Investigating the link between $$\hbox {PM}_{2.5}$$ PM 2.5 and atmospheric profile variables via penalized functional quantile regression

Identifying meteorological drivers of PM2.5 levels via a Bayesian spatial quantile regression

“…In implementing our approach, we used the cubic spline representation and removed the spatiotemporal random effect from the model. Proceeding in this fashion is roughly equivalent to implementing the methodology outlined in Russell et al (2017). Figure 5 and Table 1 provide the same summary of the posterior estimates obtained from this study as those discussed above.…”

“…A key attribute of our approach when compared to the methodology presented in Russell et al (2017) is that our method specifically accounts for the spatiotemporal dependence through (⋅, ⋅), while the existing technique does not. To examine the potential impacts of failing to acknowledge these dependencies, an additional simulation was conducted.…”

“…The air quality monitoring stations used in the analysis A first step toward predictive capability involves developing an in-depth understanding of the ways in which meteorological factors are associated with abnormally high PM 2.5 levels, and how these associations change regionally (Porter et al, 2015;Russell et al, 2017;Tai et al, 2010). To this end, we seek to identify meteorological drivers of PM 2.5 throughout the eastern United States.…”

“…In considering these data, we note that a variety of models have been proposed for the analysis of PM 2.5 data; for example, see Smith et al (2003), Porter et al (2015), Wu et al (2016), Russell and Dyer (2017), and Qi et al (2019). Although, none of these techniques are completely suitable for our purposes.…”

“…That is, these existing methods primarily focus on modeling the conditional mean, fail to acknowledge the spatially varying effects of meteorological drivers, and/or ignore the spatial (or spatiotemporal) dependence in the data. The most relevant contribution, to our knowledge, was proposed by Russell et al (2017) which saw the development of a quantile regression model with spatially varying effects. This technique can be used to estimate the entire conditional (given the meteorological drivers) distribution of PM 2.5 levels while accounting for spatially varying meteorological effects.…”