Simplified representation of atmospheric aerosol size distributions using absolute principal component analysis

摘要

Principal component analysis provides a fast and robust method to reduce thedata dimensionality of an aerosol size distribution data set. Here wedescribe a methodology for applying principal component analysis to aerosolsize distribution measurements. We illustrate the method by applying it todata obtained during five field studies. Most variations in thesub-micrometer aerosol size distribution over periods of weeks can bedescribed using 5 components. Using 6 to 8 components preserves virtuallyall the information in the original data. A key aspect of our approach isthe introduction of a new method to weight the data; this preserves theorthogonality of the components while taking the measurement uncertaintiesinto account. We also describe a new method for identifying the approximatenumber of aerosol components needed to represent the measurementquantitatively. Applying Varimax rotation to the resultant componentsdecomposes a distribution into independent monomodal distributions.Normalizing the components provides physical meaning to the componentscores. The method is relatively simple, computationally fast, andnumerically robust. The resulting data simplification provides an efficientmethod of representing complex data sets and should greatly assist in theanalysis of size distribution data.