Application of fuzzy c-means clustering for analysis of chemical ionization mass spectra: insights into the gas-phase chemistry of NO₃-initiated oxidation of isoprene

Abstract. Oxidation of volatile organic compounds (VOCs) can lead to the formation of secondary organic aerosol, a significant component of atmospheric fine particles, which can affect air quality, human health, and climate change. However, current understanding of the formation mechanism of SOA is still incomplete, which is not only due to the complexity of the chemistry, but also relates to analytical challenges in SOA precursor detection and quantification. Recent instrumental advances, especially the developments of high-resolution time-of-flight chemical ionization mass spectrometry (CIMS), greatly enhanced the capability to detect low- and extremely low-volatility organic molecules (L/ELVOCs). Although detection and characterization of low volatility vapors largely improved our understanding of SOA formation, analyzing and interpreting complex mass spectrometric data remains a challenging task. This necessitates the use of dimension-reduction techniques to simplify mass spectrometric data with the purpose of extracting chemical and kinetic information of the investigated system. Here we present an approach by using fuzzy c-means clustering (FCM) to analyze CIMS data from chamber experiments aiming to investigate the gas-phase chemistry of nitrate radical initiated oxidation of isoprene.

The performance of FCM was evaluated and validated. By applying FCM various oxidation products were classified into different groups according to their chemical and kinetic properties, and the common patterns of their time series were identified, which gave insights into the chemistry of the system investigated. The chemical properties are characterized by elemental ratios and average carbon oxidation state, and the kinetic behaviors are parameterized with generation number and effective rate coefficient (describing the average reactivity of a species) by using the gamma kinetic parameterization model. In addition, the fuzziness of FCM algorithm provides a possibility to separate isomers or different chemical processes species are involved in, which could be useful for mechanism development. Overall FCM is a well applicable technique to simplify complex mass spectrometric data, and the chemical and kinetic properties derived from clustering can be utilized to understand the reaction system of interest.