Introduction

Mineral dust, principally originating from global arid and semi-arid regions, constitutes the dominant atmospheric aerosol by mass on Earth1. The influence of this aerosol on climate is complex and yet to be fully quantified. At present, the magnitude of the radiative effect of dust aerosols remains highly uncertain2, and the specific local radiation effects are regulated by multiple factors, including particle size distribution, the iron oxide content, and surface albedo3,4. Therefore, enhancing our understanding of the shortwave absorption characteristics of dust can provide new insights into revealing the magnitude of radiation effect caused by mineral dust5. However, due to insufficient understanding of the key parameter \(k\) describing the absorption characteristics of dust aerosols and the important parameter SSA characterizing the ratio of absorption to scattering capacity, the representation of relevant microphysical properties in climate models remains highly uncertain6,7.

\(k\) for dust can be determined through various methods. Numerous studies on Saharan dust have derived representative \(k\) through laboratory measurements with theoretical calculations on field-collected samples8,9,10. In contrast, the study of Asian dust is relatively limited. The existing Asian dust \(k\) estimates are derived through dust component analysis with corrections11,12 and combined satellite observations and models13. While their \(k\) are of a similar magnitude, East Asian dust particles are finer and exhibit stronger backscattering capabilities14. Furthermore, their iron oxide content is lower than that of North African dust15. Whether these commonly used optical parameterization schemes are applicable in the East Asian region requires further investigation. In addition, past observations of EAD primarily relied on single or sparse soil samples and atmospheric data collected during dust events15. This approach failed to accurately characterize key absorbing minerals in East Asia, leading to substantial bias in radiation effect estimates.

This study utilizes a total of 22 surface soil samples obtained from East Asian dust source regions, in conjunction with meticulous model validation, to formulate a region-specific dust complex refractive index scheme for East Asia. The implementation of this scheme within a Global Climate Model (GCM) yields simulations that demonstrate enhanced congruence with ground-based station observations. The results further reveal significant discrepancies in the estimated shortwave direct radiative effect of East Asian dust aerosols compared to previous assessments.

Results

k of EAD constrained by laboratory measurements

Iron oxides in mineral dust are predominantly present as hematite or goethite, with their relative abundances depending on the pedogenic conditions of the source regions16. Other types of light-absorbing oxides co-exist in dust aerosols, but their overall content is usually negligible when considered against the significance of hematite and goethite in the short-wave spectral range for dust absorption. These minerals exhibit strong absorption capacity in the UV-visible spectral region15, making them key contributors to the radiative properties of dust aerosols despite their relatively low mass fractions. This is of particular significance when the minerals are internally mixed with other minerals that exhibit lower levels of absorption17. Therefore, accurately reflecting the optical properties of these minerals is essential to reduce the uncertainty of dust-climate feedback5. Analysis of soil samples from 22 sites across major EAD source regions—Gobi Desert, Taklimakan Desert, and Qaidam Basin (Fig. 1c), which reveals regional variability in the mass percentage of iron oxide (\({{MC}}_{{Fe}-{ox} \% }^{s}\)) (Fig. 1a). \({{MC}}_{{Fe}-{ox} \% }^{s}\) were found to be comparatively low throughout East Asia; however, higher values were observed in the western desert regions of China. These findings suggest a potential correlation with the surface material composition, degree of weathering, and climate-driven soil-forming processes at the sampling sites. The Gobi Desert exhibits the lowest mean content (0.74 ± 0.004%), the Taklimakan Desert intermediate (0.77 ± 0.002%), and the Qaidam Basin the highest (0.89 ± 0.005%).

Fig. 1: Total particulate iron oxide mass percentages, imaginary part of the complex refractive indices across East Asian dust source regions, and distribution of soil sampling sites.
figure 1

a The total particulate \({{MC}}_{{Fe}-{ox} \% }^{s}\) were calculated at 22 sampling sites. b Wavelength-dependent distribution of \(k\). Comparative data include: mineral dust complex refractive indices from the OPAC database (black dashed line), the observationally derived refractive index by Colarco and Kim (blue dashed line) and the observationally calculated refractive index by Lee (purple dashed line). c Spatial distribution of sampling sites and dust emission fluxes. All sites are proximal to dust source regions.

Full size image

Based on soil samples and the established correlation between \({{MC}}_{{Fe}-{ox} \% }^{s}\) and \(k\) in the short-wavelength band10, the wavelength-dependent variation of \(k\) was derived (Fig. 1b). It has been discovered that the absorption characteristics of EAD are weak in the short-wave spectrum (0.3–1 μm), and \(k\) range from 2.26 × 10−4 to 8.37 × 10−4, which is significantly lower than the value in the OPAC database constructed by Sahara dust measurement18. Furthermore, it is pointed out by previous studies that the numerical differences of \(k\) used in the model vary from five times to one order of magnitude in the spectral range of 0.3 to 30 μm19. It is evident that this reflects the high level of uncertainty in the current characterization of the optical properties of mineral dust. The \(k\) of East Asia obtained by combined observation is 0.00042 at 550 nm wavelength, which is over one order of magnitude below the OPAC value used by NASA/GEOS20 (Fig. 2b: Colarco et al.20) and the intermediate \(k\) value of 0.003 used by WRF-Chem model in standard experimental simulation21,22,23. This result further confirms the significant differences in the optical properties of dust between regions, and suggests that the traditional global model may systematically overestimate the radiation effect of EAD due to over-reliance on single, overestimated absorption parameters. However, it is only marginally higher than observations of long-range transported dust in Seoul15, potentially reflecting the differences between dust source area and aging effects during transport.

Fig. 2: Observation-constrained simulations effectively ameliorate the underestimation of SSA in the default model scheme.
figure 2

a–e Comparison between 550 nm SSA simulations from two experiment (Default: blue dots; New-avg: red dots) and AERONET-derived 550 nm SSA at five sites. f Comparison of observed SSA means (dots) with four experimental groups. Shaded areas denote ±1 standard deviation. Mean values for New-p90 and New-p10 are indicated by upper/lower red horizontal dashes.

Full size image

Effective improvement of optical parameters

In order to rigorously assess the degree of improvement of the new parameterization of the optical properties of dust, the five stations located at or near the dust source regions were selected for comparison according to the standard that the station was mainly affected by dust. The SSA simulation results of the Default experiment and the New-avg experiment are compared with the results of AERONET at the wavelength of 550 nm, and it is found that the latter is very consistent with the results of AERONET. This phenomenon is evidenced by two salient statistical indicators: firstly, the mean relative bias of the five sites has been diminished from 15% to 2%, and secondly, the root mean square error (RMSE) has been substantially reduced by an order of magnitude (Fig. 2a–e), with the most substantial enhancement occurring at the Beijing site, which is not only data-rich, but also exhibits a high degree of similarity to the observed data. Furthermore, the simulation outcomes of the new scheme at the SACOL and Dalanzadgad stations, which are situated within the dust source region, have been significantly enhanced, though they do exhibit a slight discrepancy when compared to the observational results. Additionally, simulations from the New-p90 and New-p10 experiment results exhibit more pronounced improvements at specific sites (Figs. S1, S2).

The statistical results indicate that the Default experiment significantly underestimated the SSA, with mean underestimations of 0.12 (SACOL), 0.16 (Beijing), 0.13 (Beijing_RADI), 0.16 (Beijing_CMAS), and 0.19 (Dalanzadgad) (Fig. 2f). In contrast, the three new schemes’ mean SSA values (red dots and short dashes) align better with observations in overall trend and variability range. The presence of at least one mean point at each site (across the three sets of experiments of the new scheme) falling within one standard deviation of the observed values (the black shaded area) indicates that the systematic bias of the new parameterization scheme has been largely contained within the dominant range of random errors.

It is vital to note the dependence of AAOD on dust extinction, which can be measured using AOD. Furthermore, the absorption proportion in this extinction can be determined by SSA24. The Default experiment (dashed lines) systematically overestimates the AAOD of 550 nm at three key sites (i.e., SACOL, Beijing, and Dalanzadgad) (Fig. 3a–c). In contrast, the observationally constrained New-avg experiment (solid lines) aligns closely with site observations (dots). This improvement stems from more accurate characterization of iron oxides, which are the key absorption components in mineral dust, and reduces the root mean square error (RMSE) by 43.2%, 55.2% and 93.2% at the three locations of AAOD, respectively.

Fig. 3: Observation-constrained simulations effectively mitigate overestimation of absorbing optical depth in the default model scheme.
figure 3

a–c Temporal evolution of AAOD simulated in two experiments: New-avg (solid lines) and Default (dashed lines), compared with AERONET observations at SACOL (blue dots), Beijing (magenta dots), and Dalanzadgad (yellow dots). upper/lower horizontal dashes in corresponding colors denote mean values from New-p90 and New-p10 experiments. d–f Spatial distribution of AAOD and its difference between the two experiments.

Full size image

However, by comparing AOD values, as shown in Fig. S3, we also conclude that this bias likely stems from issues such as dust emission, transport and deposition, as current free-running atmospheric model that is not nudged would result in biases in wind fields and precipitation fields. Additionally, the missing larger particles (>10\(\mu m\)) in the model may lead to an underestimated AOD in the visible spectrum. Moreover, when focusing on total dust optical depth (DOD), although both simulations exhibit a systematic positive bias relative to satellite retrievals, particularly pronounced over East Asian dust source regions, their spatial distribution patterns remain broadly consistent. The new parameterization scheme yields regional mean values closer to satellite observations (Fig. S4).

Finally, spatial distributions (Fig. 3d–e) further indicate significant overestimations by the Default experiment across EAD sources (e.g., TD, Mongolian Plateau) as well as their downstream regions (e.g., North China Plain). Moreover, the difference between the two groups of experiments in the western desert and the northern desert can reach more than -0.14, that is, there is a more obvious improvement in the dust source area. The Dalanzadgad station, situated in proximity to the area exhibiting substantial variations, has yielded a limited number of observations. However, these observations demonstrate a high degree of similarity with the new parameter scheme. The New-avg experiment constrains the regional mean AAOD to 0.0106 (0.0093-0.0132) by 65.4% compared with the default scheme.

Assessment of short-wave DRE in East Asia

The New-avg experiment reveals markedly distinct radiative effects compared to the Default experiment. This shift is particularly evident in the enhanced cooling effect at the top of the atmosphere (TOA): the revised scheme adjusts SW direct radiative effect (DRE) from +0.75 W/m2 (Fig. 4a) to −3.80 W/m2 (Fig. 4d). Despite numerical differences, it aligns with prior estimates of shortwave radiative cooling by EAD aerosols12,22,23. Crucially, we constrain the \(k\) to 10-4 through extensive observations, providing a more reliable data foundation compared to the commonly adopted 10-3 to 10-2 range (Table 1). Further analysis of changes in surface (SFC) direct radiation effect shows that the revised scheme diminishes surface radiative cooling by 34.6% (Fig. 4i), but this reduction is notably less pronounced than the TOA radiative effect sign reversal and the 79.5% attenuation in atmospheric absorption (Fig. 4h). Overall, it shows that the absorption capacity of EAD is weak, which may be related to the significant difference in optical characteristics between EAD and Sahara dust. The former has a strong backscattering ability, which leads to a higher negative value of shortwave radiation effect at the TOA11.

Fig. 4: The spatial distributions of dust shortwave (SW) direct radiative effect.
figure 4

a–c the Default experiment of OPAC scheme, (d–f) New-avg experiment of updated scheme and their dust shortwave radiation effect at the TOA, ATM and SFC respectively. g–i Difference between the two sets of experiments. Significant differences have occurred at the TOA, where radiative effect exhibits sign reversal. Concurrently, atmospheric absorption is substantially attenuated.

Full size image
Table 1 Comparison of shortwave direct radiation effect and complex refractive index scheme of EAD with previous studies
Full size table

Spatially, the dust radiative effect exhibits pronounced regional characteristics. Regions of intense shortwave radiative cooling at the TOA (≥−18 W/m2) are predominantly localized near dust source regions, exemplified by the Taklimakan and Badain Jaran Deserts. In these regions, the simulated maximum SW DRE at the SFC under the new parameterization scheme reaches −27 W/m2, slightly exceeding the −25 W/m2 reported in similar studies25,26. Moreover, there is also a certain improvement in the downwind region of its transmission, extending to other seas such as the Yellow Sea and the East China Sea, where radiative cooling of the TOA has occurred. Additionally, the simulated TOA shortwave radiative cooling over the Tibetan Plateau is stronger, with a regional average of −5.63 (−6.37 to −4.48) W/m2, also higher than other estimates22,23, which confirms that robust scattering processes over EAD sources exert significant influences on local climate. On the time scale, the dust cooling effect is stronger in 2010 and 2014, and the regional average of SW radiation effect at the TOA is −4.45 W/m2, which is about 22% higher than the regional average of −3.47 W/m2 in other years (Table S4). This systematic quantification provides critical scientific insights into the DRE of EAD.

Discussion

Currently, GCMs exhibit high uncertainty in both projections of future dust emissions and climate-dust feedbacks5. This leads to aerosol radiative effect as an uncertainty in current Earth energy balance studies. To address the problem of overestimation of dust aerosol absorptive properties in climate models for East Asia, this study developed region-specific k using 22 soil samples covering most of the dust source regions, combined with rigorous model validation methods. Moreover, the shortwave DRE in East Asia was reassessed, and it was found that the absorptive properties of EAD are weaker than the previous estimates. The cooling effect of SW radiation is more pronounced near the dust source area and in the downwind region.

However, in simulating particle size distribution and dust emission processes, this study employs multiple log-normal distributions for simplified estimation of \({p}_{m}(d)\) and \({p}_{f}(d)\) (Eq. 3) due to the lack of a complete soil sample dataset in the Shao04 scheme. The actual distribution varies with soil type, wind speed, and transport processes, and spatial variability is not accounted for. In addition, significant uncertainties are not only present in the characterization of surface soil conditions in dust source regions, but are also widespread in several key aspects of the transport phase. For example, the characterization of processes dust–cloud interactions, the treatment of dry and wet deposition, and the mixing state with other aerosols is still not precise enough, limiting the accuracy of model simulations of dust and its climatic effects.

Particle size distribution is considered to be an important parameter that strongly affects the DRE of dust27, but the problem of its characterization remains unresolved to date28. Therefore, although the complex refractive index of minerals itself does not change with particle size17, that is, the optical properties of minerals remain constant at different particle sizes, the SSA gradually decreases as the proportion of coarse-grained dust increases. However, there are still major challenges in the characterization of dust size distributions in current models, and related problems have not been effectively solve27. Since coarse particles of dust have a stronger absorption capacity and shorter lifetime than fine particles, we set the cut-off diameter of aerosols to 10 \(\mu {\rm{m}}\) in the CAS-ESM2 model, and the geometric diameter range (0.1–10 μm) considered in the model fails to cover the true distribution29, i.e., the contribution of coarse particles may be excluded, leading to an underestimation of the warming effect of dust2. Previous studies have attempted to incorporate spatiotemporal variations in mineral composition30, the impact of super-coarse particles31, or the effects of different mixing rules on dust optical properties7 into models. These represent areas warranting further investigation in subsequent research.

Additionally, dust exhibits a non-negligible warming effect in the longwave band32. Uncertainties in the DRE of long waves arise mainly from the complex refractive index of mineral particles, their size distribution and their vertical distribution in the atmosphere, rather than from the mineral composition itself28. This means that even if the relative content of various minerals changes, the effect on long-wave radiation is relatively weak and shows a low sensitivity. Furthermore, since iron oxides are less significant in the longwave band compared to minerals like quartz and clay28, and their imaginary part of the refractive index exhibits smaller variations in the longwave band relative to the visible spectrum33, this study focuses on shortwave radiation effects. Based on data availability and phased objectives, it lays the groundwork for integrating longwave effects in subsequent research.

The influence of dust aerosols on the atmospheric circulation34,35 and weather systems mainly stems from their perturbation of the energy balance at the surface and in the uplift layer34,36, which may not only alter the atmospheric stability37, but also affect the horizontal temperature gradient. Particularly in the springtime, when dust storms are frequent, the extension from the Taklamakan Desert to North China and other downwind areas may experience significant changes in the DRE of dust, making these regions key areas for studying dust-climate interactions. Since changes in meteorological conditions can in turn affect the dust release and transport processes37, this study improves the accuracy of the estimation of the optical properties of dust by introducing the constraint of iron oxides on the k of dust, which is more helpful for realistically simulate the radiative effect of mineral dust and its feedback role in the dust cycle.

Methods

Soil sample collection

Dunes and gravel soils represent different surface conditions of desert terrain and may be the main sources of desert atmospheric dust38. This study collected 22 soil samples from typical sand dune and gravel desert surfaces in China’s major dust source areas. Detailed information on the locations and surrounding geomorphology of the sampling sites is listed in Table S1.

Laboratory assay

The dust mobilization system generates mineral dust aerosol particles from the collected sample soil by simulating natural processes such as dust sandblasting and saltation39. The chemical composition and particle size distribution are the same as those in the field40.

Computer-controlled scanning electron microscopy (CCSEM) allows for the analysis of atmospheric particulate matter by characterizing the properties of air masses through the detection of the morphology and composition of thousands of individual particles41. The CCSEM system employs the rotating Ferret box method to measure particle diameter (Fig. S5), thereby calculating detailed morphological parameters such as aspect ratio and particle roundness42. It also determines the weight percentage of elements based on their characteristic X-ray signals43. The system can classify particles according to both morphological parameters and elemental composition, with its data complementing results from bulk analysis techniques. Traditionally, particle classification within this system relies on hierarchical rules that group particles with similar characteristics by integrating elemental concentrations and morphological parameters. The particle classification rules are listed in Table S244. The empirical formula and elemental composition information are available in the handbook of mineralogy45 and several online databases at https://duffy.princeton.edu/mineralogy-and-crystallography-databases. This study focuses on the percentage of iron oxide mass of each sample, so single-particle iron oxides need to be treated. The specific formula is expressed as follows:

$${{MC}}_{{Fe}-{ox} \% }^{s}=\frac{{\sum }_{i=1}^{n}{M}_{{Fe}-{ox}}}{{\sum }_{j=1}^{m}{M}_{{sp}}}\,\,s=0,1,2\ldots \ldots .22$$
(1)

where \(n\) is the number of single-particle iron oxides, \({M}_{{Fe}-{ox}}\) is a mass of single-particle iron oxide, \(m\) is the total count of particles in each sample, \({M}_{{sp}}\) is a total mass of all particles.

Complex refractive index of dust

Evidence indicates a pivotal role for the composition of iron oxides in determining the optical properties of dust46. Within the 0.3–1 μm wavelength range, \({{MC}}_{{Fe}-{ox} \% }^{s}\) exhibits a strong correlation with k10. In this study, the robust correlation between them is adopted. The formula used to calculate k is as follows:

$${k}_{s,\lambda }={a}_{\lambda }\times {{MC}}_{{Fe}-{ox} \% }^{s}+{b}_{{\lambda }}\,\lambda {=}{370},\ldots 950\,nm$$
(2)

Here, the coefficients a and b are defined as linear fitting parameters10 and consist of data from six different bands; s denotes the sample serial number.

CAS-ESM2

CAS-ESM2 employs a modular architecture, including several independent component models such as land and atmosphere. These components are coupled via the CPL7 coupler to exchange physical quantities and enable synchronized simulation47. The simulations follow the protocols of the Atmospheric Model Intercomparison Project, incorporating prescribed sea surface temperature, precursor emissions, and anthropogenic aerosol data. It has a horizontal resolution of \({1.4}^{^\circ }\times {1.4}^{^\circ }\) and a vertical coordinate of 35 layers of atmospheric hybrid sigma-pressure. Four sets of simulation results were used, including one control experiment (Default) utilizing the default OPAC data and three additional experiments utilizing observation-constrained new schemes based on the mean (New-avg), 90th percentile (New-p90), and 10th percentile (New-p10) values. The model runs from January 2006 to December 2015, with the initial four years designated as the spin-up period and the subsequent six years used for analysis. During the summer and autumn seasons, dust particles are influenced by East Asian monsoon precipitation and are also susceptible to mixing with anthropogenic aerosols. This weakens the dust radiation effect signal and increases uncertainty in the model decoupling. To ensure analytical accuracy and avoid making highly uncertain inferences, the simulation results for the primary dust-generating months (March to May) between 2010 and 2015 were compared with the more abundant, quality-controlled spring observations from AERONET.

The Shao04 dust emission scheme has been introduced into the Common Land Model (CoLM), thereby enabling the CAS-ESM2 model to produce a more complete representation of land-atmosphere interactions48. This parameterization scheme calculates dust emissions generated by saltation impacts across varying particle sizes. Based on the USDA soil texture classification, the soil particle size distributions (\({p}_{m}(d)\) and \({p}_{f}(d)\)) are represented as multi-lognormal combinations. This approach makes the emitted dust particle size distribution dependent on both surface soil properties and dynamic processes, yielding good agreement with field observations49. The present study configures the fifth-generation atmospheric general circulation model developed by IAP (IAP AGCM5) with CoLM, implementing the Shao04 dust emission scheme for dust process simulations. The calculation of dust emission fluxes is performed in CoLM, and these data are subsequently transferred to IAP AGCM5, where the Bulk Aerosol Module (BAM) is utilized. Aerosol size distributions are grouped into four size bins. Each bin is assumed to follow a log-normal distribution, thereby ensuring the preservation of particle size information50,51. From a climate-scale perspective, the atmospheric residence time of coarse particles is typically shorter, which often attenuates the cumulative effects of long-term climate feedback. Consequently, in our simulations, we primarily employed a universal particle size distribution (0-10 μm). The following formula is employed to calculate dust particle emission flux:

$$F\left({d}_{j},{d}_{i}\right)={c}_{y}{\eta }_{{fj}}\left(\left(1-\gamma \right)+\gamma \frac{{p}_{m}({d}_{j})}{{p}_{f}\left({d}_{j}\right)}\right)\frac{Q\left(i\right)g}{{u}_{* }^{2}}\left(1+{\sigma }_{m}\right)$$
(3)
$${\sigma }_{m}=12{u}_{* }^{2}\frac{{\rho }_{p}}{p}\left(1+14{u}_{* }\sqrt{\frac{{\rho }_{p}}{p}}\right)$$
(4)

In this equation, \({c}_{y}\) is a coefficient without dimensions, \({\eta }_{{fj}}\) denotes mass-specific gravity, \(\gamma\) represents a weighting factor linked to the particle size distribution, and \({p}_{m}({d}_{j})\) and \({p}_{f}\left({d}_{j}\right)\) are the minimum and maximum disturbance to the particle size distribution, correspondingly. The quantity \(Q\left(i\right)\) is employed to measure the flux of sand particles with a diameter of \({d}_{i}\). The gravitational acceleration constant \(g\) is also considered, while the symbol \({u}_{* }\) is used to denote the friction speed. The parameter \({\sigma }_{m}\) is defined as the efficiency of the bombardment process.

Subsequently, based on the effective radii of four size bins in the bulk aerosol module and the newly estimated k, the dust optical parameters (including mass extinction coefficient, single-scattering albedo, and asymmetry factor) for the 14 shortwave bands required by the RRTMG radiation scheme were calculated using Mie scattering theory. These parameters replaced the dust optical parameters calculated using OPAC and were simulated using the Chinese Academy of Sciences Earth System Model Version 2 (CAS-ESM2). Each shortwave band covers a specific wavelength interval to capture the critical absorption and scattering behavior of dust in solar radiation. This method is based on the spherical particle assumption and is generally suitable for radiative transport modeling of sand and dust aerosols. Furthermore, it is imperative to undertake a diagnostic calculation to ascertain the instantaneous DRE of dust aerosols. This involves removing the species from all modes that contain it. A new mode definition excluding dust aerosols is created based on the default definitions. The distinction between these two configurations is pivotal in diagnosing dust’s SW DRE. The specific formula is as follows:

$${R}_{T}{=}\Delta {{FS}}_{{TD}}-\Delta {{FS}}_{{TN}}$$
(5)
$${R}_{S}=\Delta {{FS}}_{{SD}}-\Delta {{FS}}_{{SN}}$$
(6)

In this equation, R denotes the calculated radiative effect; subscripts T and S represent the TOA and the SFC, respectively; D and N denote the default definition containing dust aerosols, as well as the new definition without dust aerosols; and \(\Delta {FS}\) represents the shortwave radiative flux at a given level.

AERONET

AERONET is a network specifically designed for the monitoring of aerosols. It was established by NASA (United States) and CNRS (France)52. The present study employs Level 2.0 dataset (high-quality assured), encompassing SSA, AAOD, aerosol optical depth (AOD), and Ångström exponent (AE) at a spatial resolution of 15 min.

Specifically, the dataset utilized includes AOD, SSA, and AAOD at the 550 nm wavelength band (denoted as \({{AOD}}_{550}\), \({{AAOD}}_{550}\), and \({{SSA}}_{550}\), respectively) and the AE over the 440–870 nm wavelength band (denoted as \({{AE}}_{440-870}\)). The \({{AOD}}_{550}\), \({{AAOD}}_{550}\) and \({{SSA}}_{550}\) values are derived through logarithmic interpolation of data from the 440 nm and 675 nm wavelength bands. As illustrated in Table S3, the study incorporates a comprehensive inventory of the AERONET sites used in this investigation, accompanied by their geographical locations and temporal coverage. To ensure data quality, the following screening criteria are applied: instances where the AOD at 440 nm is elevated (AOD440 ≥ 0.4) and the AE over the 440–870 nm band is diminished (AE440-870 < 0.6) are classified as predominantly influenced by dust aerosols on that particular day53. Although this method enhances the significance and fidelity of dust signals, residual non-dust pollution remains difficult to quantify.