2021 Keynote: Reduction of large datasets and expensive air quality model calculations through statistical analysis and machine learning.

Keynote Speaker: Daven Henze, University of Colorado, Boulder
Atmospheric chemists of our generation are frequently faced with the challenges of processing, interpreting, and generating large datasets and increasingly detailed simulations.  Statistical methods have long been used to adjust estimates from air quality models or to estimate pollutant time-series independently from a physics-based model. More recently, machine-learning based approaches are being used to supplement or augment real-time modeling capabilities. This presentation will touch on the ways in which current research has incorporated machine learning methods to reduce model bias, increase model efficiency, and augment the chemical complexity attainable within a reasonable simulation timeframe, as well as approaches used for statistical reduction of complex datasets.  In addition to an overview of the current state of research across the community in these area, I will touch on results from projects that specifically aim to (i) merge PM2.5 datasets for generating improved estimates of pollution exposure (ii) enhance the ability of 3D models to simulate the full set of heterogenous chemistry driving SOA formation from specific VOCs (iii) implement surrogate models for increasing the computational efficiency of chemical data assimilation and forecast systems, (iv) automate the construction of reduced-complexity chemical mechanisms for simulation of atmospheric chemistry, and (v) quickly identify source factors in large aerosol mass-spec datasets.


Molecule to Single Particle

Chairs: Tinja Olenius, Swedish Meteorological and Hydrological Institute & Ivan Piletic, US EPA


A Quantum Machine Learning Approach for Studying Atmospheric Molecular Cluster Formation

Presenter: Jonas Elm, Aarhus University 
Presentation Description: Molecular cluster formation is the first step in the formation new particles in the atmosphere. Using state of the art quantum chemical methods it is possible to study cluster formation up to cluster diameters around 1.5 nm. Particle counters commonly employed in the laboratory and in the field usually operates at a lower cutoff of roughly 2.0 nm. This implies that there is a gap in knowledge about the cluster formation and growth processes in between the current state-of-the-art theoretical and experimental techniques. In this work we present an efficient approach to produce high quality quantum chemical datasets for applications in cluster formation studies and how to train an accurate quantum machine learning (ML) model on the generated data.
We apply a kernel ridge regression model using a local descriptor for the representation of the molecular structures. Using the two-component sulfuric acid - water system as a proof-of-concept, we demonstrate that a ML model can accurately predict the energetics of the clusters leading to massive decrease in the number of quantum chemical calculation required. The application of the ML model leads to an enormous gain in computational efficiency allowing the efficient configurational space exploration of clusters larger than previously possible. An outlook on how the presented ML approach can be directly applied to clusters of arbitrary composition is presented.

Liquid-liquid phase separation in submicron aerosol

Presenter: Miriam Arak Freedman, Pennsylvania State University
Presentation Description: Liquid-liquid phase separation, in which two liquid phases in an aerosol particle demix, has the potential to affect heterogeneous chemistry, aerosol optical properties, water uptake, and the growth of new particles. The separation relative humidity, at which liquid-liquid phase separation occurs, depends strongly on the pH and composition of the solution. For submicron particles, liquid-liquid phase separation is inhibited at the smallest particle sizes (<~30 nm) because small particles cannot overcome the activation barrier needed to form a new phase. The size under which phase separation is inhibited depends on the drying rate and the composition of the particles. In addition, the separation relative humidity for submicron particles occurs at lower values and over a larger range of relative humidity than for supermicron particles. This talk will give an overview of my group's work in this area and discuss opportunities for development of parameterizations for submicron particles in terms of which systems undergo liquid-liquid phase separation as well as the relative humidity at which phase separation occurs.

Understanding nucleation and initial particle growth in binary vapors with controlled Laval expansion

Presenter: Chenxi Li, Shanghai Jiao Tong University
Presentation Description: Gas phase nucleation and initial particle growth play a key role in the formation and evolution of atmospheric particulate matter. As nucleation and initial growth critically depend on the properties of small molecular clusters, a complete understanding of these processes requires quantitative analysis on the molecular level. Here we combine the uniform postnozzle flow of pulsed Laval expansions, soft single-photon ionization and time-of-flight mass spectrometry to directly measure the chemical composition and the concentration of the molecular aggregates in binary nucleation systems at <100 K. Specifically, we focus on various binary CO2-containing vapors with CO2 being the more volatile species. We develop simple equations to extract monomer-cluster association rates from the evolving mass spectra of molecular clusters and apply kinetic modelling to fit key parameters in a proposed nucleation mechanism. Our analysis reveal that CO2 essentially catalyzes nucleation through a Chaperone-like mechanism, i.e. CO2 enhances the nucleation of the low vapor pressure component through the formation of transient, heteromolecular clusters. Moreover, we present data on the initial stages of particle growth and discuss CO2 condensation on the molecular aggregates.

A general parametrization for salt nanoparticle formation

Presenter: Nanna Myllys, University of Jyväskylä
Presentation Description: Acid-base chemistry plays a key role in atmospheric aerosol particle formation. Using combinations of three representative acids (sulfuric, methanesulfonic and nitric) with nine representative bases (ammonia, methylamine, dimethylamine, trimethylamine, trimethylamine-N-oxide, guanidine, monoethanolamine, putrescine and piperazine), we have studied molecular properties affecting particle formation efficiency. The best predictor of forming stable acid-base heterodimers, and thus particles, was found to be gas-phase acid and base strength. We have explored the relationship between particle formation rate and heterodimer stability over a wide range of temperatures and monomer concentrations. To yield a general parameterization of particle formation rate at any given conditions, heterodimer concentration was calculated from their Gibbs free formation energies. This parameterization was tested for the sulfuric acid-ammonia system by comparing the predicted values to experimental data and was found to be accurate within two orders of magnitude. For a more accurate model that incorporates the dependence on temperature and monomer concentration on particle formation rate, we defined a new term called normalized heterodimer concentration. While the general parameterization is not compound-dependent, the more accurate model varies as a function of acid and base strength, and thus must be used separately for different acid-base systems.

Identification of environmentally relevant isomers based on computational ion mobility predictions

Presenter: Ivo Neefjes, Institute for Atmospheric and Earth System Research, University of Helsinki, Finland 
Presentation Description: Chemical ionization atmospheric pressure interface mass spectrometry (CI-APi-TOF MS) is uniquely suited to identify and quantify the gaseous precursors of new particle formation at low ambient concentrations. Current mass spectrometry set-ups alone are, however, unable to distinguish between isomers. Isomers are ubiquitous in the atmosphere and have distinctive properties and reaction pathways. We combined nitrate CI-APi-TOF MS with a differential mobility analyzer to separate isomers based on their ion mobility. To improve the identification of isomers, we employed the Ion Mobility Software (IMoS) program to obtain computational ion mobility predictions of environmentally relevant isomer-nitrate dimers, studied using our established configurational sampling workflow. We tested the dependency of the ion mobility predictions on the choice of functional in the ab initio calculation, the conformer geometry, and the partial charge calculation method. We find that the absolute error of the ion mobility predictions cannot consistently be lowered to less than 2%. However, the errors exhibit a certain systematic behavior, suggesting that an ion mobility difference of 0.03 cm2V-1s-1 in the experiments is enough to reliably differentiate isomers based on the computational ion mobility predictions.

Atomistic molecular dynamics simulations of ion-dipole collisions in the atmosphere

Presenter: Bernhard Reischl, Institute for Atmospheric and Earth System Research / Physics, University of Helsinki 
Presentation Description: The first stage of atmospheric new particle formation is a collision of molecules or ions in the gas phase, leading to a stable dimer. As the interactions between ions and dipoles are much stronger than between neutral molecules, analytical (Langevin et al.) and parametrized models (Su and Chesnawich, Maergoiz et al.) have been developed to determine collision rate coefficients directly from the charge of the ion and the dipole moment and polarizability of the neutral. However, these approaches neglect the actual dynamic charge distribution of dipolar or charged molecules and clusters. To study these effects, we perform atomistic molecular dynamics simulations of collisions between dipolar sulfuric acid molecules or bisulfate-dimethylammonium clusters, and nitrate, bisulfate, ammonium, dimethylammonium ions or charged sulfuric acid-bisulfate clusters. From the collision probabilities observed over the relevant ranges of impact parameters and relative velocity distributions, we calculate the collision rate coefficients, and compare them to the model predictions. We find surprisingly good agreement, even for ion-dipole collisions involving small molecular clusters, validating the use of these models in cluster dynamics codes, such as ACDC, for the calculation of particle formation rates.


Process Models to Box-Model

Chairs: Thomas Berkemeier, Max Planck Institute for Chemistry & Manabu Shiraiwa, UC Irvine


Modeling the Tropospheric Multiphase Chemistry of Biomass Burning Trace Compounds Using the Chemical Aqueous Phase Radical Mechanism (CAPRAM)

Presenter: Lin He, The Leibniz Institute for Tropospheric Research (TROPOS)
Presentation Description: Biomass burning (BB) is a significant contributor to air pollution on global, regional and local scale with impacts on air quality, public health and climate. Anhydrosugars and methoxyphenols are key tracers emitted through BB. Once emitted, they can undergo complex multiphase chemistry in the atmosphere contributing to secondary organic aerosol (SOA) formation. However, their chemical multiphase processing is not yet well understood and investigated by models. Thus, the present study aimed at a better understanding of the multiphase chemistry of these BB tracers by detailed model studies with a new developed CAPRAM biomass burning module (CAPRAM-BBM). This module was developed based on the kinetic data from our laboratory measurements at TROPOS and other literature studies. The developed CAPRAM-BBM includes 2991 reactions (9 phase transfers and 2982 aqueous-phase reactions). By coupling with the multiphase chemistry mechanism MCMv3.2/CAPRAM4.0 and the extended CAPRAM aromatics (CAPRAM-AM1.0) and halogen modules (CAPRAM-HM3.0), it is being applied for residential wood burning cases in Europe and wildfire cases in the US. Our model results show that levoglucosan and vanillin are effectively oxidized under cloud conditions. Furthermore, the results demonstrate that the chemistry of BB tracers can affect the budgets of key oxidants such as H2O2, and contribute to the SOA formation especially by increasing the fraction of brown carbon and substituted organic acids.

Process-Level Modeling Can Simultaneously Explain Secondary Organic Aerosol Evolution in Chambers and Flow Reactors

Presenter: Charles He, Colorado State University 
Presentation Description: Environmental chambers and oxidation flow reactors (OFR) are used to study secondary organic aerosol (SOA) formation under a wide range of atmospheric aging times. While SOA parameters used in atmospheric models are typically developed using chamber data, there is an opportunity to incorporate OFR data to improve their representation over longer photochemical ages. In this work, we use a size-resolved chemistry and microphysics model, updated to represent nucleation, diffusion-limited gas/particle partitioning, oligomerization, wall losses, and heterogeneous oxidation, to simulate SOA formation from the photooxidation of alpha-pinene in chamber and OFR experiments. We found that by accounting for these processes, the model is able to simultaneously capture measured SOA mass concentrations and size distributions in both type of reactors using a consistent set of SOA parameters. Some key findings regarding these processes include: (1) nucleation dynamics can significantly affect SOA mass and size distribution, and for the model to capture the measured size distributions in OFR, the nucleation rate must be calculated based on a multi-generational oxidation product, as opposed to first-generation; (2) vapor and particle wall losses have minimal to moderate impact on OFR SOA formation; (3) SOA in OFR can be semi-solid which has a moderate influence on SOA formation; (4) heterogeneous oxidation significantly affects SOA mass and size distribution at higher OH exposures.

Multiphase chemistry within Arctic fog droplets explains unexpected growth of Aitken mode particles to CCN sizes

Presenter: Erik Hoffmann, Leibniz Institute for Tropospheric Research
Presentation Description: New particle formation and early growth are efficient processes generating high concentrations of potential cloud condensation nuclei (CCNs) precursors within the Arctic marine boundary layer (AMBL). Due to the low particle concentrations in the AMBL, even the smallest amount of Aitken mode particle growth is capable to significantly increase the CCN budget. During the PASCAL campaign in 2017, measurements of aerosol particles were performed and an unexpected rapid growth of Aitken mode particles was observed right after fog episodes. Detailed multiphase chemistry box model simulations with CAPRAM were performed to study the underlying processes. A new mechanism is proposed explaining how particles with d < 50 nm are able to grow into CCN size range in the Arctic. The rapid particle growth is related to chemical processes within the Arctic fog. The redistribution of semi-volatile acidic (e.g., methanesulfonic acid) and basic (e.g., ammonia) compounds between different particle sizes leads to a rapid particle growth of non-CCNs after fog evaporation enable them to grow towards CCN size. Thus, chemical in-fog processes and subsequent post-fog repartitioning are key processes contributing to the increase in the number of CCNs and cloud droplets leading to an increased albedo of Arctic clouds. Since fogs will occur more frequently in the Arctic as a result of climate change, this process and a deeper knowledge on its feedbacks is essential to understand Arctic warming.

Modeling the Role of Peroxy Radicals in Camphene Secondary Organic Aerosol Formation

Presenter: Jia Jiang, University of California, Riverside 
Presentation Description: The SAPRC mechanism generation system (MechGen) and box model were used to simulate a series of camphene chamber experiments across a range of atmospherically relevant conditions. Experiments with added nitrogen oxides (NOx) produced higher SOA yields (up to 65%) than experiments without added NOx (up to 28%); the SAPRC simulations showed that the higher SOA yields at the higher initial NOx levels were primarily due to the production of highly oxygenated organic molecules (HOMs). After the initial reaction between camphene and OH, subsequent rapid ring-opening and decomposition reactions occurred, leading to the formation of HOMs through autooxidation in the presence of NOx. In contrast, without NOx, the initial camphene peroxy radical (RO2) quickly reacted with HO2 to form terminal products with much higher volatilities. In addition, camphene SOA yields increased with SOA mass (Mo) at lower mass loadings, but a threshold was reached at higher mass loadings due to the increasing fraction of self- and cross-reactions of RO2 under higher initial [HC]/[NOx] ratios. The observed differences in camphene SOA yields here are distinguished from the suppressing effects of NOx reported in previous studies of α-pinene and β-limonene, and are largely explained by the gas-phase RO2 chemistry. This highlights an important role for chemically-detailed box models in elucidating SOA formation mechanisms and better linking gas- and particle-phase chemistry.

Exhalation kinetics of surrogate lung fluid particles using experiments and process modelling

Presenter: Liviana Klein, ETH Zurich 
Presentation Description: Respiratory viruses, such as SARS-CoV-2 and influenza, can be transmitted via virion-containing expiratory aerosol particles. Since indoor relative humidity is typically much lower than the humidity within the respiratory tract, the particles experience a rapid drying upon exhalation. This may change pH with consequences for the viability of the virions.
The aerosol composition is very complex, containing water, various salts with NaCl being the most abundant, proteins, and surfactants. Thus, their physico-chemical properties during drying are not trivial and so far, only poorly characterized. Using an electrodynamic balance, we measure the thermodynamics and drying kinetics of surrogate lung fluid particles. Our measurements indicate that at efflorescence relative humidity (ERH) the kinetics of crystallization follows two different regimes. First, a quick crystallization step followed by a slow crystal growth requiring up to hours. Modeling this kinetics using a multi-layer respiratory aerosol model allows an estimation of the effective diffusivity of the salt ions. In addition, water diffusivity can be deduced from experiments above the ERH. For estimating the pH of these particles in indoor air, we assume that the salt diffusivity represents the diffusivities of all larger molecules. This allows the model to predict the gas to particle partioning of not only water but also NH3, HNO3, and CO2 during exhalation and thus pH of expiratory aerosol particles.

Lab-to-environment modeling for bridging scales in atmospheric chemistry

Presenter: V. Faye McNeill, Columbia University
Presentation Description: A major challenge in atmospheric chemistry is bridging scales between complex molecular-level chemical mechanisms and computationally expensive large scale atmospheric models. We apply the McNeill Group photochemical box model, GAMMA, using a technique we call 'lab-to-environment modeling,' to analyze the kinetics and mechanisms of chemical processes identified in the laboratory and evaluate their potential environmental impact. This approach allows us to prioritize chemical pathways for inclusion in large scale models. Examples from our work will be given.

Modelling new particle formation, secondary aerosol dynamics and aerosol phase-state – highlighting the key role of peroxy radical autoxidation

Presenter: Pontus Roldin, Lund University 
Presentation Description: Secondary aerosol formation involving new particle formation and growth by condensation are a major but still not well-quantified source of climatically essential cloud condensation nuclei. Comprehensive smog chamber experiments are performed in many aerosol laboratories around the world to improve the knowledge about the chemical and physical pathways governing secondary aerosol formation.
In my presentation, I will focus on our multiphase-chemistry and aerosol dynamics box-modelling work, which we perform for different smog chamber facilities. With the combination of detailed process-modelling and recent advances in gas- and aerosol mass spectrometry, we are able to improve the molecule-level understanding of the key steps in secondary aerosol formation from major biogenic secondary aerosol precursors, such as terpenes and dimethyl sulfide (DMS).
I will present results from recent advances in the development of near-explicit atmospheric reaction mechanisms for monoterpenes and DMS. For monoterpenes, peroxy radical (RO2) autoxidation and the formation of highly oxygenated organic molecules (HOM) has a central role for the secondary organic aerosol (SOA) formation and SOA phase-state at different environmental conditions. In contrast, for DMS the RO2 autoxidation tends to substantially slow down and potentially suppress atmospheric new particle formation and secondary aerosol yields.

Suppression of nucleation and growth of alpha-pinene oxidation products due to isoprene

Presenter: Meredith Schervish, Carnegie Mellon University, Department of Chemistry 
Presentation Description: Monoterpenes and isoprene together are thought to dominate total biogenic emissions. The oxidation products of biogenic organic compounds contribute to new-particle formation in remote areas of the atmosphere as well as in laboratory experiments. They may have been the predominant particle formation pathway in the pre-industrial continental troposphere. Isoprene has been shown to suppress nucleation even with high concentrations of monoterpenes both in chamber studies and ambient measurements. The cause of this suppression is, however, debated. Recent experimental work suggests that isoprene suppression of nucleation is due to the suppression of the ultra-low volatility organic compounds (ULVOCs), specifically the C20 dimers, that form from alpha-pinene oxidation chemistry. In this work we explore the interactions between monoterpene and isoprene oxidation in a model employing the radical two-dimensional Volatility Basis Set (radical-VBS) and show that isoprene can strongly suppress the initial stages of new-particle formation by competing for monoterpene-derived peroxy radicals and forming higher volatility C15 compounds rather than the nucleating C20 compounds. However, as particles grow to larger sizes, the effect of isoprene is smaller as higher volatility compounds such as the C15 products are able to contribute to, and in fact dominate, the growth at these sizes.

Modelling of aerosol coagulation in turbulent flows with Direct Numerical Simulation and Population Balance Modelling

Presenter: Malamas Tsagkaridis, Imperial College London
Presentation Description: The central objective of the present study is to investigate the turbulence-coagulation interaction via direct numerical simulation (DNS) coupled with the population balance equation (PBE) (also known as the General Dynamic Equation). Coagulation is an important process in several environmental and engineering applications involving turbulent flow, including soot formation, gas-phase synthesis of nanoparticles and atmospheric processes, but its interaction with turbulence is not yet fully understood. Particle dynamics can be described by the PBE, whose Reynolds decomposition leads to unclosed terms involving correlations of number density fluctuations. The identification of the unclosed terms dates back to the early work of Levin and Sedunov (1966). One of the first attempts to model these terms was made by Rigopoulos, (2007) where the PDF approach was tested in a partially stirred reactor. It was concluded that future work should evaluate the error which occurs by neglecting the turbulent fluctuations of the PSD in particle formation/coagulation problems. In the present study, we employ a recently proposed discretisation (sectional) method (Liu and Rigopoulos, 2019), which is free of a priori assumptions regarding the particle size distribution (PSD) to solve the PBE together with flow DNS and aim to shed light on the long-standing problem of the unknown correlations.

Modeling Size-distributed Dynamic SOA Partitioning

Presenter: Rahul Zaveri, Pacific Northwest National Laboratory
Presentation Description: One of the major challenges in modeling secondary organic aerosol (SOA) formation is to accurately capture its size distribution dynamics. The size-dependent timescale of SOA partitioning crucially depends on the volatility of the condensing organic vapors as well as their diffusivity and reactivity within the bulk of the particle phase. Here we describe an efficient algorithm for dynamic SOA partitioning within the MOSAIC aerosol box-model. We apply the model to interpret chamber observations of isoprene SOA partitioning to a pre-existing bimodal aerosol consisting of an Aitken mode (potassium sulfate) and accumulation mode (aged α-pinene SOA) as a function of relative humidity (RH). We find that bulk diffusivity values corresponding to a liquid-like aerosol phase state are unable to reproduce the rapid growth of the Aitken mode. Instead, much lower bulk diffusivities in aged α-pinene SOA, consistent with viscous semisolid phase states, are needed to successfully reproduce the growth of both modes, even at ~80% RH. The low bulk diffusivity hinders the partitioning of semivolatile organic compounds (SVOCs) to large semisolid particles and thereby allows smaller, less viscous particles to effectively absorb the available SVOCs and grow much faster than would be possible otherwise. This mechanism also explains the observed rapid growth of pollution-induced nanoparticles in the Amazon and has important implications for modeling SOA partitioning in a larger realm.

Intercomparison of aerosol microphysics parameterizations in the MAM aerosol box model

Presenter: Kai Zhang, Pacific Northwest National Laboratory
Presentation Description: The representation of aerosol microphysics in global aerosol models involves a number of physical and numerical assumptions, which in turn cause large differences in the simulated aerosol physical properties and lifecycle between models. Isolating the aerosol microphysics from the resolved atmospheric dynamics and other parameterized processes (e.g., clouds and radiation) helps to identify the root cause of model simulation differences, which is difficult for general intercomparison using global models. In this work, we extend the functionality of the Modal Aerosol Module (MAM) box model and compare different aerosol microphysics treatments that are typically used in global aerosol models. We find substantial differences in the simulated aerosol physical properties due to changes in the assumption for new particle formation and growth, aerosol wateruptake, and the coupling between individual aerosol microphysical processes.

Automatic 2-D component classification and lumping for gas-particle partitioning computations

Presenter: Andreas Zuend, McGill University
Presentation Description: Near-explicit atmospheric chemistry mechanisms often simulate the concentration evolution of hundreds to millions of oxidation products after several generations of gas phase chemistry. Direct processing of such a wealth of data can lead to technical and interpretation challenges of the data. Furthermore, many of the simulated compounds will be similar in chemical structure (isomers), which tend to have similar physicochemical properties of interest in gas-particle partitioning and liquid phase mixing behavior. For a reduced-complexity representation of aerosol particle composition in box models and atmospheric chemistry models, a substantial degree of system simplification is often required or desired.
In this presentation, we will introduce a chain of tools developed to enable the processing of molecular structure and concentration outputs using cheminformatics tools with customized SMARTS/SMILES-based libraries. The SMARTS-processed outputs allow automatic lumping of the molecules to a set of surrogate compounds selected by means of a 2-D volatility vs. polarity classification framework. The framework offers different methods of gridding or clustering the data, adjustable resolution (targeted number of surrogate compounds) and choices for the two axes. We will discuss these tools, impacts of resolution and their applications for gas-particle partitioning computations with the AIOMFAC-based thermodynamic equilibrium model.


Advances in regional and global scale aerosol model developments for simulating the myriad processes affecting the properties and chemical composition of fine particles in the atmosphere

Chairs: Tzung-May Fu, Southern University of Science and Technology & Manish Shrivastava, Pacific Northwest National Laboratory


Process-based and Observation-constrained SOA Simulations in China: The Role of Semivolatile and Intermediate-Volatility Organic Compounds and OH Levels

Presenter: Qi Chen, Peking University
Presentation Description: Chemical transport models have difficulties to reproduce the variability of OA concentrations in polluted areas, hindering understanding of the OA budget. We applied both process-based and observation-constrained schemes to simulate OA in China. Comprehensive data sets of surface observations were used for model evaluation. Updates of the emissions, volatility distributions, and SOA yields of semivolatile and intermediate volatility organic compounds (S/IVOCs) are insufficient to reproduce the SOA concentrations in observations. The addition of nitrous acid sources is an important model modification, which improves the simulated surface concentrations of hydroxyl radical (OH) in winter in northern China. The increased surface OH concentrations lead to greater SOA mass concentrations by over 30%, highlighting the importance of having good OH simulations in air quality models. With all the model improvements, both the process-based and observation-constrained SOA schemes can reproduce the observed mass concentrations of SOA and show spatial and seasonal consistency with each other. Our best model simulations suggest that anthropogenic S/IVOCs are the dominant source of SOA in China with a contribution of over 50%. The residential sector may be the predominant source of S/IVOCs in winter, despite large uncertainty remains in the emissions of IVOCs from the residential sector in northern China. The industry sector is also an important source of IVOCs, especially in summer.

A 3D particle-resolved model (WRF-PartMC) for quantifying structural uncertainties in a modal model (MAM3) on the regional scale

Presenter: Jeffrey Curtis, University of Illinois at Urbana-Champaign
Presentation Description: Field measurements indicate that aerosol populations are complex distributions of particle size and composition. These microscale particle-level details are important for determining the ability of aerosols to act as cloud condensation nuclei and ice nucleating particles, and how they scatter and absorb solar radiation. At the same time, these details introduce high computational cost for regional and global scale simulations. As a compromise of accuracy for computational efficiency, aerosol models make simplifying assumptions which lead to structural uncertainty in the model, complicating an already difficult problem surrounding aerosol-cloud-radiation interactions. To investigate this problem, we have developed a particle-resolved model for the regional scale, WRF-PartMC, which simulates thousands of individual aerosol particles within each model grid cell, capturing the highly complex size and composition present in ambient aerosol populations. Here we use this model to benchmark the performance of the modal model MAM3, which is part of the WRF-Chem aerosol modeling suite. The comparison of WRF-Chem-MAM3 and WRF-PartMC simulations for a case study in California during the CARES campaign allows us for the first time to rigorously quantify the impact of aerosol representation on simulating aerosol properties within a 3D modeling framework on the regional scale. In this presentation, we will discuss the leading causes of structural uncertainty in MAM3.

Global budget of atmospheric organic nitrogen aerosols and implications for nitrogen deposition

Presenter: Yumin Li, School of Environmental Sciences and Engineering, Southern University of Science and Technology, Shenzhen, Guangdong Province, China
Presentation Description: Organic nitrogen (ON) is the major colored component of atmospheric organic aerosols and constitutes a significant fraction of atmospheric nitrogen deposition. We constructed a global simulation of ON aerosols, including primary emissions of ON and its precursors from natural and anthropogenic sources, as well as secondary formation pathways, with the goal of evaluating the atmospheric budget of ON aerosols and its impact on nitrogen deposition and climate. The global primary emission of ON is 4.9 Tg yr-1, including 1.3 Tg yr-1 from anthropogenic sources, 2.8 Tg yr-1 from biomass burning, 0.77 Tg yr-1 from the global oceans, and 45 Gg yr-1 from dusts. Secondary production of ON is 2.1 Tg yr-1, 90% of which is from the aqueous reaction of dicarbonyl with NH4+. The global atmospheric burden of ON is 0.12 Tg with a lifetime 6.3 days. Our simulated ON concentrations agree with observations in magnitudes and spatiotemporal variation. On the global scale, ON in fine particles represent 9.4% of total nitrogen wet deposition, mostly due to wet deposition of primary ON from biomass burning (39%) and secondary ON from dicarbonyls (32%). The contribution of ON to wet nitrogen deposition is significant (over 25%) in the polar regions of Asia, North America, and the tropical rain forests of South America and Africa.

Reduction of numerical diffusion in linear advection schemes with application to Eulerian transport models

Presenter: Robert McGraw, Brookhaven National Laboratory
Presentation Description: Linear advection schemes have the great advantage of preserving tracer correlation during advection of aerosol/cloud particle distributions in Eulerian models. Nevertheless, the widespread use of linearized advection, especially in regional to global scale atmospheric models is almost never recommended on account of numerical diffusion. As a result, nonlinear schemes have been developed and are in widespread use. These can succeed remarkably well in reducing numerical diffusion for the case that a single tracer, such as particle number or total particle mass, is advected, but they come at the price of significant correlation loss when interrelated sets of 2-3 or more tracers, e.g., radial moments and/or chemical compositions of mixed particle populations, are advected. In this paper we show the tradeoff between numerical diffusion and loss of tracer correlation is only apparent - one can have their cake and eat it too! A mathematical transformation, based on a simple linear program is shown to preserve tracer correlation while minimizing numerical diffusion. The new method illustrated for 1 and 2D advection cases, preserves linearity and the computational speed of linear schemes. Moreover, our approach is shown to fully preserve interrelationships between moments and other particle attributes, thus enabling atmospheric modelers to finally utilize the power of moment methods and quadrature in their simulations.

Chemistry Across Multiple Phases (CAMP): An integrated multi-phase chemistry model

Presenter: Nicole Riemer, University of Illinois at Urbana-Champaign
Presentation Description: Progress in identifying increasingly complex, atmospherically relevant mixed-phase physicochemical processes has resulted in an advanced understanding of the evolution of atmospheric systems but has also introduced a level of complexity that few atmospheric models were originally designed to handle. We present a flexible treatment for gas- and aerosol-phase chemical processes for models of diverse scale, from box up to global models. The flexibility of the model is achieved by (1) object-oriented design, which facilitates extensibility to new types of chemical processes and to new ways of representing aerosol systems (modal, sectional, particle); (2) runtime model configuration using JSON input files, which permits making changes to the chemical mechanism without recompiling the model; and (3) automated comprehensive tests to ensure stability of the codebase as new functionality is introduced. Taken together, this enables users to build a customized multiphase mechanism without having to handle pre-processors, solvers or compilers, and makes this type of modeling accessible to a much wider community, including modelers, experimentalists, and educators. The new treatment has been deployed in the particle-resolved PartMC model and in the MONARCH chemical weather prediction system for use at regional and global scales. Results from the initial deployment will be discussed, along with future extension to more complex gas-aerosol systems, and the integration of GPU-based solvers.

Global simulations of monoterpene-derived peroxy radical fates and the distributions of highly oxygenated organic molecules and accretion products

Presenter: Joel Thornton, University of Washington
Presentation Description: We add unimolecular autoxidation reactions and self-/cross reactions of monoterpene-derived (MT) peroxy radicals (RO2) to the GEOS-Chem global chemical transport model, including treatments of highly oxygenated organic molecules (HOM) and accretion product formation. We conducted a group of sensitivity simulations to evaluate the dependence of MT-RO2 fates, and of HOM and accretion product abundance and distributions, on current uncertainties in associated kinetic parameters. We compare predictions with limited available observations worldwide. Autoxidation is the dominant fate up to 6-8 km for a fraction of MT-RO2, but reaction with NO can be a more common fate, when smaller H-shift rate constants (e.g., 0.1 s-1) are used or at altitudes higher than 8 km due to the expected Arrhenius temperature dependence of unimolecular H-shifts. The major fate for MT HOM-RO2 is predicted to be reaction with other RO2 throughout most of the boreal and tropical forested regions, while reaction with NO dominates in temperate and subtropical forests of the Northern Hemisphere. Comparison with observations reveals that total HOM concentrations can be of order 0.5 to 1 ug m-3 within the PBL of MT-rich regions and seasons, although large uncertainties remain for key reaction parameters, especially the photochemical lifetime of HOM. Our simulations also indicate measured accretion products in ambient OA outside of tropical forested regions are more likely from particle-phase chemistry or very specific RO2 gas-phase reactions, rather than the newly added more general MT-RO2 self-/cross-reactions, even considering large rate constants.

GENOA: the generator of semi-explicit mechanisms for SOA modeling

Presenter: Zhizhao Wang, CEREA, Joint laboratory of École des Ponts ParisTech and EdF R&D
Presentation Description: Volatile organic compounds (VOC) undergo multi-generation oxidation in the troposphere. Their oxidized products contribute significantly to the formation and aging of secondary organic aerosols (SOA) via gas-particle partitioning. This non-linear evolution from VOC oxidation to aerosol growth can be described precisely by a detailed VOC chemistry that may involve numerous organics.
Thus, when coupling aerosol formation to other atmospheric processes (e.g., transport), 3D air quality models (AQM) are faced with the dilemma of spending an intolerable overwhelming computational capacity on the entire complexity of VOC to SOA or utilizing highly simplified parameterization (e.g., Odum 2-product, VBS, SOM), where only a few surrogates represent various organics.
To address this issue, the GENerator of the reduced Organic Aerosol mechanism (GENOA) is developed to generate user-customized semi-explicit SOA mechanisms with a number of species and reactions suitable to 3D AQM. GENOA currently uses the Master Chemical Mechanism (MCM) as a baseline and a 0D aerosol model SSH-aerosol to simulate gas-particle partitioning taking into account non-ideality and hygroscopicity.
When applied to the beta-caryophyllene, GENOA generates SOA mechanisms that reduce the number of species by up to 95% and CPU time by 90% in 0D modeling with an average error (compared to the original MCM) lower than 3%. The application to other primary emitted VOCs (e.g., limonene, 1-3 butadiene) is also tested.

Implementation of FIREX-AQ measured optical properties in WRF-Chem for estimations of secondary organic aerosol formation

Presenter: Chenchong Zhang, Washington University in St. Louis
Presentation Description: Light-absorbing organic carbon (LAOC) is one of the major components of smoke plumes from biomass burning (BB) events. The atmospheric processing of emitted LAOC during atmospheric transport could have a significant influence on regional radiative forcing and air quality. The spatiotemporal evolution of LAOC optical properties from BB events and their radiative implications are poorly understood, which limits this process-based parameterization in radiative transfer and chemical transport models. Here we synergistically integrate the insights gained from chemical transport models (CTM) and field measurements to investigate the impacts of atmospheric processing on LAOC emitted during the field study portion of the 2019 Fire Influence on Regional to Global Environments and Air Quality (FIREX-AQ) campaign. We incorporate the optical properties of LAOC measured by electron energy-loss spectroscopy into the WRF-Chem (Weather Research and Forecasting model coupled with Chemistry) which couples with the volatility basis set to account for the interplays between the atmospheric photo-oxidation processes and the yields of secondary organic aerosols. Our simulation results are further compared with ground-based measurements to improve our prediction for the atmospheric processing for LAOC. We anticipate our findings to provide better insights for improving LAOC representation in current radiative transfer codes and CTMs.

Simulating new-particle formation and the impact on cloud condensation nuclei in pristine and polluted environments of the Amazon

Presenter: Bin Zhao, School of Environment, Tsinghua University
Presentation Description: A major challenge in quantifying aerosol radiative forcing is to understand the sources of aerosols and cloud condensation nuclei (CCN) in pristine environments. Here we investigate the mechanisms of new-particle formation (NPF) and its impact on CCN over the Amazon using an updated WRF-Chem model. We incorporate in the model an advanced NPF scheme and an experimentally-constrained Radical Two-Dimensional Volatility Basis Set (R2D-VBS), which simulates the temperature-dependent formation chemistry and thermodynamics of extremely low volatility organics that drive NPF. We find that, under pristine conditions, new particles are predominantly produced at high altitudes from organic NPF pathways, especially pure-organic NPF which accounts for 65–83% of the column total NPF rate. The NPF processes contribute over 90% of the CCN in the upper troposphere and, through downward transport, contribute 25–80% of the CCN near the surface. Among the CCN attributed to NPF, over half in the upper troposphere and 80% near the surface originate from the transport and growth of remotely formed new particles. In regions downwind of a megacity in the central Amazon, our model shows that NPF usually contributes > 70% of the particle number and > 25% of the CCN. The ternary NPF of sulfuric acid with organics dominates among all NPF pathways in our model. This study improves the understanding of aerosol and CCN sources in pristine and moderately polluted environments of the Earth's atmosphere.


Emerging Modelling Techniques

Chairs: Christopher Tessum, University of Illinois & Zhonghua Zheng, Columbia University


Droplet Breakup for the Super-Droplet Method

Presenter: Emily de Jong, California Institute of Technology
Presentation Description: Lagrangian particle-based cloud microphysics schemes represent the dynamics of a population of particles such as aerosols, cloud droplets and raindrops. This class of models provides good computational efficiency and physical rigor at the particle-scale, and is becoming a key computational tool for studying the interactions between aerosol particles and clouds. However, droplet breakup presents a challenge, as the total number of simulation particles (known as "super-droplets") used to represent the system must be conserved for the method to function efficiently. In this work, we propose an algorithm for the Super-Droplet Method to perform collisional and spontaneous breakup which: (1) is consistent with existing rate formulations; and (2) preserves the number of super-droplets in the system. We present an implementation of breakup in the PySDM software package, along with results to validate the algorithm and demonstrate convergence. Implementing droplet breakup, which is often neglected as a process in atmospheric simulations, as an option in these high resolution Lagrangian microphysics schemes will make more physically realistic simulations of aerosol-cloud interactions possible.

Predicting Glass Transition Temperature and Viscosity of Organic Molecules via Machine Learning and Molecular Embeddings

Presenter: Tommaso Galeazzo, University of California, Irvine
Presentation Description: Modeling secondary organic aerosols (SOA) rely on accurate representation of physical properties of constituting semi-volatile organic species. Notably, their partitioning between the gas and particle phases is highly influenced by particle phase state and viscosity. SOA viscosity can be estimated from the glass transition temperature (Tg) of the constituting compounds. For accurate predictions of the viscosity of complex SOA mixtures, information on molecular structure and functional groups are needed.
Here, we introduce a new Tg prediction method powered by a machine learning model and “molecular embeddings. Molecular embeddings are developed using the word2vec and Morgan algorithms, and species SMILES. We have trained state-of-the-art ML models on a large database of experimental Tg data of pure organic species and their corresponding molecular embeddings. Different algorithms have been explored for accuracy in predicting Tg. The final Tg prediction method is built on top of an Extreme Gradient Boosting (XGBoost) model and it outperforms previous Tg parametrizations. The new model accounts for atom connectivity and it can predict different Tg for compositional isomers. It can also reproduce experimental viscosity data and quantify the influence of number and location of functional groups on organic compounds viscosity. This new ML powered Tg model can be exploited to predict viscosity in numerical models involving organic species, with future applications that go beyond aerosol chemistry.

Emulating an Aerosol Microphysics Model with Deep Learning

Presenter: Paula Harder, Fraunhofer Institute ITWM
Presentation Description: Aerosol particles play an important role in the climate system by absorbing and scattering radiation and influencing cloud properties. They are also one of the biggest sources of uncertainty for climate modeling. Many climate models do not include aerosols in sufficient detail. In order to achieve higher accuracy, aerosol microphysical properties and processes have to be accounted for. This is done in the ECHAM-HAM global climate aerosol model using the M7 microphysics model, but increased computational costs make it very expensive to run at higher resolutions or for a longer time. Our work shows a machine learning approach to approximate the M7 microphysics module. We investigated different approaches, neural networks as well as ensemble models like random forest and gradient boosting, with the aim of achieving the desired accuracy and computational efficiency, finding the neural network appeared to be the most successful. We use data generated from a realistic ECHAM-HAM model run and train a model offline to predict one time step of the aerosol microphysics. The underlying data distribution is challenging, as the changes in the variables are often zero or very close to zero. We do not predict the full values, but the tendencies, which requires a logarithmic transformation for both positive and negative values. After we considered different hyperparameters, a 2-layer, fully connected network to solve this multivariate regression problem is chosen. To train, validate and test our model in a representative way, we use about 11M data points from separate days spread out through the year. On the test set, an overall regression coefficient of 89% and a root mean squared error (RMSE) of 24.9% are achieved. For 75% of the variables, our model's prediction has a regression score of over 95% on the logarithmically transformed tendencies. Using a GPU we achieve a huge speed-up of 120 compared to the original model, on a single CPU we are still 1.6 faster.

Super-droplet Method to Simulate Lagrangian Microphysics of Fallout Particles

Presenter: Dana McGuffin, Lawrence Livermore National Laboratory 
Presentation Description: Nuclear detonations produce hot clouds that quickly cool and rise, forming dangerous ultrafine to coarse aerosol (fallout) that can potentially disperse over long distances and deposit. Therefore, the ability to predict aerosol size, chemical components, and location is necessary to protect the public. Quickly predicting these quantities is vital, so this work presents a computationally efficient way to model particle microphysics in a hot, turbulent cloud.
The super-droplet method (SDM) is an efficient way to model cloud microphysics without losing accuracy. SDM represents the particle size distribution by sampling it with a set of computational particles called ‘super-droplets‘ . Each super-droplet represents several real particles, and its size is adjusted as the general dynamic equation evolves.
Previously, SDM has been developed and applied to ice nucleation and cloud microphysics simulating precipitation. In this talk, we present the preliminary results of a zero-dimensional model of a homogeneous cooling fireball. We apply SDM to represent the formation and evolution of fallout including homogeneous nucleation, condensation, and coagulation.
Between 1945 and 1970, about two hundred atmospheric nuclear tests were performed by the United States. We compare size distributions for airburst nuclear tests from the SDM simulation, previous theoretical calculations, and field measurements taken during the tests.
Prepared by LLNL under Contract DE-AC52-07NA27344.

A Conservative Finite Volume Framework and Implementation for Aggregation, Breakage and Growth Problems on Arbitrary Grids.

Presenter: Daniel O'Sullivan, Imperial College London, Dep. of Mechanical Engineering 
Presentation Description: The Population Balance Equation (PBE) governs the transport and evolution of aerosol and particulate flows, and accounts for numerous physical, chemical and particulate processes. PBE-based modelling is used in various applications of aerosol science including atmospheric flows, flame synthesis of nanoparticles, industrial pigments and pharmaceutical production. The numerical solution of the PBE is a significant challenge, due to the numerous particulate processes involved. Recently, a mass-conservative finite volume scheme was derived by Liu & Rigopoulos (2019) and tested on aggregation-growth problems. In this work, the method of Liu & Rigopoulos (2019) is extended to accommodate Breakage as well as Fragmentation processes. The method works by constructing an aggregation-breakage map which allows for mass to be correctly allocated to and from finite volume cells in the particle size domain without the loss of mass or artificial conditioning of the size-distribution. Furthermore, a more accurate integration scheme is used (3-point 2D Gaussian Quadrature) to evaluate the various aggregation and breakage kernels, and this new work presents a unified framework for all particulate processes. Key advantages of this method are: Accurate prediction of size-distributions with relatively few points, applicability on arbitrary grids as well as speed and robustness in the context of coupling with full CFD simulations.
Liu, A. & Rigopoulos, S. (2019),Combustion and Flame

Aerosol parameter value tuning within an uncertainty framework

Presenter: Leighton Regayre, University of Leeds 
Presentation Description: We densely sample model uncertainty by perturbing multiple process parameters related to clouds, aerosols, radiation and precipitation. Our ensemble of model variants samples regional uncertainty in anthropogenic aerosol emissions and includes high time resolution data for optimal model-measurement comparison. We rule out observationally implausible model variants (parameter combinations) using satellite-derived values of cloud properties such as cloud droplet concentrations, albedo and cloud fraction. The remaining sample of observationally plausible model variants have narrower parameter value ranges, which guide the model tuning. Additionally, we use statistical methods to identify the relative importance of processes that cause model uncertainty in aerosol-cloud interaction forcing and related outputs. Processes that cause the remaining uncertainty after observational constraint inform the types of measurements that will be of most value in reducing the uncertainty further.

Machine Learning (ML) Clustering Algorithms as a Receptor Modelling Technique for Source Apportionment of Particulate Matter

Presenter: Manoranjan Sahu, Indian Institute of Technology Bombay
Presentation Description: Identification of aerosol sources in a particular region is a priority steps for assessing quality control and health impacts of aerosols. A source apportionment study was conducted on two PM2.5 data sets, viz. two carbon fractions and eight temperature-resolved carbon fractions collected during Cincinnati Childhood Allergy and Air Pollution Study (CCAAPS). This study's objective was to evaluate two clustering algorithms: k-Means clustering (kMC) and spectral clustering (SC) as potential receptor models for source apportionment. The application of kMC produced unsatisfactory results, but the results obtained from SC demonstrated a significant correlation with the results obtained using positive matrix factorization (PMF). The clustering results obtained were associated with practical evidence available in the literature. SC identified six source factors on analyzing two carbon fractions and seven factors from eight temperature-resolved carbon fractions. The sources (source contribution in parentheses) identified are: combustion (45.9±3.66%) and secondary sulfate (11.4±1.09%), vegetative/wood burning (17.5±1.46%), diesel (10.6±0.92%) and gasoline (3.6±0.33%) vehicles, soil/crustal (2.07±0.2%), traffic (9.3±0.81%), and metal processing (8.8±0.72%). The source profiles obtained using SC also show similarity with the profiles derived using PMF. In summary, this study presented a basic framework for applying Machine Learning algorithms for source apportionment analysis and SC as a potential receptor model technique for SA.

Machine learning for aerosols, clouds, and climate: from prototyping to model implementation

Presenter: Sam Silva, Pacific Northwest National Laboratory
Presentation Description: The activation of aerosol into cloud droplets is an important step in the formation of clouds, and strongly influences the radiative budget of the Earth. Explicitly simulating aerosol activation in Earth system models is challenging due to the computational complexity required to resolve the necessary physical processes and their interactions. Here, we explore how machine learning techniques can be used to provide accurate and fast predictions of aerosol activation.
We evaluate a set of emulators of a detailed cloud parcel model using a machine learning regression approach. We find that physically informed emulators can best reproduce the parcel model predictions at higher accuracy than many existing parameterizations currently in use. We further implement the most skillful emulator into an Earth system model and evaluate its performance and impact on the climate.

Learning coagulation processes with combinatorially-invariant neural networks

Presenter: Justin Wang, University of Illinois at Urbana-Champaign
Presentation Description: Simulating the evolution of coagulating aerosols or droplets is a key problem in atmospheric sciences. Current methods to do so require the use of computationally expensive microphysics models and assume knowledge of a coagulation kernel. Unless the aggregation process can be captured by one of the few well-known coagulation kernels (e.g., Brownian motion), the kernel may not be accurately known, and this introduces considerable uncertainties in the prediction. We present a proof of concept for modeling coagulation processes using a novel combinatorially-invariant neural network (CiNN) architecture. Using coarse-grained data from a high-detail particle-resolved aerosol simulation, we introduce two methods of training, one supervised on the directly observed coagulation events and the other using state snapshots.These two training methods out-perform a standard neural network by a factor of 20 and are competitive with a traditional state-of-the-art sectional model with a known analytical kernel. Furthermore, the introduced CiNN algorithm is physically inspired and interpretable. We will discuss applications of this approach in learning coarse-grained coagulation models for multi-species aerosols and for learning coagulation models from observed size distribution data.

Improving global aerosol models through emulation

Presenter: Duncan Watson-Parris, University of Oxford
Presentation Description: One of the most pressing climate questions we currently face is that of the effect of anthropogenic aerosol on the climate system, particularly through their interactions with clouds.
Here I will describe work using advances in machine learning (ML) to aid in climate model emulation for the reduction of this uncertainty in entire GCMs. I will introduce a general climate model emulation framework and describe how it has been applied to reduce uncertainty in both direct and indirect forcing. I will also describe ongoing work to emulate the HAM aerosol scheme, and how these two emulation approaches can be combined and extended in the future to help improve our understanding of the effect of aerosol on our changing climate.

Step-wise Hydration of Organics and Electrolytes in Atmospheric Aerosols

Presenter: Anthony Wexler, University of California, Davis 
Presentation Description: Water molecules in aqueous solutions partition between a free water pool and a bound pool that hydrates each solute. At high water activities, corresponding to high relative humidity, the solutes have a maximum number of bound water molecules. As the water activity decreases, water molecules unbind and join the free water pool. The equilibrium between water molecules in the free and bound pools is related to the Gibbs free energy of the water-water hydrogen bond in the free water pool compared to the water-solute Gibbs free energy in the bound water pool. A mathematical model of this step-wise hydration/dehydration process faithfully captures the solute and water activity in highly concentrated solutions representative of atmospheric aerosols.

AMORE: Automated Mechanism Reduction in Atmospheric Chemistry

Presenter: Forwood Wiser, Columbia University
Presentation Description: There is a need for new approaches for the systematic reduction of complex atmospheric chemical mechanisms for use in large-scale models. We have developed the Columbia University Atmospheric Chemistry Model Reduction (AMORE) algorithm, an automated tool for flexibly generating accurate condensed chemical mechanisms. Using the principles of graph theory, without running the model, AMORE analyzes a reaction network and eliminates reaction pathways with throughputs that do not meet a specified threshold. Those pathways are discarded, along with species left unconnected to the rest of the network, until the desired model size or error tolerance is reached. Species of particular interest are protected from elimination. Inputs for AMORE are sampled from a range of conditions relevant to the continental U.S., generated using GEOS-Chem. We have applied AMORE to develop an updated condensed gas-phase isoprene oxidation mechanism and tested its performance against the full mechanism of Wennberg et al. (2018) in the photochemical box model, F0AM.

Multi-phase chemistry surrogate modeling with elemental mass conservation using a Neural ODE

Presenter: Xiaokai Yang, Civil and Environmental Engineering, University of Illinois at Urbana Champaign 
Presentation Description: Modeling atmospheric chemistry and physics is computationally expensive and limits the widespread use of air quality models. This computation cost arises mainly from solving high-dimensional systems of stiff differential equations. Previous work has demonstrated the promise of machine learning (ML) to accelerate air quality model simulations but has suffered from numerical instability during long-term simulations. This may be because previous ML studies have not enforced any physical constraints on the system or used implicit or adaptive time-stepping, which are common methods for improving numerical stability in traditional dynamical systems modeling. Here we use a Neural ODE to create a multi-phase chemistry surrogate model which conserves the mass of all chemical elements. First, we couple the near-explicit gas-phase Master Chemical Mechanism (MCM) with the state-of-art Particle-resolved Monte Carlo Model for Simulation Aerosol Interactions and Chemistry (PartMC-MOSAIC). The combined multi-phase chemical mechanism can be made to conserve mass and serves as a detailed reference for emulation. Second, we use an autoencoder to learn a compressed representation of the state of the chemical system to reduce memory usage and computational expense. Third, we use a neural network—customized to conserve the mass of each chemical element—to predict the derivative of concentration which is integrated forward in time by an implicit ODE solver. This Neural ODE framework is able to stably simulate the autoregressive evolution of the chemical system over extended time periods. To generate the training dataset for machine learning, we created scenarios with Latin hypercube sampled initial conditions, and ran each scenario of the coupled model for 24 hours. The resulting ML model accelerates computation by 50 times on CPU and 700 times on GPU, as compared to the reference model. Preliminary results of training on this small dataset are promising; we are currently working to scale up our training workflow to realize the full potential of this approach.

Quantifying the structural uncertainty of the aerosol mixing state representation in MAM4

Presenter: Zhonghua Zheng, Columbia University and University of Illinois at Urbana-Champaign
Presentation Description: Aerosol mixing state is an important emergent property that affects the aerosol radiative forcing and aerosol-cloud interactions, but it has not been easy to constrain this property globally. Our study aims to verify the global distribution of aerosol mixing state represented by four-mode version of the modal aerosol module (MAM4) that is used in E3SM and CESM. We defined three aerosol mixing state indices that describe (1) the mixture of optically absorbing and non-absorbing species, (2) the mixture of primary carbonaceous and non-primary carbonaceous species, and (3) the mixture of hygroscopic and non-hygroscopic species. These mixing state indices were used to assess CESM-MAM4 as compared with benchmark simulations from the particle-resolved stochastic aerosol model PartMC-MOSAIC. A machine learning surrogate model is used to emulate the PartMC-MOSAIC results. We show that CESM-MAM4 and PartMC-MOSAIC produce up to 70% different values of mixing state index. The difference appears to be zonally structured, with the modal model predicting a more internally mixed aerosol at low latitudes, and a more externally mixed aerosol at high latitudes. Our study quantifies potential model bias in simulating mixing state in different regions, and provides insights into potential improvements to model process representation for a more realistic simulation of aerosols.


Emissions and Sources

Chairs: Kelley Barsanti, UC Riverside & Delphine Farmer, Colorado State University


Estimating emissions of sea salt aerosols in polar regions using satellite data and chemical transport modeling

Presenter: Hannah Horowitz, University of Illinois at Urbana-Champaign 
Presentation Description: The processes contributing to cold season sea salt aerosol (SSA) in polar regions and their relative importance in the Arctic vs. Antarctic are still under debate, preventing further progress in modeling polar aerosol and climate. Two main hypotheses are cracks in the sea ice, or leads, through which sea spray can release salt, and saline snow on the surface of sea ice lofted by wind, or blowing snow. Field measurements in the Arctic near Utqiaġvik suggest a dominant role for leads, while those in Antarctica support blowing snow as a major source. Both sources are affected by a changing climate and may result in additional feedbacks of sea ice loss and thinning. Here we use satellite data and the 3-D chemical transport model GEOS-Chem to estimate the potential contribution of each source to SSA emissions in polar regions. We first estimate the monthly areal extent contributing to each source in fall through spring from satellite-derived lead area fraction from AMSR-E and lead frequency from MODIS, and first-year sea ice area estimated by combining the EASE-Grid Sea Ice Age product and the NOAA/NSIDC Climate Data Record sea ice concentration. We develop a parameterization of SSA emissions from sea ice leads and calculate monthly emissions using the Harmonized Emissions Component (HEMCO) within GEOS-Chem. We compare emissions results using lead area derived from the two satellite datasets and an alternate approach defined by the MERRA-2 reanalysis sea ice fraction.

Modeling the indoor ultrafine particle dynamics for indoor episodic emission sources considering coagulation effect

Presenter: Su-Gwang Jeong, Soongsil University
Presentation Description: Human exposure to airborne particulate matter (PM) is one of the significant environmental risks threatening human health. In residential environments, particles are typically emitted with high concentrations with episodic events associated with human activities. Once particles are released from a source, the particles undergo particle dynamic processes such as coagulation, deposition, evaporation and condensation. Such aerosol processes results in time-varying changes of particle number concentrations and size distributions. This study analyzed particle size and concentration dynamics associated with indoor source events by combining an analytical aerosol dynamic model and experimental data collected from full-scale houses. The results show relative contributions of particle transformation processes (i.e., coagulation, deposition and ventilation) over the source emission and decay periods. The results reveal that coagulation and indoor surface deposition are two dominant processes responsible for temporal changes in particle size and concentration in indoor environments. The dynamic changes of coagulation, deposition and ventilation during the emission and decay period notably vary with source type and indoor activity.

Challenges in modeling emissions from fires in the past, present-day, and future

Presenter: Loretta Mickley, Harvard University
Presentation Description: Fires play a key role in shaping the chemical composition of the atmosphere, thus potentially influencing Earth's climate. Smoke particles can also adversely affect human health. As warming temperatures dry out fuel in the coming decades, fire activity and smoke exposure are expected to increase in many regions. Accurately representing fire emissions in models, however, is challenging. Estimates of past fire emissions vary greatly, depending in large part on assumptions of burning practices carried out by indigenous peoples. Inventories of present-day fire emissions, typically derived from satellite data, also show large regional discrepancies due to differences in the interpretation of satellite imagery, the emissions factors assumed for smoke components, and the adjustments made for small, short-lived, or obscured fires. Efforts to predict future smoke emissions, on the other hand, rely on incomplete knowledge of the relationships between fire activity and meteorology and between vegetation and climate change. This talk discusses new approaches to modeling fire emissions and the implications for climate and air quality. Topics include (1) improved estimates of preindustrial biomass burning in the Southern Hemisphere and the consequences for aerosol radiative forcing, (2) development of an updated emissions inventory for crop residue burning in India, and (3) application of a dynamic vegetation model to estimate future fire emissions in the western United States.

Primary and secondary aerosols in and above deciduous forests

Presenter: Allison Steiner, University of Michigan 
Presentation Description: Forest canopies can act as an aerosol generator, emitting primary biological particles such as pollen and biogenic VOC precursors to form secondary organic aerosol in the atmosphere. Most deciduous trees in the northern mid-latitudes employ anemophilous (wind-driven) pollen to disperse genetic material, which can provide a source of organic matter to the atmospheric boundary layer. While emitted pollen grains are rather large (10-50 microns), they can rupture under certain atmospheric conditions and create fine particles. Deciduous forests also emit a suite of biogenic volatile organic compounds, which play an important role in the formation of secondary pollutants in the atmosphere such as ozone and secondary organic aerosols (SOA). The use of 1-D column models can delineate where and when these precursor emissions can form local SOA above the forest canopy, and these process estimates can be used in conjunction with 3D regional chemistry models to interpret the role of local versus synoptic scale aerosol formation. Model results are compared to ground-based field observations at two sites within the continental United States to elucidate how vertical gradients of these compounds in and above the forest canopy influence the formation of secondary organic aerosols.

Predicting SOA Formation from the Athabasca Oil Sands in Northern Alberta

Presenter: Craig Stroud, Air Quality Research Division, Science and Technology Branch, Environment and Climate Change Canada
Presentation Description: The Athabasca Oil Sands are a large source of secondary organic aerosol (SOA) in Canada, with a production level similar to major urban areas, despite low concentrations of combustion markers (Liggio et al., 2016). Following recent intensive aircraft campaigns evaluating emissions and transformation of pollutants from the Oil Sands in 2013, multiple possible sources of primary semi- and intermediate-volatility organic compounds (P-S/IVOC) have been characterized, however with divergent measured volatility distributions.
This work presents chemical box modelling and 3-D chemical transport modelling results for the range of volatility distributions published in the literature and applied in the emission processing. The box modelling also allows an exploration of the dependence of predictions on SOA precursor oxidation and yields, molecular fragmentation, aging rate constants, and organic-organic phase separation. The box model was most sensitive to the parameterizations of primary IVOC oxidation, VOC oxidation and of multi-generational oxidative aging. While the sensitivities to the parameterizations for phase separation and fragmentation were weaker.
The 3-D modelling is performed with ECCC's GEM-MACH model at 2.5-km grid resolution. The model results are compared with aircraft observations from Lagrangian flight patterns following the air parcels as they age downwind of the Oil Sand facilities. In this presentation, the method selected to calculate the P-S/IVOC emissions will be described and preliminary 3-D modelling results will be shown.

Wintertime aerosol formation in mountain basins

Presenter: Caroline (Carrie) Womack, NOAA Chemical Sciences Laboratory
Presentation Description: Mountain basins in the western US are frequently subject to temperature inversions that trap emissions near the surface, leading to high levels of primary and secondary pollutants, including ozone and particulate matter. The amount and speciation of these pollutants is a function of numerous parameters that are specific to each basin, including emission profiles, NOx-to-VOC ratios, aerosol pH, temperature, the duration of the inversion events, and many others. We present here observations from several recent field campaigns, including the 2017 Utah Winter Fine particulate Study (UWFPS) and the 2012-2104 Uintah Basin Winter Ozone Studies (UBWOS), and plans for the upcoming AQUARIUS campaign, designed to elucidate the mechanisms of aerosol formation during wintertime. We show that the role of oxidants is a critical parameter in the buildup of particulate matter, particularly ammonium nitrate aerosol. We contrast these campaigns with others that have focused on particulate matter pollution in different regions of the U.S. and the world.


Air Quality Modeling for Health and Regulatory Assessments

Chairs: Ajith Kaduwela, California Air Resources Board & Ben Murphy, US EPA


Investigating Wildfire Plumes with low-cost air sensors:  Implications for grid-based photochemical Air Quality modeling.

Presenter: Ajith Kaduwela, Air Quality Research Center, University of California, Davis
Presentation Description: The grid-based photochemical modeling of wildfires is very difficult due to our current inability to accurately model emissions and plume rise. In addition, the height-specific pollutant concentrations in the 3rd dimension, needed for model evaluation, are very hard to measure. With that in mind, we have designed and built a low-cost air sensor package that can be flown on a drone.
We have measured the pollutant concentrations in a plume originated from the Dixie wildfire (and nearby fires) as a proof of concept. Concentrations measured with low-cost sensors cannot be used in an absolute sense but they could be used in a relative sense.  We will discuss the possibility of using measurements from low-cost air sensors in the evaluation of grid-based photochemical air quality models.

Integrating reactive organic carbon emissions into the Community Regional Atmospheric Chemistry Multiphase Mechanism (CRACMM)

Presenter: Havala Pye, US Environmental Protection Agency
Presentation Description: The chemical mechanism of an atmospheric chemical transport model like the Community Multiscale Air Quality (CMAQ) system contains a condensed set of reactions that describe the interactions between emitted organic compounds and nitrogen oxides as well their reaction products. The Community Regional Atmospheric Chemistry Multiphase Mechanism (CRACMM) builds on the history of the Regional Atmospheric Chemistry Mechanism, Version 2 (RACM2) and aims to couple gas- and particle-phase chemistry by treating the entire pool of atmospheric reactive organic carbon (ROC) relevant to present-day emissions. Here, we develop CRACMM species to represent the total emissions of ROC, considering the OH reactivity, ability to form ozone and secondary organic aerosol (SOA), and other properties of individual emitted compounds. Compared to RACM2, we reduce the number of traditional volatile organic carbon species and increase the number of oxygenated and semivolatile to intermediate volatility precursors in the mechanism. In addition, we add explicit hazardous air pollutants to better characterize exposures relevant for human health. We contrast emissions of ROC in 2002 and 2017 from the EPA's Air QUAlity TimE Series project and their treatment in CRACMM to illustrate how precursors to ozone, SOA, and other endpoints are expected to propagate through the system. The CRACMM mechanism species will be operationalized in chemical transport models in future work.

Particulate Matter Regulatory Modeling Applications

Presenter: Heather Simon, US Environmental Protection Agency
Presentation Description: Photochemical air quality models are an essential tool for simulating secondarily formed pollutants such as ozone and particulate matter (PM).  These models are used in regulatory assessments for a variety of purposes.  Retrospective modeling analyses can serve to provide spatially complete air quality fields for inputs to health assessment and enable the evaluation of model predictions with real-world measurements. For policy planning purposes, photochemical models can be used to predict how air pollution will change in the future as a result of implementation of air pollution management practices or other regulations to reduce emissions.  In addition, models are often used to attribute air pollution to specific sources or source categories to aid in air quality planning and development of effective regulations.  Instrumented modeling techniques and reduced form models can provide tools for more efficiently estimating expected air pollution changes due to various policy options.  Here we review some of EPA's recent regulatory and policy-focused modeling assessments with a focus on applications with a PM component.  We use these examples to highlight key research needs for PM regulatory modeling applications.

Issues with condensable organics in European PM2.5 emissions; key messages and follow-up from an expert workshop organised by EMEP MSC-W

Presenter: David Simpson, Norwegian Meteorological Institute and Chalmers University of Technology
Presentation Description: European emission reduction strategies depend strongly upon national reporting estimates as submitted to the European Monitoring and Evaluation Programme (EMEP). There are however significant differences in the way in which countries report such emissions. Some countries include "condensable" organics in their emissions, while others exclude such organics. This difference is important since inclusion of organics can sometimes increase reported PM2.5 emissions by factors of 3 or sometimes more. For some source sectors, even the national emission experts are unsure as to the degree to which condensables are included.
These differences lead to severe problems in the modelling of organic aerosol and hence PM2.5, and to imbalances in the policies recommended for emission reductions.
In March 2020 EMEP MSC-W organised an expert workshop (funded by the Nordic Council of Ministers, NMR) to discuss these issues in detail. Experts from the fields of emission inventory development, emissions measurements, atmospheric modelling, and policy agreed to a number of key messages, and outlined steps that could be taken to improve the consistency of PM2.5 reporting in Europe.
This talk will outline some of the main points from these discussions and the resulting report, as well as discuss the changes and efforts which have taken place within the EMEP system.
Report: Simpson et al., https://emep.int/publ/reports/2020/emep_mscw_technical_report_4_2020.pdf

Pathways of China's PM2.5 air quality 2015-2060 in the context of carbon neutrality

Presenter: Qiang Zhang, Department of Earth System Science, Tsinghua University
Presentation Description: Clean air policies in China have substantially reduced particulate matter (PM2.5) air pollution in recent years, primarily by curbing end-of-pipe emissions. However, reaching the level of the World Health Organization (WHO) guidelines may instead depend upon the air quality co-benefits of ambitious climate action. Here, we assess pathways of Chinese PM2.5 air quality from 2015 to 2060 under a combination of scenarios that link global and Chinese climate mitigation pathways (i.e. global 2◦C- and 1.5◦C-pathways, National Determined Contributions (NDC) pledges and carbon neutrality goals) to local clean air policies. We find that China can achieve both its near-term climate goals (peak emissions) and PM2.5 air quality annual standard (35 μg/m3) by 2030 by fulfilling its NDC pledges and continuing air pollution control policies. However, the benefits of end-of-pipe control reductions are mostly exhausted by 2030, and reducing PM2.5 exposure of the majority of the Chinese population to below 10 μg/m3 by 2060 will likely require more ambitious climate mitigation efforts such as China's carbon neutrality goals and global 1.5◦C-pathways. Our results thus highlight that China's carbon neutrality goals will play a critical role in reducing air pollution exposure to the level of the WHO guidelines and protecting public health.


Poster Presenters


Modeling atmospheric brown carbon in the GISS ModelE Earth system model

Poster Presenter: Maegan DeLessio, Columbia University
Lightning Talk Description: Brown carbon (BrC) is an absorbing organic aerosol primarily emitted by the combustion of biomass and biofuel. While field and laboratory studies have shown that BrC exhibits light absorption unique from black carbon (BC) and organic carbon (OC) aerosols, the climate forcing of BrC is still poorly understood as it is not well represented in most Earth system models (ESMs). BrC undergoes photochemical transformation, or aging, in the atmosphere, resulting in changing absorption. This makes it particularly difficult to incorporate into ESMs, as most are limited to tracers with invariant optical properties. BrC emitted from biomass burning was introduced in the GISS ModelE One-Moment Aerosol (OMA) module by creating three BrC tracers, each with different refractive indices and absorbance. Aging of BrC was simulated through mass exchange between lighter (less absorbing) and darker (more absorbing) BrC tracers. Through this modeling approach, we were able to successfully incorporate primary brown carbon with a chemical aging scheme into the GISS climate model. Sensitivity tests of BrC and OC refractive indices can be used to determine the radiative effect of including dynamic BrC tracers, rather than just attributing some absorbance to OC as previously modeled.

A predictive viscosity model for aqueous electrolytes and mixed organic-inorganic aerosol phases

Poster Presenter: Joseph Lilek, McGill University
Lightning Talk Description: Aerosol viscosity is determined by mixture composition and temperature, with a key influence from relative humidity in modulating aerosol water content. We have extended the thermodynamics-based group-contribution model AIOMFAC-VISC to predict viscosity for aqueous electrolyte solutions and aqueous organic-inorganic mixtures. For aqueous electrolyte solutions, our semi-empirical approach uses a physical expression based on Eyring's absolute rate theory, and it includes simultaneously fitted ion-specific expressions, arguably making this approach more predictive than that of other models. This also enables viscosity calculations for aqueous solutions containing an arbitrary number of cation and anion species, including mixtures that have never been experimentally studied. Predictions achieve an excellent level of accuracy while also providing physically meaningful extrapolations to extremely high electrolyte concentrations, which is essential in the context of microscopic aqueous atmospheric aerosols. For organic-inorganic mixtures, multiple methods were tested to couple the AIOMFAC-VISC electrolyte model with its existing aqueous organic model. We discuss the best performing mixing models implemented in AIOMFAC-VISC for reproducing viscosity measurements of aerosol surrogate systems. Finally, we present advantages and drawbacks of different model design choices and associated computational costs of these methods, of importance for use of AIOMFAC-VISC in dynamic simulation.

Leveraging novel higher-order sensitivity analysis to assess the predictability of background SOA concentrations with a state-of-the-science aerosol mechanism

Poster Presenter: Jiachen Liu, Drexel University
Lightning Talk Description: Secondary organic aerosols (SOAs) are generated by complex transformations of organic molecule emissions and are often a significant portion of atmospheric particulate matter (PM) concentrations. High PM concentrations can lead to regional haze and pose health risks to the public. The Regional Haze Rule was last revised in 2017 and the goals were made based on estimated background concentrations of SOAs. One approach to estimate the background concentrations is to use the sensitivity coefficients of aerosol formation with respect to precursor concentrations. Previous methods of calculating the sensitivity coefficients include the finite difference method, the direct decoupled method (DDM), and the adjoint method. The finite difference method suffers from truncation and cancellation errors, while DDM and the adjoint method are relatively difficult to implement and update in chemical transport models. Here, we demonstrate an alternative approach to calculating first- and second-order sensitivities of SOA with respect to precursor concentrations while seeking to evaluate the robustness of the SOA mechanism for such applications. This method is an operator overloading approach which calculates the exact first- and second-order sensitivity coefficients of select precursor concentrations through one single run of the model (Fike & Alonso, 2011). We applied this method in the Community Multiscale Air Quality (CMAQ) model v.5.3.2 to formulate the CMAQ-hyd model. We strive to evaluate and trace back the SOA concentrations to their background anthropogenic-free levels. The sensitivity results could guide policy makers on evaluating the current Regional Haze Rule and researchers on assessing the limitations of current multiphase chemical mechanisms.

Accelerating multiphase-chemical kinetics modelling through machine learning metamodels

Poster Presenter: Marcel Müller, ETH Zurich
Lightning Talk Description: To study the heterogeneous chemistry of atmospheric aerosols, accurate description of multiphase chemical kinetics is vital, but demands the use of kinetic multi-layer models in which mass transport and chemical reaction are coupled explicitly. Unfortunately, these models are computationally too expensive to be used as submodules in global models and their computational cost may limit inverse modelling approaches.
In our study, we generated two types of computationally less expensive surrogate models based on the KM-SUB multi-layer model of aerosol surface and bulk chemistry. A polynomial chaos expansion metamodel was built in the Matlab framework UQLab and a neuronal network metamodel was built with the Python package Keras. Both surrogate models were trained with KM-SUB results from parameter sets covering a wide range of typical atmospheric and laboratory reaction conditions. As example system, we used the well-studied system of oleic acid ozonolysis.
The surrogate models are compared in terms of accuracy and training conditions (training sample size and computation time). To demonstrate their benefit, the new metamodels are used in a global sensitivity analysis revealing a high model sensitivity to changes in the gas phase reactant concentration and in the particle radius. Additionally, a Bayesian inversion study using experimental data and Markov Chain Monte Carlo sampling is presented to illustrate how metamodels may be used to constrain model parameters.

Estimating size-specific particulate matter exposure in China based on machine learning parameterizations

Poster Presenter: Ruqian Miao, Peking University
Lightning Talk Description: Ambient particulate matter (PM) is a leading contributor to premature mortality globally. The greater health risk of submicron aerosol (PM1) compared to fine aerosol (PM2.5) indicates the need for quantifying the size-specific PM exposure, which however is still unclear in China, especially for PM1. In this study, we used machine learning techniques to simulate the mass ratio between PM1 and PM2.5 (PM1/PM2.5), for which the values vary with the impacts from physical and chemical processes that have not yet been well quantified. The machine learning techniques were trained using the simultaneous observations of meteorological parameters, mixing ratios of gas pollutants, and mass concentration of PM1 and PM2.5 as well as their chemical components that measured by aerosol mass spectrometer (AMS) in Beijing in the four seasons. The concentration of PM1 in China was derived from the machine-learned PM1/PM2.5 and the modeled PM2.5 mass concentration provided by atmospheric chemical transport model GEOS-Chem. The simulation results were evaluated against comprehensive datasets of the observed PM1 concentrations and the values of PM1/PM2.5 from AMS and optical monitors. The generally good model performance highlights that the machine learning parameterizations trained by observations in Beijing can be applied to most regions in China where the influence from dust is limited, which is useful for estimating the mortality associated with size-specific PM exposure.

Process level modeling of vertically resolved new-particle formation at the Southern Great Plains observatory

Poster Presenter: Samuel O'Donnell, Colorado State University
Lightning Talk Description: New particle formation (NPF) is a significant source of aerosol number concentrations in the atmosphere. The vertical profile of NPF across the boundary-layer (BL), including the mixed layer (ML) and residual layer (RL), has been poorly characterized, despite the upper ML and RL often having favorable conditions for NPF, such as low condensation/coagulation sinks, low temperatures, high relative humidity, rapid photochemistry, and elevated precursor species.
To study the vertical profile of nucleation, we developed a 1D column version of the SOM-TOMAS model, which simulates gas-phase oxidation reactions, thermodynamics, gas-particle partitioning, aerosol microphysics, vertical mixing, and nucleation. We simulated days from the DoE HI-SCALE campaign, where there were airborne and surface observations of 11 sustained NPF events observed at the ground, 1 of which was observed by the aircraft as starting near the top of the BL layer prior to being observed by surface instruments.
The model is able to predict the occurrence of nucleation and generally can predict the aerosol mass and number concentrations. For several of the modeled nucleation events, vertical mixing from the upper ML or RL to the surface was necessary to explain the observed aerosol size distribution. Finally, we find that vertical profiles of temperature, precursor concentrations as well as the evolution of the ML explain much of the vertical profile of nucleation in the model.

Assessment of Air Pollution Tolerance, Anticipated Performance, and Metal Accumulation Indices: Implications for Roadside Planting for Improving the Quality of Urban Air

Poster Presenter: Saif Shahrukh, University Of Dhaka 
Lightning Talk Description: In cities, roadside vegetation is exposed to air pollutants, including a wide variety of particulates-borne toxic compounds. An investigation was undertaken to assess the tolerance or sensitivity of four roadside trees (Ficus benghalensis, Ficus religiosa, Mangifera indica, and Polyalthia longifolia) towards air pollutants, including particulates. The four species were sampled from four different locations of Dhaka, Bangladesh. Air pollution tolerance index (APTI) was assessed using the total chlorophyll content, ascorbic acid content, relative water content, and the pH of the extract from the leaves of the studied plants. The studied biochemical parameters in the leaves of the selected tree species were found to vary among the sites. Air pollution tolerance index of the investigated plants ranged from 10.31 to 12.51 meaning they were either sensitive or intermediately tolerant. The results indicated that these evergreen species are good indicators of air pollution and can be used as an early warning tool for air pollution level that is harmful to human health. Anticipated performance index (API) was also calculated for all the species where some socioeconomic and biological characteristics were taken into consideration. Total metal accumulation capacities of different plants were evaluated using the metal accumulation index (MAI) and Ficus benghalensis was found to have the highest MAI value (13.60).

Capturing of Particulate Matter from Ambient Air by Four Evergreen Tree Species in Dhaka, Bangladesh

Poster Presenter: Saif Shahrukh, University Of Dhaka
Lightning Talk Description: An investigation was conducted to assess the particulate matter (PM) removal capacity of four common roadside trees (Ficus benghalensis, Ficus religiosa, Mangifera indica, and Polyalthia longifolia) grown at four locations in Dhaka, Bangladesh. Gravimetric analyses were performed separately to quantify PM in three size fractions (0.2-2.5 µm, 2.5-10 µm, and 10-100 µm) deposited on surfaces and trapped in waxes. Among the species studied, the deposited mass of PM was highest on Ficus benghalensis. The mean PM load on plant foliage was significantly greater in the polluted sites compared with the control site (p<0.05). Most of the PM accumulated on plant foliage belonged to the large fraction size (10-100 µm). Species-wise significant differences were also found among the sites with respect to total PM, surface PM, and wax-embedded PM (p<0.05). The amount of wax deposited on the leaves of plants grown in these sites also differed (p<0.05). A positive correlation was found between wax-embedded PM of diameter 0.2-2.5 µm and the amount of waxes. Ficus benghalensis was found to be the most effective with respect to total PM accumulation. On the other hand, Mangifera indica was found to be the most effective accumulator of wax-related PM and seems to be the best species for traffic-related sites, where organic substances from vehicle exhausts are present in high concentrations.

Pollution in Kyiv by PM 2.5 and 10 during 2020-2021 years: characteristics, and distribution by districts

Poster Presenter: Yuliia Yukhymchuk, Institute of Physics of the National Academy of Sciences of Ukraine, Kyiv, Ukraine; Department for Atmospheric Optics and Instrumentation, Main Astronomical Observatory, Kyiv, Ukraine
Lightning Talk Description: Aerosol in the atmosphere significantly affects human health. The quantitative characteristics of these influences still have not yet been determined properly in large cities of Ukraine. In this contest, aerosol pollution research is important, because the contamination air of particular matter (PM) is the main indicator of clean air. In Kyiv city, Ukraine, the team of scientists created four stations of AirVisual in early March 2020 to evaluate the level of PM2.5 load using the established AirVisual mini-network (https://www.iqair.com/). The station locations were selected to cover the city territory as much as possible. The last fifth station was installed in June 2020. Every station is equipped with an AirVisual Pro sensor, which is an air quality monitor that works on the principle of improved laser technology for the identification of aerosol particles and provides high-precision measurements of the concentration of PM2.5 particles. The data of AirVisual sensors were verified against the reference HORIBA analyzer by simultaneous PM measurements. The AirVisual network provides information about air quality in Kyiv in real-time. Furthermore, the two years set of data is using for analyzing the dynamic and changing of PM2.5 and PM10 particles content and CO2 concertation. The AirVisual data also were compared with Kyiv AERONET station, the satellite instruments Aqua/MODIS, and VIIRS observations for usual condition and biomass burning extremal events in the March - April 2020. This period, which indicated extraordinarily high PM2.5 aerosol contamination and low air quality for Kyiv city and Kyiv region is considered in details.