This study presents a multiparameter analysis of aerosol trends over the last 2 decades at regional and global scales. Regional time series have been computed for a set of nine optical, chemical-composition and mass aerosol properties by using the observations from several ground-based networks. From these regional time series the aerosol trends have been derived for the different regions of the world. Most of the properties related to aerosol loading exhibit negative trends, both at the surface and in the total atmospheric column. Significant decreases in aerosol optical depth (AOD) are found in Europe, North America, South America, North Africa and Asia, ranging from
As one of the key gears involved in the climate mechanism
Through their direct, semidirect and indirect effects
In order to provide realistic radiative-forcing estimates and projections, it is important for the atmospheric models to be able to capture the long-term aerosol trends caused by both natural and anthropogenic variations.
Assessing and improving the modeling of aerosols in global Earth system models is the main objective of the AeroCom project. Specific experiments are conducted within this initiative with a focus on individual aerosol species, such as dust
This study presents an overview of the aerosol trends for multiple aerosol parameters (optical and chemical) over the last 2 decades using ground-based observation network data as a reference for the evaluation of the models' skills in reproducing those trends.
To serve that purpose, this study addresses the following two questions:
What are the observed aerosol trends over the last 2 decades in the different regions of the world (Sect. Can the climate models reproduce these observed trends (Sect.
Then, having developed an understanding of the models' skills in reproducing the observed aerosol trends, the last section of this study aims to answer the following question:
What are the global aerosol trends derived from the model data (Sect.
The CAMS reanalysis dataset and output from six AeroCom models and four CMIP6 models (both model groups performed historical experiments) are evaluated in this work. CMIP6 (Coupled Model Intercomparison Project, Phase 6) is an intercomparison project organized by the WCRP (World Climate Research Programme). Participating models will contribute to the assessment of climate change in the upcoming 2024 IPCC (Intergovernmental Panel on Climate Change) report.
Figure
Global AOD computed from model historical runs (Oslo CTM3, GFDL-AM4, CanESM5, CESM2, IPSL-CM6A, ECHAM-HAM) at monthly (gray lines) and yearly (black lines) resolutions, overlaid with the number of active observation sites in the AERONET sun photometer network.
A set of nine columns and in situ surface aerosol datasets are used in this study. The observation networks and the models providing output for these parameters are reported in Table
List of observations and model datasets used in this study (see text for explanation).
For each of the parameters used in this study, data of the highest quality level provided by the different observation networks were used. Mountain sites, corresponding to an elevation above 1000 m, were excluded, mainly because global models have problems simulating the aerosol distribution in complex terrain
The Aerosol Robotic Network (AERONET) is a network that was established by NASA (National Aeronautics and Space Administration) and has been expanded by national and international collaborations. AERONET operates aerosol ground-based measurements in the different regions of the world
The particulate matter (PM) measurements are from EMEP (covering Europe) and IMPROVE (for North America). The PM data have been made available either via the EBAS database infrastructure (
The first PM measurements in EMEP started in 1996, and the number of sites increased steadily in the following decade
The IMPROVE network has been operating since 1988 at predominantly remote and rural sites across the United States. This ensures a good representativity of the measurements as some chemical species contributing to PM observations (i.e., organic carbon) can exhibit different seasonality and spatial variability. IMPROVE uses four separate modules to collect samples for speciated
The sulfate aerosol (
The data have been screened to be of satisfactory quality. Urban sites are not included and nor are sites where the surroundings have changed considerably in the period in question.
In
For the surface in situ PM measurements, the scattering and absorption coefficient measurements were accessed through EBAS database infrastructure. The level 2 data (quality controlled, hourly averaged, reported at standard temperature and pressure – STP – conditions) were used. Detailed information on the quality assurance and quality control procedures for GAW aerosol in situ data is available in
Scattering and absorption coefficients are measured by different instruments:
Scattering coefficients ( Absorption coefficients (
Due to the scarcity of stations available for long-term trend analysis (only 28), the presence of regionally nonrepresentative stations (e.g., stations located near roads, in cities), difficult to capture by global models, can have large effects on the computation of the regional-average time series. The urban stations have therefore been removed from this analysis.
Altogether the same data selection procedures (exclusion of stations, removal of outliers) and corrections (conversion to coefficients at 550 nm wavelength) were applied as in
A set of 10 climate and aerosol models and an aerosol reanalysis dataset are used in this study. Their main characteristics are reported in Table
Information on models used in this study (CAMS-Rean: CAMS reanalysis, AP3: AeroCom phase III, CMIP6: historical experiments from CMIP6).
Anthropogenic emissions (C
The CAMS reanalysis, which is the successor to the MACC (Monitoring Atmospheric Composition and Climate) reanalysis, is the latest global reanalysis dataset of atmospheric composition produced by the Copernicus Atmosphere Monitoring Service
Initiated in 2000, the AeroCom project (
In this study, we use the model outputs from the historical AeroCom experiment, whose main aim is to understand the regional trends in aerosol distribution from 1850 to 2015 and to quantify the aerosol forcing with a main emphasis on the direct aerosol effect. The models were also run in various configurations, providing different degrees of constraints on the evolving meteorological conditions, such as using monthly fixed sea-surface temperature (SST); historically evolving SSTs; and basic meteorology fields, e.g., wind for a given year.
The upcoming 2024 IPCC sixth assessment report (AR6) will feature new state-of-the-art CMIP6 models with model runs in higher resolution and with new physical processes. An overview of the experimental design and organization can be found in
Due to the nature of the processes involved in the emission and the deposition of aerosols, one can expect different trends in different regions of the world. Instead of investigating the trends obtained at each individual observation station in a given region, we resort here to the analysis of average regional time series as computed by assembling all measurements at stations in each region. One advantage of this method is that a single trend can be computed in a given region, with an associated significance and uncertainty. It is difficult, apart from in a diversity analysis, to define such an uncertainty when combining individual trends. Also, with our aggregation method, even a station that has not provided a sufficient number of data for computing a trend at its location can still contribute to the computation of a regional time series. The computation of such aggregated regional time series makes most sense in regions exhibiting similar seasonal patterns.
Seven regions are considered in this study. The definition of these regions has been determined in a pragmatic way to limit the number of geographic areas investigated, but altogether it also provides global coverage when considering the ensemble of all regions. The Americas and Africa have been separated into northern and southern sections. In order to assemble the sites most affected by Saharan dust, the North Africa region has been extended to the north beyond the continent itself. Stations located in the south of Spain, Cyprus and Greece contribute to the regional time series in the region we are calling North Africa. The regions' coordinates can be found in the Supplement.
As seen in Fig. AERONET is the most important network in terms of number of instruments. More than 1000 observation points, with more or less long time series, are found across the globe. The highest density of instruments is in Europe and in the central part of North America (USA). The lowest densities are found in southern Africa and Australia. For particulate matter, 227 instruments are used in this study and are spread mostly over Europe and North America. For For both parameters
Distribution of the observations within the different regions considered in this study. The numbers reported within each region correspond to the maximum number of stations given for the observation networks, corresponding to the five observation types found in the legend.
The regional time series are computed by combining, for each month, the valid data of all the stations in the corresponding region. In order to construct consistent and robust regional time series, some additional criteria are required to be met to provide a valid point (a station with valid measurements) for the regional time series. Stations with very short time series (e.g AERONET DRAGON campaign stations) are eliminated by requiring a minimum of 300 valid daily measurements in the whole period from 2000 to 2014, which reduces, as an illustration, the number of AERONET stations from 1015 to 437. A minimum of three valid stations is required to be present in the overall regional time series to produce a valid point. In other words, if the available time resolution is daily, at least three stations need to provide valid data for a certain day in order to produce a valid regional mean for that day. The list of the station names contributing to the computation of the regional time series can be found in the Supplement.
When all criteria are fulfilled for a given month in the regional time series, the median and the first and third quartiles are computed from all valid data points available. The quartiles provide an indication of the intraregional variability. An example of regional time series is shown in Fig.
Regional time series of AOD. The dark blue line corresponds to the median, and the light blue envelope is bound by the first and third quartiles of all valid points at the corresponding month, respectively. The blue dots correspond to the yearly averages which are used to compute the linear trend. The latter is displayed as a continuous line when the trend is significant and as a dashed line when it is not. Trend values, an error estimate and a significance value are given in each panel.
For all of the parameters, the trends are computed based on the yearly averages of the regional time series. Using the yearly averages eliminates any issues caused by the seasonal cycles (observed for most of the aerosol parameters used in this study) during the calculation of the trend slope. In order to ensure the statistical robustness of these yearly averages, the time averaging is performed step-by-step with specific time constraints. By starting at the finest time resolution available in the data, monthly, seasonal and then yearly averages are computed when the following criteria are fulfilled:
at least 5 d per month are available (when daily observations are available) at least 1 month with data per season is present (seasons defined as JFM, AMJ, JAS, OND) all four seasons are available for a given year.
These temporal constraints offer a reasonable compromise between the availability and robustness of the yearly statistics.
We use the same methodology as described by
An uncertainty is provided for each trend by combining the error in the slope calculation itself with the error in the residuals:
The trend is provided as a relative trend (% yr
The number of available points used to compute the regional time series is not constant in time. For a given observation station, the number of points available might vary in time due to the nature of the measurements. For instance, classic sun photometers only measure in the daytime and in cloud-free conditions. Due to seasonal daylight and cloud condition variations, clear seasonal cycles are observed in the number of observations of AOD. The density of the different observation networks can also change with time. The early development of the different observation networks usually coincided with an increase in the number of observation stations. More recently, primarily for funding reasons, some networks have reduced the number of stations. This variation in the number of available measurements raises the question of time representativity for the computation of the trends.
Associated with this time representativity issue is the space representativity issue. The data coverage is uneven across the different regions. Moreover, within a single region, the observation stations might be located in contrasting environments. Stations located in environments that are more urban, more rural or mostly affected by natural particles might have trends differing from the trend associated with the whole region.
Some studies have focused on the representativity of the observation stations by investigating the biases of different optical properties For Ref For Exp For Ref For Exp
The difference between the relative trends is computed for each parameter and region. In order to summarize the representativity, those differences are then converted into a score (
An example of the calculation is presented in Fig.
Three regional AOD time series and respective trends, constructed from model data (NorESM2) for the investigation of the representativity of the observational data. Panels
The trends in Europe show similar values for the time study, which means that the trend is not greatly affected by the variation in the available measurements in time. The difference is larger when considering all the grid boxes of the domain, but the overall difference in the two studies corresponds to a representativity of 76 %. In North America, the difference in the three trends is larger, particularly for the space study trend. This means that the trend obtained in the whole region is significantly different from the trend obtained when considering only the grid points where observation stations are located. It should be mentioned that the ocean grid points are not filtered out when computing the trends over the whole domain. For this reason, the regions containing a greater proportion of ocean grid points, where the trends are most likely to differ from those observed over land, will tend to have a lower spatial representativity, such as in North America.
This representativity study illustrates that the partial coverage in time and space of the observations leads, in some cases, to artificial trends. The representativity scores are discussed for each parameter in the following section together with the trend estimate results.
This section presents the trends in the observations computed for the different parameters and over the predefined regions. In order to compare the trends observed for the set of nine aerosol parameters in a consistent manner, we focus on the relative trends, with the reference set to the year 2000 as the first year of the study period. The means for the year 2000, reported in Table
Observational mean values for the year 2000, the reference year used for computing relative trends. Each value is extracted as the intercept of the linear trend computed in the 2000–2014 period, except for
The AOD is more than 3 times higher in Asia (AOD
The PM observations are primarily available from Europe and North America.
Analogous to the surface
The relative trends for the 2000–2014 period are shown in Fig.
In Europe, both columnar and surface parameters reveal statistically significant decreases. With the exception of
The representativity study reveals that the observed trends are actually representative of the whole period and region for all of the parameters. A good agreement is found with the trends obtained at individual stations reported by
In North America, similar trends to those in Europe are found for the columnar properties as were found for Europe. AOD decreases at a rate of 1.3 % yr
All of the columnar properties, except for
In North Africa, while significant decreases are found for all AOD parameters, an increase in the AE (
The AE also increases in Asia as a combination of a (not significant) increase in
No significant trends are found in Australia, although the representativity is greater than 50 % for
Regional trends in the aerosol properties computed with the observation datasets. The color of the circles corresponds to the slope, while the radius indicates the
This multiparameter trend analysis reveals a decrease in most of the parameters relating to aerosol burden, both in the total column and at the surface level. In Asia, the trends in
In order to evaluate the trends from the models, the regional time series have been computed with the model output co-located in space and time with the available observations at the station level. The model trends are computed in a similar manner to the trends for the observation datasets. The results, shown in Fig.
For AOD, the models show trends with the same sign as the observed trends over all the regions except in Asia, where the associated uncertainties are usually larger than the trend values. Some differences among the three model groups are observed when investigating the different regions:
For For For the AE, the trends are usually smaller than for AOD in the respective regions. This can mean that the number of particles is more subject to variations than the size (type) of these particles but could also illustrate that the AE is less sensitive to the change in a relative sense. This feature is visible with both observations and models.
For For For For
Regional trends in the aerosol properties computed with observations and models co-located in space and time with the observations. The error bars correspond to the uncertainty in the trend as calculated using both the uncertainty in the Theil–Sen slope and the residuals. The bold font indicates that the trends are significant at a confidence level of 95 % (
This model trends evaluation reveals some key points. First, CAMS-Rean, which assimilates AOD, performs the best for capturing the trends of this parameter. Second, a large intermodel variability is generally found over Asia, where the observed trends are also the most uncertain.
Considering the total column, the models usually perform rather well for AOD,
As discussed previously, the regional trends found are probably not always representative of the trends in the extended regions and over the whole study period. The reason for this is the partial spatial and temporal coverage of the ground-based observations. Moreover, the observation stations are obviously located on land. This does not allow for a depiction of global aerosol trends and is unfortunate as sea salt particles are among the most predominant aerosols on Earth
Unlike observations, models provide data at a global scale and for the entire study period. The completeness of these model datasets offers the opportunity to derive global aerosol trends. In order to provide an assessment of the aerosol trends at a global scale, we present, in this section, the trends computed with NorESM2 (CMIP6 group), which provides data for all of the nine parameters considered in this study. The calculation of the global trend is made by averaging the absolute trends computed at each grid point of the model and using all timestamps in the study period. In order to provide a relative trend, this absolute trend is normalized to the global average of the considered parameter for the year 2000. The global trends are reported for the nine aerosol parameters in Table
Global means and trends of aerosol parameters using NorESM2 model data. The values in parentheses are obtained by aggregating only grid points where observation stations are located while using the complete model time series. The relative trends are calculated by averaging the absolute trends within the considered grid points and normalizing this average to the global mean for the year 2000.
Global trends in aerosol properties using NorESM2 data regridded at a 5
While the observed trends in the three AOD parameters show a decrease in most of the regions of the world, the global AOD trend is actually positive (
The model also simulates an increase for the AE on a global scale, with a rate of
The trends in both
The surface
The
Table
The averaged global trend computed by NorESM2 indicates an increase in AOD in the 2000–2014 period with a rate of about 0.2 % yr
In this section, we investigate the trends in the major aerosol species simulated by NorESM2. For that purpose, the absolute trends in the individual contribution of these species to the AOD were computed, as well as the trends in the loads and the emissions. The trends in OD and loads are shown in Fig.
Absolute trends in OD and emissions of the main aerosol species computed with NorESM2. The
The relative increase in AOD of
The trends in OD do not necessarily represent the trends in the aerosol loads, which do not include associated water. The different species have different global mass extinction coefficients (calculated in this study as OD per load; dust – 1.8
The main findings of this multiparameter trend analysis are listed below:
The observations exhibit mostly negative trends regarding the extensive parameters in the different regions of the world. Significant decreases are found in Europe, North America, South America, North Africa and Asia. In Asia, the AE increases in time and is consistent with increases in Some observation networks allow for the derivation of representative trends over the whole study period. In other cases, the limited temporal and spatial coverage of the observations can induce artificial and/or highly uncertain trends when using regional time series. Among the 38 computed trends with observation data, 22 are considered as representative of the actual trends occurring in the whole region and study period. The models tend to capture observed AOD, AE, The rather good agreement of the trends across different aerosol parameters between models and observations, when co-locating them in time and space, implies that global model trends, including those in poorly monitored regions, are likely correct. The global trends computed with NorESM2 (CMIP6 group) model data give a different picture than the trends obtained when using only ground-based observations. Global positive trends are found for all of the parameters related to aerosol loading. The trends in AOD are dominated by the increase in the fine particles both in the column and at the surface. This tendency towards finer particles is consistent with the positive trend in the AE. This increase appears to be dominated by organic aerosols, of which the emissions have increased in the study period, and by
Some elements were not considered in this study, and they could be investigated in order to complete the aerosol trends picture:
Some regions are associated with strong seasonal cycles. In South America, the regional time series shows high peaks in AOD, associated with forest fires in the late summer, whose intensity greatly varies from year to year. In Africa, a strong seasonal contrast is also found due to the transport of desert dust at altitude in the summer months This study shows that the trends computed from the ground-based observation networks are not representative of the global aerosol trends due to the inhomogeneities in data spatial coverage. The satellites providing a global Earth observation could be utilized for the evaluation of the model trends in the regions lacking observations and over the oceans The trends in the meteorological parameters could be investigated in parallel with the aerosol trends because they affect the aerosol life cycle and their optical properties Several studies have linked the trends in anthropogenic aerosols to radiative-forcing variations while investigating sources of global dimming and brightening While the mountain sites were excluded from this study, it could be of interest to investigate the trends at higher altitude (which may be related to changes in long-range transport) by including the in situ and remote sensing stations higher than 1000 m (Jungfraujoch, Mauna Loa Observatory, etc.). Similarly, it may also be of interest to look at trends in smaller regions (e.g., split North America into several subregions which are more internally consistent in terms of climate and environment than the large North America region defined here or consider southern Europe as its own region rather than combining it with the North Africa region as was done here).
The observation and model data were read and co-located with the pyaerocom Python library, developed by MET Norway (
The supplement related to this article is available online at:
AM coordinated the study, was responsible for the statistical calculation and analysis, and wrote the paper. JG is the main developer of the pyaerocom library. MS provided feedback on the methods and the manuscript. WA, EA, JH and PL provided in situ data, contributed to the writing of the observation dataset section and provided feedback on the manuscript. BH is the principal investigator of AERONET. HB, MC, PG, HZ, ZK, AK, TL, GM, DN, DO, KvS, TT, RBS and ST provided model output data and feedback on the manuscript.
The authors declare that they have no conflict of interest.
This article is part of the special issue “The Aerosol Chemistry Model Intercomparison Project (AerChemMIP)”. It is not associated with a conference.
Data providers from all the regional and global networks are greatly acknowledged for sharing and submitting their
data to be used. The ECHAM-HAMMOZ model is developed by a consortium composed of ETH Zurich, Max Planck Institut für Meteorologie, Forschungszentrum Jülich, University of Oxford, the Finnish Meteorological Institute and the Leibniz Institute for Tropospheric Research and is managed by the Center for Climate Systems Modeling (C2SM) at ETH Zurich. Computing and data storage resources, including the Cheyenne supercomputer (
This research has been supported by the European Commission, H2020 Research Infrastructures (FORCeS (grant no. 821205), CRESCENDO (Coordinated Research in Earth Systems and Climate: Experiments, Knowledge, Dissemination and Outreach; grant no. 641816)), the National Science Foundation (NSF; grant no. 1852977), the Research Council of Norway (grant no. 295046) and the Copernicus Atmospheric Monitoring Service (grant no. CAMS84).
This paper was edited by Pedro Jimenez-Guerrero and reviewed by two anonymous referees.