“…They found that the proxy performance was poorer during leaf‐on than leaf‐off conditions and that many predictors exhibited regression coefficients that are not significantly different from zero. Nevertheless, when the data are conditionally sampled to extract only NPF event days, the R 2 for proxy algorithms of new particle formation rates ranged from 0.13 to 0.67 (for sample sizes of 18 to 55). - Sullivan et al [] extended this earlier work by including in situ data from five sites distributed across North America. They found that, relative to more simple proxies developed in the prior studies (e.g., based on AOD, AE, SO 2 , and UV), use of additional predictors (i.e., NO 2 , as a proxy for primary emitted particles, and NH 3 , HCHO, LAI, T, and O 3 ) increases the explained temporal variance of UFP concentrations at all sites.
…”