Potential sources and processes affecting speciated atmospheric mercury at Kejimkujik National Park, Canada: comparison of receptor models and data treatment methods

Abstract. Source apportionment analysis was conducted with positive matrix factorization (PMF) and principal component analysis (PCA) methods using concentrations of speciated mercury (Hg), i.e., gaseous elemental mercury (GEM), gaseous oxidized mercury (GOM), and particulate-bound mercury (PBM), and other air pollutants collected at Kejimkujik National Park, Nova Scotia, Canada, in 2009 and 2010. The results were largely consistent between the 2 years for both methods. The same four source factors were identified in each year using PMF method. In both years, factor photochemistry and re-emission had the largest contributions to atmospheric Hg, while the contributions of combustion emission and industrial sulfur varied slightly between the 2 years. Four components were extracted with air pollutants only in each year using PCA method. Consistencies between the results of PMF and PCA include (1) most or all PMF factors overlapped with PCA components, (2) both methods suggest strong impact of photochemistry but little association between ambient Hg and sea salt, and (3) shifting of PMF source profiles and source contributions from one year to another was echoed in PCA. Inclusion of meteorological parameters led to identification of an additional component, Hg wet deposition in PCA, while it did not affect the identification of other components.

The PMF model performance was comparable in 2009 and 2010. Among the three Hg forms, the agreements between model-reproduced and observed annual mean concentrations were excellent for GEM, very good for PBM, and acceptable for GOM. However, on a daily basis, the agreement was very good for GEM but poor for GOM and PBM. Sensitivity tests suggest that increasing sample size by imputation is not effective in improving model performance, while reducing the fraction of concentrations below method detection limit, by either scaling GOM and PBM to higher concentrations or combining them to reactive mercury, is effective. Most of the data treatment options considered had little impact on the source identification or contribution.