Table 1.

Predicted-PM_2.5 concentrations obtained from Stage-1 RF models were regressed against measured-PM_2.5 concentrations in a linear regression model. The performance was evaluated using two CV methods (OOB and 10-Fold) together with RMSE (a measure of the model error, μg/m3), intercept (μg/m3), and slope (μg/m3).

Stage-1
		OOB-CV				10-Fold CV
	R2	RMSE	Inter.	Slope	R2	RMSE	Inter.	Slope
2008	0.918	1.196	–0.417	1.033	0.707	4.954	0.803	0.886
2009	0.921	1.110	–0.390	1.030	0.791	3.996	0.410	0.937
2010	0.919	1.122	–0.401	1.029	0.843	3.496	0.043	0.983
2011	0.949	1.124	–0.266	1.019	0.902	3.439	–0.087	0.997
2012	0.942	1.058	–0.274	1.021	0.889	3.218	–0.035	0.986
2013	0.929	1.087	–0.368	1.028	0.847	3.584	0.218	0.972
2014	0.944	0.963	–0.267	1.022	0.891	3.003	–0.007	0.995
2015	0.933	0.865	–0.265	1.026	0.871	2.662	–0.003	0.983
2016	0.935	0.896	–0.251	1.025	0.885	2.654	–0.050	0.996
2017	0.939	0.828	–0.196	1.022	0.895	2.430	0.010	0.985
2018	0.928	0.791	–0.248	1.028	0.886	2.235	–0.019	0.993
Mean	0.932	1.003	–0.304	1.026	0.855	3.243	0.117	0.974