Non-reference monitoring (NRM) techniques need to be evaluated against reference measurements to ensure appropriate interpretation and application. The aim of this study was to develop a field calibration model for the DustTrak DRX, a NRM light scattering laser photometer that simultaneously measures multiple size fractions of particulate matter (PM), in order to adjust data results based on co-located measurements taken with federal reference (FRM) and equivalent (FEM) methods. We deployed five DRXs at the EPA NCORE site in Columbia, South Carolina, from January 16th through February 7th 2018 to collect co-located data with FRM/FEM PM10 and PM2.5. We evaluate results with descriptive statistics, correlation coefficients, ratios of the overall means (DRX/FEM), overall mean bias, and normalized mean bias. To adjust NRM data, we develop a field calibration model using the framework of a generalized additive model (GAM). During the co-location weather exhibited temperatures ranging from -4.4 to 21.7 °C, humidity ranging from 14 to 100% and wind speed ranging from 0 to 9.3 m/s. PM conditions, although FRM were not presently available, were perceived to be low-to-moderate as the overall average FEM PM2.5 was 6.8 (standard deviation = 5.1) u.g/m3 and the 5-95% range was 2.7 to 12.6 ug/m3. Correlations between DRX measurements and hourly FEM PM2.5 measures were strong (r= 0.90 to 0.91 and overall mean ratios (DRX/FEM) revealed under reporting of the DRXs (values between 0.64 and 0.71). Mean biases ranged from 1.9 to 2.4 ug/m3 and normalized mean biases were less than 25% (17.2 - 23.6%). Our GAM performed well as the variability in FEM measured PM2.5 was well explained (R2 = 0.82), the predictive error was relatively low (root-mean-square-error = 2.46), and mean biases of predictions were zero. Our approach allows us to detect and remove NRM instrumental biases and thus improves the comparability and interpretation of data collected using non-reference techniques.
展开▼
