کد مقاله کد نشریه سال انتشار مقاله انگلیسی نسخه تمام متن
4439495 1311022 2011 9 صفحه PDF دانلود رایگان
عنوان انگلیسی مقاله ISI
Impact of source collinearity in simulated PM2.5 data on the PMF receptor model solution
موضوعات مرتبط
مهندسی و علوم پایه علوم زمین و سیارات علم هواشناسی
پیش نمایش صفحه اول مقاله
Impact of source collinearity in simulated PM2.5 data on the PMF receptor model solution
چکیده انگلیسی

Positive Matrix Factorization (PMF) is a factor analytic model used to identify particle sources and to estimate their contributions to PM2.5 concentrations observed at receptor sites. Collinearity in source contributions due to meteorological conditions introduces uncertainty in the PMF solution. We simulated datasets of speciated PM2.5 concentrations associated with three ambient particle sources: “Motor Vehicle” (MV), “Sodium Chloride” (NaCl), and “Sulfur” (S), and we varied the correlation structure between their mass contributions to simulate collinearity. We analyzed the datasets in PMF using the ME-2 multilinear engine. The Pearson correlation coefficients between the simulated and PMF-predicted source contributions and profiles are denoted by “G correlation” and “F correlation”, respectively. In sensitivity analyses, we examined how the means or variances of the source contributions affected the stability of the PMF solution with collinearity. The % errors in predicting the average source contributions were 23, 80 and 23% for MV, NaCl, and S, respectively. On average, the NaCl contribution was overestimated, while MV and S contributions were underestimated. The ability of PMF to predict the contributions and profiles of the three sources deteriorated significantly as collinearity in their contributions increased. When the mean of NaCl or variance of NaCl and MV source contributions was increased, the deterioration in G correlation with increasing collinearity became less significant, and the ability of PMF to predict the NaCl and MV loading profiles improved. When the three factor profiles were simulated to share more elements, the decrease in G and F correlations became non-significant. Our findings agree with previous simulation studies reporting that correlated sources are predicted with higher error and bias. Consequently, the power to detect significant concentration-response estimates in health effect analyses weakens.


► We simulate collinearity in the mass contributions of three ambient particle sources.
► We analyze the speciated PM2.5 data in PMF and compare model inputs to outputs.
► The PMF solution deteriorates significantly as source collinearity increases.
► Sources with higher coefficients of variation are predicted with lower error.
► Sources with higher mean contributions are predicted with lower error.

ناشر
Database: Elsevier - ScienceDirect (ساینس دایرکت)
Journal: Atmospheric Environment - Volume 45, Issue 38, December 2011, Pages 6938–6946
نویسندگان
, , ,