ALBERT

Description: New particle formation (NPF) in the atmosphere is globally an important source of climate relevant aerosol particles. Occurrence of NPF events is typically analyzed by researchers manually from particle size distribution data day by day, which is time consuming and the classification of event types may be inconsistent. To get more reliable and consistent results, the NPF event analysis should be automatized. We have developed an automatic analysis method based on deep learning, a subarea of machine learning, for NPF event identification. To our knowledge, this is the first time that a deep learning method, i.e., transfer learning of a convolutional neural network (CNN), has successfully been used to automatically classify NPF events into different classes directly from particle size distribution images, similarly to how the researchers carry out the manual classification. The developed method is based on image analysis of particle size distributions using a pretrained deep CNN, named AlexNet, which was transfer learned to recognize NPF event classes (six different types). In transfer learning, a partial set of particle size distribution images was used in the training stage of the CNN and the rest of the images for testing the success of the training. The method was utilized for a 15-year-long dataset measured at San Pietro Capofiume (SPC) in Italy. We studied the performance of the training with different training and testing of image number ratios as well as with different regions of interest in the images. The results show that clear event (i.e., classes 1 and 2) and nonevent days can be identified with an accuracy of ca. 80 %, when the CNN classification is compared with that of an expert, which is a good first result for automatic NPF event analysis. In the event classification, the choice between different event classes is not an easy task even for trained researchers, and thus overlapping or confusion between different classes occurs. Hence, we cross-validated the learning results of CNN with the expert-made classification. The results show that the overlapping occurs, typically between the adjacent or similar type of classes, e.g., a manually classified Class 1 is categorized mainly into classes 1 and 2 by CNN, indicating that the manual and CNN classifications are very consistent for most of the days. The classification would be more consistent, by both human and CNN, if only two different classes are used for event days instead of three classes. Thus, we recommend that in the future analysis, event days should be categorized into classes of “quantifiable” (i.e., clear events, classes 1 and 2) and “nonquantifiable” (i.e., weak events, Class  3). This would better describe the difference of those classes: both formation and growth rates can be determined for quantifiable days but not both for nonquantifiable days. Furthermore, we investigated more deeply the days that are classified as clear events by experts and recognized as nonevents by the CNN and vice versa. Clear misclassifications seem to occur more commonly in manual analysis than in the CNN categorization, which is mostly due to the inconsistency in the human-made classification or errors in the booking of the event class. In general, the automatic CNN classifier has a better reliability and repeatability in NPF event classification than human-made classification and, thus, the transfer-learned pretrained CNNs are powerful tools to analyze long-term datasets. The developed NPF event classifier can be easily utilized to analyze any long-term datasets more accurately and consistently, which helps us to understand in detail aerosol–climate interactions and the long-term effects of climate change on NPF in the atmosphere. We encourage researchers to use the model in other sites. However, we suggest that the CNN should be transfer learned again for new site data with a minimum of ca. 150 figures per class to obtain good enough classification results, especially if the size distribution evolution differs from training data. In the future, we will utilize the method for data from other sites, develop it to analyze more parameters and evaluate how successfully CNN could be trained with synthetic NPF event data.

Description: Aerosol–cloud interactions (ACI) constitute the single largest uncertainty in anthropogenic radiative forcing. To reduce the uncertainties and gain more confidence in the simulation of ACI, models need to be evaluated against observations, in particular against measurements of cloud condensation nuclei (CCN). Here we present a data set – ready to be used for model validation – of long-term observations of CCN number concentrations, particle number size distributions and chemical composition from 12 sites on 3 continents. Studied environments include coastal background, rural background, alpine sites, remote forests and an urban surrounding. Expectedly, CCN characteristics are highly variable across site categories. However, they also vary within them, most strongly in the coastal background group, where CCN number concentrations can vary by up to a factor of 30 within one season. In terms of particle activation behaviour, most continental stations exhibit very similar activation ratios (relative to particles 〉 20 nm) across the range of 0.1 to 1.0 % supersaturation. At the coastal sites the transition from particles being CCN inactive to becoming CCN active occurs over a wider range of the supersaturation spectrum. Several stations show strong seasonal cycles of CCN number concentrations and particle number size distributions, e.g. at Barrow (Arctic haze in spring), at the alpine stations (stronger influence of polluted boundary layer air masses in summer), the rain forest (wet and dry season) or Finokalia (wildfire influence in autumn). The rural background and urban sites exhibit relatively little variability throughout the year, while short-term variability can be high especially at the urban site. The average hygroscopicity parameter, κ, calculated from the chemical composition of submicron particles was highest at the coastal site of Mace Head (0.6) and lowest at the rain forest station ATTO (0.2–0.3). We performed closure studies based on κ–Köhler theory to predict CCN number concentrations. The ratio of predicted to measured CCN concentrations is between 0.87 and 1.4 for five different types of κ. The temporal variability is also well captured, with Pearson correlation coefficients exceeding 0.87. Information on CCN number concentrations at many locations is important to better characterise ACI and their radiative forcing. But long-term comprehensive aerosol particle characterisations are labour intensive and costly. Hence, we recommend operating “migrating-CCNCs” to conduct collocated CCN number concentration and particle number size distribution measurements at individual locations throughout one year at least to derive a seasonally resolved hygroscopicity parameter. This way, CCN number concentrations can only be calculated based on continued particle number size distribution information and greater spatial coverage of long-term measurements can be achieved.

Description: To study the influence of regional biomass burning emissions and secondary processes, ambient samples of fog and aerosol were collected in the Po Valley (Italy) during the 2013 Supersito field campaign. After the extent of fresh vs. aged biomass burning influence was estimated from proton nuclear magnetic resonance (1H NMR) and high-resolution time-of-flight aerosol mass spectrometry (HR-ToF-AMS), two samples of fog water and two samples of PM1 aerosol were selected for ultrahigh-resolution Fourier transform ion cyclotron resonance mass spectrometry (FT-ICR MS) analysis. Molecular compositions indicated that the water-soluble organic matter was largely non-polymeric without clearly repeating units. The selected samples had an atypically large frequency of molecular formulas containing nitrogen and sulfur (not evident in the NMR composition) attributed to multifunctional organonitrates and organosulfates. Higher numbers of organonitrates were observed in aerosol, and higher numbers of organosulfates were observed in fog water. Consistent with the observation of an enhanced aromatic proton signature in the 1H-NMR analysis, the average molecular formula double-bond equivalents and carbon numbers were higher in the fresh biomass-burning-influenced samples. The average O : C and H : C values from FT-ICR MS were higher in the samples with an aged influence (O : C  =  0.50–0.58, and H : C  =  1.31–1.37) compared to those with fresh influence (O : C  =  0.43–0.48, and H : C  =  1.13–1.30). The aged fog had a large set of unique highly oxygenated CHO fragments in the HR-ToF-AMS, which reflects an enrichment of carboxylic acids and other compounds carrying acyl groups, highlighted by the NMR analysis. Fog compositions were more oxidized and SOA (secondary organic aerosol)-like than aerosols as indicated by their NMR measured acyl-to-alkoxyl ratios and the observed molecular formula similarity between the aged aerosol and fresh fog, implying that fog nuclei must be somewhat aged. Overall, functionalization with nitrate and sulfate moieties, in addition to aqueous oxidation, triggers an increase in the molecular complexity in this environment, which is apparent in the FT-ICR MS results. This study demonstrates the significance of the aqueous phase in transforming the molecular chemistry of atmospheric organic matter and contributing to secondary organic aerosol.

Description: While numerous studies have demonstrated the association between outdoor exposure to atmospheric particulate matter (PM) and adverse health effects, the actual chemical species responsible for PM toxicological properties remain a subject of investigation. We provide here reactive oxygen species (ROS) activity data for PM samples collected at a rural site in the Po Valley, Italy, during the fog season (i.e., November–March). We show that the intrinsic ROS activity of Po Valley PM, which is mainly composed of biomass burning and secondary aerosols, is comparable to that of traffic-related particles in urban areas. The airborne concentration of PM components responsible for the ROS activity decreases in fog conditions, when water-soluble species are scavenged within the droplets. Due to this partitioning effect of fog, the measured ROS activity of fog water was contributed mainly by water-soluble organic carbon (WSOC) and secondary inorganic ions rather than by transition metals. We found that the intrinsic ROS activity of fog droplets is even greater (〉 2.5 times) than that of the PM on which droplets are formed, indicating that redox-active compounds are not only scavenged from the particulate phase, but are also produced within the droplets. Therefore, even if fog formation exerts a scavenging effect on PM mass and redox-active compounds, the aqueous-phase formation of reactive secondary organic compounds can eventually enhance ROS activity of PM when fog evaporates. These findings, based on a case study during a field campaign in November 2015, indicate that a significant portion of airborne toxicity in the Po Valley is largely produced by environmental conditions (fog formation and fog processing) and not simply by the emission and transport of pollutants.

Description: The study of secondary organic aerosol (SOA) in laboratory settings has greatly increased our knowledge of the diverse chemical processes and environmental conditions responsible for the formation of particulate matter starting from biogenic and anthropogenic volatile compounds. However, characteristics of the different experimental setups and the way they impact the composition and the timescale of formation of SOA are still subject to debate. In this study, SOA samples were generated using a potential aerosol mass (PAM) oxidation flow reactor using α-pinene, naphthalene and isoprene as precursors. The PAM reactor facilitated exploration of SOA composition over atmospherically relevant photochemical ageing timescales that are unattainable in environmental chambers. The SOA samples were analyzed using two state-of-the-art analytical techniques for SOA characterization – proton nuclear magnetic resonance (1H-NMR) spectroscopy and HPLC determination of humic-like substances (HULIS). Results were compared with previous Aerodyne aerosol mass spectrometer (AMS) measurements. The combined 1H-NMR, HPLC, and AMS datasets show that the composition of the studied SOA systems tend to converge to highly oxidized organic compounds upon prolonged OH exposures. Further, our 1H-NMR findings show that only α-pinene SOA acquires spectroscopic features comparable to those of ambient OA when exposed to at least 1  ×  1012 molec OH cm−3  ×  s OH exposure, or multiple days of equivalent atmospheric OH oxidation. Over multiple days of equivalent OH exposure, the formation of HULIS is observed in both α-pinene SOA and in naphthalene SOA (maximum yields: 16 and 30 %, respectively, of total analyzed water-soluble organic carbon, WSOC), providing evidence of the formation of humic-like polycarboxylic acids in unseeded SOA.

Description: Studying the vertical distribution of aerosol particle physical and chemical properties in the troposphere is essential to understand the relative importance of local emission processes vs. long-range transport for column-integrated aerosol properties (e.g. the aerosol optical depth, AOD, affecting regional climate) as well as for the aerosol burden and its impacts on air quality at the ground. The main objective of this paper is to investigate the transport of desert dust in the middle troposphere and its intrusion into the planetary boundary layer (PBL) over the Po Valley (Italy), a region considered one of the greatest European pollution hotspots for the frequency that particulate matter (PM) limit values are exceeded. Events of mineral aerosol uplift from local (soil) sources and phenomena of hygroscopic growth at the ground are also investigated, possibly affecting the PM concentration in the region as well. During the PEGASOS 2012 field campaign, an integrated observing–modelling system was set up based on near-surface measurements (particle concentration and chemistry), vertical profiling (backscatter coefficient profiles from lidar and radiosoundings) and Lagrangian air mass transport simulations by FLEXPART model. Measurements were taken at the San Pietro Capofiume supersite (44°39′ N, 11°37′ E; 11 m a.s.l.), located in a rural area relatively close to some major urban and industrial emissive areas in the Po Valley. Mt. Cimone (44°12′ N, 10°42′ E; 2165 m a.s.l.) WMO/GAW station observations are also included in the study to characterize regional-scale variability. Results show that, in the Po Valley, aerosol is detected mainly below 2000 m a.s.l. with a prevalent occurrence of non-depolarizing particles ( 〉 50 % throughout the campaign) and a vertical distribution modulated by the PBL daily evolution. Two intense events of mineral dust transport from northern Africa (19–21 and 29 June to 2 July) are observed, with layers advected mainly above 2000 m, but subsequently sinking and mixing in the PBL. As a consequence, a non-negligible occurrence of mineral dust is observed close to the ground ( ∼ 7 % of occurrence during a 1-month campaign). The observations unambiguously show Saharan dust layers intruding the Po Valley mixing layer and directly affecting the aerosol concentrations near the surface. Finally, lidar observations also indicate strong variability in aerosol on shorter timescales (hourly). Firstly, these highlight events of hygroscopic growth of anthropogenic aerosol, visible as shallow layers of low depolarization near the ground. Such events are identified during early morning hours at high relative humidity (RH) conditions (RH  〉 80 %). The process is observed concurrently with high PM1 nitrate concentration (up to 15 µg cm−3) and hence mainly explicable by deliquescence of fine anthropogenic particles, and during mineral dust intrusion episodes, when water condensation on dust particles could instead represent the dominant contribution. Secondly, lidar images show frequent events (mean daily occurrence of  ∼  22 % during the whole campaign) of rapid uplift of mineral depolarizing particles in afternoon–evening hours up to 2000 m a.s.l. height. The origin of such particles cannot be directly related to long-range transport events, being instead likely linked to processes of soil particle resuspension from agricultural lands.

Description: Whilst general policy objectives to reduce airborne particulate matter (PM) health effects are to reduce exposure to PM as a whole, emerging evidence suggests that more detailed metrics associating impacts with different aerosol components might be needed. Since it is impossible to conduct toxicological screening on all possible molecular species expected to occur in aerosol, in this study we perform a proof-of-concept evaluation on the information retrieved from in silico toxicological predictions, in which a subset (N = 104) of secondary organic aerosol (SOA) compounds were screened for their mutagenicity potential. An extensive database search showed that experimental data are available for 13 % of the compounds, while reliable predictions were obtained for 82 %. A multivariate statistical analysis of the compounds based on their physico-chemical, structural, and mechanistic properties showed that 80 % of the compounds predicted as mutagenic were grouped into six clusters, three of which (five-membered lactones from monoterpene oxidation, oxygenated multifunctional compounds from substituted benzene oxidation, and hydroperoxides from several precursors) represent new candidate groups of compounds for future toxicological screenings. These results demonstrate that coupling model-generated compositions to in silico toxicological screening might enable more comprehensive exploration of the mutagenic potential of specific SOA components.