Eight years of sub-micrometre organic aerosol composition data from the boreal forest characterized using a machine-learning approach

Abstract. The Station for Measuring Ecosystem–Atmosphere Relations (SMEAR) II, located within the boreal forest of Finland, is a unique station in world due to wide range long-term measurements tracking Earth–atmosphere interface. In this study, we characterize composition organic aerosol (OA) at SMEAR II by quantifying its driving constituents. We utilize multi-year data set OA mass spectra measured situ with an Aerosol Chemical Speciation Monitor (ACSM) station. To our knowledge, spectral time series longest kind published date. Similarly other previously reported efforts source apportionment from multi-seasonal or multi-annual sets, approached characterization challenge through positive matrix factorization (PMF) using rolling window approach. However, existing methods extracting minor components were found be insufficient rather remote site. overcome issue, tested new statistical analysis framework. This included unsupervised feature extraction and classification stages explore large number unconstrained PMF runs conducted on spectra. Anchored these results, finally constructed relaxed chemical balance (CMB) run that resolved different observations. presented combination tools provided data-driven methodology, which case achieved robust solutions minimal subjectivity. Following extensive analyses, able divide 2012–2019 (mass concentration interquartile (IQR): 0.7, 1.3, 2.6 µg m?3) into three sub-categories – low-volatility oxygenated (LV-OOA), semi-volatile (SV-OOA), primary (POA) proving methodology was provide results consistent literature. LV-OOA most dominant type (organic fraction IQR: 49 %, 62 73 %). seasonal cycle bimodal, peaks both summer February. associated wintertime anthropogenic sources assumed biogenic influence formation summer. Through brief trajectory analysis, estimated summertime natural tens ng m?3 h?1 over forest. SV-OOA second highest contributor 19 31 43 Due SV-OOA's clear peak summer, estimate processes as main drivers formation. Unlike LV-OOA, concentrations detected stable nocturnal surface layers. Two nearby sawmills also played significant role production exemplified previous studies II. POA, taken mix two types previously, hydrocarbon-like (HOA) biomass burning (BBOA), made up (IQR: 2 6 13 Notably, quantification POA ACSM not possible following methodologies. Both peaked winter. Its appearance linked strong southerly winds. Similar wind direction speed dependence observed among types. high speeds probably enabled transport faraway relatively fresh state. event slower speeds, likely evaporated and/or aged oxidized before detection. significantly lower than spectrometer (AMS) 4 years prior measurements. While co-located black carbon supported hypothesis higher loadings year 2012, it short-term pollution plumes averaged out slow resolution combined further 3 h averaging needed ensure good signal-to-noise ratios (SNRs). Despite length set, did focus trends (nor components) sensitivity meteorological anomalies, occurrence normally distributed 8-year measurement period. realistic cycles meteorology dependences independent subtypes complemented reasonably low degree unexplained variability, believe approach performs well. Therefore, hope encourage researchers possessing several-year-long similar tackle via semi- machine-learning approaches. way method could optimized usability explored evaluated environments.