AERONET is a ground-based aerosol remote sensing network providing long-term observations of aerosol optical and microphysical properties, covering most of the continental areas around the world . The AERONET AOD observations are derived from direct solar radiation at several wavelength bands mainly ranging from 340 to 1640 nm, while other aerosol properties, including SSA and AAOD, are derived from diffuse sky radiance at four wavelengths at 440, 675, 870, and 1020 nm . The AE parameter is calculated using AOD measurements within the 440–870 nm interval . There is a series of quality-assurance strategies for AERONET Level 2.0 data that ensure an AOD uncertainty of 0.01 (visible) to 0.02 (UV) and an SSA uncertainty of 0.03 at AOD440 (AOD at 440 nm) ∼ 0.4 . However, as Level 2.0 quality assurance for inversion products requires a coincident AOD exceeding 0.4 at 440 nm, many stations do not have enough data samples to produce a long-term record. Therefore, considering both the data quality and data availability, we utilize the all-point Version 3 Level 2.0 direct measurements for AOD and AE and quality-controlled Level 1.5 almucantar inversion products (see below for the quality control scheme) for other parameters. The description and uncertainties of these parameters are detailed in Sect. 2.2.

The stations are selected primarily based on the availability of an extensive data record for the purpose of estimating the long-term trends in aerosol properties. The Level 1.5 almucantar inversion products are first screened based on all the Level 2.0 quality-assurance criteria except for the AOD threshold, such as solar zenith angle > 50°, sky error < 5 %, and coincident Level 2.0 AOD measurements. The Level 2.0 direct measurements and screened Level 1.5 almucantar inversion products are then used to calculate monthly measurements. Long-term trend analysis necessitates homogeneous time series, and outliers would influence the result. We first check the records, removing invalid and abnormally high or low values (such as SSA below 0.7 for all stations and AOD above 2.0 for low-AOD stations) from all-point measurements. Then we calculate the median of all-point measurements to represent the monthly value only if there are more than five all-point measurements on at least 3 different days for that month. To ensure adequate records and data continuity in trend analysis, we require the data to have at least 10 years of records with no less than 8 months of measurements for each year during the 2000–2022 period. Years with less than 8 months of data and seasons with less than 10 years of records are discarded due to poor annual and seasonal representation. We also remove the first or last several months from the time series of certain stations (Canberra and Ilorin) where discontinuities were identified relative to adjacent monthly records. Considering polar stations often have no monthly measurements in winter, the least number of monthly medians for each year are reduced to 4 for stations at latitudes above 65°. Specifically, the 2019–2022 data for Birdsville in Australia are eliminated for more accurate trend estimation, as these data are strongly biased due to a data-filtering artefact in the quality-assurance (QA) process of the algorithm according to , which results in a large jump in AOD (Thomas Eck, personal communication, 2024). This AOD artefact is caused by the erroneous time-stamping of the data that is greatest at some sites in Australia due to a unique data logging system utilized there. The unnatural increase in AOD for Birdsville in 2019 can be found in . As a result, 172 stations for the direct-sun observations and 72 stations for the inversion measurements are retained for trend analysis, covering all major continents in the world. Locations, trends, and time series for all the stations could be found in the Supplement. The distributions of all the selected stations as well as the number of annual mean samples at each station are presented in Fig. . Locations of stations mentioned in this article are presented in Fig. .

Here we focus on analysing AOD, SSA, and AAOD trends at 440 nm, which are noted as AOD440, SSA440, and AAOD440, respectively. Trends for parameters at the other wavelengths are very similar and thus skipped. The AE is calculated from all AOD measurements within the 440–870 nm wavelength range (typically including 440, 500, 675, and 870 nm) and are commonly denoted as AE440_870.

AOD represents the column aerosol extinction, directly reflecting the column loading and concentration of aerosols. AERONET Level 2.0 products provide very accurate AOD measurements under clear-sky conditions, with an uncertainty of 0.01 at visible wavelengths. The patterns of AOD (Fig. ) and AOD trends (Fig. ) should always be kept in mind when analysing trends in the other aerosol parameters because uncertainties of the other parameters are closely related to AOD level (see below), whose trends reflect changes in aerosol loading.

The AE parameter describes the slope of the logarithm of AOD versus the logarithm of wavelength , characterizing the wavelength dependency of AOD. AERONET AE440_870 products are calculated from the linear regression of AOD and wavelengths on a logarithmic scale within the range of 440–870 nm . The AE parameter closely correlates with aerosol particle size distribution and is an indicator of aerosol components. For example, dust particles typically have AE440_870 values of around 0.3 or lower, and the AE440_870 values for fine-mode particles that are mostly anthropogenic usually exceed 1.0 . Therefore, AE440_870 can reflect the relative fraction of fine- and coarse-mode particles. The error in AE can be estimated by the error in AOD as 1ΔAE=∑i=1nei2(n-2)∑i=1n(ln⁡λi-ln⁡λ‾)212, where ei is the error in the Ångström relationship, n is the number of wavelengths λi used to fit the Ångström relationship, and ln⁡λ‾ is the average of the logarithm of the wavelengths. ei can be estimated using the relative error in AOD (ΔAODAOD), and the uncertainty of AERONET AOD (ΔAOD) is considered to be 0.01 here. According to Eq. (), the uncertainty of AE is roughly inversely proportional to AOD, with larger errors under lower-AOD conditions. evaluated that the uncertainty of AE440_870 was 0.33 when AOD440 = 0.15, and the uncertainty would rapidly increase to 0.56 when AOD440 decreased to 0.08. also demonstrated significant variability in AE440_870 for lower AOD, largely attributed to increased relative errors in AOD at these low values. These results correspond to the inverse relationship between ΔAE and AOD. Therefore, it should be noted that AE440_870 is highly uncertain and the AE440_870 trends are less robust for sites with low AOD, even if the trends are statistically significant.

AAOD and SSA together characterize the scattering and absorbing properties of aerosols. AAOD represents the total aerosol absorption optical depth, whereas SSA reflects the relative contribution of scattering to total extinction. Therefore, the AAOD trend directly reflects changes in the amount of absorbing aerosols, while the SSA trend is related to variations in both absorbing and scattering aerosols. The relationship between the two parameters can be expressed as the following equation: 2AAOD=1-SSA×AOD.

The uncertainties of AAOD and SSA are also closely related to AOD level. AERONET implements a series of quality control criteria for Level 2.0 inversion products. Under these controls, AERONET SSA has an error of ±0.03 when AOD440 ∼ 0.4, and the error increases rapidly (exponentially) at lower AOD levels, i.e. an error of ±0.05 when AOD440 is ∼ 0.2, and of ±0.07 when AOD440 is ∼ 0.1 . Therefore, although we utilize all the Level 2.0 quality-assurance criteria except for the AOD threshold for AAOD and SSA data, many of the SSA retrievals in this study have larger uncertainties of ∼ 0.03 to ∼ 0.09 due to the low AOD level. Moreover, as SSA typically varies from approximately 0.8 to 1.0 , this error is remarkable when examining the variation in AAOD and SSA; i.e. a 0.03 error in SSA would lead to a 15 % uncertainty. Therefore, the great uncertainties of AAOD and SSA should be kept in mind when analysing their trends, especially for regions with low aerosol loadings.

Here we use Sen's slope combined with Mann–Kendall test to estimate the trend and its significance. The Mann–Kendall (MK) test is a non-parametric method to assess the significance of monotonic trends in a dataset without assuming any particular distribution. The slope of the trend k can be estimated by the median of the set of slopes : 3k=median(Yj-Yitj-ti),∀j>i, where Yi and Yj are the values of the variable at times ti and tj, respectively.

The significance of the trend could be tested by calculating the MK statistic: 4S=∑i=1n-1∑j=i+1nsgnYj-Yi, where 5sgn(x)=-1ifx<0,0ifx=0,1ifx>0, which has a normal distribution with zero mean and variance of 6σ2=n(n-1)(2n+5)-∑j=1ptj(tj-1)(2tj+5)18.

Sen's slope associated with the MK test is a robust measurement of the trend in a dataset and is not sensitive to outliers. As aerosol optical parameters do not follow a normal distribution and AERONET records often have missing data, Sen's slope is a good estimator of trends.

It should be noted that the MK test requires serially independent data, necessitating the removal of autocorrelation from the time series before calculating trends . Several pre-whitening methods are available to remove serial correlation, with providing a comprehensive comparison of these approaches. In this study, we apply the 3PW method developed by to eliminate autocorrelation before computing the trend.

Aerosol parameters typically exhibit strong seasonality, which should be taken into account in the analysis. We conduct seasonal MK tests and calculate seasonal trends on the pre-whitened time series, and we then derive the annual trend as the median of seasonal trends . The homogeneity of the seasonal trend is also tested, and the results are marked in the annual trend maps. The definition of the seasons is primarily based on regional climatic characteristics. Specifically, seasons for South Asia are divided into pre-monsoon (March–May), monsoon (June–September), post-monsoon (October–November), and winter (December–February). For the Arabian Peninsula, the seasons are categorized as pre-peak (November–February), peak (March–June), and post-peak (July–October) . In West Africa, the seasons are classified as Harmattan (November–March) and summer (April–October) . For the other regions, the standard seasonal divisions of spring (March–May), summer (June–August), autumn (September–November), and winter (December–February) are applied.

In addition to the retrieved parameters, we also classify the observations into six aerosol types using the fine-mode fraction (FMF) at 550 nm and SSA at 440 nm . AOD and fine-mode AOD at 440, 675, 870, and 1020 nm are first interpolated to 550 nm using a second-order polynomial fit on a logarithmic scale . Then the FMF550 is calculated by AOD and fine-mode AOD at 550 nm. The classification criteria for the six aerosol types (“dust”, “mixture”, and four fine-mode types), as well as the proportion of each type in the total number of quality-controlled Level 1.5 all-point records, are listed in Table . Sea salt aerosols typically having FMF550 below 0.4 and SSA440 around 0.98 (included in the “uncertain” type in Table ) are not considered in the analysis of aerosol type trends (Sect. 3.3) because most AERONET stations are located over land where sea salt is not the predominant type, and sea salt aerosols only account for a negligible proportion (about 2.5 % for the uncertain type).

Each quality-controlled Level 1.5 inversion all-point measurement is classified as a specific aerosol type according to the classification criteria in Table 1. For each aerosol type, we use coincident Level 2.0 AOD440 measurements to calculate the monthly AOD and analyse its trend.

The AOD440 trends at the 172 selected AERONET stations are presented in Fig. . Trends surpassing the 90 % significance level are marked with dots. Trends below the 90 % significance level are marked with triangles. Trends passing the seasonal homogeneity test at the 80 % confidence level are marked with magenta boundaries. The AOD440 time series at several representative sites are shown in Fig. . Significant negative AOD440 trends are found for the majority of stations all over the world, demonstrating a global reduction in aerosol loading. This result is consistent with previous studies . An increased number of stations with significant trends compared to these previous studies are observed in North America, Europe, and the Mediterranean, likely due to the spatial and temporal expansion of the network in recent years. The rates of AOD440 reduction in western Europe (typically below -0.05 per decade) are not as substantial as that reported in , which was -0.1 per decade, suggesting a decelerated aerosol reduction rate in Europe in recent years. This is also in line with the AOD440 time series at representative European sites (Fig. g, h). Stations in the Arctic also exhibit coherent negative AOD440 trends, consistent with previous studies . Since a significant proportion of aerosols in the Arctic are transported from lower latitudes, the reduction in aerosols in the Arctic is in line with the general reduction in AOD observed in the Northern Hemisphere. Strong negative AOD440 trends are identified at more than 10 stations in East Asia and Southeast Asia, which were previously reported as exhibiting no significant trends in global studies . The most considerable AOD440 reductions are observed in East China, with significant declines of -0.1 per decade. However, the trend of AOD440 in East Asia is not coherent throughout the period of 2000–2022. According to the AOD440 time series (Fig. a–d), AOD440 increased in the early 2000s and decreased rapidly in the later years since around 2008, consistent with other regional aerosol trend studies . This result also explains why found no significant AOD440 in East Asia with shorter records, as the increase in AOD440 in the early 2000s offset the reduction after 2008. When applying longer records, the continuous reduction in AOD440 after 2008 becomes dominant.

Significant positive AOD440 trends are mainly found over Solar_Village in the Arabian Peninsula and over Kanpur in northern India. The Level 2.0 AOD440 records at Solar_Village (Fig. f) ended in 2013, limiting current insights into aerosol properties in the Arabian Peninsula. Kanpur (Fig. e) has extensive records over the past 2 decades, exhibiting a positive AOD440 trend of 0.068 per decade. This value is close to the trends calculated from different periods in previous studies , indicating a steady increase in AOD440 there. Significant positive AERONET AOD440 trends over the other regions, such as Trelew in South America and some oceanic island stations, are generally weak, with magnitudes typically below 0.03 per decade. As these sites have very low AOD440 (typically below 0.1 for monthly values) as well as low AOD440 variability, the results in these stations are typically more uncertain. The positive AOD trend for Birdsville in Australia was confirmed by the independent research conducted by ; however this was a false trend resulting from a previously mentioned data screening anomaly. The positive AOD440 trends over oceanic stations worldwide suggest a widespread increase in oceanic aerosols, primarily sea salts. This result is consistent with , who also reported an increase in oceanic AOD.

Significant negative AE440_870 trends are universally found for stations across Europe, the Mediterranean, and eastern North America (Fig. , Fig. ). The Arabian Peninsula also exhibits a negative AE440_870 trend, although the trend is not significant. Stations in western North America and northern India mainly exhibit positive AE440_870 trends. The significant negative AE440_870 trends for Europe, the Mediterranean, and eastern North America are likely due to reductions in fine-mode anthropogenic aerosol and precursor emissions. The Arabian Peninsula is a well-known dust source , and the AE440_870 values are typically low (Fig. b); therefore the negative AE440_870 trend for Solar_Village is likely attributed to increased dust activity. In northern India, considering the seasonal cycle of the AE440_870 value, the positive AE440_870 trends primarily result from increased fine-mode anthropogenic emissions as well as decreased coarse-mode dust loading. These shifts in anthropogenic emissions have been assessed through satellite observations and emission inventories , and the decline in dust loading over South Asia was also verified by satellite observations and AERONET measurements . The increased AE440_870 in western North America might be partly due to both increases in biomass burning aerosols and possibly diminished dust sources. These inferences align with previous studies, as also reported positive dust trends in the Middle East and negative trends in North America, whereas and revealed increases in biomass burning emissions over western North America. Two Arctic stations located in Europe (Andenes and Hornsund) exhibit significant positive trends in AE440_870, indicating decreased coarse-mode natural source emissions. However, the results are associated with considerable uncertainty due to the low aerosol loading in the Arctic. East Asia exhibits no significant AE440_870 trends, indicating weak changes in the ratio of fine-mode and coarse-mode aerosols. Therefore, the great decrease in aerosol loading in East Asia revealed in Fig.  might be related to similar reductions in both anthropogenic fine-mode aerosols and coarse-mode dust in these areas.

The points with black borders in Figs.  and  indicate that the seasonal trends at these sites are not homogeneous at the 80 % confidence level. This is largely attributed to the seasonal patterns of aerosol emissions and meteorological conditions. However, the spatial distribution of seasonal AOD440 (Fig. ) and AE440_870 (Fig. ) trends is generally similar to that of annual results with the same monotonicity despite the magnitude of the trends potentially varying by season. Certain stations also exhibit significant trends only during particular seasons. This variation in trend magnitude and significance accounts for the seasonal heterogeneity of AOD440 and AE440_870, while the consistent monotonicity across different seasons emphasizes that the overall changes in AOD440 and AE440_870 are uniform. Specifically, in Europe and North America, a greater number of stations exhibit significant AOD440 trends in MAM (Fig. a), with these trends exhibiting greater magnitude compared to other seasons. Three Arctic stations located in North America also exhibit significant negative AOD440 trends only in MAM. Conversely, AE440_870 trends found for Europe and North America are more pronounced and exhibit greater deviation from the annual results during DJF (Fig. d). This is because aerosol concentrations are typically higher in spring and lower in winter in the Northern Hemisphere (see Supplement), allowing for more substantial reductions in spring and more significant compositional variations in winter. However, AE440_870 trends during low-AOD seasons, such as the more pronounced AE440_870 trends in winter in the Northern Hemisphere, should be treated with caution because the uncertainty in AE440_870 becomes large under low-AOD conditions. In northern India, Kanpur only exhibits significant AOD440 trends during the post-monsoon (Fig. c) and winter (Fig. d), while significant AE440_870 trends predominantly occur in the pre-monsoon (Fig. a). We can find that AE440_870 values in northern India (Fig. a) significantly exceed 1.0 in the post-monsoon and winter, suggesting the predominance of fine-mode anthropogenic aerosols in these seasons. In contrast, AE440_870 values in the pre-monsoon and monsoon start at approximately 0.2–0.3 in the 2000s, emphasizing the dominance of coarse-mode aerosols, and have risen to about 0.7 in recent years, suggesting a largely increased fraction of fine-mode aerosols. The seasonal patterns of AOD and AE in South Asia have also been verified through multi-year observations . During the pre-monsoon and monsoon seasons, higher wind speeds and stronger precipitation lead to stronger dust activity and higher wet scavenging of aerosols, whereas in the post-monsoon and winter the meteorological conditions become reversed, with weaker dust activity and less efficient wet removal of aerosols occurring . As a result, in the post-monsoon and winter, the increases in anthropogenic emissions, mainly post-monsoon crop residue burning and biofuel and fossil fuel burning in winter , have a negligible impact on changes in aerosol compositions and AE440_870 values but would lead to the significant positive AOD440 trends under less efficient wet removal. On the other hand, in the pre-monsoon and monsoon, stronger wet scavenging of aerosols makes the AOD trend less pronounced, and the dominant aerosol type, dust, is mainly affected by natural variability and exhibits a negative trend . Therefore, the increase in anthropogenic aerosols, i.e. biomass and biofuel burning emissions, fossil fuel emissions, and industry emissions , does not have a significant impact on the total AOD440 in these two seasons, but it serves to increase the fine-mode fractions, leading to the insignificant AOD440 trends and significant positive AE440_870 trends.

Similar to AOD440, significant negative AAOD440 trends (Figs. , ) are universally found for AERONET stations in the Northern Hemisphere, especially in East Asia, northern India, Europe, and North America, indicating reductions in absorbing species, mainly primary aerosols. Conversely, a significant positive AAOD440 trend is mainly found for Solar_Village in the Arabian Peninsula (Fig. d), suggesting increases in absorbing aerosols. The reductions in AAOD440 over East Asia, Europe, northern India, and North America are primarily attributed to declines in anthropogenic emissions, such as reduced black carbon (BC) and/or organic carbon (OC) emissions from fossil fuels because aerosols in these regions are mainly of the urban/industrial type . Decreased dust emissions discussed in the previous section might also be a potential contributor to the negative AAOD440 trends in northern India and western North America , but the effect might not be as substantial as that of anthropogenic emissions, since dust is not the dominant type in these regions. The significant positive AAOD440 trend for Solar_Village in the Arabian Peninsula is likely attributed to increased dust loading. As dust mainly exhibits strong absorption for short wavelengths, AAOD trends at other channels with longer wavelengths might not be that significant.

The SSA440 trends (Figs. , ) are generally opposite to the AAOD440 trends, with exceptions in some stations in central Europe and North America. Northern India, East Asia, and Europe primarily exhibit significant positive SSA440 trends, corresponding to a decrease in the fraction of absorbing aerosols over time. The increase in SSA440 in northern India is attributed to both the decrease in absorbing species and a more pronounced increase in scattering aerosols. For East Asia and Europe, the positive SSA440 trends suggest stronger reductions in absorbing species than scattering aerosols. SSA440 trends show large spatial heterogeneity in North America, and most stations do not exhibit significant trends. The positive SSA440 trends for some stations correlate with the reduction in absorbing aerosols. Negative SSA440 trends for other stations in North America and several stations in central Europe are likely attributed to a great reduction in scattering aerosols such as sulfates, thereby increasing the proportion of absorbing aerosols. This result aligns with , who also found SSA reductions in North America and central Europe through in situ measurements and attributed them to significant decreases in primarily scattering secondary aerosols. The substantial reductions in precursors of these scattering aerosols were also confirmed by satellite observations and emission inventories . The significant negative SSA440 trend for Solar_Village (Fig. b) in the Arabian Peninsula is attributed to increases in absorbing dust aerosols.

The AAOD440 and SSA440 trends also exhibit pronounced seasonality, with half of the stations failing to pass the homogeneity test. However, the spatial patterns of seasonal AAOD440 (Fig. ) and SSA440 (Fig. ) trends are still similar to those of annual results. It is notable that Kanpur in northern India exhibits stronger negative AAOD440 trends during the monsoon (Fig. b) when dust is the dominant aerosol type, further verifying that the decreased AAOD440 is partly attributed to the decline in dust loading. As for SSA440, the significant positive trends for Kanpur are stronger during the post-monsoon and winter, indicating that the increased anthropogenic emissions in northern India are mainly scattering species.

To better explain the aerosol parameter changes, we make a further attempt to classify the measurements into six aerosol types, as described in Sect. 2.4, and examine the long-term changes in the loadings for each type. The global AOD440 trends in the six aerosol types are shown in Fig. .

A significant positive trend for the dust-type AOD is found for Solar_Village, suggesting increased dust activity over the Arabian Peninsula, which is consistent with the analysis in previous sections and other studies using satellite observations and AERONET measurements . We do not find significant trends over other dust sources, as dust loading can have strong decadal variability which often does not yield monotonic trends. A dust trend can also be difficult to detect when combined with fine-mode anthropogenic aerosols. The mixture type straddles the boundary between the dust type and fine-mode types and is affected by both coarse-mode and fine-mode particles. Significant negative AOD trends in the mixture type are mainly found over East Asia and Europe (Fig. b). Since East Asia and Europe are both dominated by fine-mode aerosols , the decreased mixture aerosols are thus primarily due to reductions in fine-mode anthropogenic emissions.

The majority of stations in Europe, North America, and East Asia exhibit significant negative AOD trends in four fine-mode types (Fig. c–f), corresponding to the reduction in both absorbing and scattering anthropogenic emissions revealed by the reductions in AOD (Fig. ) and AAOD (Fig. ) in these regions. The great reduction in absorbing types (SA, MA, and HA) is also the possible reason for the increase in SSA (Fig. ). It is notable that XiangHe in East Asia exhibits a significant positive trend in non-absorbing type (NA) and even larger negative trends in the absorbing types, suggesting a great reduction in BC and/or OC emissions which might potentially lead to a shift in the predominance of aerosol type in pollution events . Several stations in North America exhibit a greater reduction in non-absorbing aerosols than that in absorbing species, thus leading to a decrease in SSA (Fig. ). Kanpur in northern India exhibits positive trends in SA aerosols and negative trends in the MA type. Compared to MA, the SA type is more scattering with a lower BC proportion. As fine-mode aerosols in Kanpur are initially absorbing types , the increase in SA loading suggests a decreased proportion of BC, making the fine-mode aerosols in this region more scattering.