PNNL

Abstract

Understanding how different processes in the atmosphere affect the formation of organic aerosols has been a persistent challenge. Inferring causal relations between organic aerosols and different chemistry and meteorological features from time series measurements is complicated since correlation does not imply causality. Here, we apply Koopman operator framework coupled with a state of the art information transfer measure to infer causal relations from time series predictions of a key component of organic aerosols, which are predicted by a high resolution three dimensional regional chemical transport model. Isoprene epoxydiol SOA (IEPOX-SOA) involves one of the most complex SOA formation pathways involving chemistry in gas- and particle-phases and is formed by the interactions between natural biogenic isoprene emissions, and anthropogenic emissions affecting sulfate, acidity, particle water. Since the regional model captures the known relations of IEPOX-SOA formation with different chemistry and meteorological features, the modeled time series implicitly includes the causal relations. We show that our causal model successfully infers the known major causal relations between IEPOX-SOA components and input features for e.g. particle sulfate is inferred to have the greatest information transfer to IEPOX organosulfates. We provide the first proof of concept that application of our causal model identifies known major causal relations from time series data. Our work has tremendous implications in the analyses of field measurements and atmospheric chemistry models, and it can identify unknown processes and features affecting aerosols and atmospheric chemistry in the Earth's atmosphere.