The association between environmental chemistry and chemometrics has deep and consolidated roots. Environment indeed, is characterized by an extreme degree of chemodiversity, with highly expressed time-variability (especially in the fluid phases), huge inhomogeneity over a massive size as compared to the laboratory scale, always claiming large datasets. Therefore, chemometric modeling is a necessary step for both understanding environmental complexity and diagnostic purposes.

The present work proposes the use Self-Organizing Maps (SOM)[1] for diagnostic purposes, i.e. to detect the influence of Saharan dust (SD) events in March 2022 in Munich, Germany, an unusual occurrence for this type of transport at these latitudes. Munich, indeed, besides being a heavily man-impacted city in southern Germany, is located beyond the Alps and is therefore rarely reached by Northern African air masses loaded in Saharan mineral dust, differently from the Mediterranean basin.

These events, however, are increasing in frequency and intensity, sometimes reaching latitudes as high as the UK, as a possible result of climate change. During these events, huge alterations of PM10 concentrations, often exceeding the EU air quality standards as well as of composition are observed leading to increased environmental and health hazard[2].

In this study, particulate matter (PM) was collected daily on quartz fiber filters from March to May 2022, and its metal composition was evaluated by Inductively Coupled Plasma Mass-Spectrometry (ICP-MS). After basic data pre-processing, data was then subjected to SOM with the aim of evaluating the enrichment of metal concentrations due to the presence of SD over Munich. Although Positive Matrix Factorization would have been more appropriate to achieve source apportioning of airborne particulate matter, it requires a huge amount of data to compute reliable models. Though many other multivariate models are usually applied in the case of limited matrices of data, SOM has been proved as a valid and reliable alternative to such methods, due to its simpler and faster computational procedure that can be carried out with any number of samples.

SOM results, also assisted by meteorological data and physical transport-based models, as backtrajectory analysis, successfully demonstrated not only the differences in trace element composition of PM in Munich due to the advent of SD, but even some differences in SD composition according to different source locations as a result of evolving atmospheric dynamics.


Licen S, Franzon M, Rodani T, Barbieri P (2021) SOMEnv: An R package for mining environmental monitoring datasets by Self-Organizing Map and k-means algorithms with a graphical user interface, Microchemical Journal, 165, 106181

Tositti L, Brattich E, Cassardo C, Morozzi P, Bracci A, Marinoni A, Di Sabatino S., Porcù F, Zappi A (2022) Development and evolution of an anomalous Asian dust event across Europe in March 2020, Atmospheric Chemistry and Physics, 22, 4047–4073