Identification of New Particle Formation Events With Hidden Markov Models

Formation of new particles in the atmosphere is a phenomenon of great importance in the Earth’s climate system. To study this phenomenon, number concentration of particles of various sizes (even nanometric) must be measured over long periods of time. Traditionally the analysis of the data requires a manual visual inspection of the records following pre-established protocols. A critical step in the analysis of the measurements is to detect those moments where the new particle formation (NPF) events actually occurred. In addition, the number of formed new particles and their particle dynamics are typically investigated and quantified. Manual analysis of the measurements makes the obtained results strongly subjective, even if the established protocols are strictly followed. Therefore, obtained results, such as the frequency of occurrence of such events, or the average new particle formation rate, can be highly variable. To decrease these uncertainties, we have developed a new methodology to automatize the NPF analysis. In this work, we present a system based on Hidden Markov Models (HMM) to automatically detect in long data series the instants where a NPF event occurs. We show that the HMM can be used to detect NPF event in an objective and effective way, with low complexity either to create the automatic classification system or to use it.