The results of this comprehensive training for estimating the abundance of PM2.5 (μg/m3) are shown in the table below. The performance of the approach we have used here is substantially better than that of the previous studies. Our worst performance has a correlation coefficient of 0.98, which is better than the best performance of the previous studies, 0.83 for the non-coincident measurements of Van Donkelaar et al. (2006).
These are the global correlation coefficients, so include the west coast and south west of the United States which are typically more challenging to reproduce (Zhang et al., 2009). Figures 2 and 3 below show the results just for the west coast state of California and the southwest state of Texas that have correlation coefficients of 0.97 and 0.94 respectively.