Atmospheric aerosols affect climate change by altering the energy balance of the atmosphere, and public health due to their variable chemical composition, size, and shape. While the formation of secondary organic aerosol (SOA) from gas phase precursors is relatively well understood, it does not account for the abundance of SOA observed during field measurements. Recently it has become apparent that in-aerosol aqueous chemical reactions likely provide some of the missing sources of SOA production, and many studies of aqueous phase processes are underway. This work explores the fates of the simplest 2-oxocarboxylic acids, glyoxylic acid (GA) and pyruvic acid (PA), under simulated solar irradiation in the aqueous phase. Field measurements have revealed that mono-, di-, and oxocarboxylic acids are abundant species present in atmospheric waters. Of particular interest are 2-oxocarboxylic acids because their conjugated carbonyl moieties result in significant UV-visible absorption above 300 nm, allowing absorption of sunlight in the lower troposphere, thereby initiating radical photochemistry and leading to formation of SOA. In Chapter 2 of this work, GA is demonstrated to primarily undergo α-cleavage, producing CO, CO2, formic acid, and the key SOA precursor glyoxal. Trace amounts of oxalic acid and tartaric acid are also quantified. Additionally, the dark thermal aging of glyoxylic acid photoproducts, studied by UV-visible and fluorescence spectroscopies, reveals that the optical properties of the solutions are altered radically by the glyoxal produced. The optical properties display periodicity during photolytic-dark cycles, reflecting behavior expected for aerosols during nighttime and daytime cycles. In contrast, Chapter 3 shows that PA photoreacts via a proton-coupled electron transfer (PCET) mechanism that produces CO2 and organic acids of increased complexity with 6 to 8 carbons. A combination of analytical techniques including 1H and 13C NMR; 13C gCOSY NMR; mass spectrometry; chromatography; and isotope substitutions allows the organic products to be identified as: 2,3-dimethyltartaric acid; 2-hydroxy-2-((3-oxobutan-2-yl)oxy)propanoic acid; and the quasi-intermediate 2-(1-carboxy-1-hydroxyethoxy)-2-methyl-3-oxobutanoic acid. In Chapter 4, PA irradiation is also shown to consume dissolved oxygen so fast that solutions become depleted within a few minutes depending on reaction conditions. This fast process directly produces the atmospheric oxidant singlet oxygen, which enhances the oxidizing capacity of the atmosphere. Additionally, PA photochemistry only proceeds under very acidic conditions (pH ≤ 3.5), like those in most atmospheric aerosols. Finally, we require a thorough understanding of the behavior of 2-oxocarboxylic acids at the air-water interface of aerosols because much of the GA and PA present in the atmosphere is produced in the gas phase and needs to partition into the aqueous phase to undergo photoreaction. Therefore, Chapter 5 uses surface sensitive online electrospray ionization mass spectrometry (OESI-MS) to demonstrate that carboxylic acids delivered from the gas phase onto the surface of aqueous microdroplets display enhanced acidities relative to bulk water solutions. This work demonstrates that aqueous photolysis is a very competitive atmospheric fate for both GA and PA. It also shows that these photoreactions are capable of contributing substantially to SOA formation by building chemical complexity and forming oxidants directly.

This thesis describes the investigation of three aspects of the formation of secondary organic aerosol (SOA): * Aerosol formation from mixed precursors * Global modelling of SOA formation * Modelling of dynamics of SOA formation based on empirical data collected from smog chamber experiments. The formation and growth processes of secondary organic aerosol were investigated using smog chamber experimentation and modelling techniques to gain a better understanding of the application of SOA yield values in modelling both SOA mass and dynamics. Published SOA yields from a range of volatile organic compounds (VOCs) are used to model SOA mass on a local, regional or global scale, based on the assumption that the SOA yield of a mixture is the sum of the yields of the components. Experimental investigations into SOA yield from mixtures of VOC revealed potential uncertainties that would result from applying these yields to systems containing multiple VOCs. SOA formation in systems of toluene or m-xylene, compared with systems of these VOCs and propene, have shown that the introduction of propene (which has a zero SOA yield) to smog chamber photo-oxidations of toluene or m-xylene delays the formation and suppresses the overall yield of SOA from 450 to 90 µg m-3 ppm-1 for the toluene system and from 325 to 125 µg m-3 ppm-1 for the mvxylene system compared with systems of individual species without propene. The SOA partitioning yield data also indicates that partitioning of species to existing aerosol is suppressed in the mixed systems. Gas-phase modelling of these experiments showed that potential SOA species were expected to be formed sooner due to the increased system reactivity provided by propene. The observed delay in SOA nucleation, similar consumption rates of toluene and m-xylene in both the single and mixed systems and the gas-phase modelling results suggest that the addition of propene to hydrocarbon SOA systems modifies the gas-phase chemistry leading to the formation of potential SOA species from toluene and m-xylene. This result calls into question the bulk and partitioning yield values that have been published for pure substances as well as the validity of applying individual VOC yields to VOC mixture. Application of SOA yields to the global scale provides estimates of annual global SOA formation, global contributions from various VOCs and regional SOA distributions. Two SOA modules, using bulk and partitioning yield methods, were added to a global atmospheric chemical transport model, MOZART-2. The bulk yield method, representing the maximum possible global SOA burden, gave an annual production of 24.5 Tg of SOA, which is slightly lower than previous estimates (30 - 270 Tg yr-1). The partitioning method, which gives a more realistic estimate of SOA formation, produced 15.3 Tg yr-1; the biogenic fraction (13.6 Tg yr-1) compares to a previous estimate of biogenic SOA of 18.5 Tg yr-1 and 2.5 to 44 Tg yr- 1 using the partitioning method. Anthropogenic SOA contributions of 1.1 Tg yr-1 from MOZART-2 compared to recent estimates of 0.05 -2.62 Tg yr-1. SOA production was found to be dependent on oxidant availability and VOC emissions in South America and Asia. The partitioning method produced significantly less SOA due to limited availability of OC. Thepartitioning method also produced a peak SOA concentration of 10 µg m-3 over South America in September and showed that SOA is at maximum production for most of the year in Asia and Europe. The two SOA formation methods also provides data to analyse the restrictions to SOA formation in particular regions, based on the maximum amount of SOA able to form (bulk yield method) and the more realistic partitioning estimate from the same region. Limitations to SOA formation in a particular region can be attributed to deficiencies in OC availability or VOC oxidant concentrations. Comparisons to limited observational and modelled data suggest that the MOZART-2 SOA model provides a good representation of global averaged SOA. SOA mass concentrations, predicted by models such as MOZART-2, can be used in part to model the dynamics of an SOA population (e.g. size of particles, number concentrations etc.). Aerosol properties such as size and number concentration can then be used to estimate their effect on climate and health. The explicit representation of the processes that affect aerosol dynamics, such as nucleation, condensation, evaporation and coagulation can be complex and use significant computational resources. Simplification of the discrete coagulation equation and empirical coagulation coefficients for continuum and non-continuum regime diffusion kinetics provided a simplified method of coagulation capable of predicting the evolution of inert sodium chloride aerosol in chamber experiments. A variable coagulation coefficient (linked to the mean particle number concentration of each experiment) was developed. This method is an empirical surrogate for the standard coefficient corrections applied to Brownian based diffusion in the continuum regime to account for the different kinetic effects within the transition and free molecular diffusion regimes. This method removes the need for calculating individual coefficients for each particle interaction. Estimates of modeluncertainty show that within uncertainty limits the model provides a good representation of experimental data. Correlation and index of agreement (IOA) calculations revealed good statistical agreement between modelled and experimental. Some experiments showed degrees of coagulation under prediction using the variable coefficient technique. Investigations into the effect of aerosol type and size, temperature and humidity may be necessary to refine the variable coefficient calculation technique. The model showed little sensitivity to model time step and is capable of high resolution representation of the aerosol. Mass concentration is conserved within the model whereas some error due to numerical diffusion within the number concentrations results from the bin sectioning technique used. The simplicity of this sectioning method over other methods and the minimal effect of numerical diffusion establishes a simplified method of modelling relative to the high resolution of the aerosol distribution the model achieves. It is suggested that the efficiency improvements introduced by the approaches used in developing this model provide an efficient ultra-fine coagulation modelling for atmospheric models. A semi-empirical model for SOA dynamics (SPLAT) incorporating coagulation, nucleation, condensation and evaporation was developed. The aim of the model and the development process was to predict, with high resolution and minimal computational expense, the formation and growth of SOA given a SOA mass input as a function of time. The average size distribution profile from chamber experimental data was used as part of the nucleation module. This technique provided an alternative method of representing the particle distribution compared to those models that assume a single diameter of nucleated particle or a fixed log-normal mode for the entire evolution of SOA. All SPLAT simulations assume organic nucleation events within the experiments modelled, although it is stilluncertain whether they occur in the atmosphere. The modelled nucleation events have produced a single nucleation burst, a result of immediate domination of condensation as soon as nucleation occurs. This deficiency is likely to be a result of the assumption of free molecular diffusion for condensation. The rate of condensation, calculated at every time step, is based on the aerosol size distributed surface area and the particle-size-dependent saturation mass concentrations. The SPLAT coagulation module was a version of the model developed in Chapter 6. Comparisons between experimental and modelled data showed good agreement. These comparisons revealed the shortcomings in the nucleation module while a statistical analysis of the modelled and experimental data has shown SPLAT to be effective in modelling a range of SOA systems. The complexity introduced in modelling aerosol dynamics in high resolution is offset in SPLAT by efficiency improvements due to the insensitivity of the model to time step size and simplified methods of bin sectioning, nucleation, coagulation, condensation and evaporation. Published SOA yields can be applied to predict SOA mass at local, regional or global scales. Although previously unreported uncertainties in these yields have been shown to exist, the MOZART-2 global chemical transport model has shown that SOA mass concentration can be predicted with reasonable quality, considering the scale of the model and limited observational data. These global scale SOA mass predictions can be used purely for global burden and occurrence, or as the input for modelling the dynamics of an aerosol population, which is significant for estimating an aerosol population's effect on climate change and health. SOA mass concentrations from chamber experiments were used as input to a SOA dynamics model. This model (SPLAT) then predicted the evolution of particle number concentrations and size within these experiments based on this mass input. Application of the dynamics model to the output of the MOZART-2 model could then provide a comprehensive global scale SOA modelling package.