Application of cluster analysis to identify sources of airborne particles

A sampling programme for determining the concentrations in particulate matter of SO₄²⁻, Fe, Pb, Zn, K, Ca, Mg, Cd, Cr and Ni was established in the city of Valladolid for 13 months. Seasonal variations and enrichment factors were determined, then source apportionment techniques were applied. A multivariate clustering Q-technique has been used to identify specific sources. Each one of the six retained clusters has been interpreted by analyzing the composition of their centroids and the common attributes of the samples belonging to each one of them. To simplify the interpretation, some external auxiliary variables were chosen: percentage of the samples corresponding to the heating season period, minimum temperature, weekends, precipitation, daily average wind speed, etc. The reliability and practical significance of the results were studied after comparing the results obtained by two different initial partitions P1 and P2.This algorithm identified a crustal source (Fe, K, Ca and Mg), residual fuel burning (SO₄²⁻, Zn, Cd and Ni) and indirect vehicle exhaust (Pb). In addition, a double source for Cr (soil and coal ashes), a pollution ‘episode’ cluster, a low pollution cluster and an average pollution cluster have been identified.