Development and evaluation of a model to correct tapered element oscillating microbalance (TEOM) readings of PM2.5 in Chullora, Sydney.

Tapered element oscillating microbalance (TEOM) monitors offer substantial benefits to air pollution regulatory bodies in regards to their reduced need for labor and their ability to provide data in “real-time” through an automated system. However, research has demonstrated that the TEOM tends to provide inaccurate particulate matter concentrations due to the operational framework of the instrument. This paper presents the results of collocated comparisons of two PM2.5 monitors, a TEOM and a beta-attenuation monitor (BAM), conducted from September 2010 to November 2012, in the greater urban area of Chullora, Sydney, Australia. The objective of this work is to define the relationship between these two monitors, and develop a model to correct the TEOM instrument to bring in into line with what is seen as the ‘gold standard’ of PM2.5 monitors, the BAM. The results show that there is a significant positive linear relationship between TEOM and BAM samplers, at an hourly and daily scale (p-value < 0.001), with the TEOM generally reporting lower PM2.5 concentrations than the collocated BAM. Local meteorological, air pollution and gas covariates were integrated into a single linear model for PM2.5 predictions, at hourly and daily intervals. Although the model significantly improved the R2 of the agreement between instruments at hourly intervals (from 0.24 to 0.43, with a 95% confidence interval of 6.97 µg/m3 and 7.58 µg/m3), results indicate autocorrelation in the residuals of the model, suggesting there is information in the residuals that should be included in computing the forecast. Hence, producing a robust hourly model remains a challenge. A model for daily predictions improved the agreement between instruments (R2 improved from 0.75 to 0.81, with a 95% confidence interval of 7.93 µg/m3 and 8.52 µg/m3). Time series cross validation demonstrated a strong statistical performance of the daily model on independent data (FAC2 = 1.00, mean bias = 0.02 µg/m3, Pearson’s correlation coefficient = 0.92). A 7-year record of hourly TEOM measurements from 2004 to 2010 were corrected, based on the equation derived from the daily 2-year collocated measurements. Although not significantly significant, the overall trend analysis combining both the adjusted TEOM and BAM measurements demonstrated 0.62% per year increase (95% confidence interval of -0.53%, 2.03%) in PM2.5 concentrations from 2004 to 2012. Only spring produced a statistically significant increase in PM2.5 concentrations from 2004 to 2012, of 4.93% per year (3.41%, 6.1%). Hence, our daily model can robustly estimate historical PM2.5 concentrations at Chullora when PM2.5 BAM measurements were not available. Implications: The robustness of the daily model means that it can be applied to correct the historical TEOM data, to examine long term-trends at this site. This technique of correction can be adapted to other sites in Sydney, serving as a stepping stone in the long term-goal of developing an Air Quality Index for New South Wales, for periods when a TEOM was the only PM2.5 sampler at a site.