Classifying Asian dust aerosols and their columnar optical properties using fuzzy clustering

Abstract

The Gustafson-Kessel fuzzy clustering algorithm is used to classify Asian dust aerosols from Aerosol Robotic Network inversions, and properties within different membership degree intervals are analyzed. Five typical Asian dust sites are selected for source data. Based on dust aerosol physics, seven key parameters are selected as clustering inputs. The two clusters, dust and nondust, are then divided by a membership degree threshold of 0.5 on a scale of 0 to 1. The membership degree denotes the level of confidence that a record belongs to a cluster. The degree of mixing is denoted by diminishing membership values; e.g., the classified dust aerosol associated with higher membership degree denotes the purer dust. The two-dimensional principle component projections, together with the properties of the clusters, show the capability of fuzzy clustering to classify a pure aerosol type based on a membership degree. Then the detailed dust properties within different membership degree intervals are analyzed. The results show that both the optical properties and volume size distributions vary significantly with membership degrees and range diversities can be reduced within each interval. The proposed method can also be used to determine weights in an aerosol fitting model to improve the parameterization. The method is most suited to air masses with multicomponent aerosol mixtures and can easily be extended to other types of aerosols.