SI#1: Scatter plots of PM2.5 concentrations vs. MODIS AOD, and NO2 concentrations vs. OMI tropospheric vertical column NO2 density.

Figure SI1: Scatter plots of PM2.5 vs. MODIS aerosol optical depth (linear relationship shown in solid line, Dec 29, 2015-Jan 5, 2016).

Figure SI2: Scatter plots of NO2 vs. OMI tropospheric vertical column NO2 density (linear relationship shown in solid line, 2015-Jan 5, 2016, satellite data missing in 01.04).

SI#2:Parametric regression model – Multiple linear regression (MLR).

(SI1-2)

where and are the intercept and slope of satellite data (MODIS AOD and OMI), is the nth selected meteorological predictor, and is the mth selected geographical predictor.

Table SI1: The coefficients of MLR model for PM2.5

Estimate / Std.error / p-value
Intercept / 394.03 / 60.09 / <0.0001
MODIS AOD / 22.17 / 3.33 / <0.0001
Wind speed / -47.03 / 2.46 / <0.0001
Relative humidity / 2.27 / 0.09 / <0.0001
Road length / 0.09 / 0.02 / <0.0001
Pressure / -0.34 / 0.05 / <0.0001

Table SI2: The coefficients of MLR model for NO2

Estimate / Std.error / p-value
Intercept / 69.25 / 17.14 / <0.0001
OMI NO2 / 0.95 / 0.02 / <0.0001
Wind speed / -12.38 / 0.67 / <0.0001
Relative humidity / 0.19 / 0.02 / <0.0001
Road length / 0.04 / 0.005 / <0.0001
Pressure / -0.03 / 0.015 / 0.09

SI#3: Parametric regression model – Linear mixed effect model (LME).

(SI3-4)

where and are the fixed intercept and slope of satellite data (MODIS AOD and OMI data), and are the random intercept and slope of satellite data; is nth selected meteorological predictor, is nth selected geographical predictor; is the error term at site i on day j.

Table SI3. The coefficients of LME model for PM2.5 (fixed effects of predictor variables).

Estimate / Std.error / p-value
Intercept / 380.82 / 45.31 / <0.0001
MODIS AOD / 37.46 / 5.84 / <0.0001
Wind speed / -30.51 / 2.03 / <0.0001
Relative humidity / 0.91 / 0.11 / <0.0001
Temperature / 1.67 / 0.42 / <0.0001
Elevation / -0.03 / 0.01 / 0.0002
Pressure / -0.28 / 0.04 / <0.0001

Table SI4. The coefficients of LME model for NO2 (fixed effects of predictor variables).

Estimate / Std.error / p-value
Intercept / 119.12 / 16.52 / <0.0001
OMI NO2 / 0.92 / 0.05 / <0.0001
Relative humidity / -0.10 / 0.03 / 0.004
Temperature / -0.66 / 0.11 / <0.0001
Wind speed / -8.11 / 0.67 / <0.0001
Road length / 0.04 / 0.005 / <0.0001
Elevation / -0.009 / 0.002 / 0.0002
Pressure / -0.07 / 0.015 / <0.0001

SI#4: Space-time covariances (empirical and theoretical) of PM2.5 and NO2concentrations (the dataset has been pre-processed before the empirical covariance was calculated, so the results correspond to the transformed covariance).

(a) (b)

Figure SI3: Space-time covariances (empirical and theoretical) of (a) PM2.5 concentrations, and (b) of NO2 concentrations.

Table SI5: Covariance model formulas and their parameters.

Space-time covariance model / / /
(m) /
(days) /
(m) /
(days)
PM2.5 / / 0.7 / 0.3 / 70,000 / 5 / 900,000 / 3
NO2 / / 0.7 / 0.3 / 50,000 / 7 / 700,000 / 3

SI#5:Comparison with Space-Time Kriging Pollutant Predictions (Figures S14 and S15 in the SI contain the BME-GBM plots from Figure 5).

Figure SI4: Daily estimatesof PM2.5 by space-time ordinary kriging (STOK), Bayesian Maximum Entropy and linear mixed effect model (BME+LME) and gradient boosting model (BME+GBM).

Figure SI5: Daily estimates of NO2 by space-time ordinary kriging (STOK); Bayesian Maximum Entropy and linear mixed effect model (BME+LME) and gradient boosting model (BME+GBM).

1