s-1) were extracted for spring (March-May) and winter–spring (December-May), and compared with ERA5-derived trends at the corresponding grid points. Stations with fewer than five years of data were excluded to ensure robust trend estimates (Relevant text: pp. 6, lines 157–166).
Response to RC1-5:
Response: We sincerely thank the reviewer for this insightful comment. We fully agree that pooling station-year pairs can potentially confound spatial differences among stations with interannual variability within each station.
On the one hand, we have clarified that Fig. S6 presents station-by-station Spearman correlation coefficients, not pooled correlations. For each of the 26 stations, the Spearman correlation between PM10 and each driver (wind speed, precipitation, soil moisture, and NDVI) was calculated separately using the full time series at that station. Because each correlation is computed entirely within a single station, this approach is not affected by spatial differences among stations.
On the other hand, to formally separate spatial and temporal covariance, we fitted linear mixed-effects models with station as a random effect, details are as follows:
To separate within-station temporal relationships from between-station spatial differences, we fitted linear mixed-effects models with station as a random effect:
PM10ij = β0 + β1 × Driverij + ui + εij (2)
where PM10ij and Driverij are the spring mean PM10 concentration and the corresponding driver (wind speed, soil moisture, NDVI, or preceding-year precipitation) at station i in year j, ui is the station-specific random intercept, and εij is the residual error. The fixed effect β1 represents the within-station temporal relationship between the driver and PM10, after accounting for spatial differences among stations. Prior to model fitting, all variables were detrended at each station to remove long-term trends. For wind speed and soil moisture, winter-spring means were used; for NDVI, spring means were used, consistent with the seasonal windows used in the main analysis (Relevant text: pp. 8, lines 246–254).
Wind speed (β = 60.08, P ≤ 0.001) and soil moisture (β = -190.45, P ≤ 0.001) show significant within-station temporal effects on PM10, confirming that year-to-year variations in these drivers modulate dust activity within individual stations, rather than merely reflecting spatial differences among stations. In contrast, NDVI (β = -165.25, P = 0.143) and preceding-year precipitation (β = -0.026, P = 0.592) show no significant within-station temporal effects, although their pooled correlations with PM10 are significant. This indicates that vegetation and precipitation primarily modulate dust activity through spatial differences among stations (e.g., more vegetated or wetter sites have lower dust levels), whereas wind speed and soil moisture act as temporal drivers of interannual dust variability. These results are consistent with our conceptual framework and confirm that the PM10-wind and PM10-soil moisture relationships are robust within-station temporal relationships, not artifacts of spatial covariance (Relevant text: pp. 17, lines 483–492).
Response to RC1-6:
Response: We thank the reviewer for this important comment. We fully agree that PM10, dust storm days (DSD), and MERRA-2 dust column mass density (DUCMASS) represent fundamentally different aspects of the dust cycle and should not be treated as interchangeable measures of "dust activity." We have revised the manuscript to clearly define each metric, avoid interchangeable usage, and provide a more careful interpretation of their apparent disagreement over the Taklimakan Desert.
Firstly, we have added an introduction to the DSD (NOAA ISD) data and the physical meanings of the three indices in Section 2.2, as follows:
Hourly surface weather observations were obtained from the NOAA Integrated Surface Database (ISD; https://www.ncei.noaa.gov/data/global-hourly/). The ISD provides quality-controlled surface meteorological observations from more than 14,000 stations worldwide. For this study, we selected stations within the study domain that had continuous records during 2001–2025 (Relevant text: pp. 5, lines 140–143).
Three complementary metrics are used to characterize dust activity. Near-surface PM10 concentration (CNEMC) reflects near-surface particulate loading. Ground-based DSD is defined as days with daily PM10 > 500 μg·m-3. NOAA DSD is defined as days when any hourly observation reports a dust-related present weather code (MW1 = 6, 7, 30–35, or 98) in the Integrated Surface Database (ISD). DUCMASS (MERRA-2) represents column-integrated mineral dust mass. These metrics are not equivalent. PM10 and DSD characterize near-surface conditions and event frequency, whereas DUCMASS assesses column-integrated dust loading (Relevant text: pp. 5, lines 144–149).
Secondly, we have provided a more detailed explanation of the comparative trends in the Taklimakan Desert, as follows:
This apparent inconsistency arises from the different physical meanings and detection principles of these metrics. DUCMASS and PM10 reflect dust loading (column-integrated and near-surface, respectively), while DSD reflects event frequency based on a specific detection criterion. The increase in NOAA DSD over the Taklimakan therefore indicates more frequent but weaker dust events, rather than an increase in total dust loading. This interpretation is consistent with the topographic shielding of the Taklimakan from Mongolian cyclone activity, the basin may experience more frequent local dust activations under weakened synoptic forcing, but these events are weaker and contribute less to column-integrated dust loading. Meanwhile, weather-code-based DSD (NOAA ISD) relies on ground observer judgment and may therefore be biased (Relevant text: pp. 13–14, lines 390–397).
Thirdly, we have revised the descriptions of the metrics throughout the manuscript to make them more precise and avoid ambiguity. For example, statements such as "dust activity increased" have been replaced with more precise descriptions ("dust loading increased" for PM10 and DUCMASS, or "dust storm occurrence increased" for DSD), depending on the metric being discussed.
Finally, we have revised the caption of Figure 3 as follows: Fig. 3: (a) Interannual variations of near-surface PM10 concentrations during 2016–2025; (b) Interannual variations of spring dust column mass density (DUCMASS) from 2001 to 2025; (c) Mean dust storm days (DSD) across the four subregions derived from ground-based PM10 observations (daily PM10 > 500 μg·m-3, 2016–2025); (d) NOAA ISD weather-code-based DSD (2001–2025) (Relevant text: pp. 13, lines 380–382).
Response to RC1-7:
Response: We thank the reviewer for this critical comment. We agree that the projection workflow needed clearer documentation and that the original cyclone attribution was not supported by an explicit diagnosis of future cyclone behaviour. We have therefore revised the projection methodology, the interpretation of results, and all related claims.
Firstly, we have added a detailed description in Section 2.3, as follows:
The same five predictors were extracted from three CMIP6 models (CESM2, MPI-ESM1-2-LR, and UKESM1-0-LL) under SSP1-2.6, SSP2-4.5, and SSP5-8.5. All CMIP6 predictors were bias-corrected against the historical mean of each predictor using mean-ratio scaling. The RF model was then applied year-by-year to generate winter-spring DUCMASS projections for 2026–2100, with uncertainty estimated as ±1 standard deviation across the three CMIP6 models and the lower bound truncated at zero because DUCMASS is physically non-negative. HAI was not included in the future projection because CMIP6 does not provide consistent land-use scenarios across the three models, and future cyclone intensity, frequency, and position were not explicitly diagnosed from CMIP6; their influence is represented indirectly through projected changes in wind speed and PGI (Relevant text: pp. 9, lines 277–284).
Secondly, we have removed all statements claiming that future DUCMASS increases are "primarily controlled by climate-driven strengthening of Mongolian cyclone activity". Such statements have been replaced with "changes in large-scale circulation" or "projected changes in wind speed and PGI". We now state that wind speed is the dominant driver of dust activity, and that its projected changes are consistent with large-scale circulation changes associated with Mongolian cyclone variability, while noting that future cyclone behaviour itself was not explicitly diagnosed.
Finally, because we modified the calculation method, the predicted values differ significantly from the previous ones, and modifications have been made in the main text, as follows:
Based on historical data, CMIP6 future scenarios (SSP1-2.6, SSP2-4.5, SSP5-8.5) were used to project winter-spring DUCMASS for 2026–2100 (Fig. 6d–f). The projected ensemble-mean DUCMASS ranges from 0.11 to 0.42 g·m-2 over the Alxa Plateau, 0.08 to 0.30 g·m-2 over the Mu Us Sandy Land, and 0.17 to 1.14 g·m-2 over the Taklimakan Desert, indicating an increasing trend in all three regions. The largest increase occurs over the Taklimakan Desert, consistent with the dominant role of wind speed in this region (67.0% of RF feature importance) and a relatively large vegetation contribution (23.0%) that remains below the effective suppression threshold. Over the Alxa Plateau, wind speed also dominates (66.2%), but precipitation (21.6%) and PGI (7.5%) contribute more strongly, partially modulating the wind-driven enhancement. The Mu Us Sandy Land shows the smallest increase, consistent with the stronger roles of precipitation (22.4%) and PGI (6.4%) in this region. Wind speed is the primary control on future DUCMASS, but its effect is modulated by vegetation greening (up to 13.2% per decade under SSP5-8.5, Table S4) and precipitation changes, which partially offset the wind-driven enhancement in the Alxa Plateau and Mu Us Sandy Land but not in the Taklimakan Desert (Relevant text: pp. 19–20, lines 542–552).
Our results differ from Zhao et al. (2023), who projected a future decline in dust activity (0.8%–11.9%), mainly due to differences in seasonal focus and the representation of large-scale circulation changes. While Zhao et al. (2023) analyzed annual mean changes, our study focuses on winter-spring, the peak dust season. Under SSP1-2.6, winter-spring wind speed increases by 1.3–2.4% per decade across the three regions, whereas under SSP5-8.5, it decreases by 0.2–0.8% per decade. Meanwhile, LAI and precipitation increase under SSP5-8.5 (8.0–13.2% and 2.8–5.4% per decade, respectively), which may partially offset wind-driven increases in dust loading over the Alxa Plateau and Mu Us Sandy Land (Table S4). Given the dominant role of wind variability in controlling dust activity in our study, the projected DUCMASS increases likely reflect the combined effects of large-scale atmospheric circulation changes and concurrent land-surface responses, rather than changes in emission pathways alone. Although future cyclone intensity, frequency, and track position were not explicitly diagnosed from CMIP6 simulations, their potential influence is indirectly represented through projected changes in wind speed and PGI. Therefore, future dust projections should be interpreted as climate-driven potential changes arising from interactions between atmospheric circulation and land-surface conditions, rather than as precise estimates of future dust emissions (Relevant text: pp. 20, lines 553–565).
Response to RC1-8:
Response: We thank the reviewer for this important comment. We agree that the Taklimakan should not be treated as a homogeneous source region, and that the spatial distribution of dust activity and wind fields within the basin needs to be explicitly shown.
We have added a spatial distribution map of DUCMASS and wind fields over the Taklimakan in Fig. S1. g and h. The interior of the Taklimakan Desert exhibits internal spatial heterogeneity in both wind speed and DUCMASS (Fig. S1 g, h). Both wind speed and DUCMASS are higher in the eastern region, with maximum values of 8.6 m·s-1 and 0.8909 g·m-2, respectively. The prevailing wind direction is northeasterly, with northwesterly winds confined to a small western subregion. Over the 25-year period, the increasing trend in DUCMASS is stronger in the central and western regions and weaker in the east. Nevertheless, the basin-scale averages used in our analysis still capture the overall trend reversal around 2019, regardless of this internal variability, and the key contrast with the Alxa Plateau-between the terrain-shielded zone and the cyclone-affected zone-remains significant (Relevant text: pp. 13, lines 372–378).
Response to RC1-9:
Response: We thank the reviewer for this important comment. We agree that our analyses are primarily observational and statistical, and that causal language was in places stronger than the evidence supports. We have systematically revised the Abstract, Results, Discussion, and Conclusions to distinguish association, prediction, and physical causation. Firstly, we have already changed the title to "Mongolian cyclone variability modulates spatially divergent dust responses across contrasting source regions in northwest China". Secondly, causal verbs such as “reshapes”, “governs”, “drives”, and “outweighs” have been replaced with associational or contributory wording such as “is associated with”, “is linked to”, “contributes to”, and “may play a larger role than previously recognized”. Thirdly, statements that wind is a “direct reflection” of cyclone forcing have been revised to describe wind as a surface expression of large-scale circulation variability. Thirdly, claims that large-scale circulation “outweighs” land-surface controls have been revised to state that circulation variability is the dominant statistical contributor, without asserting physical dominance. Also, we have added a limitation statement noting that a full causal attribution would require numerical simulations with and without cyclone forcing, which we identify as future work (Relevant text: pp. 1, lines 1–2; pp. 1, lines 21–22; pp. 2, lines 52; pp. 11, lines 342; pp. 12, lines 368–370; pp. 20 , lines 567; pp. 21, lines 582; pp. 22, lines 605–607).
Response to RC1-10：
Response: Thank you for your suggestion. In this paper, the term refers to the period since 2021. We have revised "recent years" to "Since 2021" in the manuscript (Relevant text: pp. 1, lines 13).
Response to RC1-11:
Response: We thank the reviewer for this comment. We have revised the sentence so that the knowledge gap now covers both land-surface changes and circulation changes. Specifically, we now state that previous studies have largely attributed dust variability to land-surface changes and broad-scale circulation anomalies, but the relative contribution of these two sets of controls, and the spatially divergent responses of different dust-source regions to circulation changes, remain poorly understood (Relevant text: pp. 1, lines 15–17).
Response to RC1-12:
Response: We thank the reviewer for this comment. The Taklimakan Desert is included as a topographically enclosed end-member for comparison with the exposed Alxa Plateau, which is important for evaluating how land-surface conditions modulate the dust response to circulation. We have clarified this in the Abstract by describing the three regions as contrasting dust-source settings (Relevant text: pp. 1, lines 18–19).
Response to RC1-13:
Response: We thank the reviewer for this check. Our objective cyclone tracking shows no statistically significant southward shift of Mongolian cyclone tracks, consistent with Mu and Fiedler (2025). We have therefore removed the “southward shift” from the Abstract and now describe the cyclone changes only in terms of frequency and intensity (Relevant text: pp. 1, lines 20; pp. 10, lines 295–296; pp. 12, lines 364; pp. 18, lines 510; pp. 19, lines 532; pp. 21, lines 572).
Response to RC1-14:
Response: We thank the reviewer for this comment. Mongolia is introduced because it is the upstream source region for a substantial fraction of dust affecting China, and Mongolian cyclones are the primary synoptic system that transports this dust into China. The increasing Mongolian contribution therefore directly motivates our focus on Mongolian cyclone variability as a key control on dust activity within China. We have revised the passage to make this link explicit (Relevant text: pp. 2, lines 48–50).
Response to RC1-15:
Response: Thank you for your suggestion. We have revised this passage accordingly, as detailed below:
China accounted for nearly 75% of East Asian dust emissions during spring 2018–2020 (Zhang et al., 2024). However, a recent study based on source apportionment suggested that Mongolia has become an increasingly dominant source, contributing about 43% of East Asian dust in the early 2000s, about 53% in recent years, and as much as 62% in 2022 (Xing et al., 2025). Anomalous warming in the northwest Atlantic combined with eastern Pacific cooling (La Niña) has reduced precipitation and dried soils in Mongolia, enhancing dust emissions (An et al., 2018; Hu et al., 2023). Because this Mongolian dust is transported into China mainly by Mongolian cyclones, this increasing Mongolian contribution provides the motivation for focusing on Mongolian cyclone variability as a key control on dust activity within China. Approximately 88% of severe dust storms in China are linked to Mongolian cyclones (Gao et al., 2024), a proportion much higher than that associated with cold fronts or other systems (Wu et al., 2016; Yu et al., 2023). Mongolian cyclones are the dominant synoptic system driving severe dust storms in China, but their influence on dust activity varies across source regions depending on topography and regional circulation (An et al., 2018; Luo et al., 2026). Since 2021, the intensity and frequency of dust storms in China have risen noticeably (Filonchyk, 2022; Wu et al., 2025). These contrasting findings raise a central question: how the Mongolian cyclone activity contributes to the spatial variations of dust storm occurrence in dust sources in China (Relevant text: pp. 2, lines 40–53).
Response to RC1-16:
Response: We thank the reviewer for this comment. We agree that land-surface properties and Mongolian cyclone activity should not be framed as two equivalent controls. We have revised the sentence so that cyclone activity is presented as the primary driver, while land-surface conditions are described as modulators of its surface expression, as detailed below:
Although dust occurrence is fundamentally linked to Mongolian cyclone activity (Yu et al., 2023), its surface expression is strongly modulated by land-surface conditions (Pu and Jin, 2021; Wu et al., 2022; Zan et al., 2025) (Relevant text: pp. 2, lines 54–55).
Response to RC1-17:
Response: Thank you for your suggestion. We have removed the relevant statements (Relevant text: pp. 2, lines 55).
Response to RC1-18:
Response: We thank the reviewer for this comment. We agree that dust storm increases and vegetation greening are not directly contradictory, because they reflect different processes. We have removed the word “contradicts” and now explain that vegetation mainly suppresses local dust emission, whereas severe dust storms are linked to large-scale circulation that can transport dust from remote sources such as the Gobi and Mongolian Plateau. The post-2021 resurgence therefore suggests that circulation-driven transport may have outweighed local vegetation suppression, which directly motivates our focus on Mongolian cyclone activity (Relevant text: pp. 3, lines 65–70).
Response to RC1-19:
Response: We thank the reviewer for this comment. We agree that the final sentence did not follow logically from the preceding discussion of environmental controls. We have revised it so that it now links directly to the preceding content: after reviewing the main environmental controls on dust, the text now notes that most of these controls have been examined for individual dust storm events rather than over multi-decadal time scales, and that their relative importance over 2001–2025 remains unclear. This provides a direct transition to the objectives of the present study (Relevant text: pp. 3, lines 81–83).
Response to RC1-20:
Response: We thank the reviewer for this comment. We have replaced the vague phrase “Building on this foundation” with an explicit reference to the preceding evidence, namely that Mongolian cyclone activity is a primary driver of severe dust storms in China and that its influence may differ among dust source regions (Relevant text: pp. 3, lines 84–85).
Response to RC1-21:
Response: We thank the reviewer for this comment. The Mu Us Sandy Land is now explicitly included as one of the three study regions, consistent with its earlier introduction. Regarding the Mongolian Plateau, we have clarified that it is an important upstream dust source more directly exposed to Mongolian cyclone activity, but that the present study focuses on dust-source regions within northwest China and does not aim to quantify the complete upstream cyclone-dust-source pathway. Incorporating Mongolian dust-source observations and event-based cyclone tracking is identified as a direction for future work (Relevant text: pp. 3, lines 91–92; pp. 3–4, lines 99–102).
Response to RC1-22:
Response: We thank the reviewer for this comment. The second hypothesis is our own and should not be attributed to a previous study. We have removed the citation to Yu et al. (2023) from this sentence. We have also clarified the comparison as being between dust-source interiors and their surrounding areas (Relevant text: pp. 3, lines 88–89).
Response to RC1-23:
Response: We thank the reviewer for this comment. We agree that the basic physical controls on dust emission are relatively well established. We have therefore narrowed the statement: instead of referring to a broad “significant controversy”, we now state that the mechanisms driving the post-2021 dust resurgence and the spatial differences in dust responses across source regions remain debated (Relevant text: pp. 3, lines 89–91).
Response to RC1-24:
Response: Thank you for your comment. We have checked the formatting here (Relevant text: pp. 4, lines 105).
Response to RC1-25:
Response: We thank the reviewer for this comment. The “four dust-storm centers” refer to the traditional source centers in northern China from previous studies (Liu et al., 2020, 2025): the Hexi Corridor, the Yinshan-Hushandake area, the area along the Great Wall, and the Shanxi-Ningxia region. These describe the climatological distribution of dust storms and are not the same as the four analytical subregions used here. We have added the regionalization and its basis to the manuscript, and have drawn a map of the study area, as detailed below:
The study area is located in northwest China (Fig. 1). Most areas in northwest China are arid and semi-arid regions, and dust events occur frequently in spring (Liu et al., 2020). Previous studies have identified four major dust-storm centers in northern China: the Hexi Corridor in Gansu Province, the northern slope of the Yinshan Mountains together with the Hunshandake. While useful for characterizing the spatial distribution of dust emissions, they are not designed to resolve the full source-transport-sink chain of dust activity. Because the present study aims to examine the spatially divergent response of dust activity to Mongolian cyclone variability along the entire transport pathway, we reorganize the domain into four subregions according to synoptic exposure, position in the dust source-transport-sink chain, and data availability: Source 1 (Taklimakan Desert; 6 cities), Source 2 (Alxa Plateau and Mu Us Desert; 8 cities), the transport pathway (Hexi Corridor; 5 cities), and the downwind urban areas (8 cities; Fig. 1, Fig. S1a). Distinct wind regimes exist across these subregions. The Tarim Basin generally exhibits northeasterly near surface winds during the study period, although mesoscale and thermally driven circulations can locally modify wind patterns, while northwesterly winds prevail over the remaining areas (Fig. S1a, b, and d). Such wind field differences fundamentally determine the regional dust transport routes (Mu and Fiedler, 2025). These regions represent a gradient of exposure to cyclone-related circulation, from directly influenced open terrain to topographically sheltered basins (Relevant text: pp. 4, lines 105–121).
Response to RC1-26:
Response: We thank the reviewer for this suggestion. We agree that the observational evidence for the dust rebound should be clearly presented before it is linked to circulation changes. We have revised the manuscript accordingly.
Firstly, we removed the forward reference to the “recent rebound of dust storm activity” from the opening of Section 3.1. Section 3.1 now objectively describes the circulation anomalies and the objective cyclone-tracking results, without presupposing a dust response (Relevant text: pp. 10, lines 291).
Secondly, we added a bridging statement at the end of Section 3.1 to explicitly indicate that such atmospheric circulation provides an atmospheric background for dust variations (Relevant text: pp. 12, lines 348–351).
Thirdly, we revised the opening of Section 3.2 so that the post-2019 dust rebound is first established there, using PM10, DUCMASS, and DSD, before it is interpreted in relation to the circulation changes (Relevant text: pp. 12, lines 353).
We have retained the order of Sections 3.1 and 3.2 because this order follows a “forcing-response” logic: Section 3.1 first establishes the large-scale circulation and objectively tracked cyclone variability, and Section 3.2 then examines whether and how dust activity responded to these changes. Presenting the circulation background first allows the subsequent dust changes to be evaluated against a well-defined synoptic context, rather than interpreting dust variability without knowing the associated atmospheric conditions. We believe this structure is logically coherent and, with the forward reference removed, no longer introduces the dust rebound before its observational evidence.
Response to RC1-27:
Response: We thank the reviewer for this constructive suggestion. We acknowledge that positive relative vorticity indicates cyclonic circulation but does not necessarily identify a cyclone centre, and that Mu and Fiedler (2025) reported a different trend using an explicit tracking algorithm.
Firstly, the specific cyclone identification and tracking methods have been supplemented in the Methods section, as follows: Cyclone identification and tracking. We identified and tracked Mongolian cyclones using 6-hourly ERA5 mean sea-level pressure (MSLP) fields during 2001–2025. Cyclone centers were detected as local minima in the MSLP field that were at least 4.0 hPa lower than the surrounding environment (computed as a 15×15 grid-point background) and with central pressure below 1006 hPa. Cyclones were tracked using a nearest-neighbor algorithm: at each time step, each active track was matched to the nearest detected center within a search radius of 600 km, with a maximum allowed displacement of 700 km between consecutive time steps. Tracks with a lifetime shorter than 18 hours, a total displacement less than 2.5°, or a predominantly westward movement were excluded. This procedure yielded 27–63 cyclone tracks per spring season, with a mean of approximately 42 tracks per year, consistent with previous cyclone-tracking studies over this region (Mu and Fiedler, 2025). Cyclone intensity was defined as the minimum central pressure along each track, and the annual mean intensity was calculated as the mean of all track minimum pressures in each year. Cyclone candidates were restricted to the Mongolian Plateau and adjacent regions (90–130 °E, 40–58 °N), where cyclones affecting northern China dust events are typically generated or intensified (Relevant text: pp. 6–7, lines 182–193).
Secondly, we revised the interpretation of relative vorticity: We now present relative vorticity anomalies as an indicator of large-scale cyclonic circulation rather than as a direct identification of cyclone centers. The revised text reads: "A positive relative vorticity anomaly intensified from 3.18 × 10-5 s-1 (at 106.5°E, 48.8°N) to 11.90 × 10-5 s-1 (at 95.0°E, 44.5°N, Fig. 2b), indicating enhanced cyclonic circulation over the Mongolia-Gobi region. However, relative vorticity anomalies reflect large-scale circulation conditions rather than objectively identified cyclone centers." (Relevant text: pp. 10, lines 293–296)
Lastly, The objective tracking results show that cyclone frequency increased after 2019 and cyclone intensity strengthened after 2019, but the mean latitude shows no statistically significant southward trend. This is consistent with Mu and Fiedler (2025).
Response to RC1-28, 29:
Response: We thank the reviewer for this comment. The values cited in the text were originally derived from the upper tail of the wind speed difference distribution rather than from the region mean, which we agree is not appropriate for describing region-average changes. We have therefore recalculated the values as region-averaged differences in near-surface wind speed between 2021–2025 and 2016–2020, using ERA5 10 m wind fields. The details are as follows:
Correspondingly, region-averaged near-surface wind speeds during 2021–2025 were 0.32 m·s-1 higher over the Hexi Corridor and 0.63 m·s-1 higher over the Alxa Plateau than during 2016–2020. These increments are about 2–3 times greater than the increase (+0.17 m·s-1) over the Taklimakan Desert (Fig. S1c), highlighting a distinct spatial divergence in wind responses to cyclone activity (Relevant text: pp. 11, lines 321–324).
Response to RC1-30:
Response: We thank the reviewer for this careful check. Mu and Fiedler (2025) do not directly demonstrate that topographic shielding weakens Mongolian cyclone influence on the Taklimakan. They note that the mechanisms require further investigation, and they identify low-level jet breakdown as another possible driver.
We have therefore removed the claim that Mu and Fiedler (2025) demonstrated a topographic shielding effect. The revised text now states that Mongolian cyclones can still influence dust activity in the Taklimakan through an indirect pathway. Cold-air intrusion in the cyclone wake triggers nocturnal low-level jets, whose breakdown enhances near-surface winds and promotes dust mobilization and increases atmospheric dust loading (Mu and Fiedler, 2026) (Relevant text: pp. 11, lines 333–336).
We have also checked the entire manuscript for consistency and replaced all instances of “topographic shielding” and “topographically sheltered” with more neutral wording (“topographic modulation”, “topographically enclosed”). The Abstract, Sections 3.1, 3.2, 3.4, and the Conclusions have been revised accordingly.
Response to RC1-31:
Response: We thank the reviewer for this comment. The Arctic forcing and dust-radiation feedback are the macroscopic background of the Mongolian cyclone. We agree that they are not directly evaluated in this study, and that presenting them as part of the main analysis was inappropriate. So we have revised this passage in place. Specifically:
The interannual and decadal variability of Mongolian cyclone activity has been linked to Arctic thermal anomalies in previous studies. Rapid Arctic warming and declining Barents sea ice can polar vortex and generate circulation anomalies favourable for cyclogenesis over the Mongolian-Lake Baikal region, potentially increasing the frequency and strength of Mongolian cyclones and contributing to dust events over Northwest China (Wang et al., 2025a; Yin et al., 2022). In turn, radiative heating by suspended dust aerosols has also been proposed to modify regional horizontal temperature gradients and to intensify Mongolian cyclones, forming a possible positive feedback (Chen et al., 2023). These large-scale processes provide background context for the cyclone–dust relationship examined here. To quantify the relative contributions and nonlinear responses of multiple environmental drivers to dust activity, we next examine the statistical relationships between PM10 concentrations and a suite of meteorological and land surface variables. Such atmospheric circulation provides an atmospheric background for dust variations (Relevant text: pp. 12, lines 342–351).
Response to RC1-32:
Response: We thank the reviewer for this important comment. We agree that the statement "the entire Tarim Basin is dominated by northeasterly winds" was too general, and that the northern rim of the basin is an active dust-source region with different wind regimes.
We have revised the overly general statement:
However, the interior of the Taklimakan Desert exhibits internal spatial heterogeneity in both wind speed and DUCMASS (Fig. S1g and h). Both wind speed and DUCMASS are higher in the eastern region, with maximum values of 8.6 m·s-1 and 0.8909 g·m-2, respectively. The prevailing wind direction is northeasterly, with northwesterly winds confined to a small western subregion. Over the 25-year period, the increasing trend in DUCMASS is stronger in the central and western regions and weaker in the east. Nevertheless, the basin-scale averages used in our analysis still capture the overall trend reversal around 2019, regardless of this internal variability, and the key contrast with the Alxa Plateau (between the terrain shielded zone and the cyclone affected zone) remains significant (Relevant text: pp. 13, lines 372–378).
Response to RC1-33:
Response: We thank the reviewer for this comment. We have tested the statistical significance of the inter-period changes for both PM10 concentrations and dust storm days (DSD), using a linear mixed-effects model with city as a random effect and verifying the results with a paired t-test across cities. The details are as follows:
After 2021, the Tarim Basin (Source 1) experienced a 6.28% (P = 0.35) decline in PM10 concentrations (Fig. 3a; Fig. S3a). In contrast, notable rises were detected in Source 2 (+17.68%, P ≤ 0.01), transport pathways (+24.22%, P ≤ 0.01) and downwind cities (+3.29%, P = 0.48; Fig. 3a; Fig. S3b–d) (Relevant text: pp. 12, lines 360–362).
Ground-based PM10 observations for 2021–2025 reveal that compared with the 2016–2020 period, dust storm days (DSD) increased sharply in Source 2 (+49.68%, P ≤ 0.01), transport pathways (+37.32%, P ≤ 0.01) and downwind urban areas (+30.42%, P ≤ 0.05), while a decrease of -9.91% (P = 0.18) was observed over the Taklimakan Desert (Fig. 3c; Fig. S4a–d) (Relevant text: pp. 13, lines 383–386).
Response to RC1-34:
Response: We agree with your idea; therefore, we conducted a study on the interior of the Taklimakan, as detailed below:
However, the interior of the Taklimakan Desert exhibits internal spatial heterogeneity in both wind speed and DUCMASS (Fig. S1g and h). Both wind speed and DUCMASS are higher in the eastern region, with maximum values of 8.6 m·s-1 and 0.8909 g·m-2, respectively. The prevailing wind direction is northeasterly, with northwesterly winds confined to a small western subregion. Over the 25-year period, the increasing trend in DUCMASS is stronger in the central and western regions and weaker in the east. Nevertheless, the basin-scale averages used in our analysis still capture the overall trend reversal around 2019, regardless of this internal variability, and the key contrast with the Alxa Plateau (between the terrain shielded zone and the cyclone affected zone) remains significant (Relevant text: pp. 13, lines 372–378).
-
AC2: 'Reply on RC1', Wenhui Zhang, 15 Oct 2026
-
CC1: 'Comment on egusphere-2026-3940', Guoming Zhang, 19 Aug 2026
- General comments：
The manuscript investigates the spatially divergent effects of the intensification and southward shift of Mongolian cyclones on dust activity in northwest China. It uses ground-based PM₁₀ observations, MERRA-2, ERA5, MODIS, land-use, and CMIP6 data, with the multiple data sources providing mutual corroboration. Trend analysis, segmented regression, and random forest models are combined to compare the Taklimakan Desert, Alxa Plateau, and Mu Us Sandy Land. The topic is of practical significance. Overall, the research framework is generally sound, and the main results are reasonably clear. I recommend that the manuscript be considered for publication after minor revision.
The manuscript should be further focused on its central theme. In Section 3.1, the discussion of Arctic warming, Barents Sea ice, polar-vortex perturbations, and dust-radiative feedback extends beyond the processes directly analyzed in this study. The current wording may lead readers to interpret these mechanisms as findings derived from and validated by the present study. I recommend shortening this paragraph and clearly identifying these processes as possible background mechanisms proposed in the literature, or moving the paragraph to the end of the Discussion as a direction for future research.
Key methodological details should be added, particularly regarding the processing of multisource datasets. The manuscript does not adequately describe the criteria used to identify Mongolian cyclones, the methods used to determine cyclone centers and calculate cyclone intensity, the definition of the PGI, the source and basic attributes of the NOAA dust-storm-day data, or the procedures used to train and validate the random forest model.
- Specific comments:
Line 52: “The Alxa Plateau” and “the Mu Us Sandy Land” are two distinct geographical regions. The expression “the Alxa Plateau (including Mu Us Sandy Land)” may incorrectly imply that one encompasses the other, and the two regions are analyzed separately later in the manuscript. The wording should therefore be standardized as “the Alxa Plateau and the Mu Us Sandy Land.”
Line 69: This paragraph refers to the study period as the “past 45 years,” whereas the main dataset analyzed in the manuscript covers 2001-2025, a period of only 25 years. This should be corrected.
Lines 84-88: The description of the study area is too brief. The authors should provide the latitudinal and longitudinal extent of the study area, the boundaries of each subregion, the number of monitoring stations, and the criteria used for regional classification. These features could be clearly presented in a study-area map.
Lines 90-94: The ground-based PM₁₀ data cover 2016-2025, whereas the caption of Fig. 2C (Lines 229-231) states that the data cover 2001-2025. This is an obvious inconsistency and should be verified and corrected. The authors should also describe the treatment of missing station data, quality-control procedures, calculation of daily mean values, and criteria used to identify valid observation days.
Lines 181-185, 219-224, and 389-392: The Results describe the change in cyclone latitude as occurring “after 2020,” whereas the Abstract and Conclusions state “after 2019.” These expressions should be standardized. Alternatively, the authors should clarify that 2019 represents the statistical breakpoint, whereas 2020 marks the onset of the pronounced southward shift.
Fig. 5 and Lines 352-370: I recommend replacing the three-dimensional pie charts in panels A-C with two-dimensional charts, as the three-dimensional presentation is not effective. For panels D-F, the multi-model ensemble mean, model spread, or 95% confidence intervals should be provided. The figure caption should also clarify whether the vertical axis represents spring mean or annual mean PM₁₀ and whether the curves represent the three-model ensemble results.
Line 395: The precipitation threshold is reported as 144 mm in the Conclusions but as 143 mm in the main text (Line 284). These values should be made consistent.
- Technical corrections:
Line 443: “Inner Monglia” should be corrected to “Inner Mongolia.” The authors should also check the entire manuscript for consistent formatting of PM₁₀, m·s⁻¹, g·m⁻², m³·m⁻³, and all subscripts and superscripts.
Citation: https://doi.org/10.5194/egusphere-2026-3940-CC1 -
AC4: 'Reply on CC1', Wenhui Zhang, 15 Oct 2026
Specific comments:
Response to CC1-1:
Response: We thank the reviewer for this correction. The Alxa Plateau and the Mu Us Sandy Land are indeed two distinct geographical regions, and the original wording incorrectly implied that one encompasses the other. We have revised the sentence to present them as separate regions, consistent with the separate analyses later in the manuscript. The revised sentence now reads: The Taklimakan Desert, Alxa Plateau, the Hexi corridor, and Mu Us Sandy Land provide a useful comparative framework for examining the spatial heterogeneity of dust responses to large-scale atmospheric circulation variability. The Taklimakan Desert is located within the enclosed Tarim Basin and is surrounded by major mountain systems, which can substantially modify the transmission of synoptic-scale circulation to the surface and generate spatially heterogeneous wind regimes. Mongolian cyclones can still influence this basin, but primarily through indirect pathways such as cold-air intrusion and low-level jet breakdown. In contrast, the Alxa Plateau, located along the southern margin of the Mongolian Plateau, has relatively open terrain and is more directly exposed to synoptic-scale northwesterly circulation associated with Mongolian cyclone activity than the topographically enclosed Taklimakan Basin. The Hexi Corridor is a channel for sediment transportation. The Mu Us Sandy Land represents a transitional dust-source region where large-scale atmospheric circulation interacts with different hydrothermal and vegetation conditions (Han et al., 2025; Yang et al., 2014) (Relevant text: pp. 2, lines 55–65).
Response to CC1-2:
Response: Thank you for your comment. We have revised "past 45 years" to "over the 2001–2025 period." (Relevant text: pp. 3, lines 82)
Response to CC1-3:
Response: We thank the reviewer for this suggestion. We have added a study-area map (Fig. 1) showing the geographical extent, the distribution of monitoring cities, and the four subregions. In addition, we have described in the text the criteria used for regional classification, the number of monitoring cities in each subregion, and the constituent cities of each subregion, so that the information is available independently of the figure. The details are as follows.
The study area is located in northwest China (Fig. 1). Most areas in northwest China are arid and semi-arid regions, and dust events occur frequently in spring (Liu et al., 2020). Previous studies have identified four major dust-storm centers in northern China: the Hexi Corridor in Gansu Province, the northern slope of the Yinshan Mountains together with the Hunshandake. While useful for characterizing the spatial distribution of dust emissions, they are not designed to resolve the full source-transport-sink chain of dust activity. Because the present study aims to examine the spatially divergent response of dust activity to Mongolian cyclone variability along the entire transport pathway, we reorganize the domain into four subregions according to synoptic exposure, position in the dust source-transport-sink chain, and data availability: Source 1 (Taklimakan Desert; 6 cities), Source 2 (Alxa Plateau and Mu Us Desert; 8 cities), the transport pathway (Hexi Corridor; 5 cities), and the downwind urban areas (8 cities; Fig. 1, Fig. S1a). Distinct wind regimes exist across these subregions. The Tarim Basin generally exhibits northeasterly near surface winds during the study period, although mesoscale and thermally driven circulations can locally modify wind patterns, while northwesterly winds prevail over the remaining areas (Fig. S1a, b, and d). Such wind field differences fundamentally determine the regional dust transport routes (Mu and Fiedler, 2025). These regions represent a gradient of exposure to cyclone-related circulation, from directly influenced open terrain to topographically sheltered basins (Relevant text: pp. 4, lines 105–121).
Response to CC1-4:
Response: We thank the reviewer for catching this inconsistency. The ground-based PM10 data cover 2016–2025, and the caption of Fig. 2C has been corrected accordingly. The 2001–2025 period applies only to the MERRA-2 DUCMASS and NOAA ISD datasets, not to the CNEMC PM10 data.
We have also expanded the data description in Section 2.2 to include the quality-control procedure, the treatment of missing values, the calculation of daily means, and the criteria used to define valid observation days. Specifically, hourly PM₁₀ values were first screened for negative values and values exceeding the instrument upper limit; daily means were calculated only when at least 18 valid hourly records were available for that day; and days with fewer than 18 valid hours were treated as missing. Station-years with fewer than 300 valid days were excluded from the analysis. These criteria follow the standard practice for CNEMC data processing and are now described in the revised manuscript (Relevant text: pp. 5, lines 125–132).
Response to CC1-5:
Response: We thank the reviewer for pointing out this inconsistency. We have standardized the terminology throughout the manuscript. The year 2019 is used as the statistical breakpoint in the segmented regression analysis, and all references to “after 2020” have been changed to “after 2019” for consistency.
We have also re-examined the cyclone latitude results. Our objective tracking shows no statistically significant trend or trend reversal in cyclone latitude (F-test P = 0.478). We have therefore removed the claim of a “pronounced southward shift” and only focuses on the spatial differences in the impact of the Mongolian cyclone on the northwest region of China.. We added a cyclone-active versus cyclone-inactive year composite analysis to examine whether dust responses differ along this gradient of cyclone influence. Over the Mongolia-Gobi region (cyclone genesis and direct impact zone), DUCMASS is significantly higher in active years than in inactive years (+20.7%, P ≤ 0.05; Fig. 2d). Over the Alxa Plateau (downstream of cyclone tracks), DUCMASS is also significantly higher in active years (+12.1%, P ≤ 0.05; Fig. 2e). In contrast, over the topographically sheltered Taklimakan Desert, DUCMASS is slightly lower in active years (-6.1%, P > 0.05; Fig. 2f), though this change is not statistically significant. These results indicate that cyclone activity is associated with enhanced atmospheric dust loading primarily in exposed regions, whereas the Taklimakan exhibits a weaker response likely due to the combined effects of topographic constraints, basin-scale circulation, and local mesoscale processes (Relevant text: pp. 11, lines 314–320).
The methods are as follows: Cyclone-active vs inactive year classification. To test whether cyclone activity modulates dust emissions in a spatially divergent manner, we classified years based on the mean intensity of tracked cyclones (defined as the mean minimum central pressure of all cyclone tracks in each year). Years with the strongest 40 % of cyclone intensity (lowest central pressure) were designated as "cyclone-active years" and years with the weakest 40 % were designated as "cyclone-inactive years". The remaining 20% of years were excluded from the comparison. The 40 % threshold was selected as a compromise between maximizing cyclone intensity contrast and retaining sufficient sample size for statistical comparison. This yielded 10 active years (2001, 2003–2007, 2009, 2012, 2021, 2023) and 10 inactive years (2008, 2010, 2011, 2015–2017, 2019, 2020, 2022, 2024). We defined three regions representing a gradient of cyclone influence: the Mongolia-Gobi region (cyclone genesis and direct impact zone), the Alxa Plateau (downstream of cyclone tracks), and the Taklimakan Desert (topographically sheltered, weak cyclone influence). For each region, we extracted spring DUCMASS from MERRA-2 and calculated regional means. We then compared regional mean DUCMASS between active and inactive years using a two-sample t-test and the non-parametric Mann-Whitney U test (Relevant text: pp. 7, lines 194–205).
Response to CC1-6:
Response: We thank the reviewer for these constructive suggestions. We have revised Fig. 5 and its caption accordingly.
Firstly, we have replaced the three-dimensional pie charts in the Fig. 5 A–C with two-dimensional charts. In Fig 5D–F, we added 95% confidence intervals and stated in the figure captions that these are DUCMASS values for winter and spring, because we modified the calculation method, the predicted values differ significantly from the previous ones, and modifications have been made in the main text, as follows:
Based on historical data, CMIP6 future scenarios (SSP1-2.6, SSP2-4.5, SSP5-8.5) were used to project winter-spring DUCMASS for 2026–2100 (Fig. 6d–f). The projected ensemble-mean DUCMASS ranges from 0.11 to 0.42 g·m-2 over the Alxa Plateau, 0.08 to 0.30 g·m-2 over the Mu Us Sandy Land, and 0.17 to 1.14 g·m-2 over the Taklimakan Desert, indicating an increasing trend in all three regions. The largest increase occurs over the Taklimakan Desert, consistent with the dominant role of wind speed in this region (67.0% of RF feature importance) and a relatively large vegetation contribution (23.0%) that remains below the effective suppression threshold. Over the Alxa Plateau, wind speed also dominates (66.2%), but precipitation (21.6%) and PGI (7.5%) contribute more strongly, partially modulating the wind-driven enhancement. The Mu Us Sandy Land shows the smallest increase, consistent with the stronger roles of precipitation (22.4%) and PGI (6.4%) in this region. Wind speed is the primary control on future DUCMASS, but its effect is modulated by vegetation greening (up to 13.2% per decade under SSP5-8.5, Table S4) and precipitation changes, which partially offset the wind-driven enhancement in the Alxa Plateau and Mu Us Sandy Land but not in the Taklimakan Desert (Relevant text: pp. 19–20, lines 542–552).
Response: We thank the reviewer for these constructive suggestions. We have revised Fig. 5 and its caption accordingly.
Firstly, we have replaced the three-dimensional pie charts in the Fig. 5 A–C with two-dimensional charts. In Fig 5D–F, we added 95% confidence intervals and stated in the figure captions that these are DUCMASS values for winter and spring, because we modified the calculation method, the predicted values differ significantly from the previous ones, and modifications have been made in the main text, as follows:
Based on historical data, CMIP6 future scenarios (SSP1-2.6, SSP2-4.5, SSP5-8.5) were used to project winter-spring DUCMASS for 2026–2100 (Fig. 6d–f). The projected ensemble-mean DUCMASS ranges from 0.11 to 0.42 g·m-2 over the Alxa Plateau, 0.08 to 0.30 g·m-2 over the Mu Us Sandy Land, and 0.17 to 1.14 g·m-2 over the Taklimakan Desert, indicating an increasing trend in all three regions. The largest increase occurs over the Taklimakan Desert, consistent with the dominant role of wind speed in this region (67.0% of RF feature importance) and a relatively large vegetation contribution (23.0%) that remains below the effective suppression threshold. Over the Alxa Plateau, wind speed also dominates (66.2%), but precipitation (21.6%) and PGI (7.5%) contribute more strongly, partially modulating the wind-driven enhancement. The Mu Us Sandy Land shows the smallest increase, consistent with the stronger roles of precipitation (22.4%) and PGI (6.4%) in this region. Wind speed is the primary control on future DUCMASS, but its effect is modulated by vegetation greening (up to 13.2% per decade under SSP5-8.5, Table S4) and precipitation changes, which partially offset the wind-driven enhancement in the Alxa Plateau and Mu Us Sandy Land but not in the Taklimakan Desert (Relevant text: pp. 19–20, lines 542–552).
Lastly, regarding the methods, the curves represent the ensemble results of three models. The specific methods have been added to Section 2.3, as follows:
The same five predictors were extracted from three CMIP6 models (CESM2, MPI-ESM1-2-LR, and UKESM1-0-LL) under SSP1-2.6, SSP2-4.5, and SSP5-8.5. All CMIP6 predictors were bias-corrected against the historical mean of each predictor using mean-ratio scaling. The RF model was then applied year-by-year to generate winter-spring DUCMASS projections for 2026–2100, with uncertainty estimated as ±1 standard deviation across the three CMIP6 models and the lower bound truncated at zero because DUCMASS is physically non-negative. HAI was not included in the future projection because CMIP6 does not provide consistent land-use scenarios across the three models, and future cyclone intensity, frequency, and position were not explicitly diagnosed from CMIP6; their influence is represented indirectly through projected changes in wind speed and PGI (Relevant text: pp. 9, lines 277–284).
Response to CC1-7:
Response: Thank you for your comment. We have unified the precipitation threshold to 144 mm (Relevant text: pp. 16, lines 450).
Technical corrections:
Response to CC1-8:
Response: Thank you for your suggestion. We have corrected “Inner Monglian” to “Inner Mongolia,” and have checked the formatting of PM10, m·s-1, g·m-2, m3·m-3, and all other superscripts and subscripts throughout the manuscript (Relevant text: pp. 23, lines 631).
-
RC2: 'Comment on egusphere-2026-3940', Anonymous Referee #2, 30 Aug 2026
General comments:
This manuscript investigates how changes in Mongolian cyclone activity influence dust activity variability across Northwest China, with particular emphasis on the contrasting responses of the Taklimakan Desert, Alxa Plateau, and Mu Us Sandy Land. The regional comparison is interesting and may provide useful insight into the spatial heterogeneity of East Asian dust activity. However, several aspects of the methodology and interpretation require substantial clarification and revision. In particular, the definition and use of the Mongolian cyclone, dust-related indicators, and the consistency between the Methods and Results needs to be clarified. Therefore, I recommend Major Revision.
Major comments:
- Mongolian cyclone variability is the central component of this study, yet the Methods do not provide sufficient information on how Mongolian cyclones were identified and quantified. The manuscript reports changes in mean cyclone intensity, mean cyclone latitude, and a southward displacement, but it is unclear how individual cyclone events or centers were detected, how cyclone intensity was defined, and how these quantities were averaged over time. Please provide more details regarding the cyclone identification or tracking criteria, definition of cyclone center and intensity, and the procedure used to calculate the annual mean intensity and position. Without this information, the results concerning cyclone intensification and southward migration are difficult to reproduce and evaluate.
- PM10, DUCMASS, and DSD represent different aspects of dust activity and should not be treated interchangeably. The manuscript uses near-surface PM10 concentration, MERRA-2 dust column mass density (DUCMASS), and dust storm days (DSD) as indicators of “dust activity.” However, these variables represent fundamentally different aspects of the dust cycle: PM10 describes near-surface particulate concentration and may include non-dust aerosols, DUCMASS represents column-integrated mineral dust loading, and DSD represents event frequency based on a specific detection criterion. This distinction is particularly important because the Taklimakan Desert shows increasing DSD but decreasing DUCMASS. The manuscript should clearly define the physical meaning of each indicator and avoid treating them as interchangeable measures of dust activity. The apparently inconsistent trends among these indicators should also be discussed more carefully.
- There is an inconsistency between the PM10-based definition of dust storms and the PM10 values used to derive the reported thresholds. The Methods define a dust-storm occurrence as a day with PM10 > 500 μg m⁻³. However, the PM10 values used in the segmented-regression analyses in Fig. 3E, G, and H are predominantly within approximately 50–350 μg m⁻³, i.e., below the manuscript's dust-storm criterion. Nevertheless, the resulting breakpoints are subsequently interpreted as a “dust emission threshold” (4.63 m s⁻¹) or “dust suppression thresholds” for soil moisture and NDVI. These values appear instead to represent statistical breakpoints in relationships with PM10 concentration and should not automatically be interpreted as physical thresholds for dust-storm occurrence or dust emission. In addition, please clarify the PM10 metric used in each panel (daily, monthly, March mean, spring mean, or station-year mean), why the PM10 ranges differ among Fig. 3E, G, and H, and whether identical samples were used in these analyses.
- The response variable used in the Random Forest analysis is inconsistent between the Methods and Results. Section 2.3 states that Random Forest regression was used to quantify the relative contributions of environmental factors to DUCMASS. However, the caption of Fig. 5 and the corresponding results describe the RF importance values as contributions to PM10 concentrations. DUCMASS and PM10 are fundamentally different quantities, and therefore this inconsistency needs to be resolved.
- The methodology used to project future PM10 concentrations is insufficiently described. The manuscript states that three CMIP6 models under SSP1-2.6, SSP2-4.5, and SSP5-8.5 were used to project spring PM10 concentrations for 2026–2100. However, it is not explained how CMIP6 climate-model output was converted into PM10 concentrations. The manuscript also attributes future PM10 increases to strengthening Mongolian cyclone activity; therefore, it should be clarified whether future cyclone intensity, frequency, or position was explicitly diagnosed from the CMIP6 simulations.
Minor comments:
- Introduction: The relationship between PM10 and mineral dust should be better introduced. The Introduction mainly discusses dust and dust storms, but then shifts directly to PM10. I suggest that the authors briefly review the relationship between PM10 and dust and explain why PM10 is an appropriate proxy for dust activity in this study. The manuscript currently introduces PM10 mainly as a particulate-matter metric rather than establishing its relevance to mineral dust.
- Line 92: The manuscript defines a dust storm as a day with PM10 > 500 μg m⁻³. Please provide an appropriate reference or observational standard supporting this threshold and explain why this value was selected.
- Line 103: Please specify the soil-moisture layer used from ERA5 (e.g., 0-7 cm).
- Section 2.2: Please clarify how datasets with substantially different spatial resolutions were harmonized for the statistical analyses.
- Line 116: The definition and calculation of HAI should be provided when it is first introduced. The description currently presented around Lines 156–163 would be better moved to this part of the Methods.
- Line 153: PGI should be clearly defined before it is used in the Random Forest analysis.
- The manuscript alternates between “after 2019” and “after 2020” when describing the timing of the observed changes. Please clarify the breakpoint or transition year and use it consistently throughout the manuscript.
- Line 284 and Line 395: The manuscript uses both 143 mm and 144 mm for the preceding-year precipitation breakpoint. Please use one consistent value throughout the text and figures.
Citation: https://doi.org/10.5194/egusphere-2026-3940-RC2 -
AC3: 'Reply on RC2', Wenhui Zhang, 15 Oct 2026
Major comments:Response to RC2-1:Response: We thank the reviewer for this important comment. We agree that the Methods section lacked sufficient detail on cyclone identification, tracking, and quantification. In the revised manuscript, we have added a detailed description in Section 2.3, which includes:First, the cyclone identification criteria, methods, and definitions have been added to the Methods section, as follows: Cyclone identification and tracking. We identified and tracked Mongolian cyclones using 6-hourly ERA5 mean sea-level pressure (MSLP) fields during 2001–2025. Cyclone centers were detected as local minima in the MSLP field that were at least 4.0 hPa lower than the surrounding environment (computed as a 15×15 grid-point background) and with central pressure below 1006 hPa. Cyclones were tracked using a nearest-neighbor algorithm: at each time step, each active track was matched to the nearest detected center within a search radius of 600 km, with a maximum allowed displacement of 700 km between consecutive time steps. Tracks with a lifetime shorter than 18 hours, a total displacement less than 2.5°, or a predominantly westward movement were excluded. This procedure yielded 27–63 cyclone tracks per spring season, with a mean of approximately 42 tracks per year, consistent with previous cyclone-tracking studies over this region (Mu and Fiedler, 2025). Cyclone intensity was defined as the minimum central pressure along each track, and the annual mean intensity was calculated as the mean of all track minimum pressures in each year. Cyclone candidates were restricted to the Mongolian Plateau and adjacent regions (90–130 °E, 40–58 °N), where cyclones affecting northern China dust events are typically generated or intensified (Relevant text: pp. 6–7, lines 182–193).Second, the annual mean intensity is the average of the minimum central pressures of all tracks in each year. The annual mean latitude and longitude are the averages of the positions of all track points in each year.Response to RC2-2:Response: We thank the reviewer for this important comment. We fully agree that PM10, DUCMASS, and DSD represent fundamentally different aspects of the dust cycle and should not be treated as interchangeable measures of "dust activity". We have revised the manuscript to clearly define each metric, avoid interchangeable usage, and provide a more careful discussion of their apparently inconsistent trends over the Taklimakan Desert.Firstly, we have added an introduction to the DSD (NOAA ISD) data and the physical meanings of the three indices in Section 2.2, as follows:Hourly surface weather observations were obtained from the NOAA Integrated Surface Database (ISD; https://www.ncei.noaa.gov/data/global-hourly/). The ISD provides quality-controlled surface meteorological observations from more than 14,000 stations worldwide. For this study, we selected stations within the study domain that had continuous records during 2001–2025 (Relevant text: pp. 5, lines 140–143).Three complementary metrics are used to characterize dust activity. Near-surface PM10 concentration (CNEMC) reflects near-surface particulate loading. Ground-based DSD is defined as days with daily PM10 > 500 μg·m-3. NOAA DSD is defined as days when any hourly observation reports a dust-related present weather code (MW1 = 6, 7, 30–35, or 98) in the Integrated Surface Database (ISD). DUCMASS (MERRA-2) represents column-integrated mineral dust mass. These metrics are not equivalent. PM10 and DSD characterize near-surface conditions and event frequency, whereas DUCMASS assesses column-integrated dust loading (Relevant text: pp. 5, lines 144–149).Secondly, we have provided a more careful discussion of the inconsistent trends over the Taklimakan Desert, as follows:This apparent inconsistency arises from the different physical meanings and detection principles of these metrics. DUCMASS and PM10 reflect dust loading (column-integrated and near-surface, respectively), while DSD reflects event frequency based on a specific detection criterion. The increase in NOAA DSD over the Taklimakan therefore indicates more frequent but weaker dust events, rather than an increase in total dust loading. This interpretation is consistent with the topographically enclosed setting of the Taklimakan, the basin may experience more frequent local dust activations under weakened synoptic forcing, but these events are weaker and contribute less to column-integrated dust loading. Meanwhile, weather-code-based DSD (NOAA ISD) relies on ground observer judgment and may therefore be biased (Relevant text: pp. 13, lines 390–397).Thirdly, we have revised the descriptions of the metrics to avoid interchangeable usage of these metrics. For example, statements such as "dust activity increased" have been replaced with more precise descriptions ("dust loading increased" for PM10 and DUCMASS, or "dust storm occurrence increased" for DSD), depending on the metric being discussed.Finally, we have revised the caption of Figure 3 as follows: Fig. 3: (a) Interannual variations of near-surface PM10 concentrations during 2016–2025; (b) Interannual variations of spring dust column mass density (DUCMASS) from 2001 to 2025; (c) Mean dust storm days (DSD) across the four subregions derived from ground-based PM10 observations (daily PM10 > 500 μg·m-3, 2016–2025); (d) NOAA ISD weather-code-based DSD (2001–2025) (Relevant text: pp. 13, lines 380–382).Response to RC2-3:Response: We thank the reviewer for this careful and constructive comment. We fully agree that the breakpoints identified in Fig. 3E, G, and H should not be interpreted as physical thresholds for dust storm occurrence or dust emission, and that the PM10 metric and sample composition need to be clearly specified. We have revised the manuscript accordingly.Firstly, we confirm that Fig. 3E and Fig. 3G use identical samples (winter-spring), while Fig. 3H uses a different (spring) sample. The PM10 ranges differ slightly among panels because of the different seasonal windows. Moreover, because the segmented regression analysis used monthly average PM10 concentrations rather than daily values, daily extreme values were smoothed out. Therefore, the PM10 values used in these analyses were mainly concentrated in the range of 50–350 μg·m-3, which is below the daily dust storm threshold (PM10 > 500 μg·m-3). We have clarified the composition of each data set in Section 2.3, as follows:The segmented regression analyses use winter-spring monthly means of wind speed, soil moisture, and PM10 (n = 60), spring monthly means of NDVI and PM10 (n = 30), and preceding-year annual precipitation versus winter-spring dust storm days (n = 231), all averaged across all stations. Because monthly means smooth out daily extremes, the resulting PM10 values fall below the daily dust-storm criterion (PM10 > 500 μg·m-3) (Relevant text: pp. 8, lines 236–239).Secondly, we have supplemented the segmented regression with a linear mixed-effects model to account for spatial differences among stations, as follows:To separate within-station temporal relationships from between-station spatial differences, we fitted linear mixed-effects models with station as a random effect:PM10ij = β0 + β1 × Driverij + ui + εij (2)where PM10ij and Driverij are the spring mean PM10 concentration and the corresponding driver (wind speed, soil moisture, NDVI, or preceding-year precipitation) at station i in year j, ui is the station-specific random intercept, and εij is the residual error. The fixed effect β1 represents the within-station temporal relationship between the driver and PM10, after accounting for spatial differences among stations. Prior to model fitting, all variables were detrended at each station to remove long-term trends. For wind speed and soil moisture, winter-spring means were used; for NDVI, spring means were used, consistent with the seasonal windows used in the main analysis (Relevant text: pp. 8, lines 246–254).The result as follow: To separate within-station temporal relationships from between-station spatial differences, we applied linear mixed-effects models with station as a random effect (Table S3). Wind speed (β = 60.08, P ≤ 0.001) and soil moisture (β = -190.45, P ≤ 0.001) show significant within-station temporal effects on PM10, confirming that year-to-year variations in these drivers modulate dust activity within individual stations, rather than merely reflecting spatial differences among stations. In contrast, NDVI (β = -165.25, P = 0.143) and preceding-year precipitation (β = -0.026, P = 0.592) show no significant within-station temporal effects, although their pooled correlations with PM10 are significant. This indicates that vegetation and precipitation primarily modulate dust activity through spatial differences among stations (e.g., more vegetated or wetter sites have lower dust levels), whereas wind speed and soil moisture act as temporal drivers of interannual dust variability. These results are consistent with our conceptual framework and confirm that the PM10-wind and PM10-soil moisture relationships are robust within-station temporal relationships, not artifacts of spatial covariance (Relevant text: pp. 17, lines 483–492).Lastly, we have revised the interpretation of the breakpoints. We agree that the breakpoints identified by segmented regression are statistical breakpoints in the relationship between monthly mean PM10 and each driver, not physical thresholds for dust-storm occurrence or dust emission. We have replaced "dust emission threshold" and "dust suppression threshold" with "statistical breakpoint" throughout Sections 3.3 and 3.4. We have also added an explicit statement in Section 2.3, as follows:The breakpoints identified by segmented regression represent statistical transitions in the monthly mean PM10-driver relationship. They indicate the environmental conditions under which PM10 becomes more sensitive to the driver, but should not be interpreted as physical thresholds for dust emission or dust storm occurrence, which would require wind tunnel or process based modeling validation (Relevant text: pp. 8, lines 239–243).Response to RC2-4:Response: We thank the reviewer for catching this inconsistency. The Random Forest analysis was indeed performed on DUCMASS, not PM10, and the inconsistency was a writing error in the Results section and Fig. 6 caption. We have corrected all instances where the RF response variable was incorrectly described as PM10 (Relevant text: pp. 19, lines 529–531).Response to RC2-5:Response: We thank the reviewer for this important comment. We agree that the projection methodology required clearer documentation and that the original attribution to future cyclone strengthening was not supported by an explicit diagnosis of future cyclone behaviour, and we have revised both accordingly.We have first clarified that the projection target is DUCMASS (column-integrated mineral dust mass), not PM10. The two metrics are not interchangeable, and because the Random Forest (RF) model was trained on DUCMASS, all future projections and uncertainty estimates are for DUCMASS.Secondly, we have then added a detailed description of the workflow in Section 2.3. The same five predictors were extracted from three CMIP6 models (CESM2, MPI-ESM1-2-LR, and UKESM1-0-LL) under SSP1-2.6, SSP2-4.5, and SSP5-8.5. All CMIP6 predictors were bias-corrected against the historical mean of each predictor using mean-ratio scaling. The RF model was then applied year-by-year to generate winter-spring DUCMASS projections for 2026–2100, with uncertainty estimated as ±1 standard deviation across the three CMIP6 models and the lower bound truncated at zero because DUCMASS is physically non-negative. HAI was not included in the future projection because CMIP6 does not provide consistent land-use scenarios across the three models, and future cyclone intensity, frequency, and position were not explicitly diagnosed from CMIP6; their influence is represented indirectly through projected changes in wind speed and PGI (Relevant text: pp. 9, lines 277–284).Thirdly, because we modified the calculation method, the predicted values differ significantly from the previous ones, and modifications have been made in the main text, as follows:Based on historical data, CMIP6 future scenarios (SSP1-2.6, SSP2-4.5, SSP5-8.5) were used to project winter-spring DUCMASS for 2026–2100 (Fig. 6d–f). The projected ensemble-mean DUCMASS ranges from 0.11 to 0.42 g·m-2 over the Alxa Plateau, 0.08 to 0.30 g·m-2 over the Mu Us Sandy Land, and 0.17 to 1.14 g·m-2 over the Taklimakan Desert, indicating an increasing trend in all three regions. The largest increase occurs over the Taklimakan Desert, consistent with the dominant role of wind speed in this region (67.0% of RF feature importance) and a relatively large vegetation contribution (23.0%) that remains below the effective suppression threshold. Over the Alxa Plateau, wind speed also dominates (66.2%), but precipitation (21.6%) and PGI (7.5%) contribute more strongly, partially modulating the wind-driven enhancement. The Mu Us Sandy Land shows the smallest increase, consistent with the stronger roles of precipitation (22.4%) and PGI (6.4%) in this region. Wind speed is the primary control on future DUCMASS, but its effect is modulated by vegetation greening (up to 13.2% per decade under SSP5-8.5, Table S4) and precipitation changes, which partially offset the wind-driven enhancement in the Alxa Plateau and Mu Us Sandy Land but not in the Taklimakan Desert (Relevant text: pp. 19–20, lines 542–552).Our results differ from Zhao et al. (2023), who projected a future decline in dust activity (0.8%–11.9%), mainly due to differences in seasonal focus and the representation of large-scale circulation changes. While Zhao et al. (2023) analyzed annual mean changes, our study focuses on winter-spring, the peak dust season. Under SSP1-2.6, winter-spring wind speed increases by 1.3–2.4% per decade across the three regions, whereas under SSP5-8.5, it decreases by 0.2–0.8% per decade. Meanwhile, LAI and precipitation increase under SSP5-8.5 (8.0–13.2% and 2.8–5.4% per decade, respectively), which may partially offset wind-driven increases in dust loading over the Alxa Plateau and Mu Us Sandy Land (Table S4). Given the dominant role of wind variability in controlling dust activity in our study, the projected DUCMASS increases likely reflect the combined effects of large-scale atmospheric circulation changes and concurrent land-surface responses, rather than changes in emission pathways alone. Although future cyclone intensity, frequency, and track position were not explicitly diagnosed from CMIP6 simulations, their potential influence is indirectly represented through projected changes in wind speed and PGI. Therefore, future dust projections should be interpreted as climate-driven potential changes arising from interactions between atmospheric circulation and land-surface conditions, rather than as precise estimates of future dust emissions (Relevant text: pp. 20, lines 553–565).Minor comments:Response to RC2-1:Response: We thank the reviewer for this suggestion. We have added a brief explanation of why PM10 is an appropriate proxy for dust activity: mineral dust is dominated by coarse particles, and PM10 is the standard metric for dust-related particulate pollution. During dust events, PM10 increases much more strongly than PM2.5, making it a sensitive indicator of near-surface dust loading (Relevant text: pp. 3, lines 71–73).Response to RC2-2:Response: We thank the reviewer for this comment. The threshold of PM10 > 500 μg·m-3 follows the quantitative classification of dust weather based on PM10 mass concentration, in which 500 μg·m-3 represents the lower bound of the “dust storm” category (> 500 μg·m-3), distinct from lighter dust conditions (200–500 μg·m-3; Hoffmann et al., 2008). This threshold is also consistent with the operational practice in East Asian dust monitoring, where PM10 concentrations above 500 μg·m-3 are treated as indicative of dust storm conditions. We have added this reference to the revised manuscript (Relevant text: pp. 5, lines 130–132).Response to RC2-3:Response: We thank the reviewer for this comment. The soil moisture data used in this study correspond to the first soil layer of ERA5, which represents the volumetric soil water content in the 0–7 cm layer. This is the standard surface soil moisture layer provided by ERA5 and is commonly used in dust-related studies. We have specified this in the revised manuscript (Relevant text: pp. 5, lines 151).Response to RC2-4:Response: We thank the reviewer for this comment. For the statistical analyses, all gridded datasets were area-averaged over each subregion on their native grids. To confirm that this approach is robust, we coarsened ERA5 from 0.25° to 0.5° and recomputed the region means. The differences are within 1.4% for all four subregions, well below the interannual variability. This indicates that the subregional means are insensitive to the native grid resolution, and the harmonization is therefore not sensitive to the resampling method (Relevant text: pp. 9–10, lines 285–288).Response to RC2-5:Response: We thank the reviewer for this comment. HAI is first introduced in Section 2.2 as a derived land-use variable, and its definition and calculation are provided in Section 2.3. To avoid forward reference without definition, we have added a cross-reference at the first mention of HAI (“see Section 2.3 for definition and calculation”), so that readers can immediately locate the formula (Relevant text: pp. 6, lines 176–177).Response to RC2-6:Response: We thank the reviewer for this comment. We have added a definition and calculation of the Pressure Gradient Index (PGI) before its first use in the Random Forest analysis. The PGI is defined as the sea-level pressure difference between two latitude bands over the study region, which represents the large-scale meridional pressure gradient driving northwesterly synoptic flow associated with Mongolian cyclones (Relevant text: pp. 9, lines 273–276).Response to RC2-7:Response: We thank the reviewer for this comment. The two time points reflect two different analyses. For the long-term series (2001–2025), segmented regression identifies 2019 as the statistical breakpoint. For the shorter PM10 record (2016–2025), we compare the two five-year periods 2016–2020 and 2021–2025, because the record is too short for a reliable breakpoint analysis. Descriptions of the PM10-based changes now refer to “2021–2025 relative to 2016–2020”. The phrase “after 2020” has been removed throughout.Response to RC2-8:Response: Thank you for your comment. We have unified the precipitation threshold to 144 mm (Relevant text: pp. 16, lines 450).
-
AC1: 'Comment on egusphere-2026-3940', Wenhui Zhang, 15 Oct 2026
Dear Marco Gaetani:
We sincerely thank the editor and all three reviewers for their careful reading and constructive comments on our manuscript "Mongolian cyclone variability modulates spatially divergent dust responses across contrasting source regions in northwest China" (Manuscript ID: EGUSPHERE-2026-3940). We have carefully revised the manuscript accordingly.
We have also revised the title from "Mongolian cyclone intensification reshapes dust activity in northwest China" to "Mongolian cyclone variability modulates spatially divergent dust responses across contrasting source regions in northwest China", to better reflect the revised focus on cyclone variability and spatially divergent dust responses rather than a simple intensification narrative.
The major revisions include: (1) implementing an objective Mongolian cyclone tracking algorithm and related diagnostics; (2) adding a cyclone-active vs inactive year composite analysis to examine regional differences; (3) performing independent validation of ERA5 surface wind trends against NOAA station observations; (4) applying linear mixed-effects models to disentangle spatial and temporal covariance in station-year datasets; (5) explicitly distinguishing different dust metrics (PM10, DUCMASS, DSD) and clarifying their physical meanings; (6) softening overly strong causal language; and (7) revising the CMIP6-based future projection workflow and removing unsupported claims about future cyclone intensification.
In addition, we have updated the funding information in the Acknowledgements section. We have also carefully checked the entire manuscript for language, formatting, figure captions, and references to ensure consistency with the requirements of Atmospheric Chemistry and Physics. All modifications are marked in the revised manuscript using track changes, and our responses to each referee comment are provided in red italic font.
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 405 | 211 | 56 | 672 | 115 | 42 | 37 |
- HTML: 405
- PDF: 211
- XML: 56
- Total: 672
- Supplement: 115
- BibTeX: 42
- EndNote: 37
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
The manuscript addresses an interesting and timely question: whether recent changes in Mongolian cyclone activity can explain spatially divergent dust variability across northwest China. However, I do not think the central conclusion is currently supported by the analysis. My main concern is the Taklimakan–Alxa comparison. The manuscript uses these two regions to diagnose contrasting responses to Mongolian cyclone forcing, yet the Taklimakan is described as topographically sheltered and only weakly influenced by Mongolian cyclones. However, synoptic-scale cyclone forcing may interact with the basin topography and generate mesoscale and local wind systems in the Tarim Basin. While the Mongolia–Gobi source region, which is much more directly linked to Mongolian cyclone activity, is not included. A second concern is the attribution of cyclone change. The manuscript largely infers cyclone intensification and southward displacement from relative vorticity, and then interprets concurrent changes in wind, dust and land-surface variables as a causal pathway. However, there is no objective cyclone tracking or cyclone-conditioned dust analysis. A third concern is the attribution framework. The manuscript interprets concurrent changes in Mongolian cyclones, surface wind, dust, soil moisture and vegetation as evidence of a causal pathway, but the analysis mainly demonstrates statistical associations and synchronized trends. It does not directly quantify how cyclone activity modifies these environmental variables, nor does it test cyclone-conditioned dust responses. Addressing these issues would require substantial re-framing and additional analysis beyond a normal major revision. I therefore recommend rejection.
Major comments
1. The Taklimakan–Alxa comparison does not isolate a Mongolian-cyclone effect.
Relevant text: pp. 2–3, lines 50–80
The manuscript itself argues that the Taklimakan is topographically sheltered and less directly affected by Mongolian cyclones, whereas Alxa is more exposed to northwesterly synoptic flow. Observing opposite dust trends in two regions governed by fundamentally different circulation regimes does not by itself demonstrate a spatially divergent response to Mongolian cyclone forcing. The Taklimakan also contains strong mesoscale (Aoki et al., 2005), thermally driven winds (e.g., low-level jets and haboobs, Chen et al., 2026) and topographically mediated winds, and the northern/eastern basin margins should not be represented by a single prevailing northeasterly regime. Also, the monitoring stations are concentrated along the northern rim of the basin rather than the northeastern edge. Importantly, the Mongolia–Gobi source region, which is more directly linked to Mongolian cyclone activity, is not included in the comparison. The study therefore needs a stronger justification for the regional framework and should distinguish direct cyclone influence from topographically mediated and local circulation effects.
2. Mongolian cyclone activity is not objectively defined or tracked.
Relevant text: p. 1, lines 15–25; pp. 6–7, lines 165–200; p. 12, lines 385–410.
The manuscript infers cyclone intensification and a southward shift mainly from sea-level-pressure and relative-vorticity anomalies. A positive relative-vorticity anomaly is not itself equivalent to an objectively identified Mongolian cyclone, and the paper does not clearly explain how cyclone centres, tracks, frequency and intensity are detected. This is especially important because published cyclone-tracking studies may report different spatial tendencies.
3. Coincident post-2019 changes are treated as causal attribution.
Relevant text: pp. 6–8, lines 165–225; p. 12, lines 390–415.
A central argument is that cyclone intensity/latitude, near-surface wind and dust all changed around 2019–2021. Synchronised changes are not sufficient to establish that cyclone changes caused the dust reversal, particularly over such a short post-break period. Also some statistics are needed to clarify the reverse.
4. ERA5-based long-term surface-wind trends require independent validation.
Relevant text: pp. 3–5, lines 100–155
The main mechanism depends on relatively small long-term changes in near-surface wind derived from ERA5. Reanalysis is useful for circulation diagnostics, but surface-wind trends can be sensitive to models. The claimed wind recovery should therefore be independently evaluated using meteorological-station observations.
5. Pooled station-year correlations may mix spatial and temporal covariance.
Relevant text: p. 5, lines 125–155; pp. 9–10, lines 280–330.
The manuscript pools 26 stations over 10 years (n = 260 station-year pairs) for correlation and segmented-regression analyses. This can confound spatial differences among stations with interannual variability within each station. A positive relationship may therefore partly reflect that dustier locations are also windier, rather than that year-to-year wind anomalies drive year-to-year dust anomalies.
6. Dust metrics are not equivalent and their disagreement needs more careful treatment.
Relevant text: pp. 3–4, lines 90–120; pp. 7–8, lines 205–245.
The manuscript combines PM10, PM10-defined dust-storm days, MERRA-2 DUCMASS and visibility-based dust-storm records. These metrics represent different aspects of aerosol loading, emission and transport and can respond differently to observing practice and background pollution. Over the Taklimakan, the metrics even show contrasting behaviour. Interpreting this disagreement specifically as 'more frequent but weaker dust events' is plausible but not uniquely demonstrated.
7. The future PM10 projection is insufficiently documented and over-interpreted.
Relevant text: p. 4, lines 120–125; p. 12, lines 390–420.
The manuscript projects very large PM10 increases through 2100 using three CMIP6 models, but the workflow linking CMIP6 climate variables to PM10 is not described in sufficient detail. It is unclear how model bias, spatial-scale mismatch, future NDVI/HAI, PGI, soil moisture and cyclone activity are treated or validated. In particular, the claim that future PM10 is primarily controlled by climate-driven strengthening of Mongolian cyclone activity requires explicit diagnosis of future cyclone behaviour. Without this, the future attribution and the very large projected increases are not sufficiently supported.
8. Spatial heterogeneity within the Taklimakan is under-resolved.
Relevant text: pp. 7–8, lines 210–225.
Again, treating the entire Tarim Basin as one homogeneous source region may obscure substantial differences among the northern, eastern and southern basin margins, where dust activation and wind regimes differ. And I did not see any maps reporting this spatial distribution.
9. Causal language throughout the manuscript is stronger than the evidence supports.
Relevant text: p. 1, lines 15–25; p. 13, lines 420–425.
The manuscript frequently moves from association or predictive importance to causal statements, including claims that Mongolian cyclone intensification ‘reshapes’ dust activity, that wind is a direct reflection of cyclone forcing, and that large-scale circulation outweighs land-surface controls. The analyses presented are primarily observational and statistical. Unless stronger causal tests are added, the Abstract, Results, Discussion and Conclusions should consistently distinguish association, prediction and physical causation.
Other comments:
Page 1, Lines 12–14: Please specify what is meant by “recent years.” Dust activity has generally been reported to decrease over a longer period, so the recent increase may partly reflect interannual or decadal variability.
Page 1, Lines 15–16: The previous sentence mentions both land-surface changes and circulation changes, but here the knowledge gap refers only to circulation changes. Please clarify the role of land-surface changes in the identified gap.
Page 1, Lines 16–18: The Taklimakan Desert is relatively weakly affected by Mongolian cyclones, as the authors also acknowledge later in the manuscript. Therefore, the rationale for investigating Mongolian cyclone impacts on the Taklimakan should be clarified.
Page 1, Lines 18–20: Please double-check the reported southward shift after 2019. Mu and Fiedler (2025) appear to report a different, or northward, tendency for Mongolian cyclone activity.
Page 2, Lines 38–43: I do not fully understand why the increasing contribution of Mongolia as a dust source is introduced here, as this study mainly investigates dust activity within China. Please clarify its relevance to the study objective.
Page 2, Lines 45–48: The preceding sentences describe different mechanisms, but it is unclear how they lead directly to the stated central question about Mongolian cyclone control. The logical connection could be strengthened.
Page 2, Lines 49–50: I am not sure that land-surface properties and Mongolian cyclone activity should be framed together in this way.
Page 2, Lines 50–52: The manuscript later examines the effects of wind, precipitation, soil moisture and vegetation on dust activity, rather than explicitly quantifying how Mongolian cyclones modulate each of these variables. Please revise this statement accordingly.
Page 2, Lines 55–58: I do not think an increase in dust storms and a greening trend necessarily constitute a direct contradiction, as they describe different processes. In addition, much of the post-2021 dust resurgence in northern China might originate from the broad Gobi region rather than the Taklimakan. Please clarify how this motivates the focus on Mongolian cyclones.
Page 3, Lines 65–69: The first part of this paragraph discusses environmental controls on dust, whereas the final sentence shifts to the lack of long-term studies versus individual dust-storm events. I do not see a clear logical connection between these points; they may need to be separated or better linked.
Page 3, Line 70: “Building on this foundation” is unclear. Does “this foundation” refer to the divergent dust behaviour between the Taklimakan and Alxa regions? Please specify.
Page 3, Lines 70–72: Why is Mu Us Sandy Land not included here, given that it was introduced earlier as one of the study regions? Also, please clarify why the Mongolian Plateau itself is not considered.
Page 3, Lines 71–73: The second hypothesis is presented as a hypothesis of this study but is immediately supported by a citation to another study, which is confusing. I think the intended comparison may instead be between the Tarim Basin interior and regions outside the basin; please clarify.
Page 3, Lines 73–74: It is not clear how the authors establish that there is “significant controversy” regarding the primary driving mechanisms. There is certainly debate regarding dust trends, but the basic physical controls on dust emission might be relatively well established. Please clarify what specific controversy is meant.
Page 3, Line 84: Please check the letter/article formatting here.
Page 3, Lines 85–89: Four dust-storm centres are mentioned, but the study subsequently focuses on only two major source regions. Please clarify the relationship between these four centres and the regions actually analysed.
Page 6, Lines 165–167: The “recent rebound of dust storm activity” is introduced before the corresponding dust records are presented. It may be clearer to present the observational evidence for the dust rebound first and then investigate the associated circulation changes.
Page 6, Lines 168–171: I am not convinced that the change in relative vorticity alone is sufficient to identify an intensified and southward-shifted Mongolian cyclone. Positive relative vorticity indicates cyclonic circulation but does not necessarily identify a cyclone centre. In addition, Mu and Fiedler (2025), using an explicit cyclone-tracking algorithm, report a different trend. A more robust cyclone identification/tracking method or additional circulation diagnostics would be helpful.
Page 7, Line 175: Could the authors provide the corresponding figure or analysis supporting this value?
Page 7, Lines 176–178: The text states that the regional wind-speed differences are shown in Fig. S1A, but I cannot clearly identify these details in that figure. Please check the figure reference or make the differences more explicit.
Page 7, Lines 190–194: Please double-check whether Mu and Fiedler (2025) actually demonstrate that topographic shielding weakens Mongolian cyclone influence on the Taklimakan in the way stated here. My understanding is that they identify this as an issue requiring further investigation rather than establishing it directly.
Page 7, Lines 195–202: This discussion introduces Arctic forcing and dust-radiative feedbacks, but no analysis or figure in this study directly evaluates these mechanisms. At present, these statements rely mainly on previous studies and appear somewhat detached from the analyses presented here.
Page 8, Lines 213–216: The statement that the entire Tarim Basin is dominated by northeasterly winds may be too general. The northern rim of the basin is also an active dust-source region associated with different wind regimes (e.g., Chen et al., 2026, ACP). Regional differences within the basin should be considered.
Page 8, Lines 216–219: The text gives percentage changes between the two periods, but were these changes tested for statistical significance? Earlier sections state that the study aims to “quantify” these changes, so significance or uncertainty should also be reported.
Page 8, Lines 220–223: It would be useful to show the spatial pattern of the DUCMASS trends rather than only basin-mean values. I suspect that the eastern and northern parts of the Tarim Basin may exhibit different trends, which could be obscured by averaging over the whole basin. .
Page 9, Lines 247–250: Why is March used for the long-term wind analysis? Dust activity and circulation can differ substantially between spring months (Wang et al., 2026), so the choice of March requires justification, or the authors could examine the full spring season and individual spring months separately.