Multi-year Cluster Analyses of CALIPSO Aerosol Layer Data

The Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO) mission, a collaboration between NASA and Centre National d'Études Spatiales (CNES), was launched in April 2006 to provide vertically resolved measurements of cloud and aerosol distributions. The primary instrument on the CALIPSO satellite is the Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP), a near-nadir viewing two-wavelength polarization-sensitive instrument. The unique nature of CALIOP measurements make it quite challenging to validate backscatter profiles, aerosol type, and cloud phase, all of which are used to retrieve extinction and optical depth. We exploit the large data set generated by CALIPSO between 2006 - 2013 to find groups in the data from which we can infer aerosol type. We use classical clustering analyses (k-means) to group the data based on five Level 2 data parameters: integrated attenuated backscatter (532 nm), volume depolarization ratio, integrated attenuated color ratio, layer base altitude, and layer top altitude. The data are screened using standard CALIPSO quality assurance flags, cloud aerosol discrimination (CAD) scores, overlying features and layer properties. Furthermore, the data were standardized to a 'z score' in order to remove weighting bias due to arbitrary units. The k-means algorithm forms clusters by minimizing the Euclidian distance between each data point and its respective cluster centroid independently for each parameter. Silhouette coefficients, a dimensionless measure of internal cluster interrelation or consistency and external cluster separation or uniqueness were used to determine the validity of clusters. These investigations yielded optimally distinct clusters when six groups were formed similar to the result in Omar et al, 2005 which form the basis for the current CALIPSO aerosol subtypes. While the number of clusters in both studies coincide, the clusters do not match with all aerosol subtypes. Analysis of the data points within each cluster along with geographical map overlays shows that the clusters formed contained combinations or mixtures of the six types in CALIPSO Level 2 data. These results suggest a larger number of aerosol types, possibly involving combinations of the six types in the current CALIPSO subtyping algorithms, or a smaller number of base species, e.g., sulfate, black carbon, seasalt and dust.