Multivariate receptor modeling with widely dispersed Lichens as bioindicators of air quality

Biomonitoring studies evaluating air quality via airborne element accumulation patterns in lichens typically control variability by focusing on narrow geographic regions and short time windows. Using samples of the widespread "rock-posy" lichen sampled across the Intermountain Region of the United States, we investigate whether accumulation patterns of generic pollution sources are detectable on broad geographic and temporal scales. We develop a novel Bayesian multivariate receptor modeling (BMRM) approach that sharpens detection and discrimination of candidate pollution sources through (i) regularization of source contributions to each sample and (ii) incorporating estimated lichen secondary chemistry as a factor. Through a simulation study, we demonstrate a distinct advantage in shrinking contributions when they are truly sparse, as would be expected with heterogeneous samples from dispersed collection sites. We contrast analyses employing both standard and sparse BMRMs, and positive matrix factorization (PMF). The sparse model better maintains source identity, as specified though informative prior distributions on elemental profiles. We advocate quantitative profile matching, which reveals that PMF primarily captures variations of the baseline profile for lichen secondary chemistry. Both PMF and BMRM results suggest that the most detectable signatures relate to aeolian dust deposition, while spatial patterns hint at sporadic anthropogenic influence.