Understanding Sources and Drivers of Size-Resolved Aerosol in the High Arctic Islands of Svalbard Using a Receptor Model Coupled with Machine Learning

Abstract

Atmospheric aerosols are important drivers of Arctic climate change through aerosol–cloud–climate interactions. However, large uncertainties remain on the sources and processes controlling particle numbers in both fine and coarse modes. Here, we applied a receptor model and an explainable machine learning technique to understand the sources and drivers of particle numbers from 10 nm to 20 μm in Svalbard. Nucleation, biogenic, secondary, anthropogenic, mineral dust, sea salt and blowing snow aerosols and their major environmental drivers were identified. Our results show that the monthly variations in particles are highly size/source dependent and regulated by meteorology. Secondary and nucleation aerosols are the largest contributors to potential cloud condensation nuclei (CCN, particle number with a diameter larger than 40 nm as a proxy) in the Arctic. Nonlinear responses to temperature were found for biogenic, local dust particles and potential CCN, highlighting the importance of melting sea ice and snow. These results indicate that the aerosol factors will respond to rapid Arctic warming differently and in a nonlinear fashion.

Supplementary material to "Enrichment of calcium in sea spray aerosol through bulk measurements and individual particle analysis during the R/V Xuelong cruise over the Ross Sea, Antarctica"

“…In multiple linear models, two of the most common measures of model fit are the residual standard error and proportion of variance explained (R 2 ), by using least squares fit. In contrast, random forest with nonlinear multiple regression has been widely applied to predict and reproduce the importance of factors, by building multiple decision trees (Lundberg et al, 2020;Zhang et al, 2021;Song et al, 2022).…”

Measuring the Emission Changes and Meteorological Dependence of Source‐Specific BC Aerosol Using Factor Analysis Coupled With Machine Learning

“…After estimating S dew for each BC source, we built another RF model to relate the meteorologically driven contribution (S meteo = S − S dew ) with the aforementioned meteorological-related variables. The contributions of meteorological variables to BC sources were estimated using the SHAP analysis (Hou et al, 2022;Song et al, 2022). The core of SHAP analysis is the Shapley value, which is an additive feature attribution method derived from coalitional game theory created by Shapley in 1953(Shapley, 1953.…”

An intercomparison of weather normalization of PM2.5 concentration using traditional statistical methods, machine learning, and chemistry transport models

“…70% of the datasets were used to build the model and the remaining 30% of datasets were used to test the model. In line with previous studies 36,38,53,73,74 , the settings below were used to train the RF model: the number of the tree (ntree) = 300; the number of variables that may split at each node (mtry) = 3; the minimum size of terminal nodes (min.node.size) = 5. In addition to the default settings for the RF model, these parameters were also tuned by random search with 5-fold crossvalidation after 100 times evaluation.…”