Table 4.
Predicted-PM2.5 concentrations obtained from Stage-3 RF models were regressed against Stage-1 measured/predicted-PM2.5 concentrations in a linear regression model. The CV–R2 (how well the model described the PM2.5 variability in new locations) described in three different patterns (overall, spatial, and temporal), RMSE (μg/m3), intercept (μg/m3), and slope (μg/m3).
| Stage-3 | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Overall | Spatial | Temporal | ||||||||||
| R2 | RMSE | Inter. | Slope | R2 | RMSE | Inter. | Slope | R2 | RMSE | Inter. | Slope | |
| 2008 | 0.704 | 4.547 | –1.251 | 1.064 | 0.486 | 2.698 | –0.749 | 1.026 | 0.760 | 3.677 | 0.000 | 1.074 |
| 2009 | 0.742 | 4.247 | –1.104 | 1.042 | 0.680 | 2.255 | –0.203 | 0.982 | 0.762 | 3.593 | 0.000 | 1.055 |
| 2010 | 0.709 | 4.330 | –1.424 | 1.075 | 0.627 | 2.342 | 0.137 | 0.972 | 0.738 | 3.628 | 0.000 | 1.102 |
| 2011 | 0.821 | 4.421 | –0.898 | 1.029 | 0.733 | 2.280 | –0.509 | 1.003 | 0.843 | 3.756 | 0.000 | 1.035 |
| 2012 | 0.786 | 4.354 | –0.749 | 1.027 | 0.661 | 2.527 | 0.073 | 0.966 | 0.823 | 3.552 | 0.000 | 1.043 |
| 2013 | 0.764 | 4.305 | –1.093 | 1.047 | 0.637 | 2.616 | –0.565 | 1.013 | 0.791 | 3.604 | 0.000 | 1.061 |
| 2014 | 0.784 | 4.140 | –1.044 | 1.051 | 0.632 | 2.292 | –0.145 | 0.983 | 0.815 | 3.478 | 0.000 | 1.062 |
| 2015 | 0.736 | 3.792 | –1.194 | 1.072 | 0.579 | 2.139 | –0.026 | 0.969 | 0.776 | 3.127 | 0.000 | 1.095 |
| 2016 | 0.781 | 3.702 | –0.980 | 1.050 | 0.725 | 1.964 | –0.532 | 1.010 | 0.796 | 3.149 | 0.000 | 1.061 |
| 2017 | 0.816 | 3.343 | –0.933 | 1.041 | 0.746 | 1.720 | –0.406 | 0.994 | 0.834 | 2.860 | 0.000 | 1.055 |
| 2018 | 0.790 | 3.275 | –1.030 | 1.046 | 0.726 | 1.776 | –0.745 | 1.015 | 0.807 | 2.775 | 0.000 | 1.056 |
| Mean | 0.767 | 4.042 | –1.064 | 1.049 | 0.658 | 2.237 | –0.334 | 0.994 | 0.795 | 3.382 | 0.000 | 1.063 |