Development and application of a three-dimensional aerosol chemical transport model,PMCAMx

A three-dimensional chemical transport model (PMCAMx) is used to simulate PM mass and composition in the eastern United States for a July 2001 pollution episode. The performance of the model in this region is evaluated, taking advantage of the highly time and size-resolved PM and gas-phase data collected during the Pittsburgh Air Quality Study (PAQS). PMCAMx uses the framework of CAMx and detailed aerosol modules to simulate inorganic aerosol growth, aqueous-phase chemistry, secondary organic aerosol formation, nucleation, and coagulation. The model predictions are compared to hourly measurements of PM_2.5 mass and composition at Pittsburgh, as well as to measurements from the AIRS and IMPROVE networks. The performance of the model for the major PM_2.5 components (sulfate, ammonium, and organic carbon) is encouraging (fractional errors are in general smaller than 50%). Additional improvements are possible if the rainfall measurements are used instead of the meteorological model predictions. The modest errors in ammonium predictions and the lack of bias for the total (gas and particulate) ammonium suggest that the improved ammonia inventory used is reasonable. The significant errors in aerosol nitrate predictions are mainly due to difficulties in simulating the nighttime formation of nitric acid. The concentrations of elemental carbon (EC) in the urban areas are significantly overpredicted. This is a problem related to both the emission inventory but also the different EC measurement methods that have been used in the two measurement networks (AIRS and IMPROVE) and the actual development of the inventory. While the ability of the model to reproduce OC levels is encouraging, additional work is necessary to confirm that that this is due to the right reasons and not offsetting errors in the primary emissions and the secondary formation. The model performance against the semi-continuous measurements in Pittsburgh appears to be quite similar to its performance against daily average measurements in a wide range of stations across the Eastern US. This suggests that the skill of the model to reproduce the diurnal variability of PM_2.5 and its major components is as good as its ability to reproduce the daily average values and also the significant value of high temporal resolution measurements for model evaluation.     

Atmospheric hydrogen peroxide: Evidence for aqueous-phase formation from a historic perspective and a one-year measurement campaign

H₂O₂ is produced in the atmospheric gas phase only through a single pathway, the HO₂ radical recombination. Its main role has been identified in oxidizing SO₂ dissolved in hydrometeors to sulphate. Thus aqueous-phase chemistry has been considered to be a main sink (apart from dry deposition and scavenging) but rarely a source of H₂O₂ despite early findings of its heterogeneous and aqueous-phase production. The aim of this paper is to discuss the atmospheric budget of H₂O₂ from the multiphase chemistry approach with special emphasis on new sources other than gas-phase HO₂ recombination. After providing a brief historic view on H₂O₂ chemistry, often unknown to young atmospheric chemists but important for a complete understanding, the results of a one-year study of simultaneous measurements of H₂O₂ in rain and air are presented that show strong evidence for aqueous-phase H₂O₂ formation. Implications for future changes in atmospheric chemistry are discussed from the viewpoint of an “interfacial chemistry”.     

On the oxidation of SO2 to sulfate in atmospheric aerosols

This paper presents the results of a study to investigate the atmospheric oxidation of sulfur dioxide (SO₂). A detailed model of gas-phase chemistry, aerosol thermodynamics and aerosol chemistry is employed to simulate atmospheric sulfate formation. The calculations indicate that, in addition to the gasphase oxidation by hydroxyl (OH) radicals, SO₂ oxidation in aqueous aerosols may also contribute significantly to sulfate formation. Reactions of SO₂ with hydrogen peroxide (H₂O₂) and O₂ (catalyzed by Fe³⁺ and Mn²⁺) are identified as principal aqueous-phase oxidation mechanisms. The results of this study confirm the conclusions drawn from the analysis of ambient aerosol data qualitatively. However, some discrepancies also exist between the results of our modeling study and field data. Such discrepancies emphasize the need for the collection of ambient data for a more rigorous and quantitative evaluation of atmospheric aerosol models.     

Air Quality Modeling in the South Coast Air Basin of California: What Do the Numbers Really Mean?

Abstract

This study evaluates air quality model sensitivity to input and to model components. Simulations are performed using the California Institute of Technology (CIT) airshed model. Results show the impacts on ozone (O₃) concentration in the South Coast Air Basin (SCAB) of California because of changes in: (1) input data, including meteorological conditions (temperature, UV radiation, mixing height, and wind speed), boundary conditions, and initial conditions (ICs); and (2) model components, including advection solver and chemical mechanism. O₃ concentrations are strongly affected by meteorological conditions and, in particular, by temperature. ICs also affect O₃ concentrations, especially in the first 2 days of simulation. On the other hand, boundary conditions do not significantly affect the absolute peak O₃ concentration, although they do affect concentrations near the inflow boundaries. Moreover, predicted O₃ concentrations are impacted considerably by the chemical mechanism. In addition, dispersion of pollutants is affected by the advection routine used to calculate its transport. Comparison among CIT, California Photochemical Grid Model (CALGRID), and Urban Airshed Model air quality models suggests that differences in O₃ predictions are mainly caused by the different chemical mechanisms used. Additionally, advection solvers contribute to the differences observed among model predictions. Uncertainty in predicted peak O₃ concentration suggests that air quality evaluation should not be based solely on this single value but also on trends predicted by air quality models using a number of chemical mechanisms and with an advection solver that is mass conservative.     

The sensitivity of a 3-dimensional Eulerian model to uncertainties in emissions and dry deposition velocities

A 3-dimensional STEM-I Eulerian, regional scale transport/chemistry model for SO₂ and sulfate is used to investigate the relationships between model predictions and the resolution and uncertainty in the emissions and dry deposition velocities. Model results for conditions of 4 July 1974 where the emissions are time modulated, where the emissions are increased and decreased, and where the dry deposition velocities are increased and decreased are presented and discussed.     

Analysis of emission databases for regional models

General procedures for adapting emission inventories to regional models and for studying the impact of differences in inventories on model predictions are outlined. To illustrate the methods, analysis of two inventories which are still being validated is presented. The inventories together satisfy current requirements for the NCAR regional acid deposition model (RADM). These include anthropogenic emissions of SO₂, sulfate aerosol, NO, NO₂, NH₃ and volatile organic compounds (VOC) in 10 reactivity classes, from United States and Canadian point and area sources on 80-km grid resolutions, for weekend and weekday seasonally representative days on a diurnal basis during the 1980–1982 period. Application of checking procedures, designed by our group to screen for subtle anomalies not identified at previous stages of quality assurance employed by the inventory developers, resulted in adjustments primarily to VOC emissions. Comparisons of the modified inventories, which provide an indication of uncertainties in emissions due to variations in inventory development procedures, revealed differences in the eastern United States total daily emissions to be at most on the order of 5 % for SO_x, and NO_x, 20% for VOC and 85% for NH₃. Studies of the impact of inventory differences on predictions of RADM were conducted for the 22–24 April 1981 period, which was monitored as part of the Oxidation and Scavenging Characteristics of April Rains program. Event total wet sulfate deposition differed by 10% or less while midday O₃ concentrations differed by 1% or less for individual grids over the modeling domain.     

Simulation of aerosols and gas-phase species over Europe with the Polyphemus system. Part II: Model sensitivity analysis for 2001

This paper presents a multi-pollutant sensitivity study of an air quality model over Europe with a focus on aerosols. Following the evaluation presented in the companion paper, the aim here is to study the sensitivity of the model to input data, mathematical parameterizations and numerical approximations. To that end, 30 configurations are derived from a reference configuration of the model by changing one input data set, one parameterization or one numerical approximation at a time. Each of these configurations is compared to the same reference simulation over two time periods of the year 2001, one in summer and one in winter. The sensitivity of the model to the different configurations is evaluated through a statistical comparison between the simulation results and through comparisons to available measurements. The species studied are ozone (O₃), nitrogen dioxide (NO₂), sulfur dioxide (SO₂), ammonia (NH₃), coarse and fine aerosol particles (PM_c and PM_2.5), sulfate, nitrate, ammonium, chloride and sodium.For all species, the modeled concentrations are very sensitive to the parameterization used for vertical turbulent diffusion and to the number of vertical levels. For the other configurations considered in this work, the sensitivity of the modeled concentration to configuration choice varies with the species and the period of the year. O₃ is impacted by options related to boundary conditions. PM_c is sensitive to sea-salt related options, to options influencing deposition and to options related to mass transfer between gas and particulate phases. PM_2.5 is sensitive to a larger number of options than PM_c: sea-salt, boundary conditions, heterogeneous reactions, aqueous chemistry and gas/particle mass transfer. NO₂ is strongly influenced by heterogeneous reactions. Nitrate shows the highest variability of all species studied. As with NO₂, nitrate is strongly sensitive to heterogeneous reactions but also to mass transfer, thermodynamic related options, aqueous chemistry and computation of the wet particle diameter. While SO₂ is mostly sensitive to aqueous chemistry, sulfate is also sensitive to boundary conditions and, to a lesser extent, to heterogeneous reactions. As with nitrate, ammonium is largely impacted by the different configuration choices, although the sensitivity is slightly lower than for nitrate. NH₃ is sensitive to aqueous chemistry, mass transfer and heterogeneous reactions. Chloride and sodium are impacted by sea-salt related options, by options influencing deposition and by options concerning the aqueous-phase module.     

The potential of statistical state space models in urban ozone forecasting

State space models for tropospheric urban ozone prediction are introduced and compared with linear regression models. The linear and non-linear state space models make accurate short-term predictions of the ozone dynamics. The average prediction error one hour in advance is 7 μg/m³ and increases logarithmically with time until it reaches 26 μg/m³ after 30 days. For a given sequence of solar radiation inputs, predictions converge exponentially with a time scale of 8 hours, so that the model is insensitive to perturbations of more than 150 μg/m³ O₃. The slow increase of the prediction error in addition to the uniqueness of the prediction are encouraging for applications of state space models in forecasting ozone levels when coupled with a model that predicts total radiation. Since a radiation prediction model will be more accurate during cloud-free conditions, in addition to the fact that the state space models perform better during the summer months, state space models are suitable for applications in sunny environments.     

Distribution of reactive nitrogen species in the remote free troposphere: data and model comparisons

The available reactive nitrogen measurements from the global free troposphere obtained during the period of 1985–1995 have been compiled and analyzed. The species of interest are NO, NO_x (NO+NO₂), NO_y, PAN, HNO₃ and O₃. Data extending to 13 km have been gridded with a 5°×5° horizontal and 1 km vertical resolution. The data have been divided into two seasons, namely “Winter” and “Summer” depending upon the time and location of the observations. Data described here as well as additional analysis have also been archived and are accessible on-line through the World Wide Web at: http://george.arc.nasa.gov/∼athakur. Global maps of the reactive nitrogen species distribution are produced in a form that would be most useful for the test and evaluation of models of tropospheric transport and chemistry. Limited comparisons of the observed reactive nitrogen species data with predictions by 3-D global models were performed using three selected models. Significant model to model as well as data to model differences were frequently observed. During summer, models tended to underpredict NO (−25 to −60%) while significantly overpredicting HNO₃ (+250 to +400%) especially in the upper troposphere. Similarly, the seasonal HNO₃ variations predicted by some models were opposite to those observed. PAN was generally overpredicted, especially in the upper troposphere, while NO_y was underpredicted. Ozone on average was better simulated but significant deviations at specific locations were evident. By comparing model predictions with observations, an overall quantitative assessment of the accuracy with which these three models describe the global distribution of measured reactive nitrogen species is provided. No reliable trend information for any of the reactive nitrogen species was possible based on the presently available data set. The reactive nitrogen data currently offer only a limited spatial and temporal coverage for the validation of global models.     

Evaluation of the community multiscale air quality (CMAQ) model version 4.5: Sensitivities impacting model performance; Part II—particulate matter

This paper is Part II in a pair of papers that examines the results of the Community Multiscale Air Quality (CMAQ) model version 4.5 (v4.5) and discusses the potential explanations for the model performance characteristics seen. The focus of this paper is on fine particulate matter (PM_2.5) and its chemical composition. Improvements made to the dry deposition velocity and cloud treatment in CMAQ v4.5 addressing compensating errors in 36-km simulations improved particulate sulfate (SO₄²⁻) predictions. Large overpredictions of particulate nitrate (NO₃⁻) and ammonium (NH₄⁺) in the fall are likely due to a gross overestimation of seasonal ammonia (NH₃) emissions. Carbonaceous aerosol concentrations are substantially underpredicted during the late spring and summer months, most likely due, in part, to a lack of some secondary organic aerosol (SOA) formation pathways in the model. Comparisons of CMAQ PM_2.5 predictions with observed PM_2.5 mass show mixed seasonal performance. Spring and summer show the best overall performance, while performance in the winter and fall is relatively poor, with significant overpredictions of total PM_2.5 mass in those seasons. The model biases in PM_2.5 mass cannot be explained by summing the model biases for the major inorganic ions plus carbon. Errors in the prediction of other unspeciated PM_2.5 (PM_Other) are largely to blame for the errors in total PM_2.5 mass predictions, and efforts are underway to identify the cause of these errors.     

Characterization and source presumption of wintertime submicron organic aerosols at Saitama,Japan, using the Aerodyne aerosol mass spectrometer

The chemical composition and size distribution of submicron aerosols were analyzed at a suburban site at Saitama, Japan, in the winter of 2004/2005, using an Aerodyne aerosol mass spectrometer. Although organics and nitrate were the dominant species during the sampling period, a large fraction of sulfate was observed at the accumulation mode when mass loading was low and wind speed was high. The size distributions of m/z 44 (mostly CO₂⁺) and sulfate aerosols during periods of high wind speed showed remarkable similarities in the accumulation mode, indicating that oxygenated organics were aged aerosols and internally mixed with sulfate. Ozone concentrations were also increased during these high wind speed periods although nighttime (e.g., 12/17 2004), indicating that the oxygenated compounds were strongly influenced by transported and aged air masses. The diurnal profiles of ultrafine-mode organics and hydrocarbon-like organic aerosols (HOA) were similar to NO_X derived from traffic and other combustion sources. The temporal variation of oxygenated organic aerosols (OOA) agreed well with that of nitrate as a secondary aerosol tracer, and the diurnal profile of the OOA fraction of organics increased during the day associated with higher UV light intensity. The result of time and size-resolved chemical composition of submicron particles indicated that the OOA is associated with both photochemical activity and transboundary pollution, and ultrafine-mode organic and HOA aerosols are mainly associated with combustion sources.     

Sulfur and Carbon Cycling in a Flue Gas Desulfurization Sludge Disposal Site

Abstract

Products of a power plant flue gas desulfurization scrubber are discharged into a pond as sludge consisting of calcite (initial δ¹³C 3.2–3.8‰), gypsum (initial δ³⁴S 7.6–8.6‰), and aqueous solution. Reducing conditions exist below a boundary that appears to move vertically as a function of changes in pond water level. Under reducing conditions, bacteria partially reduce aqueous sulfate to low-δ³⁴S sulfide, consuming organic carbon and generating low-δ¹³C bicarbonate. Under oxidizing conditions, sulfide is converted to sulfate, leading to calcite dissolution, gypsum precipitation, and isotopic re-equilibration of remaining calcite with dissolved bicarbonate near the pond surface. The gypsum has δ³⁴S near 6‰, and calcite has δ¹³C as low as -1.7‰; the changes from initial values correspond to predictions based on isotopic balance and reaction stoichiometry. The pond largely contains the products of bacterial reduction. After the pond is abandoned, these products may adversely affect attempts to revegetate the site. Future bacterial reduction may be best controlled by dewatering and limiting the supply of organic matter in percolating surface water.     

Acidity of raindrop by uptake of gases and aerosol pollutants

Below-cloud raindrops acidification simulated with a simple model incorporating gas–liquid equilibriums, gas-phase mass transfer, and catalyzed SO₂ oxidation in aqueous phase with uptake of gases and scavenging of particles. Ionic contents of various species in raindrops of different size and pH are computed using one-dimensional time-variant model. The model results are based on SO₂ and NH₃ absorption and collection of calcium aerosols by raindrops with various collection mechanisms. Aqueous concentrations of (SO₂)_l and (NH₃)_l and their ionic components in raindrops are found to be increased with the fall distance from cloud base and decrease of drop size. The overall magnitude of pH enhances with the increase in drop size and transient position of raindrops in the atmosphere below the cloud base. The elevated ionic calcium in raindrops by impaction of calcium aerosols of higher inertia neutralizes the acidic components. Acidic ion contents in smaller droplets are found to be significant and resulted pH of raindrop increases with the size and neutralizing potential of alkaline species. The pH values of rainwater contents of predominant size raindrops in bulk samples corresponding to various rainfall intensities are higher as against the individual non-evaporating smaller raindrops. Results are important in view of the impact of showers on earth surfaces during rain containing large number of smaller droplets as compared to the acidification studies of bulk rainwater.     

Partitioning of nitrate and ammonium between the gas and particulate phases during the 1997 IMADA-AVER study in Mexico City

The partitioning of nitrate and ammonium between the gas and particulate phases is studied combining available equilibrium models and measurements taken in Mexico City during the 1997 IMADA-AVER field campaign. Based on this analysis, there are no significant differences in model predictions, but some discrepancies exist between predictions and observations. The inclusion of crustal elements in the modeling framework improves agreement of model predictions for particulate nitrate against measurements by approximately 5%. Although some equilibrium aerosol models do not explicitly treat crustal elements, these species can be treated as equivalent concentrations of sodium. Atmospheric equilibrium models predict daily average PM_2.5 nitrate concentrations within 20% of the IMADA-AVER measurements at the MER site. Six-hour average PM_2.5 nitrate concentrations are predicted within 30–50% on average except for the afternoon sampling periods (12:00–18:00 h). Investigating the possible sources of these discrepancies, it appears that a dynamic instead of an equilibrium approach is more suitable in reproducing aerosol behavior during these afternoon periods. By applying the Multicomponent Aerosol Dynamic Model (MADM), model performance in predicting concentrations of particulate nitrate significantly improves during the afternoon periods.     

Evaluation of a Gaussian and a Lagrangian model against a roadside data set,with emphasis on low wind speed conditions

The evaluation of the high percentiles of concentration distributions is required by most national air quality guidelines, as well as the EU directives. However, it is problematic to compute such high percentiles in stable, low wind speed or calm conditions. This study utilizes the results of a previous measurement campaign near a major road at Elimäki in southern Finland in 1995, a campaign specifically designed for model evaluation purposes. In this study, numerical simulations were performed with a Gaussian finite line source dispersion model CAR-FMI and a Lagrangian dispersion model GRAL, and model predictions were compared with the field measurements. In comparison with corresponding results presented previously in the literature, the agreement of measured and predicted data sets was good for both models considered, as measured using various statistical parameters. For instance, considering all NO_x data (N=587), the so-called index of agreement values varied from 0.76 to 0.87 and from 0.81 to 1.00 for the CAR-FMI and GRAL models, respectively. The CAR-FMI model tends to slightly overestimate the NO_x concentrations (fractional bias FB=+14%), while the GRAL model has a tendency to underestimate NO_x concentrations (FB=−16%). The GRAL model provides special treatment to account for enhanced horizontal dispersion in low wind speed conditions; while such adjustments have not been included in the CAR-FMI model. This type of Lagrangian model therefore predicts lower concentrations, in conditions of low wind speeds and stable stratification, in comparison with a standard Lagrangian model. In low wind speed conditions the meandering of the flow can be quite significant, leading to enhanced horizontal dispersion. We also analyzed the difference between the model predictions and measured data in terms of the wind speed and direction. The performance of the CAR-FMI model deteriorated as the wind direction approached a direction parallel to the road, and for the lowest wind speeds. However, the performance of the GRAL model varied less with wind speed and direction; the model simulated better the cases of low wind speed and those with the wind nearly parallel to the road.     

Analysis of air pollution data at a mixed source location using boosted regression trees

This paper explores the use of boosted regression trees to draw inferences concerning the source characteristics at a location of high source complexity. Models are developed for hourly concentrations of nitrogen oxides (NO_X) close to a large international airport. Model development is discussed and methods to quantify model uncertainties developed. It is shown that good explanatory models can be developed and further, allowing for interactions between model variables significantly improves the model fits compared with non-interacting models. Methods are used to determine which variables exert most influence over predicted concentrations and to explore the NO_X dependency for each. Model predictions are used to estimate aircraft take-off contributions to total concentrations of NO_X and determine how these predictions are affected by annual variations in meteorological conditions and runway use patterns. Furthermore, the results relating to the aircraft contributions to total NO_X concentration are compared with those from a more detailed independent field campaign. Finally, we find empirical evidence that plumes from larger aircraft disperse more rapidly from the point of release compared with smaller aircraft. The reasons for this behaviour and the implications are discussed.     

Application of positive matrix factorization in source apportionment of particulate pollutants in Hong Kong

An advanced algorithm called positive matrix factorization (PMF) in receptor modeling was used to identify the sources of respirable suspended particulates (RSP) in Hong Kong. The compositional data obtained from the Hong Kong Environmental Protection Department from 1992 to 1994 were analyzed. The species analyzed in this study are Al, Ca, Mg, Pb, Na⁺, V, Cl⁻, NH₄⁺, SO₄²⁻, Br⁻, Mn, Fe, Ni, Zn, Cd, K⁺, Ba, Cu, and As. Unlike the conventional receptor modeling algorithm, factor analysis PMF only generates non-negative source profiles. To eliminate sulfate from such factors where it is not physically plausible, special penalty terms were included in the model so that sulfate concentrations could be selectively decreased in specified factors. A 9-factor model containing non-zero sulfate concentrations in three factors gives the most satisfactory source profiles. Ammonium sulfate, chloride depleted marine aerosols and crustal aerosols are the three non-zero sulfate sources. Other factors are marine aerosols, non-ferrous smelters, particulate copper, fuel oil burning, vehicular emission and bromide/road dust. The last two sources can be combined as a single source of vehicle/road dust. The compositional profiles of these factors were also developed. The mass profiles obtained can be improved by further refinement of distribution of sulfate in the sources.     

Modeling smog chamber measurements of incremental reactivities of volatile organic compounds

A series of experiments performed at the GM chamber facility provided useful data for the evaluation of two current chemical mechanisms used in airshed models (SAPRC97 and SAPRC93 mechanisms) and a test of their predictions of maximum incremental reactivities which describe the change in ozone caused by adding a small amount of a compound to a polluted urban mixture under high-NO_x conditions. In general, the SAPRC97 detailed mechanism performed well in simulating the volatile organic compound (VOC) reactivity experiments for most test species; however, it had a tendency to underpredict incremental reactivities. For base-case runs containing a nine-component urban-surrogate mixture under high-NO_x conditions, where maximum concentrations of either O₃ or the smog produced (SP=the initial NO oxidized plus the ozone produced) were not attained during a 12-h irradiation, the SAPRC97 performed well while the SAPRC93 underestimated SP or O₃ significantly. Under low-NO_x conditions where SP or O₃ maximums were attained, the SAPRC97 as well as the SAPRC93 underpredicted SP or O₃ for runs containing the urban-surrogate mixture. Simulations of incremental reactivity experiments and special chamber runs showed that the SAPRC97 mechanism performed poorly for n-octane and some aromatic isomers such as ethylbenzene and p-xylene, while it performed well for other aromatic isomers such as toluene, m-xylene and 1,3,5-trimethylbenzene. Although, additional chamber data for aromatic isomers is needed to further clarify the parameterized chemical mechanisms for aromatic isomers, the newer SAPRC97 mechanism appears to be much improved over the older SAPRC93 mechanism for simulating aromatic chemistry.     

Nitrate-contaminated groundwater remediation by combined autotrophic and heterotrophic denitrification for sulfate and pH control: batch tests

Groundwater remediation was evaluated for combined autotrophic and heterotrophic denitrification under high (154 mg/L as CaCO₃) and low (95 mg/L as CaCO₃) alkaline conditions. Two levels of acetate (47 and 94 mg/L) and ethanol (24 and 48 mg/L) were added to the reactors. Obtained denitrification rates were 2.89, 2.58, 3.55, 1.96, and 2.0 mg-N/L?·?h for high alkaline conditions, whereas under low alkaline conditions has given 2.36, 1.94, 2.47, 2.74, and 2.29 mg-N/L?·?h for control, 47 and 94 mg/L acetate, and 24 and 48 mg/L ethanol, respectively. Nitrite was accumulated for controls but reactors with acetate and ethanol did not accumulate nitrite. Acetate and ethanol addition decreased sulfate to nitrate ratios in the range of 4.5–7.58 for high alkaline conditions (12.77 for control) and 4.43–6.78 for low alkaline conditions (7.90 for control). Acetate was more efficient compared with ethanol in controlling sulfate production and pH maintenance.     

Analysis of air quality within a street canyon using statistical and dispersion modelling techniques

The dispersion model, ADMS-Urban, alongside the statistical modelling technique, generalized additive modelling, have been used to predict hourly NO_x and nitrogen dioxide (NO₂) concentrations at a busy street canyon location and the results compared with measurements. Generalized additive models (GAMs) were constructed for NO₂ and NO_x concentrations using input data required to run ADMS-Urban. Bivariate polar plots have been produced from the wind flow (speed and direction) and pollution data (measured and predicted concentrations) to provide further information regarding the complex wind-pollutant interactions in an urban street canyon. The predictions made with the GAMs show excellent agreement with measured concentrations at this location, reproducing both the magnitude of NO_x and NO₂ concentrations and also the wind speed-wind direction dependence of pollutant sources within the canyon. However, the predictions made with ADMS-Urban under-estimated the measured NO_x by 11% and NO₂ by 21% and there are clear differences in the bivariate polar plots. Several sensitivity tests were carried out with ADMS-Urban in an attempt to produce predictions in closer agreement to those measured at Gillygate. Increasing the primary NO₂ fraction in ADMS-Urban (from 10% to 20%) had a considerable effect on the predictions made with this model, increasing NO₂ predictions by ∼20%. However, the bivariate plots still showed major differences to those of the measurements. This work illustrates that generalized additive modelling is a useful tool for investigating complex wind-pollutant interactions within a street canyon.