Skip to main content
. Author manuscript; available in PMC: 2021 Feb 1.
Published in final edited form as: Atmos Environ (1994). 2019 Nov 14;222:117130. doi: 10.1016/j.atmosenv.2019.117130

Table 7. Overall performance of exposure assessment methods by level of observed PM concentration.

Correlation coefficient between PM2.5 concentration predictions and observed PM2.5 concentration in μg/m3 stratified by PM2.5 concentration level. Groupings are based on: whether PM2.5 is less than 6 μg/m3, 6–12 μg/m3 and greater than or equal to 12 μg/m3 (12 μg/m3 is the EPA standard). The methods considered are: raw CMAQ output, ordinary least squares (“OLS”), inverse distance weighting (“IDW”), universal Kriging(“UK”), downscaler, random forests (“RF”), support vector regression (“SVR”) and Neural networks (“NN”). Methods use either CMAQ and/or other geographic covariates (“Covs”). The last line reports the mean and standard deviation, in parenthesis, for the observed PM2.5 concentration within each category (low, medium or high).

Method Low Med High
CMAQ 0.37 0.26 0.21
OLS (CMAQ) 0.33 0.33 0.31
OLS (Covs) 0.40 0.43 0.42
OLS (CMAQ+Covs) 0.32 0.36 0.36
IDW 0.41 0.56 0.66
UK (CMAQ) 0.50 0.62 0.65
UK (Covs) 0.49 0.61 0.64
UK (CMAQ+Covs) 0.45 0.55 0.64
Downscaler (CMAQ) 0.49 0.62 0.65
RF (CMAQ + Covs) 0.37 0.42 0.42
SVM (CMAQ + Covs) 0.37 0.55 0.46
NN (CMAQ + Covs) 0.37 0.44 0.54
PM2.5 4.14 (1.21) 8.66 (1.71) 17.48 (5.81)