Harmonized aerosol size distribution, cloud condensation nuclei, chemistry and optical properties at 10 sites

Abstract

A harmonized data set of aerosol particle number size distributions and cloud condensation nuclei (CCN) concentrations at 10 US Department of Energy (DOE) Atmospheric Radiation Measurement (ARM) sites is presented. Additional complementary data including total particle number concentration, aerosol chemical composition and aerosol optical properties are included where available. Calculated parameters related to particle activation, (e.g., activated fraction, hygroscopicity parameter and critical diameter) are provided. DOE/ARM data are acquired and processed following DOE/ARM protocols which are based on international recommendations. Here, the data have undergone additional stages of quality assessment including various closure tests. This data set may be useful for studying aerosol processes and, with other data, assessing aerosol-cloud- interactions. High-resolution (5-min) data is provided along with 1-h averaged data, making the data set suitable for exploring the importance of new particle formation (NPF) events on CCN concentrations. Further applications include model and satellite evaluation of aerosol size distributions and CCN concentrations, and exploring relationships among different aerosol properties which may be useful for improving model parametrizations.

Background & Summary

Atmospheric aerosols impact climate, air quality and visibility, and human health (e.g.^1,2). Aerosols affect the Earth’s radiation balance directly by scattering and absorbing sunlight and indirectly through their role as cloud condensation nuclei (CCN)^3,4,5,6. To date, radiative forcing through aerosol–cloud interactions (ACI) constitutes the least understood anthropogenic influence on climate⁷. It remains a significant challenge to reduce these uncertainties and to thereby increase our confidence in predictions of global and regional climate change^7,8.

One leading cause of the large uncertainty in ACI is detailed knowledge about how particles evolve to become effective CCN under different conditions/environments. CCN are emitted to the atmosphere directly from both anthropogenic⁹ and biogenic sources¹⁰, and over 50% of CCN are thought to be formed by condensational growth of smaller particles¹¹. These smaller particles may originate from atmospheric new particle formation (NPF), anthropogenic combustion or other emissions. NPF is the formation of thermodynamically stable clusters from multi-component systems followed by condensational growth of these clusters to a detectable size¹². Ultimately, the newly formed particles can grow to sizes in the CCN range by coagulation and condensation of more vapors¹³. The occurrence of NPF events has been observed to take place almost everywhere in the atmosphere¹⁴. The total particle number concentration (N_tot) in regional background conditions, as well as in the global troposphere, is very likely to be dominated by NPF events¹⁵. The contribution of aerosols formed via NPF to overall CCN concentration is still poorly understood (e.g.¹⁶). Many modeling studies suggest that NPF events can have an impact on the abundance of the global CCN (e.g.^{11,17,18,19,20,21,22,23}) but the magnitude of the NPF contribution to CCN differs substantially in different models and for different environments^11,18,21.

Combining information about aerosol microphysics (number concentration and size distribution) and aerosol composition with information about the CCN activity of those aerosols can help to improve our understanding of the particles that can form clouds. Undertaking such a harmonization across a variety of locations and aerosol types has the potential to reduce uncertainty in ACI for different environmental conditions. While satellite retrievals are important for understanding ACI processes across different scales, the retrievals are currently only capable of assessing highly uncertain proxies for CCN activity for limited conditions²⁴. Those CCN proxies require evaluation with in-situ observations of CCN across different environments in order to assess when and whether they provide accurate enough estimates of CCN to be used in ACI studies^24,25. In an enormous effort, Schmale et al.^26,27 provided a first overview of co-located Particle number size distribution (PNSD) and CCN measurements at 11 observatories (8 in Europe, 2 in Asia and 1 in USA), based on 1-h time resolution data. Today, additional collocated measurements of CCN and PNSD exist which can enhance global coverage. In particular, the Atmospheric Radiation Measurement program of the US Department of Energy (https://www.arm.gov/) has extensively performed these measurements both as a long-term monitoring effort as well as on a short-term/campaign basis²⁸.

Here we present a harmonized data set of US Department of Energy (DOE) Atmospheric Radiation Measurement (ARM) measurements that includes surface CCN concentrations (N_CCN) as a function of supersaturation (SS) and PNSD at 10 sites along with collocated, complementary particle chemical composition and aerosol optical properties where available. We provide both high resolution (5-min averages) and hourly averaged data for these measurements. The data averaging was done as part of our processing (see data handling section below). The high resolution dataset is useful for NPF related investigations while the hourly averaged data is comparable to the Schmale et al.²⁶ dataset and is useful for applications such as global model evaluation²⁹. We also provide CCN-related data products including the hygroscopicity parameter (κ_CCN³⁰), critical diameter (D_crit) and activated fraction (AF), derived by combining measurements from different instruments at each site.

Methods

In the methods section, we first describe the overall study design and provide a general overview of the individual sites along with the associated campaign(s) if applicable. There is then a section that briefly describes the DOE/ARM data archive along with individual sections for each measurement type. Each measurement type section includes the original DOE/ARM citation(s) for the measurement and a description of the operating principles of the instrument(s) used to make the measurement. The data handling section presents information on how we processed the original ARM files to generate the final harmonized data files presented here. Following that, we describe the methodology we used to calculate some CCN parameters using the finalized quality-assured data sets. Finally, in the Technical Validation section, we discuss the closure tests and other data checks we used to assess the quality of the resulting harmonized data sets.

Study design

The benchmark data set presented here consists of multiple collocated measurements from various DOE/ARM campaigns or long-term sites which are relevant for improving our current knowledge on CCN spatio-temporal variability and CCN sources and for exploring (a) the relationship between NPF and CCN and the factors impacting that relationship and (b) aerosol-cloud interaction studies and measurement-model comparisons, among others. Specifically, for each site we include CCN and PNSD data as well as an independent measure of total particle number concentration. These data, in their primary form, are already available from DOE/ARM’s data archive (https://arm.gov/data/) but here we have put them under further review, harmonized the data (as described below), calculated related data products and provided them together so the next researcher may find the data more easily and in a consistent format. Many of the sites measured other relevant parameters (i.e., aerosol composition and aerosol optical properties) so those data, where available, are included in this harmonized data set.

Sites

All sites were DOE/ARM deployments operating either on a campaign basis (e.g., via the ARM Mobile Facility, AMF) or as part of ARM’s long-term monitoring efforts. Table 1 provides general site information, while Figs. 1, 2 show a map of site locations and the instrument operating periods at each site, respectively. As a prerequisite for being considered in this study, only data from sites where collocated and contemporaneous sub-micron PNSD and CCN measurements were available in the ARM archive are included in the harmonized data set. We also required the CCN measurements to scan multiple SS values between 0.1 and 1. One site initially considered (Cape Cod, MA during the Two-Column Aerosol Project (TCAP)³¹) met both these criteria but was eliminated due to irregularities with CCN supersaturation scanning. We also considered data from two other DOE/ARM deployments in Houston, Texas³² and San Diego, California³³; unfortunately these sites needed to be removed from the final dataset after DOE/ARM identified a leak in the CCN sample line. Figure 1 shows that the measurements occurred primarily in the western hemisphere (Europe and the Americas) and, combined with the coverage of similar measurements described in Schmale et al.²⁷, highlights the need for these types of measurements across Asia and the global south.

Table 1 General site information.

Fig. 1

Map of DOE/ARM sites with collocated simultaneous CCN and SMPS data. Site type is indicated with different colors; if the outline is different than the fill color the site could be described by more than one type (e.g., polar and marine). MOS is a mobile deployment so the location shown on the map is approximate. See Tables 1, 2 for more details.

Fig. 2

Data coverage of the instruments used in this study for all the sites considered. The main objective is the analysis of CCNC and SMPS measurements so, although there may be longer CPC, ACSM and optics measurements, these have not been taken into account.

The PNSD and CCN data meeting these criteria were available from 10 stations. All of these stations provided additional aerosol chemical composition and/or optical measurements. Measurements at most sites were performed following standard operational procedures established by DOE/ARM, making the measurements among sites comparable and the data formats consistent. However, at some sites some of the instruments were operated by external DOE-funded principal investigators (PIs) to support the science needs during specific campaigns. The format of the PI data sets are different than the DOE/ARM operational data sets and are found in DOE/ARM’s intensive operating procedure (IOP) archive. Individual characteristics of each site are briefly described below and in more detail in the provided references if available. The sites are alphabetically ordered by their three letter identification.

ANX, Andenes, Norway – COMBLE

The Cold-Air Outbreaks in the Marine Boundary Layer Experiment (COMBLE) was an AMF deployment designed to study boundary layer convection and air mass transformation during Arctic cold air outbreaks over a 6 month period (Dec. 2019 - May 2020)³⁴. The ANX sampling site was located on the coast of northern Norway.

ASI, Ascension Island, British Overseas Territories –LASIC

The overarching goal of the Layered Atlantic Smoke Interactions with Clouds (LASIC) campaign was to investigate absorbing aerosol and their interactions with clouds. The AMF was installed on Ascension Island and made measurements between June 2016 - October 2017³⁵. The AMF was deployed approximately 1.5 km inland on the windward side of the island, in the path of the biomass burning plume that typically flows off the African continent in the March-May timeframe.

COR, Córdoba, Argentina – CACTI

The Cloud Aerosol and Complex Terrain Interactions (CACTI) campaign was a 7 month deployment of the AMF in Argentina from September 2018 - April 2019. CACTI was designed to investigate cloud life cycle in complex terrain in order to improve parameterizations of cloud and aerosol processes in models³⁶. Measurements were made both at the surface and in an instrumented airplane, but we focus on the surface measurements. The surface sampling site was located on the eastern side of the Sierras de Córdoba mountain range and likely sampled biogenic, biomass burning and/or anthropogenic aerosol depending on the synoptic patterns³⁷.

ENA, Azores, Portugal - Eastern North Atlantic

Measurements at the ENA site on Graciosa Island began as one of the AMF deployments in 2009–2010, however our focus is on more recent data after ENA became one of DOE/ARM’s long-term atmospheric observatories. The data used covers the time period between November 2022 - December 2023. The aerosol instrumentation is located on the north side of the island approximately 1 km inland³⁸.

GUC, Colorado, USA – SAIL

The Surface Atmosphere Integrated field Laboratory (SAIL) was an AMF deployment near Crested Butte, Colorado. The campaign went from September 2021 - June 2023 and was designed to investigate the water cycle and energy budget in complex terrain³⁹. The aerosol measurements were originally at the campaign’s main facility (M1) but were moved about 6 weeks into the campaign to the S2 location. In this study, we focus on data from the S2 location which covers the time period October 2021 - June 2023. The S2 site was located at the Crested Butte ski resort in the Rocky Mountains.

MAO, Manacapuro, Brazil – GOAMAZON

The Observations and Modeling of the Green Ocean Amazon (GOAMAZON) was a 23 months deployment (2014–2015) in Brazil to study aerosol and cloud life cycles in the Amazon basin⁴⁰. The AMF and a new DOE/ARM mobile aerosol observing system (MAOS) were both deployed to this site with facility designators M1 and S1, respectively. These two facilities made some duplicate measurements (e.g., optical properties, N_tot). We utilized the data from the S1 facility for all instruments except the CCN counter (CCNC) for which there was only data available from the M1 facility. Both facilities were located in a pasture 70 km downwind of the city of Manaus and sampled both the urban plume and regional aerosol from the surrounding rainforest. A large amount of NPF and CCN analysis from this campaign and IOP measurements has already been published^{41,42,43,44,45,46,47}.

MOS, Arctic Ocean – MOSAiC

The Multidisciplinary drifting Observatory for the Study of ArctIc Climate (MOSAiC) campaign was a deployment of the AMF on an icebreaker frozen into and moving with the ice. The goal was to collect data to better understand climate change in the Arctic and to characterize the cryosphere-atmosphere relationship in the region⁴⁸. The 1 year campaign began in October 2019.

SBS, Colorado, USA – STORMVEX

Storm Peak Lab Cloud Property Validation Experiment measured cloud and aerosol properties in complex terrain at multiple sites over 6 months (Nov. 2010-Apr. 2011)⁴⁹. One focus of the campaign was to investigate the role of different types of aerosols on cloud and precipitation processes. There were two sites with CCN and PNSD measurements on the mountain at the Steamboat Springs ski resort, one at Christy Peak (CP) and one at the University of Utah’s Storm Peak Laboratory (SPL) (https://atmos.utah.edu/storm_peak_lab). We refer to these two sites as SBS-CP and SBS-SPL, respectively. PI datasets for raw CCNC and condensation particle counter (CPC) from SPL and raw SMPS data from SPL and CP are in the DOE/ARM IOP archive.

SGP, Oklahoma, USA - Southern Great Plains

This is the longest set of measurements and the most complete as it is where DOE/ARM tests instruments before deploying further afield. For comparability reasons, at SGP we limited the analysis to the operational measurements from 2017 to 2023 at central facility ‘E13’. Note however that there are CCN and PNSD data from the SGP C1 facility prior to this, e.g.^50,51, albeit in different format data files and with a more limited auxiliary data set. The central facility site is located in a rural agricultural region more than 100 km from any large cities (e.g., Wichita, Oklahoma City, Tulsa).

DOE/ARM Datastreams and measurements

In this section we first briefly provide some details about the DOE/ARM data archive files. These files were the starting point for our evaluation and the further processing that we did to generate the final harmonized dataset. We then describe the individual instruments used for each measurement included in the harmonized data set; these instrument descriptions include information on the data processing performed by DOE/ARM. For some instruments we note where we had to replicate the DOE/ARM processing in order to have a consistent starting part for further data treatment. The additional data processing and evaluation that we performed (beyond that performed by DOE/ARM) to generate the harmonized data set is described in the data handling and data validation sections below. Table 2 summarizes the information about specific instruments used for each parameter on a site by site basis.

Table 2 Summary of site specific instrument information for measurement types included in harmonized data set.

DOE/ARM Datastreams

The starting page for finding and downloading data from the DOE/ARM data archive is: https://arm.gov/data/. The DOE/ARM data archive stores both ‘operational’ and principal investigator (PI) data files from ARM field campaigns and permanent measurement locations. The operational data files are consistent across the sites in terms of data acquisition, QA and formatting. These operational files became available in the 2014–2015 time frame. Prior to that the DOE/ARM data sets were collected using data acquisition, processing and formatting procedures, often by entities external from DOE/ARM.

DOE/ARM uses a consistent three component naming convention for their operational data files. Each file name starts with a three letter identifier representing the site (stn). The site identifier is followed by the name of the datastream and then a second location identifier indicating the specific facility/building (fac) at the site where the measurement was made. The second part of the file name indicates data processing level and the third part of the file name is a date string which is the start time of the data contained in the file. For example, the file name sgpaosnephdry1mE13.b1.20221213.000030.nc represents data from the SGP site at the E13 facility. The datastream is for the aerosol observing system (AOS) dry nephelometer measurements averaged to 1 min. The data level is ‘b1’ which indicates the data have been quality controlled, reviewed and standard corrections have been applied. The data are for the day of 13 December 2022. To find the original DOE/ARM data files used in this study we recommend searching by datastream name (e.g., in this example, aosnephdry1m on the DOE/ARM data archive (https://arm.gov/data/)). Table 3 provides the names of the operational DOE/ARM datastreams and data citations for the files used to generate the harmonized dataset. More information on DOE/ARM’s file naming and data levels are available here: https://www.arm.gov/guidance/datause/formatting-and-file-naming-protocols.

Table 3 List of DOE/ARM filenames for operational data sets, with links to data citation.

In addition to the actual data files for each operational datastream, a text file containing data quality reports (DQRs) accompanies downloaded data. These DQRs are generated by each DOE/ARM instrument mentor (the person responsible for reviewing the data) when the mentor identifies a problem with the data (e.g., that it’s missing, that it’s invalid due to an instrument issue or that it’s potentially suspect). Users of DOE/ARM data need to incorporate review of the DQRs into their analysis to ensure they are using the best possible data. Typically DOE/ARM automatically sends additional DQR information to people who have downloaded the data. Based on our experience with the Houston and San Diego campaign data, where that did not happen, we recommend doing a final check of data availability/quality before publishing results using DOE/ARM data.

Additionally, as noted above, the ARM data archive includes links to measurement data generated by DOE-funded PIs to support the science needs at individual sites. Such data sets utilize PI-specific file naming conventions, processing, and formats and are found on the IOP section of the ARM data archive. Where applicable, Table 4 lists the DOE/ARM IOP data archive locations for PI datastreams. PI data does not come with DQRs, although there is often a README file.

Table 4 List of IOP archive locations for PI formatted files.

Particle number size distribution

The submicron PNSD is measured using a Scanning Mobility Particle Sizer (SMPS) spectrometer. Table 2 lists the various PNSD instruments that operated at each site. The working principle of the instrument is based on the counting of particles of different sizes that are selected based on their electrical mobility. The instrument consists of a bipolar diffusion charger (neutralizer), a differential mobility analyzer (DMA), and a CPC in series. The bipolar diffusion charger brings the particles into a near bipolar charge equilibrium. By scanning or stepping the voltage in the DMA, particles of certain electrical mobility are selected, and counted with the CPC (the CPC is often referred to as the detector). Depending on configuration the instrument has the capability of scanning from 10 - 1000 nm, although that full size range was not employed in the data sets here, and it scans 64 bins per diameter decade. Each full scan takes the SMPS 5 minutes to complete. Humidity in the sample line is maintained below 40% using a Nafion™ dryer. Data acquisition, instrument control and data inversion are accomplished through the manufacturer’s software (Aerosol Instrument Manager, TSI).

The SMPS system undergoes calibration prior to deployment, including DMA sheath flow, voltage control, and particle sizing verification. DMA particle sizing is verified with polystyrene latex particle size standards (150 nm). The CPC characterization includes verifying the inlet flow rate and determining the size-dependent particle counting efficiency. The DOE/ARM SMPS handbook describes the SMPS and operations in more detail⁵².

The two SMPS data sets from the STORMVEX campaign at SBS were not from the DOE/ARM operational measurements but rather were from an individual PI (see Table 4) so more effort was required to identify, process and QA the data because the formats and processing differed from those of the ‘operational’ instruments. As described in the data handling section, similar evaluation and validation analyses were applied to the PI PNSD data as were used for the DOE/ARM operational PNSD measurements.

Cloud condensation nuclei concentration

The N_CCN as a function of SS is measured with a CCNC. There are two modes of operation for the CCNC: monodisperse operation mode in which particles are size selected by means of a DMA prior to entering the CCNC (providing size-resolved CCN concentrations) and polydisperse mode in which activation is measured for the whole aerosol population. In this study, all CCN measurements were performed in the polydisperse operation mode. Over the years DOE/ARM has operated two different versions of the Droplet Measurement Technology (DMT) CCNC. Table 2 lists the version of the CCNC that operated at each site. The DMT1C (i.e., model DMT100), a single column CCNC, typically scanned across 4–6 SS values over the course of an hour providing N_CCN at SS = 0.1% on the low end to approximately SS = 1% on the high end. More recently, the DMT2C (i.e., model DMT200), a dual column CCNC, has been deployed. For DOE/ARM deployments of the DMT2C, one column ‘column A’ typically scans across 4–6 SS values over the course of an hour and the second column ‘column B’ measures at constant SS (typically 0.4%). In our data handling the single column of the DMT1C is considered column A. For both the DMT1C and DMT2C over the course of an SS scan, the CCNC typically spends about 10 min at a given SS (6.5 minutes of measurement and 3.5 minutes for transition between SS). The DOE/ARM CCNC handbook describes the CCNC and operations in more detail⁵³. Below we describe generally how the CCNC works which is same for both column A and column B so we do not specify column. We do however specify the column in the data handling section where there might be some confusion and in the harmonized data files.

Briefly, the operating principle of the CCNC is based on a cylindrical continuous-flow thermal-gradient diffusion chamber where constant temperature gradients are applied, generating different SS conditions⁵⁴. Inside the instrument, the aerosol sample flow is guided through the center of the cylinder by a particle-free laminar sheath flow. The centerline SS is generated by applying a controlled (and constant) streamwise temperature gradient at the cylinder wall; if the inner wall is kept wet, heat and water vapor continuously diffuse towards the center of the tube. Because water vapor has a lower molecular weight than moist air, diffusion of water vapor is faster than heat and the centerline becomes supersaturated⁵⁴. Particles that activate at a critical SS lower than the measurement SS form droplets. At all sites, the CCNC instruments were operated at similar conditions, with an aerosol to sheath flow ratio of 10 as recommended by the manufacturer (volumetric aerosol flow 45 cm³min⁻¹, volumetric sheath flow 450 cm³min⁻¹).

Operationally, the user chooses a specific SS value or values for the CCNC to measure at and also sets the timing of the measurements if the instrument is scanning across multiple SS. These specified SS values are referred to as the setpoint SS (SS_setpoint). Ideally, SS_setpoint would correspond to the actual supersaturation at the centerline of the CCNC column but in reality this is not the case. Actual SS (SS_act) in the CCNC depends on the instrument calibration and operating conditions (e.g., flows and temperature gradient) and can differ significantly from SS_setpoint⁵⁵. Figures 3, 4a,b show the instrument SS_setpoint and SS_act for the measurements at ASI, MOS and GUC, as examples. While SS_act can be similar to the SS_setpoint there is a fair amount of variability in SS_act for each scan and it’s also clear from these figures that the relationship between SS_setpoint and SS_act can change with time. At most sites the percent difference between SS_setpoint and the SS_act was less than 20%. The largest difference was observed for column B at MOS (Fig. 3c,d), where differences of up to 40% between the instrument SS_setpoint,B and the SS_act,B were observed, likely due to variations in column temperature gradient related to instrument operating conditions or calibration issues. The difference between SS_setpoint and SS_act varied among sites and also across time for some sites. Several supersaturation values are included in the harmonized data files (see Data Handling and Data Records sections for definition of the SS provided and the description of the harmonized data files).

Fig. 3

CCNC supersaturation time series for ASI (a,b) and MOS (c,d). Plots a) and c) show the SS_setpoint time series for columns A (blue dots) and B (orange dots) of the CCNC at both sites. Plots b) and d) display the SS_act time series for columns A and B.

Fig. 4

(a) Time series for SS_setpoint,A (blue dots) and SS_setpoint,B (orange dots); (b) Time series for SS_act,A and SS_act,B; (c) example time series of N_CCN,A for one SS_setpoint,A with horizontal dashed lines indicating the 25th and the 75th percentiles of N_CCN,A. (d) Example of CCN interpolation for one column A SS scan. The blue dots are the mean measured N_CCN,A between the 25th and 75th percentile at the mean SS_act,A (Fig. (c)), the blue line is the two degree polynomial fit line based on the blue dots, the yellow dots are the interpolated (i.e., on the fit line) N_CCN,A at the mean SS_act,A and the pink dots are the N_CCN,har values corresponding to each SS_har value. All figures are for CCNC measurements at GUC.

Ideally CCNC calibration occurs before and after the deployment and/or approximately annually. Calibration of the CCNC includes flow and pressure calibrations and procedures to determine the thermal characteristics of the instrument. The flow, pressure and thermal characteristics are then used to derive the SS_act⁵⁵. All DOE/ARM level b1 operational CCN data sets (i.e., aosccn2cola.b1, aosccn2colb.b1, and aos2colaspectra.b1) used in this study provide the SS_act, which takes into account the thermal characteristics of the column⁵⁵ and the instrument calibrations. In contrast, the two SBS sites, CP and SPL, report N_CCN referenced to the instrument SS_setpoint rather than the SS_act. SBS-CP only had the data level a1 datastream available which does not include SS_act and the CCN dataset from SBS-SPL was sourced from a PI rather than DOE/ARM operational measurements (see Table 4). These data were processed, quality-assured and put into the same harmonized format as we used for the DOE/ARM operational CCNC data, albeit with some assumptions related to SS values (see data handling section for specific details).

Particle number concentration

A CPC measures the total number concentration (N_tot,CPC) of particles above a certain minimum particle size. Table 2 lists the model of CPC (TSI 3010 or TSI 3772) that operated at each site. The manufacturer’s specifications for these two CPC models indicate that they have the same minimum size detection limit of 10 nm.

The general principle of a CPC instrument is that it enables the operating fluid (typically butanol) to condense onto particles in the sample flow, creating aerosol droplets large enough to be detected efficiently using a light-scattering technique. Upon entering the CPC instrument, the sample air stream passes through a saturation block where the operating fluid evaporates into the sample stream, saturating the flow with vapor. The sample air stream then passes into a condenser tube cooled by thermoelectric coolers. Here, the operating fluid vapor supersaturates and condenses onto the particles. The minimum diameter of the original particles that will grow into droplets depends on the temperature difference between the saturator and the condenser. Droplets leaving the condenser tube pass one at a time through a single particle counting optical detector. Although the CPC may suffer counting issues at extremely high concentration (>10⁴ for the TSI 3010 and 3772 models), these data have not been removed or flagged in the database. The CPC datastream is dried to low RH prior to entering the instrument. The DOE/ARM CPC handbook describes the CPC and operations in more detail⁵⁶.

The CPC dataset from SBS-SPL was sourced from a PI rather than DOE/ARM operational measurements archive (see Table 5). The same processing and quality assurance criteria as were applied to the operational CPC instruments were applied to the SBS-SPL CPC.

Table 5 Uncertainties for different measurements based on literature and DOE/ARM instrument and uncertainty handbooks.

Sub-micron aerosol chemical composition

Aerosol chemical composition was measured by a Quadrupole Aerosol Chemical Speciation Monitor (Q-ACSM, hereafter referred to as ACSM) at many of the sites contributing to the harmonized data set. The ACSM measures sub-micron particle mass and chemical composition in real time. Specifically, it measures mass concentrations of particulate organics, sulfate, nitrate, ammonium, and chloride (M_organics, M_sulfate, M_nitrate, M_ammonium, and M_chloride) from 40 nm to 1 µm in aerosol aerodynamic diameter and a mass range (m/z) of 10 to 200 amu. DOE/ARM operations and protocols for the instrument are found in the ACSM instrument handbooks^57,58.

The main ACSM inlet system has a flow rate of 3 lpm from which a sample flow rate of approximately 0.1 lpm is picked-off to create near isokinetic sampling conditions. There is a Nafion dryer upstream of the ACSM at each site to keep the sample RH lower than 30%. The ACSM sample flow is focused by an aerodynamic lens, creating a 1 mm diameter aerosol stream that impacts a hot (600 °C) vaporizer. The vaporized aerosol is then ionized to produce the ions which are analyzed by the quadrupole mass spectrometer. The ACSM uses a naphthalene source for reference during calibrations and to determine an effective ion transmission efficiency through the quadrupole mass spectrometer. This ion source operates continuously resulting in background mass that must be subtracted from the particle mass spectra to accurately determine the particle mass composition.

The ACSM does not detect all particles in the air stream but rather a fraction of aerosol particles. This fraction is defined by a parameter known as the collection efficiency (CE) that should be applied to account for this incomplete detection. A primary cause of a CE less than 1 is the particles bouncing off the vaporizer impaction surface prior to vaporization and dectection⁵⁹.

One approach to deal with this undersampling is to assume a constant CE, e.g., CE = 0.5 to indicate only half of the particles are detected, e.g.⁶⁰.

More recently composition-dependent collection efficiencies (CDCE) have been used to express CE as function of chemical composition. CDCE helps account for the phase properties of ambient aerosol particles. Particles that exhibit more liquid-like properties tend to stick onto the vaporizer resulting in a CE closer to 1.0, whereas solid particles bounce off the vaporizer resulting in lower CE values⁶¹. For the DOE/ARM ACSM operational data, DOE/ARM data level c files (datastream acsmcdce) apply a CDCE parameterization^58,61, while the level b files (datastream aosacsm) assume a constant CE of 1.0. For most sites (ASI, COR, ENA, GUC, and SGP) the level c files were available. For MAO, only ACSM level b data files were available. We derived CDCE values for this level b data using the CDCE methodology^58,61, and checked our processing against another site which had both level b and level c data.

Aerosol optical properties

Measurements of aerosol scattering and absorption coefficients were made at most sites included in this harmonization project. The aerosol scattering coefficients were measured by an integrating nephelometer, specifically a TSI 3563 integrating nephelometer which measures aerosol scattering (σ_sp) and backscattering coefficients (σ_bsp) at three wavelengths (450, 550, and 700 nm). The nephelometer measures aerosol scattering by detecting light scattered by the aerosol and subtracting light scattered by the gas, the walls of the instrument and the background noise in the detector. Due to instrument angular limitations the TSI nephelometer only measures scattering between 7° and 170°, requiring the so-called truncation correction to account for the angles not measured. Anderson et al.⁶² and Anderson and Ogren⁶³ provide an overview of the nephelometer and the truncation correction, respectively. In the DOE/ARM operational files (aosnephdry1m) the nephelometer data are reported at standard temperature and pressure (STP, T_std = 0 °C and P_std = 1013 hPa) and have already been corrected for angular truncation and lamp non-idealities⁶³.

The aerosol absorption coefficients (σ_ap) were obtained from the three wavelength (464, 529, and 648 nm) Radiance Research (RR) Particle Soot Absorption Photometer (PSAP⁶⁴). This is a filter-based instrument which relates the decrease in light transmission through a filter due to particle loading to the aerosol absorption coefficient. Bond et al.⁶⁴ note that several corrections are required to transform the change in transmittance to an σ_ap value, including accounting for the aerosol scattering artifact, properly measuring the size of the filter spot, and a calibrated flow rate. Virkkula et al.⁶⁵ additionally account for the single scattering albedo of the sampled aerosol. We utilized the DOE/ARM developed, value added product datastream [stn]aoppsap1flynn1m[fac].c1.[timestamp].nc⁶⁶ which is produced operationally. These files contain three different versions of corrected PSAP data (i) average of Virkkula^65,67 and Bond/Ogren^64,68 correction; (ii) Virkkula correction alone, (iii) Bond/Ogren correction alone. Note that DOE/ARM recommends the average of the two types of corrections “due to beneficial cancellation of offsetting correction artifacts” so that is the value included in the harmonized data set⁶⁶. The DOE/ARM nephelometer and PSAP handbooks describe these two instruments and their operations in more detail^69,70.

At all sites studied here, both the nephelometer and PSAP were downstream of a switched impactor system. Over the course of an hour the size cut for the σ_sp and σ_ap measurements switched between 10 µm and a 1 µm in diameter, referred to here as the PM₁₀ and PM₁ size cut, respectively. This enables quantification of the contribution of coarse mode aerosol to σ_sp and σ_ap, complementing information derived from the spectral dependence of σ_sp and σ_ap. The timing of the switching between PM₁₀ and PM₁ was designed to sample both size cuts over the course of an hour, though not necessarily for the same amount of time. For example, a typical repeating three hour pattern was the first hour measured 7 min on PM₁₀ and 52 min on PM₁, the second hour measured 7 min on PM₁₀, 45 min on PM₁ and then 7 more min on PM₁₀ and the third hour measured 7 min on PM₁ and 52 min on PM₁₀. Despite the datastream name (aosnephdry1m), the nephelometer sample stream was not actively dried for many of these deployments. This resulted in a sample RH for the scattering measurements that varied significantly depending on the both ambient temperature and RH conditions and the temperature conditions inside the measurement trailer and instrument. RH conditions were part of our further assessment as described in the data handling section.

Data handling

To simplify sharing of original and harmonized data, all data sets from DOE/ARM were downloaded to a server at the University of Utah to which team members had access. Individual members of team were responsible for evaluating data from specific instruments. Once the data were further quality controlled and averaged the processed files were uploaded to a different directory on the Utah server and were then available for technical validation (see next section). In what follows, we first outline the additional data processing (beyond that performed by DOE/ARM) and quality control procedures we followed in order to develop the final harmonized data. We then discuss data averaging, and finally we describe our calculation of some key CCN parameters from the harmonized CCN and PNSD data sets.

Data review

Reliability and comparability of the operational data sets presented here were assured by the application of the DOE/ARM technical standards and protocols^{52,53,56,69,70} and by further data checks we performed as described in the data handling and validation sections. The PI data sets used were unprocessed. For the SMPS and CPC PI data we performed the same processing and evaluation as was applied by DOE/ARM to the DOE/ARM operational data sets. For the CCNC PI data we had to make some assumptions as not all necessary information was available to replicate the DOE/ARM processing (see description of CCNC data handling below for more details).

The DOE/ARM operational data files come with instrument-dependent flags tied to data quality. Here we’ve taken the approach that if any of the DOE/ARM data flags were set for a given timestamp then the data for that timestamp were removed prior to our own QA processing and averaging. Additionally, data identified as invalid or suspect in the separate DQRs provided when the data files were downloaded were removed. There are a few exceptions to this as noted below. In the supplemental materials we have included sections for each instrument that list the standard parameters which might cause an automatic instrument flag to be set in the data file.

In addition to using the DOE/ARM provided flags we applied some additional constraints and processing. For example, all instruments are adjusted to and reported at STP in the harmonized data files, if not already reported at STP in the DOE/ARM files. Additionally, in order to ensure consistency across instruments, if the measurements were made at RH higher than 40% they were considered invalid and not included in the harmonized dataset (this only impacted some of the optical measurements). The SMPS and CCNC data required the most processing effort, so we discuss those data sets first and then describe the processing and constraints applied to the other instrument data sets.

SMPS data handling

The SMPS software allows the user to change the size range over which the instrument measures. While processing different data sets it became clear that different sites used slightly different size ranges and, sometimes, the size range setting changed over the course of time for individual sites. To harmonize the datasets we generated files with columns for the entire range of possible diameter bins for each site (113 bins, ranging from a minimum diameter of 9.14 nm to a maximum diameter of 514 nm). If the SMPS measurements were shifted to larger (or smaller) diameters and did not include particles in the smallest (or largest) size bins, those bins were filled with missing value codes. Thus, the harmonized SMPS data sets will always have the same number of columns, with each size bin column always representing a consistent diameter.

CCNC data handling

One of our end goals for the CCNC data is to generate files containing mean N_CCN values interpolated to the same harmonized SS (SS_har) for all the sites, regardless of the SS_act. We chose SS_har values of 0.1, 0.2, 0.4, 0.6, 0.8 and 1%. This harmonization required a three-step process: (1) calculating the mean N_CCN and mean SS_act for each SS_setpoint in a scan; (2) deriving a fit to those values for each scan so that mean N_CCN can be interpolated to different mean SS_act; (3) calculating N_CCN,har by interpolating mean N_CCN to the harmonized SS (SS_har). Step 1 is done for both column A and column B, while steps 2 and 3 can only be done for column A. For most sites steps 1 and 2 are done by the DOE/ARM processing with the mean N_CCN, SS_act and fit parameters available in the operational data files, while step 3 is done by us to get the mean N_CCN values interpolated to our chosen SS_har. For a few sites (ANX, MAO, SBS-CP and SBS-SPL), we performed all three steps in the harmonization process as the needed files or other information was not available from DOE/ARM or the data provider. We describe these steps in more detail below.

At most sites we only used DOE/ARM’s CCN spectra file (either aos1colaspectra or aos2colaspectra, depending whether it is a 1 column or 2 column instrument). The spectra file provides the mean value of N_CCN,A between the 25th and 75th percentile (i.e., the interquartile range, IQR) for each SS and the mean value of various parameters (SS_setpoint,A, SS_act,A and N_tot,CPC) coincident with the N_CCN,A falling within interquartile range. Figure 4c is a graphical example of how step 1 in the harmonization process is performed. The figure shows a 6.2 minute time series of 1-sec N_CCN,A data for one SS_setpoint,A value during a SS scan, with the horizontal dashed lines indicating the 25th and 75th percentiles of the plotted data. The mean N_CCN,A is the mean of the data points falling between the two dashed lines and the mean SS_act,A is the mean of the SS_act,A associated with N_CCN,A values contributing to the mean N_CCN,A. The mean SS_act,AB associated with N_CCN,B are determined the same way, with the column B data being temporally matched with the mean N_CCN,A.

The spectra file also includes DOE/ARM’s fit parameters that describe the relationship between mean N_CCN,A values and the mean SS_act,A values for each hourly scan across the range of SS. Figure 4 shows a fit to the mean N_CCN,A and mean SS_act,A for a single SS scan and provides a graphical example for steps 2 and 3 in the harmonization process. The step 2 fit is performed using a two degree polynomial (Eq. 1):

$${N}_{{CCN}}={{aSS}}_{{act}}^{2}+{{bSS}}_{{act}}+c$$

(1)

where a, b, and c are the fit parameters.

The fit parameters are arrived at iteratively. For each scan, the fitted N_CCN values are compared against individual mean N_CCN,A data for each SS_act,A and, if the difference is >20% for a given SS_act,A, the corresponding mean N_CCN,A and SS_act data is removed and a new fit is performed on the remaining mean N_CCN,A and SS_act,A data points. This calculation is repeated until no more N_CCN,A outliers are observed but there are still 3 or more mean N_CCN,A and SS_act,A values available to perform the fit. This procedure allows better data quality assessment than looking at N_CCN,A for individual SS values because looking at the SS spectra permits identification (and exclusion) of spectra during periods when instrument conditions were unstable.

Figure 4d shows an example of the step 3 interpolation procedure for one SS scan at GUC. The blue dots correspond to the mean measured N_CCN,A between the 25th and 75th percentile at the mean SS_act,A (calculated in step 1). The blue line is the two degree polynomial fit line based on the blue dots (calculated in step 2). The yellow dots are the interpolated (i.e., on the fit line) N_CCN,A at the mean SS_act,A and are all within 20% of the blue dots. The pink dots are the N_CCN,A,har values, i.e., the mean N_CCN,A values interpolated to SS_har (interpolation done in step 3). If the interpolation led to unreasonable N_CCN,A,har values (e.g., negative concentrations, or lower concentrations at higher SS), the whole scan is reported as invalid.

We developed code to reproduce the ARM processing (steps 1 and 2) so that we could generate harmonized CCN data for sites where the spectra file was unavailable (ANX, MAO, and both SBS sites). To check our code, we compared the ARM operational spectra files with the spectra files we generated to be sure the results were identical. For MAO and ANX we use the level b1 CCN operational file (aosccn1col for MAO and aosccn2col for ANX) which includes 1-sec resolution data of N_CCN concentration as well as housekeeping parameters such as SS, thermal efficiency information, temperatures, pressure and flows. For SBS-CP we use the level a aosccn100 files. For SBS-SPL we use the 1-sec PI file. Note that the CCNC data sets for the two SBS sites only reported instrument SS_setpoint. Due to the limited information for the SBS sites (no thermal characterization) it was not possible to calculate the SS_act within the instrument so the harmonization procedure is performed assuming SS_setpoint was equal to SS_act for those two sites. To reflect this in the harmonized files for the two SBS sites, the columns in the data file corresponding to the SS_act are reported as missing data (see Data Records section for harmonized datafile descriptions).

At some sites due to narrower SS_setpoint,A ranges the SS_har also covered a narrower SS range to minimize interpolation errors. The site specific SS_setpoint,A values and resulting range of SS_har for each site are given in Table S1. Note that the column B SS cannot be interpolated to a harmonized SS value as this column only measured at a single SS.

It should be noted that in the case of high N_CCN (>5000 cm⁻³), the SS and droplet sizes can decrease due to higher water vapor depletion⁷¹. The water vapor depletion can result in lower N_CCN. In the harmonized data set, the water depletion effect has not been taken into account. At SBS-SPL, ANX and ASI there are no N_CCN values >5000 cm⁻³. At COR, ENA, MOS, GUC, SBS-CP and EPC, the percentage of N_CCN measurements >5000 cm⁻³ is smaller than 0.1%. For SGP, less than 0.8% of measurements exceed this threshold, while at MAO, this percentage is 2%.

It is important to highlight that, during our initial data review, several of the ARM CCN data sets were identified as suspicious and required reprocessing due to incorrect instrument calibrations being applied. The ARM CCNC instrument mentors revised and updated the datasets in the ARM data portal and those are the datasets used here.

CPC data handling

The original CPC measurements are reported at 1 second resolution. In all cases they have been converted to STP in order to be consistent with the other measurements which are also at STP. We used temperature and pressure information from the SMPS or the CCN to do the STP correction as described in the Data Records section.

ACSM data handling

The DOE/ARM ACSM data were provided at different temporal resolutions, representative of the average number of sets of sampling cycles. All operational ACSM datasets were converted to 1-h resolution for simplicity due to different averaging time periods within each dataset. The original ACSM datasets were not corrected to STP conditions and the ACSM files did not include sample and measurement temperature or pressure. To do the ACSM STP adjustment, the sample temperature and pressure measurements from the collocated SMPS were used to convert ACSM measurements to STP at each site.

Optical data handling

Optical property aerosol measurements where the sample RH was greater than 40% is reported as missing in the harmonized data set. We use the nephelometer sample RH as the RH constraint for both the nephelometer and PSAP because the PSAP does not have its own measurement of RH and the PSAP requires the nephelometer scattering measurement in its corrections. This RH constraint is based on the Global Atmospheric Watch (GAW) programme recommendation to make aerosol measurements at low relative humidity (RH < 40%). It was often challenging to meet this criteria at many of the DOE/ARM sites. At ASI there is no harmonized optical data available due to this RH constraint, and some sites have limited optical data, particularly in warmer months when it is more difficult to control RH.

Other constraints and processing

Below is an itemized list of some exceptions to or issues that arose with the general QC described above

• At ASI, COR, MAO, and SBS (CP and SPL), sample air RH was not available for the PNSD measurements because the SMPS instrument model (TSI 3936) used at those sites did not include a sample air RH sensor. As noted previously an upstream Nafion™ drying system was used to lower the SMPS sample RH, however, the resulting sample RH was not recorded (A. Singh, pers. comm., 2024). Because the dryer was used, we assume that these PNSD measurements were also made at low (<40%) RH.

• At ASI, the SMPS measurement range changed from 11–461 nm to 22–461 nm. For the time period when the minimum diameter started at 11 nm, most bins had zero particle counts up to 22 nm (possibly indicating a problem with the DMA). This results in small concentration values for the nucleation mode and many AF > 1, with no results for D_crit and κ_CCN. However, the closure study with the CPC looks acceptable (see discussion and figures in Technical Validation section).

• At SGP, most SS_act values for CCNC column B (fixed SS) are missing during 2019 and 2020. Because the measured column temperatures are available, the SS_act is calculated by considering the thermal efficiency table and equation 16 in Lance et al.⁵⁵.

• At ANX, we did not apply bit19 of the DOE/ARM quality control flag to the CCN files. This flag is related to suspect instrument thermal efficiency. If the QC flag were to be applied, all CCN measurements for the site would be marked as invalid. The instrument mentor noted that bit19 is just an indicator to look more closely at the data. Due to this bit being set in the operational files, all the DOE/ARM operational spectra files also report missing value codes for the fitted values of SS_act and N_CCN. At this site we started with the basic CCN operational file aosccn2cola to generate the final harmonized files. Closure tests utilizing the harmonized ANX CCNC data, described in the technical validation section below, look good.

• At ANX there was also a period where CCN column A data was described as suspect in a DQR due to a possible internal leak in the sheath air line. The closure comparison of both columns at 0.4% SS is acceptable (Fig. S4 in supplemental materials) so we have not flagged this data as invalid.

• For both MAO and SBS-SPL there is a period where the CCNC counter made both increasing SS and decreasing SS scans. Due to limitations of our interpolation processing scheme, the decreasing SS scans are not considered.

• At GUC, we did not apply bit17 (ambient_pressure < ambient_pressure_min _warning) of QC for CPC which occurred due to low ambient pressures (station was at 3137 masl, ambient pressure was approximately 670 mb). Takegawa and Sakurai⁷² suggest for 20 nm and higher particles, butanol-based CPCs should have same counting efficiency at this pressure as at sea level.

• Evaluation of the MAO ACSM data after the CDCE was applied using closure studies and comparison with other collocated datasets suggest the data were problematic. As the DQR for this data set also suggest it is suspect, MAO ACSM data have not been further considered in this study. Users interested in chemistry data from MAO are recommended to instead download PI aerosol mass spectrometer (AMS) data from 2 shorter IOP deployments available in the ARM IOP archive directory: /arm-iop/2014/mao/goamazon/T3/alexander-ams.

Data averaging and uncertainties

Following Schmale et al.²⁶, we have provided the data at hourly resolution for all instruments studied. These data can be used for Earth system model evaluation^73,74 and climatological/phenomenological analyses^{27,75,76,77,78}. Additionally, we have provided 5-min averages of PNSD, CCN, N_tot, and scattering and absorption coefficients which have the added advantage of being useful for investigating faster time scale processes such as NPF^16,29,79.

For both averaging times the timestamp for the data represents the start of the averaging period. The 1-h and 5-min files for a given instrument file have the same format. Where data from different instruments is merged the measurements represent the same time period. Note: we cannot provide 5-min averages of CCNC column A because the SS scanning does not nicely line up with 5-min intervals. Similarly, the sampling cycles for the ACSM did not lend themselves to 5-min averaging.

It is also important to provide information on measurement uncertainties. For the DOE/ARM instruments in this study, Sisterson⁸⁰ provides information on measurement uncertainty primarily related to instrument calibration. Calibration uncertainty relates to the measurement error determined from testing the instrument with reference standards. Schmale et al.²⁶ also provide general uncertainty information for PNSD, CCN and chemical composition measurements along with citations for more in-depth uncertainty information for the SMPS, CCNC and ACSM. Sherman et al.⁸¹ included detailed information about uncertainty calculations for the TSI nephelometer and PSAP in their supplemental materials. Providing individual uncertainty calculations for the 6 instruments and 10 sites are outside the scope of this manuscript as they depend not only on the individual instruments, but also environmental conditions (e.g., atmospheric aerosol loading) which varies from location to location. Based on the references listed above we have included some very general estimates of uncertainty for the individual instruments along with comments and further references in Table 5.

Calculated parameters

From the processed and harmonized CCN and PNSD data, we have calculated the activated fraction, critical diameter and hygroscopicity parameter for each site. These calculations are described below.

The AF_inst is the ratio of the CCN concentration to the total number concentration (Eq. 2):

$${{AF}}_{{inst}}({SS})={CCN}({SS})/{N}_{{tot},{inst}}$$

(2)

where _inst indicates the instrument providing the value of N_tot (either SMPS or CPC). AF is calculated for each SS. In the data files we provide AF values based on N_tot from both the CPC (AF_CPC) and the SMPS (AF_SMPS).

Critical diameter for each SS was estimated by integrating the PNSD backwards from largest SMPS diameter (D_max) to smaller diameters. The D_crit for a given SS is the diameter when the integrated number from the PNSD is equal to the CCN concentration at that SS (Eq. 3). This approach has been used previously, e.g.^82,83. Calculating D_crit in this way provides a step-function activation cut-off diameter (all particles larger than this size act as CCN) assuming an internal mixture.

$${N}_{{CCN}}\left({SS}\right)={\int }_{{D}_{{crit}}({SS})}^{{D}_{\max }}\frac{{dN}}{{dlogD}}{dlogD}$$

(3)

The κ_CCN value is calculated using κ-Köhler theory⁸⁴. This is done in an iterative approach: D_crit is first calculated as explained above, then D_crit is put into the Köhler Eq. (4), and then κ_CCN is varied until the corresponding critical SS equals the SS value at which N_CCN was measured.

$$S=\frac{{D}^{3}-{D}_{{crit}}^{3}}{{D}^{3}-{D}_{{crit}}^{3}(1-{\kappa }_{{CCN}})}{\exp }\left(\frac{4{\sigma }_{s/a}{M}_{w}}{{RT}{\rho }_{w}D}\right)$$

(4)

where S is the saturation vapor pressure of water over an aqueous solution droplet of diameter D, σ_s/a is the surface tension at the surface/air interface, M_w is the molecular weight of water, R is the universal gas constant, T is temperature and ρ_w is density of water. We used the surface tension of pure water in all of our calculations since at the point of activation we assume that the solution is diluted. Equation 4 can be expressed in terms of the SS, which is defined as SS = (S-1)*100. These CCN-derived κ_CCN values quantify the effective hygroscopicity of activated particles in the CCNC and show a dependence on SS.

We should note that DOE/ARM provides a value-added product (VAP) data stream (aosccnsmpskappa) for many of these locations that includes calculated κ_CCN values. We do not use this data stream because in it the D_crit and κ_CCN are calculated for SS values different than our SS_har values.

In order to have a more complete representation of aerosol mass concentration in our validation analysis, we also calculated the black carbon mass concentration M_BC,PSAP from the PSAP σ_ap measurements (Eq. 5):

$${M}_{{BC},{PSAP}}={\sigma }_{{apG},1}/{MAC}$$

(5)

where σ_apG,1 is the green wavelength (529 nm) PM₁ absorption coefficient measured by the PSAP and MAC is the mass absorption cross section which was assumed to be 10 m² g^−1 85.

Data Records

We developed 6 types of aerosol data files in the course of the harmonization process. Table 6 provides an overview of all the harmonized files (file names and contents), while Tables 7–12 provide a more detailed description of what is in each file. Each of these files is also briefly described below. The file names have the following format: SiteID_Instrument_avgt.csv where SiteID is the three letter identifier given in Table 1, Instrument is the instrument type and avgt is the time resolution where appropriate (i.e., 5 min or 1 h for 5-min averaged and 1-h averaged), respectively. For most sites the 1-2 years of data are concatenated into a single file and no year is included in the file name. For SGP we have included the year in the file name (SGP_Instrument_avgt_Year.csv) as there are seven years of data and we’ve provided the data for SGP split by year. For the MOS dataset there is a separate file containing latitude and longitude as those measurements were taken aboard a moving ship. The MOS position data file was generated from the campaign navigation files^{86,87,88,89,90} which we averaged over 5-min and 1-h to match our harmonized data files. For all time series files the primary time stamp is given in the form: YYYY-MM-DD hh:mm:ss. We also provide README files for each site and instrument that briefly describes the variables and contain additional metadata (e.g., instrument model, site location, etc). All harmonized data files are available in the figshare repository⁹¹.

Table 6 Summary of harmonized data files.

Table 7 Description of SiteID_smps_avgt.csv file. ¹Midpoint diameter for each size bin is given in header for each. dN_i/dlogD_i column, e.g., if the midpoint diameter is 101.8 nm then the column header is D_101_8.

Table 8 Description of SiteID_ccn_cola.csv file.

Table 9 Description of SiteID_ccn_colb_avgt.csv file.

Table 10 Description of SiteID_acsm.csv file. Sample pressure and temperature used for STP conversion were taken from SMPS.

Table 11 Description of SiteID_neph_avgt.csv file.

Table 12 Description of SiteID_psap_avgt.csv file.

The harmonized SMPS files (see Table 7) contain N_tot,SMPS, the particle number concentrations for each individual size bin i (dN_i/dlogD_i), some housekeeping information (T, P, RH), N_tot,CPC, and several closure flags for number closure, scattering closure and mass closure (see Technical Validation section). The harmonized file name for the SMPS datasets is SiteID_smps_avgt.csv. The N_tot,CPC value is the mean between the 25th and 75th percentile of the instrument’s measurement period in the file. Here we used the SMPS measure of sample T and P to correct N_tot,CPC to STP.

There are two types of files for the harmonized CCN measurements - one for column A data (see Table 8) and one for column B data (see Table 9). The harmonized file name for the CCN datasets are SiteID_ccn_cola.csv and SiteID_ccn_colb_avgt.csv. If the CCNC was a one column instrument (i.e., at MAO and both SBS sites) then only the column A file was generated, with NaN values for fields related to column B values. The SiteID_ccn_cola.csv file contains N_CCN,A, N_CCN,AB, N_tot,CPC and N_tot,SMPS in (cm⁻³) and SS_act,A and SS_har values as well as some housekeeping parameters and flags indicating results of the quality checks (see Technical Validation section) and calculated parameters. As with the SMPS files, the N_tot,CPC value is the mean between the 25th and 75th percentile of the instrument’s measurement period for each time stamp in the file (for example, in column A file, the CPC value given is the mean value at each SS time). Here we used the CCNC measure of sample T and P to correct N_tot,CPC to STP.

The SiteID_ccn_colb_avgt.csv file contains similar information, but only for the second CCNC column (column B) which is held at constant SS. It should be noted that the averaging time for the column B parameters is different in the two CCN files. The column B parameters included in the SiteID_ccn_cola.csv file (e.g., N_CCN,AB, and SS_act,AB are calculated from values temporally matched to the filtered column A parameters (i.e., they are matched to N_CCN,A between the 25th and 75th percentile for each SS_act,A). In the SiteID_ccn_colb_avgt.csv file, the mean N_CCN,B represents the average of the high resolution N_CCN,B between the 25th and 75th percentile for the averaging time of the file (5-min or 1-h). The other mean parameters (SS, flows, etc.) in the column B file are calculated from values temporally matched with the filtered high resolution N_CCN,B.

The harmonized ACSM files (see Table 10) contain the mass concentration values (µg m⁻³) of the particle organics, sulfate, ammonium, nitrate, chloride, from the ACSM and black carbon derived from the PSAP data. Total mass concentrations calculated from the ACSM, SMPS, and combined ACSM and PSAP are included, as well as a quality assurance flag (qc_ACSM_SMPS) indicative of how well the combined mass concentration of the ACSM and PSAP compares with the mass concentration derived from the PNSD data (see Technical Validation section). The harmonized file name for the ACSM datasets is SiteID_acsm.csv.

The harmonized nephelometer data sets (see Table 11) include the truncation-corrected scattering and backscattering coefficient values (in Mm⁻¹) at the three measurement wavelengths (450, 550 and 700 nm) for the two measurement size cuts (PM₁₀ and PM₁). Additionally, they include nephelometer housekeeping values (T, P, RH) and the number of nephelometer data points making up each average value. Nephelometer RH > 40% was the primary reason scattering data are reported as missing in the harmonized data file. The harmonized file name for the nephelometer datasets is SiteID_neph_avgt.csv.

The harmonized PSAP data sets (see Table 12) include the corrected absorption coefficient values (in Mm⁻¹) at the three PSAP measurement wavelengths (464, 529 and 648 nm) for the two measurement size cuts (PM₁₀ and PM₁). The PSAP files also include the nephelometer sample RH value as nephelometer RH > 40% was primary reason absorption data were excluded. The PSAP correction requires the nephelometer data, so if the nephelometer data were invalid due to RH > 40%, then the PSAP data were also marked as invalid. The harmonized file name for the PSAP datasets is SiteID_psap_avgt.csv.

Technical Validation

Following the initial data review and handling, a variety of checks and closure experiments were performed to assess the data quality and further screen the data as described below. Successful closure experiments indicate that the data sets are internally consistent and are commonly used as part of the QA/QC process^{26,92,93,94,95,96}. Only sites and periods of measurements passing our detailed QA/QC process and closure tests as described below are included in the final harmonized data sets presented in this paper. We have also included flags (described below) that the user may want to apply as further data constraints. We use the suite of measurements from ASI, MOS and GUC to provide an example for some of our time series analyses (e.g., Figs. 3, 4a,b) and for GUC for each closure test (Fig. 5). Similar plots for all the sites can be found in supplemental materials.

Fig. 5

Example of data closure studies for GUC. (a) Number closure; (b) Scattering closure; (c) PM₁ mass closure; (d) CCN column closure; (e) CCN number closure at 1% SS; (f) CCN number closure at 0.4% SS for column A; (g) Same as f but for column B. The solid line is the 1:1 line and the dashed lines indicate +/−50% deviation from the 1:1 line. Plots a, b, c and g use 1-h averaged data, while plots d, e, and f used 1-sec data when both CCN columns overlapped (d) or when SMPS and CCN column A overlapped (e and f).

Time series analysis

Plots of the full data time series for each parameter, monthly and diurnal patterns and interannual variability (where possible) were used to identify potential irregularities in the data. For example, time series analysis of the wavelength dependence of the nephelometer scattering data at SBS-CP led to removal of this data set from consideration. Specifically, total scattering coefficient at the red wavelength was often higher than total scattering coefficient for the blue and green wavelengths which is unrealistic. Additionally, at that site the backscattering coefficient values were often similar to and sometimes greater than the total scattering coefficient values; for atmospheric aerosols measured at low RH the backscattering coefficient is normally 10–20% of the total scattering coefficient⁸¹. Additional time series plots of instrument specific diagnosis variables such as supersaturations (e.g., Fig. 3 for ASI and MOS and Fig. 4a,b for GUC), flow ratio, temperatures, RH, and so on, were also inspected to identify possible malfunctioning of each instrument.

Number closure

For each site included in the analysis, the collocated total particle number concentration (N_tot) from a CPC (N_tot,CPC) was compared with that calculated by integrating PNSD (N_tot,SMPS). Ideally, the integrated PNSD should be highly correlated and within a few percent of the total number concentration. Here we flagged the PNSD data when the comparison between the integrated PNSD and the independent N_tot,CPC was greater than 50%. The flag in the harmonized files (SiteID_smps_avgt.csv) for this closure test is Qc_CPC_SMPS. Because such a comparison does not definitively indicate which measurement is problematic (SMPS or CPC), other closure experiments with independent measurements were also performed in order to isolate the problem measurement. Figure 5a for GUC shows the scatter plot of the total particle number concentration measured with the SMPS and the CPC for 1-h time resolution (this closure test was also done with the 5-min data for all the sites). The results show very good agreement between both instruments, with most of the data confined within the + /−50% deviation from the 1:1 line. The slope is <1 indicating that the CPC instrument typically reported higher N_tot than the SMPS. Figure S1 in the supplementary material presents number closure plots for all the sites analysed. All the sites show a slope < 1 except ENA (Fig. S1d) and MOS (Fig. S1g). This may be due to instrument issues that were not caught by the automatic QC or mentor DQR reports. Alternatively, it could also be due to problematic sampling conditions; for example, at MOS the system developed for excluding ship emissions did not work, necessitating development of a post-processing pollution detection algorithm⁹⁷. That contamination masking algorithm is not applied here, but should be considered when the data are used.

Scattering closure

Another test of the PNSD data quality was performed by calculating the aerosol scattering coefficient from the PNSD using Mie theory⁹⁸ and comparing to the sub-micron green wavelength scattering (σ_spG,1) measured by the integrating nephelometer. For all sites we assumed the particles were spherical and had a constant index of refraction of 1.55 + 0.01i. If the number closure for a time period looked problematic but the scattering closure was good then the problem is likely with the CPC measurements rather than the PNSD measurements. Scattering closure was only performed for hourly averaged data when there was a PM₁ size cut for scattering and the nephelometer sample RH < 40%. These two constraints limit where scattering closure can be done for some locations and time periods. The closure flag in the harmonized SMPS data files (SiteID_smps_avgt.csv) is noted as Qc_Neph_SMPS. The scattering closure result for GUC is shown in Fig. 5b. This test shows very good results at GUC, with a slope near 1 and high correlation coefficient (R = 0.9), indicating good agreement between the SMPS and the nephelometer. The good scattering closure and number closure tests confirm the good performance of the SMPS at GUC. Figure S2 in the supplementary materials presents scattering closure plots for the other sites. For all sites the correlation coefficient is higher than 0.8 except at ENA (Fig. S2d), with a value of R = 0.64. The data dispersion observed for this closure test, together with the dispersion for the number closure test, leads us to believe that the SMPS measurements at ENA may have a problem.

PM₁ mass closure

A final PNSD check was performed by performing PM₁ mass closure tests. The use of the level c CDCE-corrected ACSM data is expected to result in a better mass closure result than if a constant CE is used⁶¹. For this closure test, the sub-micron mass calculated from the PNSD (M_SMPS) was compared with the PM₁ mass from the ACSM plus M_BC,PSAP estimated from the PSAP absorption coefficient measurement (M_ACSM+PSAP). If the number and scattering closure tests appeared to be good but the mass closure was questionable then this suggests a problem with the ACSM measurements. A mass concentration was calculated from the PNSD for aerosols larger than 40 nm to remain consistent with the aerosol range of the ACSM (40 nm to 1 µm)⁵⁷. The density used to calculate (M_SMPS) was a weighted density based on the densities listed in the ACSM instrument handbook⁵⁷ and the mass concentrations of the species measured by the ACSM. The densities of aerosol organics, sulfate, nitrate, ammonium, and chloride are 1.2, 1.77, 1.72, 1.77, and 1.52 g cm⁻³, respectively⁵⁸. The density of BC used in this study is 1.7 g cm^−3 99. In the harmonized ACSM files (SiteID_acsm.csv), we flagged ACSM values (flag = Qc_ACSM_SMPS) when M_ACSM+PSAP differed from M_SMPS by more than 50%. This flag serves as an additional quality control criterion to ensure the accuracy and agreement of ACSM and SMPS measurements. It should be noted that the SMPS measurements are referenced to aerosol mobility diameter while the ACSM measurements are reported for aerosol aerodynamic diameter. We have not accounted for the difference in diameter type in this closure test.

Correlation coefficients between the two mass values are also included as an additional criterion for quality assurance. The result of the PM₁ mass closure for GUC is shown in Fig. 5c. The closure results for GUC show good agreement (slope = 0.91, R = 0.95) between M_ACSM+PSAP and M_SMPS, suggesting a good performance of both instruments during the measurement period. Figure S3 in the supplementary materials presents the PM₁ mass closure plots for the other sites. Again, all the sites show a good agreement between measurements except ENA (Fig. S3d) (slope = 0.34 and R = 0.48). The three closure studies applied to the SMPS data for all the sites suggest a good agreement for all the sites except at ENA, for which a further filtering of the SMPS data based on our closure test flags in the harmonized data sets is recommended.

CCN column closure

As noted above, some sites operated a two column CCNC. When this was the case, we compared N_CCN,B which operated at constant SS_setpoint,B to N_CCN,A at the same SS_har from column A. This comparison is expected to show consistency between the two columns but will not be perfect because while the column SS_setpoint values are identical the SS_act in each column differ. For column A we used N_CCN,har while for column B we used N_CCN,act,B (recall N_CCN,B cannot be harmonized to a specific SS because column B was operated at a single SS). If the N_CCN for the two columns are not consistent this suggests an issue with the calibration of one or both columns. The closure flag in the SiteID_ccn_cola.csv files is Qc_column_AB. Figure 5d for GUC shows the results of this comparison. In the case of GUC, the agreement is exceptional, with a slope of 0.99 and R = 1. Figure S4 in the supplemental materials presents the column closure results for all the sites analysed. The excellent agreement observed at GUC is not always found for the other sites. Note: at MOS (Fig. S4g), the SS_setpoint,B value is 0.4% SS but, as the corresponding SS_act,AB mean values are approximately 0.6%, this column is compared with the column A measurements SS_har at 0.6%.

CCN number closure at high SS

CCN number closure was performed by comparing N_tot,SMPS with the N_CCN,A measured at the highest supersaturation value. The CCN concentration should always be equal to or less than N_tot as there cannot be more CCN than there are particles. At 1% SS most particles are likely to be activated, although this is highly dependent on the aerosol properties, especially on the aerosol hygroscopicity and size. The presence of very small particles can negatively impact the comparison as they can be too small to activate in the CCN counter, so this comparison was performed only for particles with diameters larger than 30 nm (N₃₀). The closure flag in the SiteID_ccn_cola.csv file is Qc_CCN1_N30. Figure 5e for GUC and Fig. S5 of the supplemental material for all the sites analysed show the results of this closure. For most sites, the results are consistent and the slopes of the comparison are >1 (except ASI, where all the particles activate). The largest slope is obtained for MOS (Fig. S5g, respectively) indicating the significant contribution of non-CCN active particles at these three sites.

CCN number closure at 0.4% SS

We found some cases in which the two columns of the CCNC did not agree (as in SGP where some points were outside the 50% bounds (Fig. S4j)). In order to elucidate which column is performing properly, we performed an additional CCN number closure test with the PNSD. In this case, we compared the CCN concentration at 0.4% (which is the common SS in both columns) against the N₈₀ (particle number concentration of particles with diameter above 80 nm from the SMPS). This comparison is not definitive since the 80 nm threshold might be more suitable for some sites than for others. However, it offers additional insight into the behaviour of both columns and the activation properties at the different sites. Data were flagged if the difference was greater than 50%. The closure flag for this test in the harmonized files (SiteID_ccn_cola.csv) is Qc_CCN04_N80. Results of this quality check are shown for GUC in Fig. 5f,g for columns A and B, respectively, and in Figs. S6, S7 of the supplementary material for all the sites analysed. For GUC, the agreement of both columns with N₈₀ is very good, and the slopes are near 1 indicating that at 0.4% the critical diameter is around 80 nm. Comparing slopes for the different sites, the clean marine sites (ANX, ASI) have slopes < 1, likely associated with critical diameters less than 80 nm.

Usage Notes

The harmonized csv format and consistent date stamp (YYYY-MM-DD hh:mm:ss) across instrument files should make the files readily usable in most software packages (R, Python, Matlab, IDL, Igor…) as well as more basic spreadsheet tools like Microsoft Excel. As noted above, each file type for each station comes with a README file describing the data and includes other metadata about the site and the instruments. The harmonized file names and contents are summarized in Table 6.

The time series files of the variables are designed so that they can be used without further processing. We have also provided calculated values used in the closure analyses (e.g., mass_SMPS_STP and mass_ACSM_PSAP_STP) as well as parameters related to CCN (AF, D_crit, and κ_CCN) to simplify analyses related to CCN processes. The multi-year SGP data set enables some assessment of inter-annual variability and climatology. For the rest of the locations the shorter time period of measurements means the user should assess the climatological representativity of the data by determining whether conditions were ‘normal’ or ‘exceptional’ during the time period of the available data using meteorological and other data not included in this harmonized dataset. Tables S3, S4 provide information on how much data would be removed from the SMPS and CCN datasets depending on which closure flags are applied. We recommend the user consider their application when deciding what closure flags to apply. For example, SMPS and CPC data are often flagged during NPF events when CPC concentrations are 50% higher than the SMPS total particle number concentration. This is likely due to the different cut-off sizes of the SMPS and CPC, as well as the effect of the CPC’s counting efficiency curve, rather than a malfunction of the instruments.

We encourage the user to use the data for the following types of analyses (this list should be considered a starting point rather than being comprehensive):

• Phenomenological studies of CCN across a wide variety of site types: It could be especially interesting to combine the data set described here with a similar dataset developed by Schmale et al.²⁶ and extend the analyses reported on in Schmale et al.²⁷. The two datasets together provide significantly more global coverage.

• Development of a machine learning (ML) algorithm for NPF classification using the data sets with 5 min temporal resolution: Most NPF classification is done visually and can be quite time consuming (e.g.¹⁰⁰), so an ML approach could be useful. This data set, which covers a variety of aerosol and environment types would provide a useful training and evaluation data set.

• Quantification of the impact of NPF on CCN and identification of the method or methods that work best to describe the impact of NPF on CCN: There are various methods for doing so^{29,101,102,103}, but, to our knowledge, no previous comparison of these methods exists.

• Improvement of processes relating to PNSD, NPF, and CCN in models: This could be done for global climate models such as was done for CCN⁷³, PNSD¹⁰⁴ and NPF^17,105 or for more regionally focussed models such as WRF-Chem.