Characterization of variability of air particulate matter size profiles recorded by optical particle counters near a complex emissive source by use of Self-Organizing Map algorithm

In the present study we propose the application of a procedure of data analysis based on the Self-Organizing Map algorithm and k-means clustering in series (1st level and 2nd level abstraction respectively) as a strategy to identify recurrent ambient air particulate matter (PM) size profiles starting from the elaboration of high frequency data recorded by an Optical Particle Counter (OPC). We tested the strategy on data deriving from a three months survey at a residential site in proximity to an integral cycle steel plant in Trieste (NE Italy). We were able to identify four clusters representing recurrent PM class profiles whose meaning has been interpreted and confirmed by correlation to “external data”, i.e. data not used to build the model, registered by other devices (meteorological and pollutant monitoring stations). The four clusters were found to be related to two different plant type of emissions (sources) and to two different site background profiles, respectively. The powerful visualization features of SOM map allowed to describe and characterize the variability of size distribution of PM in a concise form. The clustered SOM being built for one measuring station, proved to be helpful for the analysis of OPC data collected at another location close to the industrial plant. Moreover, occasional episodes of Saharan dusts could be identified as outliers with respect to local particulate and discussed in terms of size distribution. Eventually, by means of an animated graph, we propose a method to visualize the PM experimental data evolution during the day using the PM cluster profiles as a legend. © 2019