A global analysis of climate-relevant aerosol properties retrieved from the network of GAW near-surface observatories

Aerosol particles are essential constituents of the Earth’s atmosphere, impacting the earth radiation balance directly by scattering and absorbing solar radiation, and indirectly by acting as cloud condensation nuclei. In contrast to most 105 greenhouse gases, aerosol particles have short atmospheric residence time resulting in a highly heterogeneous distribution in space and time. There is a clear need to document this variability at regional scale through observations involving, in particular, the in-situ near-surface segment of the atmospheric observations system. This paper will provide the widest effort so far to document variability of climate-relevant in-situ aerosol properties (namely wavelength dependent particle light scattering and absorption coefficients, particle number concentration and particle number size distribution) from all sites 110 connected to the Global Atmosphere Watch network. High quality data from almost 90 stations worldwide have been collected and controlled for quality and are reported for a reference year in 2017, providing a very extended and robust view of the variability of these variables worldwide. The range of variability observed worldwide for light scattering and absorption coefficients, single scattering albedo and particle number concentration are presented together with preliminary information on their long-term trends and comparison with model simulation for the different stations. The scope of the 115 present paper is also to provide the necessary suite of information including data provision procedures, quality control and analysis, data policy and usage of the ground-based aerosol measurements network. It delivers to users of the World Data Centre on Aerosol, the required confidence in data products in the form of a fully-characterized value chain, including uncertainty estimation and requirements for contributing to the global climate monitoring system.


5
Providing reliable observations of aerosol properties relevant to climate studies at spatial and temporal resolution suited to users is essential. For example, a measured decrease in pollutant concentrations would be the ultimate indicator of a successful policy to reduce emissions. However this requires long-term production and delivery of science-based data of known quality in terms of precision, accuracy and sufficient density of data points over the region of interest for the 165 measurements to be representative. Similarly, evaluating model performances from comparisons with observations requires that sets of high quality data are made available in comparable formats, with known uncertainties so that comparisons are meaningful. Current modelling tools are suited to the diversity of applications required by the disparate spatial and temporal scales of atmospheric impacts on climate, human health and ecosystems. There is still a need for accurate representation of observed aerosol which remains challenging, leading to considerable diversity in the abundance and distribution of aerosols 170 among global models. Capacity exists to deliver information products in a form adapted to climate policy applications in particular, but models need to be validated against measured atmospheric composition both in the short-and long-term (Benedetti et al., 2018).
One major aspect of aerosol forcing on climate is linked to its multi-variable dimension: optical properties of an aerosol 175 particle population are closely linked to its chemical, physical and hygroscopic properties and also to the altitudedependency of these parameters, which undergo significant short-term (diurnal) temporal variations. The effects of aerosol on climate are driven by both extensive and intensive aerosol properties. Aerosol extensive properties depend on both the nature of the aerosol and the aerosol particle concentration. In contrast, intensive properties are independent of particle concentration and instead relate to intrinsic properties of the aerosol particles (Ogren, 1995). Table 1 lists properties relevant  180 to the determination of aerosol climate forcing. We use the terminology proposed by OSCAR (https://www.wmosat.info/oscar/) and Petzold et al. (2013) for the specific case of black carbon. Some of the aerosol properties in Table 1 are recognized as aerosol Essential Climate Variables (ECVs) products for climate monitoring in the Global Climate Observing System (GCOS). The WMO/GAW Report No. 227 (2016) provides a synthesis of methodologies and procedures for measuring the recommended aerosol variables within the GAW network. The report identifies a list of comprehensive 185 aerosol measurements to be conducted as a priority as well as core measurements to be made at a larger number of stations.
It is clear that neither a single approach to observing the atmospheric aerosol nor a limited set of instruments can provide the data required to quantify aerosol forcing on climate in all its relevant dimensions and spatial/temporal scales (Kahn et al., 2017;Anderson et al., 2005). Observations from space through remote sensing methods are providing key information to accurately document extensive properties but are still not sufficient to provide information with the required degree of spatial 190 and temporal resolution needed for many applications. Further, remote sensing retrievals have only limited capabilities for determining aerosol chemistry, aerosol particle light absorption, particle size number distribution, Condensation Nuclei (CN), Cloud Condensation Nuclei (CCN) and Ice Nuclei (IN) (Kahn et al., 2017). Instead, in situ observations from stationary surface observatories, ships, balloons, and aircraft provide very detailed characterizations of the atmospheric https://doi.org/10.5194/amt-2019-499 Preprint. Discussion started: 11 February 2020 c Author(s) 2020. CC BY 4.0 License. 6 aerosol, often on limited spatial scales. Non-continuous mobile platforms such as aircraft and balloons provide the vertical 195 dimension, however, with limited temporal resolution. The current availability and accessibility of ground-based datasets on climate relevant aerosol properties vary substantially from place to place. An aerosol observing system for climate requires that all the types of observations are combined with models to extrapolate measurement points to large geographical scales against which satellite measurements can be compared (e.g., Anderson et al., 2005.

200
The in-situ segment of atmospheric observations is very complex and involves multiple partners, some are organized in measurement networks, active at regional or global scales, some are working almost independently. Networks support consistent, long-term measurements of atmospheric variables in order to detect trends and assess reasons for those trends. MPLNET, principally ADNET in Asia and MPLNET. Other lidars (CLN, CORALNET, ALINE) contribute to 220 GALION goals but are not at the same level of maturity or are solely regional in extent.
• Networks for the detection of in-situ aerosol properties, mainly divided into contributions from NOAA's Federated Aerosol Network (NFAN), encompassing sites primarily in North America but also including sites in Europe, Asia, and the southern hemisphere, including Antarctic sites (NFAN, Andrews et al., 2019) and ACTRIS (https://actris.eu) in Europe, but also including sites other WMO regions (https://cpdb.wmo.int/regions). In Europe, 225 the European Monitoring and Evaluation Programme' EMEP (https://www.emep.int), and, in the US, the IMPROVE network (http://vista.cira.colostate.edu/Improve/) are also providing key information on aerosol in-situ variables (Tørseth et al, 2012). Additional networks contributing to the provision of in-situ aerosol properties are https://doi.org/10.5194/amt-2019-499 Preprint. Discussion started: 11 February 2020 c Author(s) 2020. CC BY 4.0 License.

the Canadian Air and Precipitation Monitoring Network (CAPMoN), the Acid Deposition Monitoring Network in
East Asia (EANET) and the Korea Air Quality Network (KRAQNb) 230 Finally specific contributions are brought by the vertical profiles to in-situ observations routinely performed by IAGOS (Inflight Atmospheric Observing System), a contributing network to GAW and by additional ground-based observations operated outside the GAW context, such as SPARTAN (https://www.spartan-network.org).

Scope of the paper 235
The scope of the present paper is to provide the necessary suite of information to define a fully traceable ground-based aerosol measurements network, and to give an overview of the state of the operation in the network for a reference year. The paper should deliver to users of the World Data Centre on Aerosol (WDCA), the required confidence in data products in the form of a fully-characterized value chain, including uncertainty estimation and requirements for climate monitoring. 240 The paper is limited to a subset of the climate-relevant aerosol variables. It focuses on variables that are measured or derived from near-surface measurements, thus excluding all columnar and profile variables, despite their strong climate relevance. A second criteria for discussion in the paper is connected to the fact that long-term information is available at sufficient sites across the globe to derive trends and variability with sufficient robustness. Clearly, for many of the variables listed in Table  245 1, information is only available from a number of stations that are either almost exclusively documenting one single region (i.e. measurements of aerosol chemical properties with online aerosol mass spectrometers in Europe only) or not numerous enough to provide a robust assessment. In the case of EC/OC observations for example, information exists for many sites in different WMO regions but many of them no longer documented at the WDCA.

250
Finally, the last criteria is connected to the quality, intercomparability and accessibility of measurements worldwide, meaning that all information used in the paper must be well documented with rich metadata, traceable in provenance and quality, and accessible for all. This clearly limits the scope of the paper to the four independent climate-relevant variables mentioned above: i) particle light scattering coefficient, ii) particle light absorption coefficient, iii) particle number concentration, and iv) particle number size distribution. 255 For this set of variables, there has been, in the last decades, a significant international effort to harmonize the practice and methodologies across the frameworks, and strengthen systematic observations through different networks, or research infrastructure in the case of Europe, operating with a certain degree of interoperability. All networks jointly defined standard operation procedures (SOPs), conduct data collection in a timely and systematic manner, and promote open access and 260 https://doi.org/10.5194/amt-2019-499 Preprint. Discussion started: 11 February 2020 c Author(s) 2020. CC BY 4.0 License. 8 exchange of data without restriction through a unique data hub, the WDCA, hosted by NILU in Norway (https://www.gawwdca.org/). Operators from these networks perform joint assessments and analyses of data resulting in scientific publications that are discussed below.
This paper then provides a full-characterization of the value chain for these four aerosol variables that will serve for defining 265 the fiducial reference network in the future. It also provides an overview of the variability of the variables, and of some additional derived variables from the collection of data for the reference year 2017. The present paper is jointly written with companion papers, three of which one (Collaud Coen et al., (submitted), Gliβ et al., (submitted) and Mortier et al., submitted) are submitted in parallel with this paper. Gliβ et al., (submitted) and Mortier et al., (submitted) Gliβ et al. (submitted) and Mortier et al. (submitted) are presented in this paper. Additional manuscript are in preparation to further investigate variability of the optical and physical properties. This paper is integrated into a larger initiative called SARGAN (in-Situ AeRosol GAW Network) that will serve as the equivalent for GALION for the near-surface observations of aerosol variables. It is intended to support a future application 285 of SARGAN, and possibly other components of the GAW network, to become a GCOS associated network (https://gcos.wmo.int/en/networks). This requires the definition of threshold, breakthrough and goals for spatial and temporal resolutions that may be used for designing an operational aerosol in-situ network suited to global monitoring requirements in GCOS. Finally, this paper documents all elements required for establishing the GCOS network by addressing 1) the procedures for collecting and harmonizing measurements, data, metadata and quality control, 2) procedures for curation and 290 access to SARGAN data, 3) the available harmonised surface observations within SARGAN and status of the station network, 5) the present-day distribution of SARGAN aerosol properties and 5) requirements for using SARGAN for global climate monitoring applications. 9

Procedures for collecting and harmonizing measurements, quality control, and data curation and access
Controlling and improving data quality and enhancing their use by the scientific community is an essential aim within 295 observational networks. Procedures are continuously evolving as new instruments become commercially available and because efforts from the scientific community have resulted in more appropriate operation procedures for monitoring purposes. In the last decade, significant progress has been made in the harmonization of measurement protocols across the different networks and to ensure that all information is made readily available in a coordinated manner 300 In the GAW program, the individual station and its host organization are scientifically responsible for conducting the observations according to the standard operating procedures. This responsibility includes quality assurance of the instruments, as well as quality control of the data after measurement. In quality assurance, the stations collaborate with dedicated calibration centers, usually by sending their instruments for off-site calibration in regular intervals, and by station audits performed by relevant GAW Calibration centers 305

Harmonization of measurement protocols in SARGAN
Improving data quality and enhancing data use by the scientific community is an essential aim within GAW and the contributing networks. The measurement guidelines and standard operating procedures (SOPs) used for aerosol in situ measurements within GAW are discussed and prepared by Scientific Advisory Group (SAG) on "Aerosol" and accepted by the scientific community through peer-reviewed processes. The SOPs provide guidelines for good measurement practice and 310 are listed in WMO/GAW report #227 (2016) and connected reports.
The knowledge of the aerosol effect on climate and air quality as well as the techniques used for the determination of the essential aerosol variables to be monitored at ground-based sites have evolved considerably in the last decade. The methodologies, guidelines and SOPs are often elaborated and tested within the regional networks such as NFAN or the 315 European research infrastructure ACTRIS, and transferred to the GAW program to be adopted as Guidelines or more operational SOPs. SOPs are now available for almost all aerosol climate-relevant measurements, including for some of the most recent aerosol instruments.
The general guidelines for in-situ aerosol measurements in GAW are given in the general WMO/GAW report #227 (2016) 320 and in specific GAW reports such as WMO/GAW Report #200 (2011) for particle light scattering and absorption coefficients. Some of the recommended procedures are also adopted at a level of recommended standards by other bodies, such as EMEP under the UNECE, CEN (Center for European Normalization). This is the case for the measurement of the particle number concentration with condensation particle counters (CEN/TS 16976) as well as for the particle number size distribution with mobility particle size spectrometers (CEN/TS 17434). 325 10 In SARGAN, measurements of the particle light scattering coefficient are performed using integrating nephelometers, while measurements of the particle light absorption coefficient utilize various filtered-based absorption photometer instruments.
Both particle light scattering and absorption coefficients are dependent upon the size, shape, and composition of the particles as well as the wavelength of the incident light. Measurements of the particle light scattering and absorption coefficients 330 ideally would be performed at various wavelengths at a defined relative humidity. In GAW and the contributing networks, in-situ microphysical and optical aerosol measurements should be performed for a relative humidity (RH) lower than 40%, although some stations allow measurements up to 50%. Furthermore, information on the relative amounts of particle light scattering vs. absorption is required for radiative forcing 335 calculations and is defined by the aerosol single scattering albedo, ω0, which is the ratio of the particle light scattering coefficient over the particle light extinction coefficient, as defined in Table 1: ω0 = σsp/(σsp + σap). In this article, ω0 is computed for one specific λ (550 nm). The scattering Ångström exponent, AE, defined by the power-law σsp∝C0λ -AE , describes the wavelength-dependence for scattered light and is an indicator of particle number size distribution, and, thus, on the type of aerosol such as anthropogenic, mineral dust or sea salt. The scattering Ångström exponent can be directly derived 340 from the measured particle light scattering coefficients at different wavelengths. Müller et al. (2011) performed an intercomparison exercise for integrating nephelometers to propose procedures for correcting the non-ideal illumination due to truncation of the sensing volumes in the near-forward and near-backward angular ranges and for non-Lambertian illumination from the light sources. Müller's work expanded the initial findings of 345 Anderson and Ogren (1998), which were for a specific nephelometer model. Additionally, measurements of the dependence of the particle light scattering coefficient on the relative humidity are essential for the calculation of aerosol radiative effects in the atmosphere. This enhanced particle light scattering due to water take-up is strongly dependent on the particle number size distribution and the size-resolved particle composition. However, such measurements require an additional instrumental set-up, which has been implemented at only at very few stations and, with few exceptions, only on a campaign basis (Burgos 350 et al., 2019;Titos et al., 2016). Petzold and Schönlinner (2004) developed the filter-based Multi-Angle Absorption Photometer (MAAP), which can determine the particle light absorption coefficient directly, considering the light attenuation through and the backscattering above the filter. For other filter-based absorption photometers, the particle light absorption coefficient is determined from the 355 light attenuation through the filter, considering scattering cross-sensitivities and loading effects. The procedures to correct for scattering cross-sensitivity in Particle Soot Absorption Photometer (PSAP) instruments are described in Bond et al. (1999) and Ogren (2010). Several correction procedures for Aethalometers are given in Collaud Coen et al. (2010). Recently, https://doi.org/10.5194/amt-2019-499 Preprint. Discussion started: 11 February 2020 c Author(s) 2020. CC BY 4.0 License. 11 the ACTRIS community developed a harmonized factor for the AE31 to determine the particle light absorption coefficient, based on long-term intercomparison between Aethalometers and the MAAP for different environments and aerosol types 360 (WMO/GAW report #227, 2016).
The physical aerosol particle properties reported in this article are derived from the particle number concentration and number size distribution limited to the ultrafine and fine range. These measurements are performed using condensation particle counters (CPC) and mobility particle size spectrometers (MPSS). Wiedensohler et al. (2012) describes procedures 365 for long-term MPSS measurements and for their quality assurance. Since measurements of particle number size distributions are mainly restricted to ACTRIS sites and at a few other stations, a global assessment on aerosol physical properties can be only derived for the particle number concentration. For sites, where only MPSS data are available, the particle number concentration is determined from the integral over the particle number size distribution measured by the MPSS (see section 5.2 for discussion). Table 2 below summarizes all technical information related to the measurements of aerosol optical and 370 physical properties in SARGAN.

Curation and access to SARGAN data
In the management of data throughout its lifecycle, data curation is the activity that collects, annotates, verifies, archives, publishes, presents, and ensures access to all persistent data sets produced within the measurement framework and program. 375 The main purpose of data curation is to ensure that data are reliable and accessible for future research purposes and reuse. To this end, SARGAN data should be traceable to the original raw observational data, include version control and identification in case of updates, and include rich metadata going beyond discovery metadata (e.g., variable and station information) to use metadata (instrument description, operating procedures, station setting, calibration and quality assurance measures and uncertainties). SARGAN data are archived at WDCA, which is the data repository for microphysical, optical, and chemical 380 properties of atmospheric aerosol for the WMO/GAW programme.
To ensure traceability of data products, WDCA uses a system of 3 data levels: • Level 0: annotated raw data, all parameters provided by instrument, parameters needed for further processing, format is instrument model specific format, "native" time resolution. 385 • Level 1: data processed to final parameter, calibrations applied, invalid and calibration episodes removed, format is property specific, "native" time resolution, conversion to reference conditions of temperature and pressure (273.15 K, 1013.25 hPa). • Level 2: data aggregated to hourly averages, atmospheric variability quantified, format is property specific. 390 https://doi.org/10.5194/amt-2019-499 Preprint. Discussion started: 11 February 2020 c Author(s) 2020. CC BY 4.0 License.

12
Each higher data level is produced from the respective lower level as specified by the pertaining operating procedure. The templates for data level and instrument are published on the WDCA homepage and pages referenced from there, together with references to the relevant operating procedures. The templates indicate the metadata and data elements (discovery and use metadata) expected when submitting data to WDCA, which have been specified in collaboration with the GAW scientific advisory group (SAG) for aerosol and the GAW World Calibration Centre for Aerosol Physics (WCCAP) to 395 ensure that relevant and useful metadata are collected.
Stations report data to WDCA on an annual basis. After quality control, the station submits the data to WDCA via an online, web-based submission tool: https://ebas-submit-tool.nilu.no. In this process, the tool gives immediate feedback on syntax errors, and performs checks on semantics and sanity of both metadata and data. During curation at WDCA, the data files are 400 inspected both automatically and manually for metadata completeness and consistency, while the data are inspected for outliers, spikes, and sanity. Issues discovered in the process are reported back to the station, and the station asked to take corrective action and resubmit the data. The same applies for issues discovered after data publication.
By joining the GAW programme, stations commit to reporting their observations in a fully and manually quality controlled 405 version (level 2) on an annual basis, with a deadline of 31 December of the year following the data year to be reported.
WDCA encourages stations to report their data in a traceable way, i.e. to include data level 0 and 1 with their submissions.
GAW guidelines for quality control have developed and improved over the lifetime of the programme. At the beginning, quality control reflected the GAW objective of providing observations of atmospheric compositions with large scale 410 representativity. For this reason, observations influenced by local and regional emissions, or by regional phenomena, were flagged invalid during quality control and excluded from being archived. Later, it was acknowledged that atmospheric composition data serves multiple purposes and applications. This is reflected by the recommendation to only remove data affected by instrument issues or contamination during quality control, and indicate local or regional influence with a flag that leaves the data valid. This implies, for any application of WDCA data, filtering the data according to purpose is the first step. 415 When using WDCA data, this shift in quality control approach, which may vary among stations due to their scientific independence, needs to be taken into account.
The Global Atmosphere Watch, and the affiliated networks have agreed on a FAIR-use data policy encouraging an unlimited and open data policy for non-commercial use, provided without charge, unless noted otherwise. Users of WDCA are 420 encouraged to contact and eventually offer co-authorship, to the data providers or owners whenever substantial use is made of their data. Alternatively, acknowledgement must be made to the data providers or owners and to the project name when these data are used within a publication. All data related to the present article are available at the WDCA. https://doi.org/10.5194/amt-2019-499 Preprint. Discussion started: 11 February 2020 c Author(s) 2020. CC BY 4.0 License. 13 4 Procedures for collecting and harmonizing measurements, quality control, and data curation and access 4.1 A short history of aerosol monitoring networks 425 The first network designed to make long-term measurements of climate-relevant aerosol properties was the Geophysical Monitoring for Climate Change (GMCC) program, formed by NOAA in the early 1970's. GMCC was "designed to establish and maintain a program of observation and analysis of data representative of the global background of selected gases and aerosols" This focus on establishing a global background climatology meant that the stations were located at remote sites, far from human emission sources, in order to ascertain the extent to which human activities caused changes in climate-relevant 430 aerosol properties. The four initial GMCC stations were chosen to sample representative latitudes within both hemispherespolar, mid-latitude, and tropical, and were located at South Pole, Antarctica; Point Barrow, Alaska; Mauna Loa, Hawaii; and Cape Matatula, American Samoa. Two additional locations were initially planned, on the west coast of the USA and on or eastward of the east coast of the USA, but were not established until much later. As a consequence of the site selection criteria, the GMCC stations were not positioned to characterize the climate-forcing properties of aerosols in the regions 435 where the climate forcing was large, a weakness that was not addressed until the 1990's when NOAA established stations in and downwind of the continental USA and the GAW network was founded.
Aerosol particle number concentration was the first aerosol property measured at the GMCC stations, initially with manual expansion-type, water-based instruments and later with automated versions. The rationale for the choice of this variable was 440 that these very small particles "are present in all forms of combustion [products], such as those from automobiles, coal or oilburning power plants, and other human activities, it is essential to monitor the background tropospheric aerosol concentration in order to assess man's possible impact on his global environment". Recognizing that aerosols may play an important role in the global radiation balance, because they influence the heat budget and scatter or absorb both incoming solar radiation and outgoing terrestrial radiation, multi-wavelength measurements of aerosol particle light scattering 445 coefficient using integrating nephelometers were added at the four GMCC stations in the mid-to late-1970's. Although measurements of aerosol particle number concentration and light scattering coefficient were made during multiple, short-term field studies and in long-term studies at individual field stations (e.g., Gras, 1995), the next network to be established for these measurements was the IMPROVE (Interagency Monitoring of Protected Visual Environments) network 450 in the USA, which was initiated in 1985 to monitor visibility degradation in US National Parks and Wilderness Areas.

An overview of recent studies of variability and trends of aerosol in-situ optical and physical properties
The pioneering works of Bodhaine (1983;, Delene and Ogren (2002) for US sites, and Putaud et al. (2004), and Van Dingenen et al. (2004 for European sites are the first studies documenting variability of climate-relevant aerosol properties using long term observations performed at the network scale. Using long term observations performed at several 480 sites across the US, Delene and Ogren (2002) investigated the systematic relationships between aerosol optical properties and aerosol loadings that can be used to derive climatological averages of aerosol direct radiative forcing. The work of Putaud et al. (2004 and2010) and Van Dingenen et al. (2004) gathered information from long and medium term observations from rural, near-city, urban, and kerbside sites in Europe to highlight similarities and differences in aerosol characteristics across the European network. As more sites provided access to longer data sets, the next series of papers 485 (2010 up to present) addressed the issues of regional variability and trends with more robust statistical approaches and providing a comprehensive view of the aerosol variability to be used for model constraints.
Variability for the in-situ climate-relevant aerosol properties relevant to SARGAN are documented for many GAW stations.
Generally, the seasonal variability of number concentration, and of the scattering and absorption coefficients, is much larger 495 than diurnal variability at all sites (Sherman et al., 2015;Asmi et al., 2011) except at mountain observatories where meteorology plays a key role (Andrews et al, 2011;Collaud Coen et al., 2018). Typically, changes in aerosol intensive properties can be related to known sources. Timing of their maximum impact leads to well-defined seasonality that varies widely from site to site with the peak occurring at different times of year worldwide (e.g., Schmeisser et al., 2018). In Europe, some aerosol properties at non-urban/peri-urban sites can be divided into different typologies connected to large 500 geographical areas (i.e. Central Europe, Nordic, Mountain, Southern and Western European), for the different properties: carbonaceous aerosol concentration Zanatta et al., 2016;Crippa et al. 2014); optical properties (Pandolfi et al., 2018); number concentration (Asmi et al., 2011); number of cloud condensation nuclei (Schmale et al., 2017) or chemical composition (Zhang et al., 2007;Crippa et al. 2014). This feature was used by Beddows et al. (2014), to propose a representation of aerosol number size distribution in Europe with a total of nine different clusters for the whole continent. 505 Two recent studies addressed variability for specific areas, using measurements from Arctic stations (Dall'Osto et al., 2019) and mountain stations (Sellegri et al., 2019). Interestingly, none of the studies detected statistically significant regional workweek or weekday related variation for any of the aerosol variables, indicating that the stations are relatively free from local emissions and that regional effects dominate over local effects.

510
Time series longer than a decade are generally required to derive trends and a lesser number of studies are available, in particular those integrating information from large sets of stations. Statistically significant trends in σsp (decreasing), were found at 2 sites of NFAN in the US (analyzing trends from mid 90's to 2013) (Sherman et al., 2015). Similar results for a more globally representative set of sites were obtained for a comparison period of up to 18 years 1992-2010 (although less for some sites) by Collaud Coen et al. (2013); for mostly European sites by Pandolfi et al., (2018) for aerosol optical 515 properties (comparison period ending in 2015) and Asmi et al. (2013) for aerosol number concentration. Whenever a trend was detected, it was generally decreasing for the majority of the sites for almost all aerosol extensive variables. Exceptions (increasing trends) were found at several sites that could be explained by local features or by influence of emissions from the Asian continent. Decreasing trends have been reported in the literature for columnar AOD as well (e.g., Yoon et al., 2016;Zhao et al., 2017;Ningombam et al., 2018;Sogacheva et al., 2018). Decreasing trends in number concentration are 520 explained by reduction of anthropogenic emissions of primary particles, SO2 or some co-emitted species, as also shown by https://doi.org/10.5194/amt-2019-499 Preprint. Discussion started: 11 February 2020 c Author(s) 2020. CC BY 4.0 License. Aas et al., (2019) for sulfur species and Tørseth et al (2012) for PM10, PM2.5 and sulphate. In particular, Tørseth et al. (2012) show strong decreases, ca 50%, in the period 2000 to 2009 in PM10 and PM2.5. Decreasing trends (of the order of a few %/year for all variables were more pronounced in North America than in Europe or at Antarctic sites, where the majority of sites did not show any significant trend (e.g., Collaud Coen et al., 2013). 525 The difference in the timing of emission reduction policy for the Europe and North American continents is a likely explanation for the decreasing trends in aerosol optical parameters found for most American sites compared to the lack of trends observed in Europe. In fact, the decreasing trends in Europe for aerosol optical variables were more detectable in These studies did not find a consistent agreement between the trends of N and particle optical properties in the few stations with long time series of all of these properties; this is partly explained by the fact that aerosol light scattering coefficient is dominated by a different part of the aerosol size distribution than number concentration, and hence the two parameters are likely to have different sources.

535
The analysis of trends in aerosol properties needs to be regularly revisited as longer homogeneous time series become available at more sites, providing better spatial and temporal coverage. As shown in previous studies, trend and variability studies of aerosol properties still face some limitations due to heterogeneous time series, local effects that can only be addressed by some degree of redundancy among GAW stations, etc. It is also important to note that trends in terms of both statistical significance and sign are very sensitive to the period and the methodology used for the calculation. The fact that 540 different aerosol variables show opposite trends at some sites also suggests that further analysis is needed to better understand how the different aerosol parameters are connected to each other in the long term. These studies highlight the fact that other than in Europe and North America, and a few Antarctic stations, no trends can be derived due to lack of data from many areas in the world, as mentioned by Laj et al. (2010) 10 years ago! 545 Several studies have recently used in-situ measurements from, among others, the GAW network for a broad evaluation of the models, in particular in the framework of the AeroCom initiative (https://aerocom.met.no/): • Particulate organic matter concentration: Tsigaridis et al. (2014) have found for 31 AeroCom models, compared to remote surface in-situ measurements in 2008-2010, a median normalized mean bias (NMB) underestimate of 15% for particulate organic carbon mass and an overestimate of 51% for organic aerosol mass. This would indicate 550 OA/OC ratio in the models is too high, however, it is generally rather low and close to 1.4. While the bias values are robust at the sites investigated, it is assumed that the measurement data available at the time were not representative enough to provide robust global bias estimates for the models in question.

An overview of networks and organisations contributing to SARGAN 575
As mentioned previously, the data provision is organized independently resulting in a rather complex system where data originates from WMO/GAW Global, Regional, and contributing partner stations which themselves belong to one or more networks, depending on the station history and funding schemes. For example, many stations are labelled simultaneously as GAW, ACTRIS and EMEP in Europe, or GAW and NOAA in the US. Information on station status can be found in the GAW information system (GAWSIS). Registration to GAW does not exclude participation in other networks, either 580 contributing to GAW or not. WMO/GAW report #207 (2012), reviewed the situation with respect to the different aerosol networks operating globally. Although data for the report were collected in 2009-2010, the current situation is quite similar to 10 years ago.
According to the GAW information system (GAWSIS, http://www.wmo.int/gaw/gawsis/), as of June 2019 the GAW aerosol 585 network consists of 33 'Global' Stations', which are encouraged to participate in all the GAW measurement programmes and approximately 250 regional or contributing stations. Not all GAW stations are able to measure all aerosol variables listed in https://doi.org/10.5194/amt-2019-499 Preprint. Discussion started: 11 February 2020 c Author(s) 2020. CC BY 4.0 License. Table 1 and SARGAN is, therefore, a subset of stations in GAW. Contributors to SARGAN consist primarily of these international networks and research infrastructures: • Historically, there has been limited interaction among the different networks Worldwide, as mentioned in the WMO/GAW 610 report #207 (2013). However, on the specific issues of monitoring short-lived climate forcers, the main contributing networks to GAW have managed to integrate many pieces of the data value-chain, from SOPs, to QA/QC and data access.

An overview of networks and organisations contributing to SARGAN 615
All sites are established with the intention of operating in the long term. For registration to GAW (Global or Regional status) a period of successful performance of typically three years is required before a new site is added. All sites are long term in nature and, for most, adhere to rigorous siting criteria that aim to avoid local sources as much as possible. Sites have been and continue to be selected to answer pressing scientific questions, which evolve with time, and to detect and attribute changes in climate and climate forcing. 620 https://doi.org/10.5194/amt-2019-499 Preprint. Discussion started: 11 February 2020 c Author(s) 2020. CC BY 4.0 License.

19
Currently, 89 different sites worldwide are contributing to the provision of at least one SARGAN variable. These sites are indicated in Figure 1 and Table 3. Note that they are potential additional collocated sites not used in this study. All information used to compile information for this study is directly derived from NOAA-FAN and ACTRIS/EMEP with additional contributions from providers listed in Table 2. Except for a few sites, measurements from all sites comply with 625 the quality assurance and data reporting criteria defined in Section 3.1 and 3.2. If the sites are part of a contributing network, inclusion is straightforward in that the contributing network will already have met the GAW quality control and data reporting criteria. We have allowed a few exceptions for some sites located in WMO regions I, II, III and IV to ensure the widest geographical coverage as possible.

630
Because of the specific purposes for which NOAA-FAN and ACTRIS/EMEP were established, the nature of the sites is clearly biased to provide information relevant on the regional scale. This is why urban and peri-urban sites are underrepresented in SARGAN and that a majority of sites are sampling in environments far from local emission sources, with a station footprint that is generally quite large. The issue of spatial representativeness of observing stations has been addressed in many papers (e.g., Wang et al., 2018;Sun et al., 2019), and in particular related to air quality monitoring (e.g., Joly and 635 Peuch, 2012). Representativeness of a site describes how the measurements can be used to derive information for a given time or spatial scale, or for a given kind of environment. This information is key whenever ground-based observations are used to compare with space-based measurements or for evaluating models. However, defining station representativeness is not unambiguous and several papers exist with different definitions (Joly and Peuch, 2012).

640
Station representativeness is very often addressed using density plots identifying the most probable origin of air mass trajectories terminating at the station over a certain time (typically 3 to 6 days). Many stations in SARGAN can provide such analyses often performed to discriminate source areas influencing the site for climatological studies. Schutgens et al. (2017) discussed representativeness of ground-based observations both in terms of spatial and temporal averaging showing that significant errors may remain even after substantial averaging of data. Joly and Peuch (2012) developed a methodology to 645 build a classification of European air quality monitoring sites, mostly based on regulated pollutants.
In this paper, site characterization is made with a two-criteria approach: 1) a criterion describing the main geographical setting (e.g., polar, continental, coastal, mountain) and 2) a criterion providing indications about the dominant footprint (e.g., forest, rural, desert, urban, pristine, regional background, mixed). Additional details on some of these categories are 650 warranted. Mountain sites are not classified solely based on elevation (for example, high plateaux such as SPO and SUM are not considered mountain sites) but rather on the fact that the station is located higher than the surrounding environment.

20
For the air mass footprint, "Mixed" is used whenever no dominant air-mass footprint criterion is identified. This is often the case, for example, for mountain sites where air sampled during night differs from air sampled during day, due to local 655 orographic effects. "Pristine" is used whenever the site is located far away from any anthropogenic or natural sources.
Obviously, no simple site characterization can completely capture the influences on a location and we are aware of the shortcomings of this classification. In the context of the paper, this simplistic scheme was considered the easiest way to organize the statistical results. It should be mentioned that site characterization relies on authors' knowledge of the sites, along with indications by the corresponding PIs. 660

Evolution of data provision in SARGAN
In their 2013 papers, Collaud Coen et al. (2013) and Asmi et al. (2013), evaluated trends in aerosol optical and physical properties based on times series extending from 1993 to 2010. At that time, 24 sites worldwide had the capacity to provide a The present article provides an updated overview of the distribution of aerosol properties based on the information available in EBAS from sites listed in Table 3. The analysis is based on data collected in 2017 to provide the most updated view of measurements worldwide. The analysis is restricted to a very basic statistical overview (yearly and seasonal median, percentiles, average) that is completed, for some stations, by the trend analysis performed as part of Collaud Coen et al.
(submitted). To perform this analysis, we preferentially used data collected in 2017. In case the coverage for 2017 was 705 insufficient (see criteria below), data from 2016 was used. This is indicated in tables SM1 and SM2.
All sites contributing to SARGAN in 2017 were included in the analysis. The analysis is based on hourly data of σsp, σap and PNSD. Only validated measurements were used, i.e. data following the curation described in section 3.2, and, for an aerosol parameter, the datasets from the different stations were further harmonized (e.g. to ensure that the time-vectors and data were 710 of the right format and comparable with each other). Prior to the calculation of the summary statistics, a few problematic data points were also removed, following communication with the PI. For each site, annual and seasonal summary statistics were computed (median, 10th and 90th quantiles); the results were included only if 75% of the hourly data was available over the statistics reference period (with the exception of BRW, MLO and SPO whose respective coverage for each aerosol property is detailed in tables SM1 and SM2). In cases where the 2017 coverage was not sufficient (i.e. <75% for all seasons) 715 for an aerosol parameter (e.g., due to instrument failure or natural disaster impacting the station), the 2016 data was considered for that parameter. In cases where the coverage for that aerosol property was insufficient also for 2016 (i.e. <75% for all seasons), the site was discarded from the analysis for that aerosol property.

725
As mentioned in Table 3, many sites are actually influenced by different air-mass types, and some of them are influenced by anthropogenic sources. For most sites, data from all air masses are included in the statistical analysis. For BRW, MLO and SPO, the data included in this overview do not include all valid measurements collected at these three sites, but only the data corresponding to clean air masses. Clearly, in that case, the coverage criteria indicated above do not apply. This screening protocol, performed by the institutes operating the instruments, results in a lower annual data coverage and in a bias towards 730 lower levels but ensures data consistency with the multi-decadal data available from these sites. Absorption Photometer 3-wavelengths (PSAP-3W, Radiance Research Inc). It is important to note that data from Aethalometer AE33 (Magee Scientific, USA) were not used in this study as a unique value for converting the measured 740 attenuation coefficient to particle light absorption coefficient (σap) has not been fixed. The MAAP provides absorption at 637 nm (Mueller et al, 2011), the CLAP at 461, 522 and 653 nm (Ogren et al., 2017), the AE31 at 370, 470, 520, 590, 660, 880, 950 nm (Hansen et al., 1984)  scattering coefficients associated with a sample relative humidity less than or equal to 50% were used; this threshold, slightly higher than the prescribed 40%, allowed for more sites to be included, and was consistent with Pandolfi et al. (2018).

755
Single scattering albedo was computed at 550 nm using the optical properties closest to 550 nm for all multiple wavelength instruments. For σap by MAAP the data was adjusted to 550 nm assuming a constant AAE = 1.
For both σap and σsp, the effect of the difference in the instrument wavelength on the comparability of the data used for the summary statistics was considered negligible; the only exception was for the estimate of σap at 637 nm by AE16 and of σap at 760 550 nm by MAAP, for which a constant AAE = 1 was assumed.

Global variability of optical properties
The variability of aerosol absorption and scattering coefficient medians is presented in Figures 5a and 5b and in Tables SM1 and SM2 along with other main summary statistics. The range of variability of both σap and σsp is high, spanning several orders of magnitude, with variability at least partly explained by a few main drivers: site latitude, site geographic 765 location/footprint and the distance from the main anthropogenic sources. Globally the spatial variability of scattering and absorption has large similarities, being both featured by largest variability at mountain sites and minimum variability at urban polluted sites (e.g. LEI, IPR). Within the mid latitudes, absorption and scattering tend to increase from sites with a rural or forest footprint towards those in mixed and urban conditions. Polar sites, both in the Arctic and Antarctic, exhibits the lowest σap and σsp, occasionally below instrumental level of detection (LOD) for absorption. Besides polar sites, lowest 770 σap and σsp values are generally observed at mountain sites, e.g. JFJ, ZSF and MLO (whose data is screened for clean air sector and may partly explain the low value), along with the Pacific coastal background site of CGO. A similar situation is observed for the lowest σsp which, besides for pristine sites, are observed for mountain sites. Interestingly, the mountain site of JFJ, in Switzerland has a median σap and σsp lower than a few polar sites, i.e. ALT, BRW, PAL, ZEP, and ALT, BRW, NMY respectively. 775 The variability is generally higher at sites with low σap and σsp, reflecting the contrasting transport, in the case of pristine sites between the very low background values and the increase to advection of less clean air masses, and for mountain sites, the contrasting diurnal or seasonal transport patterns. A very good example is TIK, showing the largest medians among polar sites, where σap spans over one order of magnitude, reflecting the collection of both clean and polluted air masses, most 780 likely affected by biomass burning in the high latitudes.

24
The highest values and the smallest variability in both σap and σsp are observed for urban/peri urban sites (e.g. LEI, UGR, IPR). It is interesting to note that occasionally the rural stations as AMY (East Asia) and KOS (Central Europe) have median and range values of σap similar to urban sites, despite being located in rural areas far from local sources. PDI and BKT, both 785 mountain sites in Southeast Asian tropical forests, exhibit large medians for both σap and σsp compared to other forest/mountain sites due to recurrent impact by biomass burning (Bukowiecki et al., 2019). Similarly, biomass burning events related to anthropogenic emission from mainland China also affect via regional transport both LLN, another mountain site in SE Asia, and AMY.

790
At mountain sites in Southern Europe (MSA, HAC and CMN), a large scattering and absorption range is observed, comparable to that at rural background sites. This variability is partly due to the mixed nature of the sites, to long-range transport events (e.g., Saharan dust outbreaks, coal burning from Eastern Europe) and biomass burning both from forest fires in summer and domestic heating in winter. Saharan dust transport events partly explain the variability observed in other Southern European sites, e.g. FKL. 795 The seasonality of σap and σsp is presented in Figures 6a and 6b. The variability of the season median is much lower than the yearly variability reflecting the importance of transport in the variability. The most pronounced annual seasonality is observed at high mountain sites due to the seasonal variation of the boundary layer height and the local circulation induced by thermal winds that follow the ground temperature cycle. In the case of mountain sites, the seasonality is also reflecting the 800 index of boundary layer influence as defined by Collaud Coen et al. (2018). Generally, seasonality is largest at sites in an urban setting (e.g. UGR, NOA, LEI-M) and at those recurrently influenced by transport of either local or distant anthropogenic emissions (e.g. IPR, GSN). Also biomass burning can have large influence on absorption seasonality and on absolute levels, e.g., the Asian sites of GSN, LLN and AMY. In general, the seasonal variations are very clearly observed at remote sites, for example at ALT and TIK, where the seasonality of air mass origin bringing high levels of aerosol during 805 some parts of the year dominate the very minimal local emissions.

Global variability of single scattering albedo
For stations providing simultaneous measurements of scattering and absorption coefficients, it is possible to derive the single scattering albedo which is done at 550 nm. Overall, ω0 is computed for 31 stations and presented in Figure 7. Median ω0 810 values range from slightly less than 0.8 to almost purely scattering particles with ω0 close to 1. The highest values are found at coastal and polar sites clearly influenced by inorganic salts and sulfur-rich particles. The lowest ω0 are observed at sites in southern Europe (IPR and UGR), which are impacted by desert dust, biomass burning and local emissions. Only 6 sites have median ω0 below 0.9 but only the coastal, mountain and polar sites exhibit 25th percentiles constantly above 0.9. Variability of ω0 is strongly connected to air mass characteristics with, for a single station, a typical range of variability (25th-75th 815 https://doi.org/10.5194/amt-2019-499 Preprint. Discussion started: 11 February 2020 c Author(s) 2020. CC BY 4.0 License. percentile) of approx. 0.05 units of ω0. The variability at sites characterized as "Mixed", and in particular the mountain sites, is not higher than at other sites. The switch from free tropospheric air to boundary layer for the mountain sites does not appear to significantly affect ω0.

Comparison with AeroCom model outputs for optical properties 820
The AeroCom initiative has focussed since 2002 on the evaluation of global aerosol models with observations (aerocom.met.no). The integration of emission sources and aerosol processing leading to radiative effects requires complex models, which are increasingly coupled in high detail to general circulation models. Quantifying the climate forcing from aerosols requires a range of parameterised processes and derived properties of the global aerosol, which must be constrained by observations. The atmospheric dispersion of the aerosol, their optical properties, the attribution to natural and 825 anthropogenic sources, the potential of particles to influence clouds, and temporal trends -all these components need to be understood to quantify the radiative effect of aerosols. A network of in-situ aerosol measurements, well calibrated and available for long-term trend characterisation will provide important insights into the ability of models to realistically compute these radiative effects.

830
The recent generation of AeroCom models has been asked to provide additional diagnostics on dry scattering and absorption Overall, the performance of the model ensemble varies greatly as a function of station location, for both scattering and 840 absorption coefficients. Figure 8 compares observations and model ensemble results for the grid point corresponding to the station location. It shows a normalised mean bias of, on average, -28% between scattering by AeroCom models and observations, pointing to regional deficiencies in aerosol models. The normalised mean bias for absorption is lower (-18%) but still showing an underestimate by the AeroCom models. Obviously, there is, for both scattering and absorption, a large station-to-station variability in the bias, showing either good agreement, under-or over-prediction depending on the site. 845 There is also a significant variability of the normalised mean bias between models and observations when calculated for each season. This is also the conclusion of Gliβ et al. (submitted) which quantified the biases to -44% and -32% for scattering and absorption, respectively and listed possible causes for the biases such as overestimate of scattering enhancement due to https://doi.org/10.5194/amt-2019-499 Preprint. Discussion started: 11 February 2020 c Author(s) 2020. CC BY 4.0 License. hygroscopic growth and the differences in the treatment of absorption optical properties of black carbon, dust and organic aerosol. At this stage, additional investigations are needed to identify what accounts for the observed differences between 850 model and observations.

Observed and modelled trends of aerosol optical properties
The issue of long-term trends for the aerosol in-situ optical properties is specifically addressed in Collaud Coen et al. Analysis of the long-term information provides evidence that the aerosol load has significantly decreased over the last two decades in the regions represented by the 52 stations. Currently, scattering and backscattering coefficients trends are mainly decreasing in Europe and North America and are not statistically significant in Asia. Polar stations exhibit a mix of 875 increasing and decreasing trends. A few increasing trends are also found at some stations in North America and Australia.
Absorption coefficients also exhibit mainly decreasing trends. Generally, these decreases in aerosol burden are expected to be a direct consequence of decreases in primary particles and particulate precursors such as SO2 and NOx due to pollution abatement policies.

27
The single scattering albedo is one of the most important variables determining the direct radiative impact of aerosol so that its trend analysis -derived for the first time from a large number of stations -has the largest climatic relevance. The global picture is nuanced with ss positive trends mostly in Asia and Eastern Europe and ss negative trends in Western Europe and North America leading to global positive median trend of 0.02%/y. 15 stations exhibit a positive single scattering albedo trend (relatively more scattering) while 9 stations exhibit a negative trend (relatively more absorption). 885 Trends in scattering and absorption coefficients are also estimated by Mortier et al. (submitted) using AeroCom and CMIP6 models that have simulated the historical evolution of aerosol properties. For both variables, simulated trends are in agreement with SARGAN derived trends suggesting significant decreases found over North America and Europe, although the number of models providing trends in σap and σsp remains limited. Comparison with observations is also restricted to sites 890 below 1000 m asl which further reduces data points for comparisons. However, decreasing trends in AOD and sulphate are observed for North America and Europe for both model and observational data. Asian in situ surface data are too sparse to derive a regional trend for that region but it is worth indicating that not statistically significant AOD and sulphate trends are and may also demonstrate the relatively higher reduction of BC-rich emission in some regions, which will affect aerosol forcing estimates. https://doi.org/10.5194/amt-2019-499 Preprint. Discussion started: 11 February 2020 c Author(s) 2020. CC BY 4.0 License.

Data Handling
Data collected at 57 sites contributing to SARGAN were analysed to provide an overview of the condensation nuclei in the atmosphere. Measurements are performed with condensation particle counters (CPC) and mobility particle size spectrometers (MPSS); note that when both CPC and MPSS were concurrently run at a site, only MPSS data were included in the analysis, as it allowed additional investigation of the PNSD. For MPSS measurements, data inversion was performed 920 by the institutes operating the instruments, and, for both CPC and MPSS, particle number concentrations were reported in particles per cubic centimetre at STP, i.e., T = 273.15 K and P = 101 300 Pa, following the recommendations from Wiedensohler et al. (2012). As discussed in the overview of European PNSD and CN conducted by Asmi et al. (2011), the diameters associated with MPSS data correspond to the geometric mean diameter of the size intervals used in the inversion.
MPSS measurements are moreover usually representative of dry aerosol properties, as the operating procedures described in 925 Wiedensohler et al. (2012) indicate that the relative humidity of the sample air should be kept below 40%. In total, after excluding the datasets with insufficient data availability (with respect to the criteria reported in Section 5.1), CPC measurements collected at 21 stations and MPSS data from 36 sites were included in the analysis (Table SM3 in the   Supplementary).

930
To allow for the comparison of CN values derived from both instrument types, particle concentration in the range between 10 and 500 nm was inferred from MPSS measurements and assimilated to total CN (hereafter referred to as Ntot). This size range was selected as it was common to most of the MPSS included in this study. In addition, the lower end of this size range is comparable to the lower cut-off diameter of 14 of the 21 CPCs involved in the comparison (10 or 11 nm), and we assumed that particles larger than 500 nm only contributed little to Ntot. The legitimacy of this approach was supported by the 935 fair agreement between Ntot derived from collocated CPC and MPSS measurements at several sites. Moreover, using available MPSS data, we found that, on average, particles in the range between 10 and 11 nm contributed less than 1% to Ntot (90th percentile of the contribution: 5%), suggesting that such small cut point difference was not a major issue for Ntot.
However, the influence of a larger difference in lower cut points could not be discounted; this was, for instance, the case for ETL, ARN and GSN, where particles down to 2.5 nm were accounted for in Ntot (CN data were collected with a CPC TSI 940 3776 at these sites).
Results in the next section are discussed with respect to the classification of the stations reported in Table 2, including both the geographical and footprint criteria. Also, in order to describe the time evolution of CN and PNSD across the year, observations are categorized by seasons. Diurnal variations were not studied here, but would be expected to be strong for 945 certain site types and conditions (e.g., mountain upslope/downslope, urban local traffic, etc.). https://doi.org/10.5194/amt-2019-499 Preprint. Discussion started: 11 February 2020 c Author(s) 2020. CC BY 4.0 License.

Global variability of physical properties at SARGAN sites
As shown in Figure 10 and Table SM3, the lowest particle concentrations are typically observed under conditions of minimal anthropogenic influence, at polar sites, where yearly medians of Ntot are of the order of 10 2 cm -3 . Overall, as discussed earlier by Asmi et al. (2011), these stations also display a very clear seasonal cycle compared to other geographical categories, with 950 a summer maximum of Ntot likely resulting from both enhanced secondary aerosol formation, including new particle formation (NPF), and transport (Croft et al., 2016;Nieminen et al., 2018).
In contrast with polar sites, stations located in urban areas, both continental and coastal, exhibit the highest Ntot, with yearly medians in the range 10 3 -10 4 cm -3 . These sites, all located in Europe, also display a less pronounced seasonal variation 955 ( Figure 11). Slightly greater median values are, nonetheless, observed during summer, when the atmospheric boundary layer (ABL) height is also increased relative to colder seasons. This suggests the presence of an additional source of aerosols in summer which compensates for the ABL height dilution effect, as recently discussed by Farah et al. (submitted) who moreover suggested a photochemical or biogenic source. The overall weak seasonality observed in lowland urban areas is likely related to the contribution of very local sources which do not have any strong seasonal cycle (e.g., traffic). The local 960 nature of the observations collected at urban sites is supported by the differences between the measurements performed at neighbouring sites (e.g., LEI and LEI-E).
Remaining sites, including mountain and non-urban continental and coastal stations, do not exhibit as clear a common behaviour as the sites located at high latitudes or in urban areas. They display, on average, intermediate Ntot, with yearly 965 medians of the order of 10 2 -10 3 cm -3 . The signature of their dominant footprint is clear, with lower concentrations and stronger seasonal contrast observed in forested areas compared to rural background stations, while the distinction between the different geographical categories is in contrast less evident. Nonetheless, in agreement with previous observations from Asmi et al. (2011), particle concentrations measured at mountain sites tend to be lower compared to nearby lowland sites (e.g. SNB vs KOS). Mountain sites, and in specific those characterized by mixed footprints, tend to exhibit somewhat more 970 pronounced seasonality relative to lowland stations. This likely results from the strong impact of ABL height variability which, together with the topography of the sites, governs the concentration of particles and their precursors transported at high altitudes (Collaud Coen et al., 2018). Specifically, the summer enhancement of Ntot observed at most of the mountain sites is certainly tightly connected to the increased frequency of ABL injections during this time of the year (e.g., Herrmann et al., 2015). Apart from the lower concentrations, observations collected at non-urban continental and coastal sites display 975 similar seasonal variations as in urban areas, which are again likely explained by the concurrent variability of particle sources and ABL dynamics. https://doi.org/10.5194/amt-2019-499 Preprint. Discussion started: 11 February 2020 c Author(s) 2020. CC BY 4.0 License.

30
In short, particle concentrations are overall higher during warmer seasons at all sites as a result of enhanced sources, in connection with ABL dynamics for mountain sites. In addition, based on available MPSS data, the major contribution of 980 Aitken mode particles (30-100 nm) to the total particle number concentration also appears as a common feature of all environments. In contrast, the magnitude of the seasonal cycle of Ntot, together with the variations of the PNSD, exhibits some distinctive behaviour for the different geographical categories and footprint classes, with additional site-dependent characteristics. However, among other factors (including the nature and proximity of the particles sources), the level of anthropogenic influence appears to strongly affect the observations. 985

Using SARGAN for global climate monitoring applications
Climate observations are fundamental to many aspects related to prediction of future environmental changes and to meet the requirements of the UNFCCC and other conventions and agreements. The establishment of a global network of observations for assessment of atmospheric composition changes, adaptation to climate change, monitoring the effectiveness of policies 990 for limiting emission of pollutants and/or developing climate information services must define the specific observational requirements for efficiently addressing these issues.

Response of SARGAN to GCOS principles
Measurement harmonization procedures allowing for direct comparison of data provided, together with the quality control and quality analyses performed all through the data provision chain have considerably improved the value of SARGAN as an 995 essential piece of the in-situ segment of Earth Observations for its specific climate-relevant variables. SARGAN addresses to all 10 basic principles of the WMO-IOC-UNEP-ICSU Global Climate Observing System (GCOS). GCOS is designed to meet the requirements for climate observations which are essential to climate monitoring and support implementation of UNFCCC and other climate conventions and agreements.

1000
Considering the importance of aerosol properties in the Earth Climate system, it is important to define the GCOS requirements for a number of variables that are, or may be in the future, defined as essential climate variables. Today, there are four aerosol GCOS ECV products: AOD, Single-Scattering Albedo, Aerosol Extinction Coefficient Profile and Aerosol Layer Height. Only Single-Scattering albedo is directly connected to SARGAN although the GCOS aerosol variables are currently being revised to include ECVs connected to aerosol size, composition and hygroscopic properties. In its current 1005 state SARGAN is able to address the ten basic GCOS Climate Monitoring Principles as follows (Table 4): These requirements must include the spatial and temporal resolution of the observations, and their accuracy, precision, and long-term stability. For each requirement, one additional specification is required to identify 1) Threshold or minimum https://doi.org/10.5194/amt-2019-499 Preprint. Discussion started: 11 February 2020 c Author(s) 2020. CC BY 4.0 License. requirement defined as the value that has to be met to ensure that data are useful, 2) Goal or maximum requirement defined 1010 as the value above which further improvement gives no significant improvement in performance or cost of improvement would not be matched by a corresponding benefit likely to evolve as applications progress. In between « Threshold » and « Goal », « Breakthrough » is defined as an intermediate level that would lead, if implemented to a significant improvement for the specific application.

1015
It is clear that requirements are defined for specific application areas, in this case climate monitoring applications as defined in OSCAR (https://www.wmo-sat.info/oscar/applicationareas). The Climate Monitoring application area is defined as such: "The WMO-IOC-UNEP-ICSU Global Climate Observing System (GCOS) is an internationally coordinated network of global observing systems for climate, designed to meet the requirements for climate observations, which are essential to climate monitoring. Climate observations are fundamental to detect, model and assess climate change, support adaptation to 1020 climate change, monitor the effectiveness of policies for mitigating climate change, develop climate information services, promote sustainable national economic development and meet other requirements of the UNFCCC and other conventions and agreements". Observational requirements for other application areas have been recently published (Benedetti et al., 2018) or are currently underway as part of the WMO/GAW activities.

Response of SARGAN to GCOS principles 1025
With the specific definition, and considering the results presented in this paper, in companion SARGAN papers and in previous studies, the following requirements can be defined for SARGAN variables.
The threshold for spatial requirements in the horizontal scale for SARGAN can be defined as the distance between two observing points above which no redundancy is observed when measurements are performed in parallel. A few papers have 1030 addressed this issue by investigating the autocorrelation function between time series for different aerosol properties (Anderson, 2003;Sun et al., 2019) and they both lead to similar results related to observations at the ground: temporal variations of an intrinsic aerosol variable observed at the ground are no longer statistically correlated when stations are located more than several hundred km apart. To be more specific, Sun et al. (2019) suggest that correlation of absorption coefficient time series from stations located 500 km apart is still approximately 0.5. A similar result is found for particle 1035 number in the 200-800 nm range, while distance for a similar correlation of 0.5 for particles in the lower size range (10-30 nm) is of the order of 100 km. This, of course, depends on several parameters including the intensity of emissions surrounding the station, and efficiency of removal rates (dry and wet deposition). Interestingly, similar temporal correlations It is fair to consider that two stations located more than 1000 km apart will, therefore, for aerosol variables relevant to SARGAN, provide very little redundancy in their observations, especially if the stations are located over land. Assuming an advection velocity of 20 km h -1 , 1000 km would correspond to approximately 2 days, which is shorter than the aerosol typical lifetime over continents. For observations over the oceans, it is clear that a larger threshold could be considered, 1045 corresponding to a turn-over time of approximately a week (i.e. several thousands of km). The threshold for the observation of climate-relevant parameters in SARGAN can, therefore, reasonably be set at 1000 km, while breakthrough and goals for the spatial resolution can, accordingly, be set at 500 km and 100 km, respectively. A 100 km spatial resolution would serve the purpose of deriving radiative forcing estimates at scales typical of a large urban area, together with providing information extremely relevant for model and space-based observations. These indicated horizontal requirements for threshold, 1050 breakthrough and goal would require models to provide information on approx. 0.5°x0.5° degree resolution grids for goal, which is now often achieved.
Considering a total land-area in Europe of approx. 10 M km 2 (thus only including the Russian territory in geographical Europe), and 63 measurement stations in operation (see Table), the measurement density in Europe is close to requirements 1055 for « breakthrough ». It is even close to the « goal » level if Russia is not considered. In North America, it is close to «threshold» (28 stations for 24 M km 2 ) and between recommended values for threshold and breakthrough for US territory only, including Alaska (21 stations over approx. 10 M km 2 ). For all other regions of the World, the situation is below that recommended for minimal sampling, illustrating the huge gaps in network density.

1060
Because SARGAN is based on individual observation points at the surface, the issue of vertical resolution is not relevant.
However, the value of measuring both in the boundary layer and in the free troposphere is clear for many applications.
Requirements for temporal resolution can be derived in a simpler way, considering that time-series datasets are often provided on a month-by-month variation in climate over long-time periods. Monthly data sets allow many variations in climate to be studied and can be considered as threshold as long as the data is generated by representative original data sets. 1065 Information provided with a temporal resolution of one-day are suitable for addressing issues related to cloud cover, precipitation, impact of temperature, emissions, etc… and can be considered as breakthrough while the 1-hour resolution is a requirement for many applications such as estimating aerosol fluxes or radiative impact of aerosol plumes.
The maximum time lag between observations and the data being freely available is, for most applications, of the order of one 1070 year (threshold), although data providers are more and more requested to provide information on shorter timescale, with 24 hour delay and near-real-time (6 hour delay) corresponding to « breakthrough » and « goal » levels, respectively.
The definition of requirements for GCOS also asks to establish a level of uncertainty which accounts for all quantifiable uncertainties. In the case of in-situ aerosol variables, requirements for the measurement uncertainties can be derived from the 1075 https://doi.org/10.5194/amt-2019-499 Preprint. Discussion started: 11 February 2020 c Author(s) 2020. CC BY 4.0 License. 33 observed variability on the different temporal scales, which is quite large. We have used suggested uncertainties provided in Table 2 for CN, σsp and σap. Uncertainties of ω0 is proposed following procedures of Sherman et al., (2015).
Stability is defined as the maximum permissible cumulative effect of systematic changes of the measurement system to allow long-term climate records compiled from assorted measurement systems. would not be detectable with higher stability values. Carslaw et al., (2010) have estimated the change in aerosol radiative 1085 forcing due to climate feedbacks in emission of aerosol precursors from natural systems. They show that a radiative perturbation approaching 1 Wm −2 is possible by the end of the century. Detecting and attributing changes to a climate feedback due to changing natural emissions (wildfires, biogenic organic volatile compounds) would require a much lower uncertainty than currently achieved for CN, σsp and σap and consequently ω0. At this stage, without more information on trends, we are recommending values for stability of 1%/yr for breakthrough and 0.5%/yr for goal for all variables. 1090 Requirements for the GCOS application area for σsp, σap, CN and ω0 are summarized in Table 5.

Conclusions and future challenges
The present article must be seen as the foundational framework for the observation of aerosol properties collected nearsurface from ground-based stations Worldwide, in the context of GAW. SARGAN completes a ground-based aerosol observing system composed additionally of the GAW associated networks GALLION and PFR. SARGAN relies on its 1095 regional constituents in the different WMO regions, of which ACTRIS in Europe and NOAA-FAN in the US are the principal contributors.
Although not fully implemented and operational, SARGAN sites share common methodological approaches for measurement and data quality control, and a common objective to open access for all data, that are all defined as part of the 1100 Global Atmosphere Watch Scientific advisory group on aerosol. Data provision is currently operational with some sites providing information for more than several decades. The very strong motivation in the early 2000s to develop observations of aerosol climate-relevant parameters led to a substantial increase in operating ground-based stations and availability of data time-series with the required level of quality. We consider that the degree of integration of the different providers to SARGAN has reached a mature level which has resulted in more and more users of the data worldwide. Europe. Open access to the SARGAN database should enhance the potential for many other applications. Analysis of trends for number concentration is already under way but we assume that SARGAN data can be efficiently used to support many types of studies, related to aerosol impact on air quality, health or climate, quantification of emission sources or for the 1115 development of early-warning services.
The SARGAN initiative is currently limited to four variables that are directly observed. They are the only four climaterelevant aerosol variables measured near-surface for which a relatively consistent coverage exists worldwide. Providing constraints on radiative forcing estimates would obviously require knowledge of trends and variability for other variables, 1120 such as aerosol chemical composition or number concentration of cloud condensation nuclei. Unfortunately, very few sites are currently including these variables in their observation program and they are mostly located in Europe as part of ACTRIS. It is clearly a huge and key challenge for the community to extend observations to additional variables, in particular for sites located outside Europe.

1125
The distribution of sites providing information to SARGAN confirms the analysis made in many earlier reports and in Laj et al. (2010): a very strong bias still exists in the World data coverage, with Europe and the US well-represented and observations lacking in many other regions, in particular over WMO region III (Africa) and IV (Latin American and Caribbean), Russia, and large parts of Asia. Causes may be connected to difficulties making data accessible through the World Data Centers in some cases, but for many areas of the World, it is directly related to lacking measurements. Detecting 1130 atmospheric trends of key atmospheric compounds requires long (>10 years) high quality records and, despite many initiatives, only a very few stations have managed to maintain operations for observing composition changes over more than a decade. Laj et al. (2019) have recently proposed a series of recommendations to support atmospheric observations in emerging 1135 economies. Demonstrating how climate data/ information have direct relevance to policy making and explaining the local benefits that monitoring atmospheric composition changes bring to the country in terms of socio-economic impacts, in both the short and longer terms may help engage national stakeholders to commit to maintain and develop observation sites.
Stimulating the demand for climate observations/ climate information of the kind provided by SARGAN at the user level in the countries concerned would be absolutely important. The European concept of Atmospheric Research Infrastructures, 1140 35 such as ACTRIS, was key to securing the necessary long-term engagement in the EU countries to support SARGAN observations. Similar approaches can be proposed, adapted to the different WMO regions.
In a recent comment in Nature, Kulmala (2018) suggested the establishment of 1,000 or more well equipped ground stations around the world tracking environments and key ecosystems, thus sampling beyond the observation of atmospheric 1145 composition only. Establishing observation sites with core measurement capabilities documenting key atmospheric components (greenhouse gases, reactive gases, aerosol properties) together with basic meteorology, operated by skilled personnel and providing access to measurement data in countries where this is still lacking would require a large scale coordinated effort that is far from being out of reach. Investments for atmospheric monitoring would be anywhere between 0,5 and 1 M US$ and annual operations between 50 and 100 kUS$ and 2-3 FTEP per site. 1150 There is a growing number of multilateral climate finance initiatives designed to help developing countries address the challenges of climate change and air quality. They have a role in capacity building, research, piloting and demonstrating new approaches and technologies and are perfectly suited to be used for developing the needed atmospheric component of a global Earth observing system. A "One Nation, One Station" approach to establish at least one reference stations in each 1155 country where information is lacking would definitely add essential information to large-scale modelling but also support local research, national policymakers, and promote business development for environmental services such as early warnings for extreme weather and atmospheric hazards.  The volumetric cross-section for light extinction is commonly called the particle light extinction coefficient (σep), typically reported in units of Mm -1 (10 -6 m -1 ). It is the sum of the particle light scattering (σsp) and particle light absorption coefficients (σap), σep = σsp + σap. All coefficients are spectrally dependent.

Acknowledgements
AOD 1,2 Aerosol optical depth, defined as the integral over the vertical column of the aerosol particle light extinction coefficient.
The aerosol particle single-scattering albedo, defined as σsp/σep, describes the ratio of particle light scattering coefficient to the particle light extinction coefficient. Purely scattering aerosol particles (e.g., ammonium sulphate) have values of 1, while very strong absorbing aerosol particles (e.g., black carbon) may have values of around 0.3 at 550 nm.

AAOD
The absorption Aerosol optical depth is the fraction of AOD related to light absorption and is defined as AAOD=(1−ωo)×AOD.

g, β
The asymmetry factor g is the cosine-weighted average of the phase function, ranging from a value of -1 for entirely backscattered light to +1 for entirely forward-scattered light. The upscatter fraction β gives the fraction of sunlight scattered in the upwards direction (back to space), which depends on the solar zenith angle as well as the size distribution and chemical composition of the particles.

AE (or Å)
The extinction (scattering) Angstrom exponent is defined as the dependence of AOD (or (σsp)) on wavelength (λ), e.g., AOD∝C0λ -AE where Co denotes a wavelength-independent constant. The Angstrom exponent is a qualitative indicator of aerosol particle size distribution. Values around 1 or lower indicate a particle size distribution dominated by coarse mode aerosol such as typically associated with mineral dust and sea salt. Values of about 2 indicate particle size distributions dominated by the fine aerosol mode (usually associated with anthropogenic sources and biomass burning).

MSCi, MACi
The mass scattering cross-section (MSCi) and mass absorption cross-section (MACi) for species i, often calculated as the slope of the linear multiple regression line relating σsp and σap, respectively, to the mass concentration of the chemical species i, is used in chemical transport models to evaluate the radiative effects of each chemical species prognosed by the model. This parameter has units of m 2 g -1 .

f(RH), g(RH)
f(RH) is the functional dependence of components of the aerosol particle light extinction coefficient (σep, σsp, σap) on relative humidity, expressed as a multiple of the value at a low reference RH (typically <40%). g(RH) is analogous to f(RH) but describes the change in size of particles as a function of RH PNSD 1 The particle number size distribution describes the number of particles in multiple specified size ranges. The PNSD can provide information about formation processes such as new particle formation, aerosol transport as well as aerosol types.

1620
CN, CCN, IN The particle number concentration (CN) refers to the number of particles per unit volume of air (cm -3 ). The Cloud Condensation Nuclei (CCN) number concentration is the number of aerosol particles which can activate to a cloud droplet at a given supersaturations of water. The Ice Nuclei (IN) is the number of aerosol particles onto which water freezes following various processes. CCN is often indicated as a percent of the total CN for specific supersaturation typical of atmospheric cloud formation. CCN number concentration is sometimes approximated using the fraction of particles larger than a given diameter from the particle number size distribution Fz(σep) 1,2 The profile of the particle light extinction coefficient is the spectrally dependent sum of aerosol particle light scattering and absorption coefficients per unit of geometrical path length.        Figure 10 : Yearly median of the total particle number concentration (Ntot). The markers represent the median of the data and the lower and upper edges of the box indicate the 25th and 75th percentiles, respectively. The length of the whiskers represents 1.5 interquartile range. Different markers and box colors indicate geographical categories and footprint, respectively, according to