In this project, we adapted our previously defined 5 aerosol optical typology scheme (Hamill et al. 2016) to result in a more discriminating 8 aerosol typology scheme (Giordano 2019). Previously we presented an aerosol classification based upon AERONET level 2.0 almucantar retrieval products from the period 1993 to 2012. In the initial phases of this research, we opto-physically identified five major types of Bulk Columnar Aerosol (BCA) based solely upon intensive optical properties of spectral Single Scattering Albedo (SSA), spectral Indices of Refraction (real – RRI and imaginary – IRI), and two Angstrom Exponents (extinction – EAE and absorption – AAE). These BCA were classified as Maritime Aerosol, Dust Aerosol, Urban Industrial Aerosol, Biomass Burning Aerosol, and Mixed Aerosol. The classification of a particular observation as one of these aerosol types is determined by its five-dimensional Mahalanobis distance (MD) to the centroid of each reference cluster (itself a 5-D hyperellipsoid). To retain a greater number of AERONET sites in the study (200+), we kept the variable space to 5-D. To generate reference clusters, we only retained data points that were found to lie within 2 MD from the data centroid. Our typology is based on AERONET retrieved quantities, which do not include low optical depth values (AOD440nm < 0.4 as per AERONET criteria for almucantar scan inversion). 

The classifications obtained are made available to be used in interpreting aerosol retrievals from satellite-borne instruments and as input for regional climate models. A major result of this aerosol typology is a dataset describing the types of aerosol particles that are distinct from one another in optical properties and a geographic distribution of those aerosol types. We used the typology scheme upon the qualifying AERONET data archive and produced seasonal aerosol climatologies by aerosol type for each of the AERONET sites included in the study, regional aerosol climatology maps, and a time-integrated global aerosol climatology map based entirely upon ground-based photometric data (Giordano 2022). An internally hyperlinked compendium of the individual AERONET site aerosol climatologies was produced to contain the results of the first phase of this work [available at https://ars.els-cdn.com/content/image/1-s2.0-S1352231016304265-mmc1.pdf]. Each of these original five aerosol types (Hamill et al. 2016, Giordano 2019) was further discriminated into specific sub-types by this same scheme to achieve an 8-aerosol typology (Giordano 2019 Chapter 2). For example, optical discrimination into specific sub-types of Biomass Burning aerosol may provide insight into sources exhibiting spectrally distinct smoke properties. Here we segmented the Biomass Burning Aerosol type into the sub-types of Flaming (BMF) and Smoldering (BMS) using the centroid separation method and the MD criteria for in-class inclusion was adjusted to 1.5 MD. Similarly, we found great confidence in discriminating the MIXED aerosol type into two distinct regimes which we simply labeled as MIXEDtype1 (MIXED1) and MIXEDtype2 (MIXED2). These can be visually verified by examining any one of many possible renditions of 3-D optical spaces noting their 5-D centroids are separated by a distance of 3.47-3.85 MD [Giordano 2019 Chapter 2]. Likewise, the Urban Industrial Aerosol class was further discriminated into European Urban Industrial (EURO UI) and North American (NA UI), whose 5-D centroids are separated by a distance of 2.60–3.08 MD. We then used the previously employed mathematical strategies to sort the global AERONET data retrievals into the aerosol types classified against their reference standards. We believe the strategies regarding aerosol differentiation using polarization data (Hamill, Piedra and Giordano 2020)  are  an additional method useful for analysis of the newer AERONET version 3 data retrievals, and data collected from the deployment of newer CIMEL sun-photometers (with enhanced polarization measurement capabilities) to the network. The resulting AERONET-based 8-aerosol optical typology, in a 5-D basis is useful for applications in aerosol optics, including direct forward modeling of radiative transfer to determine the effects of aerosol absorption and/or scattering on vertical heating profiles and ground received irradiance quantities, for input into more complicated remote sensing algorithms, used as calibration/validation values for in-situ and laboratory experimental studies, and evaluating radiative forcing calculations in atmospheric models.

[Work related to an 8-aerosol typology in 6-D, 8-D, 9-D and 10-D optical property bases, and their files, are to be published subsequently as a different database project in 2023.]

Methodology of Data collection:

We started from a complete fresh download of the entire AERONET set of all sites for the V2 Inversion product [now in V3 available at [https://aeronet.gsfc.nasa.gov/new_web/download_all_v3_inversions.html], capturing the Level 2.0 Almucantar product that includes all points for each site -- not averages. The original package was obtained as a zip archive named in this fashion: "INV_PFN_Level2_All_Points_V3.tar.gz".  The package was unzipped to make available all records for each AERONET site.

Data Cleaning: 

The complete set of files is operated on by a simple C-Shell script to remove all individual retrievals from each site that contained records of either "NAN" or "- 99999999" for any field other than the "aeronet site" name as a text field.  Subsequently, we imposed our minimum number of complete records criteria such that any AERONET station that did not report a minimum of 200 complete retrievals as removed. These generally refer to as qualifying sites. NOTE: This same strategy could be performed manually in an editor on each individual file to arrive at the same end result. An example of a C-shell script that can be modified for BASH or any other environment is included as "cleanup.csh".

Data Reduction:  Finding the minimum number of AERONET inversion retrieval variables that span the greatest variance in the global data set

This optimization problem was done using methods of factor analysis, to result in the final working set of files containing complete records of on which to develop our robust global aerosol typology. The details of the intricacies and history of this part of the project are detailed in Giordano 2019.

Data Processing: Determining the number of unique aerosol types globally identifiable through bulk columnar aerosol (BCA) Sun photometry

This problem of analyzing the global data to determine the individually identifiable unique aerosol "groups" was done by invoking various mathematical unique strategies generally thought of as "cluster analysis."  The main goal of this approach is to achieve separability between the individual "groups" while  preserving a maximal "homogeneity" within any individual group.  The details of the intricacies and history of this part of the project are detailed Giordano 2019 and Hamill et al. 2016.

Regarding Uncertainty measures in the reported values:

Regarding the calculated uncertainty values that accompany these data, the author offers the summary of both the centroid values for each of the 5 optical properties, and the tabulated centroid values for each of the 8 reference aerosol types to be used as standards to classify measurements of the global aerosol.  These are below as Figure1 and Table1 respectively.

Table 1:  uncertainty values for the 5 optical properties in the typological basis

The original author/creator of this database [M.E. Giordano] holds and retains all rights to its use [as it was a copy-protected student work in a Doctoral dissertation], and shares these rights via Dryad's Creative Commons License with the new readers of this work.  The author only respectfully requests that proper citation of the work be given, and derivative works attribute the original author. The only code file attached to this work [Cleanup.csh] is an example and may be used as such.

Only the file named "8AerosolOpticalPropertiesTypology5Dbasis_CENTROIDS.xlsx"  requires specific software such as MS Excel^TM, LibreOffice^TM, or OpenOffice^TM to be opened and examined.   The remaining files are in comma-separated value CSV (UTF8 comma delimited), space-separated values (.txt), and tab-separated values (.dat).

Some suggested uses:

• This data is likely to be used by those importing the specific aerosol reference standard centroid value to be compared with aerosol measurements at a specific location during a specific season, to classify them into robust global aerosol category.
• The aerosol reference standard centroids can be used for studies involving feeding values of specific sets of aerosol optical properties into radiative transfer codes to arrive at values for various radiative forcing in seasonal and geo-referenced locations.
• The centroid reference values can be used to kernel sets of synthesized aerosol optical properties by distributing centroid values statistically in a function such as a lognormal distribution, or a Gaussian about that centroid value.

If one is using a Python environment, the following might be useful to manipulate the data in this database:

• Numpy
• SciPy
• xlrd
• openpyxl
• PANDAS
• XARRAY