2023 Conference Agenda

Patrick Campbell, Irena Ivanova, Paul Makar, Zachary Moon, Wei-Ting Hung, Barry Baker, Margaret Marvin, Beiming Tang, Quazi Rasool, Youhua Tang, Fanglin Yang, Raffaele Montuoro, and Rick Saylor

Using an experimental version of NOAA’s next-generation, operational air quality forecasting system, known as “Online-CMAQ”, here we demonstrate that advancing beyond the widely used "big-leaf" approach to include both in-canopy photolysis attenuation and vertical turbulent transport can improve the model’s near-surface ozone predictions. Accounting for in-canopy effects reduces the surface ozone mean bias against ground-based U.S. EPA AirNow observations by up to 50% over the Contiguous U.S. The impact of the in-canopy effects on precursor gasses and aerosols both near and downstream from wildland fires in the U.S are also examined, as well as the resulting changes in PM2.5 concentrations. Second, we introduce the “canopy-app” (https://github.com/noaa-oar-arl/canopy-app), which computes vertically resolved parameterizations of in-canopy effects (e.g., canopy wind flow, vertical turbulence, light attenuation, and biogenic emissions) based currently on prescribed plant-shape distribution functions applied to various vegetation types. Two canopy-app uses are showcased: i) offline, using point data and gridded atmospheric analyses/forecasts, and ii) online, integrated into box models and NOAA’s Unified Forecast System (UFS). Along with its stand-alone analysis capability, the ultimate goal is to make the canopy-app available as a model component at NOAA, which can be used for atmospheric composition applications at various spatial and temporal scales.

Hantao Wang, Marc Serre, Kazuyuki Miyazaki, Jason Sun, Xiang Liu, Antje Inness, Zhen Qu, J. Jason West

Ground-level ozone is a significant air pollutant known for its detrimental effects on human health and agricultural productivity. Accurately assessing ozone exposure and its health implications remains a crucial objective. Recently, various global ozone datasets have employed diverse methodologies, such as chemical reanalysis, machine learning, and geostatistics. This study aims to comprehensively compare and evaluate six global ozone datasets for health-relevant metrics, aiming to understand differences and identify inconsistencies and biases. Specifically, we compare three chemical reanalysis datasets, two machine learning datasets, and the geostatistics dataset developed at UNC. We analyze monthly average DMA8 (daily maximum 8-hour average) and the yearly health-related metric OSDMA8 (ozone season daily maximum 8-hour average). Ozone datasets are compared with one another spatially (at 0.1° × 0.1° resolution) focusing on 2016, for temporal trends of ozone among datasets, and using statistical indicators between datasets, such as correlations. We also evaluate the performance of each ozone mapping product with respect to ground-level observations from the latest Tropospheric Ozone Assessment Report (TOAR) II database. The findings reveal significant differences between the various ozone mapping products, surpassing our initial expectations. The analysis of annual mean trends in populated areas reveals consistent upward temporal trends across five datasets, despite a maximum span of 10 ppb in global average concentrations. It is noteworthy that datasets incorporating ground-based observations as input exhibit lower mean concentrations compared to those that do not. Significant differences emerge in the rainforest regions of Africa and South America. These disparities can be attributed to variations in input data and fusion methodologies. Generally, areas with higher ground-level ozone concentrations and greater monitoring site coverage demonstrated better agreement between ozone map products and ground-based data (TOAR-II). This study underscores that global ozone maps created using different methods yield different results, and are not yet converging to agreement.

T. Nash Skipper, Emma L. D’Ambro, Barron H. Henderson, Colleen B. Baublitz, Forwood Wiser, V. Faye McNeill, Glenn M. Wolfe, Jason M. St. Clair, Thomas Hanisco, Havala O.T. Pye

Formaldehyde (HCHO) is a hazardous air pollutant and is an indicator for total reactive organic carbon (ROC) abundance since it is directly emitted and formed as a secondary product from oxidation of many ROC species. HCHO also plays a role in the formation of criteria pollutants such as ozone and secondary particulate matter. HCHO simulated by the Community Multiscale Air Quality (CMAQ) chemical transport model is typically biased low compared to surface observations. Here, the representation of HCHO in the Community Regional Atmospheric Chemistry Multiphase Mechanism (CRACMM) is evaluated and potential improvements are explored. The bias in HCHO is characterized using a combination of ground-based HCHO observations from surface monitoring networks, satellite-based observations of HCHO tropospheric column abundance from TropOMI, and observations from the NASA DC-8 aircraft during the summer of 2019. In addition, the potential role of missing secondary production of HCHO is explored by benchmarking CRACMM against the Master Chemical Mechanism (MCM) and by implementing an updated version of the Automated MOdel REduction (AMORE) isoprene chemistry representation.

Yu Morino1, Kei Sato1, Yosuke Sakamoto1,2, Jiaru Li1,2, Satoru Chatani1, Hikari Shimadera3, Kyo Kitayama1, Yoshizumi Kajii1,2,4

1 National Institute for Environmental Studies, Japan

2 Kyoto University, Japan

3 Osaka University, Japan

4 Qingdao University, PR China

Atmospheric O3 concentration has not decreased over the past 20 years in Japan despite substantial reduction in emissions of NOx (50%) and VOC (52%). It has been reported in China that the reduction of atmospheric aerosol can be a reason for the increase of O3 during the 2010s, because heterogeneous uptake of peroxy radicals (HO2) by aerosol particles suppresses O3 production.

In this study, we evaluated the effect of HO2 uptake by aerosol on atmospheric O3 concentrations by performing numerical simulations of CMAQ v5.3.2 over East Asia in 2000, 2010, and 2018. Simulations with and without consideration of HO2 uptake showed that the daily maximum 8-h average O3 concentrations were reduced by 3 – 5 ppbv over East Asia during spring and summer. The simulated aerosol concentrations were substantially reduced over East Asia from 2010 to 2018, and the effect of HO2 uptake on surface O3 was consequently reduced during this period.

In addition, we preliminarily estimated the spatial distributions of HO2 uptake coefficients by constructing emission data of transition metals (copper and iron) and calculating a resistance model with the simulated spatial distributions of aerosol composition (e.g., copper concentration and pH). The estimated HO2 uptake coefficients are higher over land areas than over marine areas. Further refinement of the HO2 uptake coefficients is necessary for accurate estimation of aerosol-oxidant interactions.

Golam Sarwar, Christian Hogrefe, Barron Henderson, Rohit Mathur

Recent field and laboratory studies suggest that aerosol nitrate (pNO_3^-) can undergo photolysis to generate gaseous nitrous acid (HONO) and nitrogen dioxide (NO2). We use the Community Multiscale Air Quality (CMAQv5.4) model to examine the potential impact of pNO_3^- photolysis on air quality over the Northern Hemisphere. Similar to other air quality models, CMAQv5.4 does not consider pNO_3^- photolysis; however, it does consider the photolysis of nitric acid (HNO3). We calculate the photolysis frequency of pNO_3^- by multiplying the photolysis frequency of HNO3 with an enhancement factor that varies between 10 and 100 and depends on pNO_3^- and sea-salt aerosols concentrations. We perform simulations without and with pNO_3^- photolysis for 2018. Preliminary results suggest that pNO_3^- photolysis decreases pNO_3^- concentrations and increases HONO, NO2, and subsequently ozone (O3) mixing ratios. Mean surface O3 mixing ratios averaged over the entire Northern Hemisphere increase in each month and range between 2.9-6.2 ppbv with the minimum enhancement occurring in July and the maximum in April. However, larger enhancements occur over some areas. For example, annual mean O3 mixing ratios increase by 8-10 ppbv over a large portion of the western U.S. We compare model NO2 and O3 data with the Ozone Monitoring Instrument (OMI) retrievals and find that the seasonal mean model NO2 and O3 column data without pNO_3^- photolysis are lower than the OMI retrievals while the data with the pNO_3^- photolysis agree better with the OMI retrievals. We compare model O3 with available surface observed data from the U.S., Japan, the Tropospheric Ozone Assessment Report – Phase II, and OpenAQ. The model without pNO_3^- photolysis underestimates the observed data in winter and spring seasons and the model with pNO_3^- photolysis improves the comparison in both seasons largely removing the pronounced underestimation in spring. Compared to measurements from the western U.S., model O3 mixing ratios with pNO_3^-photolysis agree better with observed data in all months due to the persistent underestimation of O3 without pNO_3^- photolysis. Model comparisons with observed data in other seasons provide mixed results deteriorating the comparison in some areas while improving the comparison in other areas. We also compare model predictions with ozonesonde measurements and find that model O3 mixing ratios with pNO_3^-photolysis agree better with observed data than the model results without pNO_3^-photolysis. The presentation will provide a detailed comparison of model predictions with observed pNO_3^-, HONO, NO2, and O3 data from available surface, satellite retrievals, and ozonesonde measurements.

Disclaimer: The views expressed in this paper are those of the authors and do not necessarily represent the views or policies of the U.S. EPA.

Chi-Tsan Wang, Patrick Campbell, Siqi Ma, Irena Ivanova, Paul Makar, Daniel Tong, Bok H. Baek

Atmosphere-biosphere interactions have important effects on air quality. Historically, chemical transport models (CTMs) have been well developed to account for simulating the bulk effects (e.g., biogenic emissions, dry deposition, etc.) of the overlaying and vegetative canopy top processes on fluxes of heat, momentum, and scalars, which are critical to modeling the sources and fate of atmospheric trace gasses important to regional-scale air quality modeling. However, most CTMs continue to rely on the approximate, but useful “big-leaf” model for the air-canopy interface, with little to no representation of the underlying in-canopy processes. The in-canopy airspace varies significantly in its radiative, chemical, and dynamical environment compared to the atmosphere above. Such in-canopy effects, for example, are not fully considered in the widely used Community Multiscale Air Quality (CMAQ) model. In this work, we develop and employ explicit, but simplified parameterizations for the effects of in-canopy photolysis attenuation and modulated vertical turbulence/eddy diffusivity in a variant of the George Mason University (GMU) air quality modeling system, which uses NOAA’s Global Forecast System version 16 meteorology to drive the CMAQ version 5.3.1 model with integrated Process Analysis (PA) output and assessment tools. The CMAQ-PA allows for a more robust assessment of the impacts of such in-canopy processes on atmospheric chemistry and the partitioning between in-canopy photolysis and turbulence effects. Preliminary CMAQ results for an August 2019 simulation show that the in-canopy processes lead to substantial changes in summer hourly NO2 concentration of -5 to 11 ppb in contiguous canopy regions (e.g., Eastern U.S.), with hourly O3 concentration changes between -16 to 12 ppb. Overall, the in-canopy parameterizations slightly improve model performance for near-surface O3 and NO2 over the contiguous U.S. when compared to U.S. EPA AirNow ground stations. The preliminary results of CMAQ-PA for the integrated process rates (IPR) and integrated reaction rates (IRR) using python-based PA tools (i.e., Python-Process Analysis, pyPA, and Python Environment for Reaction Mechanisms and Mathematics, PERMM), quantify more detailed chemical and physical processes for critical species, including NOx, O3, formaldehyde, and fine particulate matter (PM2.5) changes due to the in-canopy parameterizations. The NOx PA results show that the reduced in-canopy vertical transport (-20%) and photolysis (-70%) in the canopy can synergistically increase net NO2 chemical process (+95%) and decrease odd Oxygen formation (-70%). Therefore, the canopy increases the NO2 and decreases O3 concentration in the PBL. Our presentation will provide more details of CMAQ-PA regarding the in-canopy effects on chemistry and transport at different heights, as well as the impact of increased model vertical resolution on the canopy effects.

Jiachen Liu, Shannon Capps

CMAQ has helped researchers and policymakers to comprehend the complexities associated with aerosol formation processes. In policy-related scenarios, the first- and second-order partial derivatives of output variables, such as criteria pollutant concentrations, with respect to input variables, such as emissions, are often of interest to ascertain expected changes due to emissions control strategies. Several methods have been employed to address this challenge, such as the higher-order direct decoupled method (CMAQ-HDDM) and the adjoint method (CMAQ-adjoint). We recently developed a novel, augmented version of CMAQ (CMAQ-hyd) capable of computing numerically exact first- and second-order sensitivities of all modeled species concentrations with respect to select emissions.
    
The EPA developed new emissions profiles to account for emissions from VCPs (Seltzer et al., 2021) in the atmosphere. The emissions have been incorporated into the newly developed EPA’s Air QUAlity TimE Series Project (EQUATES) dataset. We exploit the new emissions profiles and employ CMAQ-hyd to calculate sensitivities and evaluate the health impacts of VCP-related aerosol formation across the continental United States. This study demonstrates the novel application of the hyperdual-step method in CMAQ to enhance our understanding of aerosol formation from VCPs and the potential of the novel model being applied to answer other critical policy-related questions.

HyeonYeong Park and SeogYeon Cho

The Global to Mesoscale Air Quality Forecast and analysis (GMAF) system was developed using WRF-CMAQ framework in this work. We modified the wet scavenging modules, the particle uptake coefficients of N2O5, and the pcVOC emission rates to improve the model performance. More importantly, we implemented grid nudging based on four-dimensional data assimilation (FDDA) to CMAQ of which the application had been limited to atmospheric model such as WRF until the present work. 

The Copernicus Atmosphere Monitoring Service (CAMS) forecast and reanalysis dataset was used as a gridded global atmospheric composition field to be used in grid nudging based on FDDA by CMAQ. The CAMS dataset provides concentrations of 14 gaseous species (formaldehyde (HCHO), hydrogen peroxide (H2O2), isoprene, methane, nitrogen dioxide (NO2), ozone (O3), propane, sulfur dioxide (SO2), carbon monoxide (CO), ethane, hydroxyl radical (OH), nitric acid (HNO3), nitrogen monoxide (NO), and peroxyacetyl nitrates (PAN)) and 9 aerosol species (ammonium, dust, hydrophilic organic matter, hydrophobic organic matter, hydrophilic black carbon, hydrophobic black carbon, nitrate, sea-salt, and sulfate). Among them, the SO2, CO, NO, NO2, isoprene, O3, dust, and sea-salt were selected as nudged variables.

In our previous work (Cho et al., 2021), we tested our approach for simulation of PM2.5 in Korea in 2016. In this study, we look at more closely on the performance of the grid nudging FDDA to CMAQ and conduct the model simulation from January to March in 2018 when severe PM2.5 pollution occurred in Korea. And we focused on simulations of not only PM2.5 but also gaseous species such as SO2, NO2, and O3. The model showed moderate performances with R values around 0.8 for PM2.5 and NO2 simulations, around 0.7 for SO2 and O3 simulations. 

George Delic, HiPERiSM Consulting, LLC, P.O. Box 569, Chapel Hill, NC 27514

This presentation continues with a review of thread parallel performance results for CMAQ 5.3 for the full-year simulation CY2016 (376 days). Attention is focused on the Rosenbrock and EBI solvers in the Chemistry Transport Model (CTM), while results for the Gear solver are in progress. Both FSparse [1], and the legacy JSparse [2] algorithms are compared. After HiPERiSM cluster system upgrades to a total of 330 cores, this year’s presentation reports model results for 24 MPI processes and 12 OpenMP parallel threads in the CTM (CHEM) and the horizontal advection science procedures (HADV). Data for the CONUS domain was provided by the U.S. EPA [3]. Both the legacy (EPA) JSparse and the FSparse thread parallel versions are compared in a hybrid MPI+OpenMP version on a heterogeneous cluster of 19 nodes using a total of 288 cores. Observed speedup with the FSparse version was 1.24 for EBI and 1.3 for the Rosenbrock solver, respectively, showing a strong improvement over previous reports with 8 threads.

[1] G. Delic, Modern Environmental Science and Engineering, Vol. 5, Nr.9, 2019, pp. 775-791. Full text available at: https://www.researchgate.net/publication/338581080_A_Thread_Parallel_Sparse_Matrix_Chemistry_Algorithm_for_the_Community_Multiscale_Air_Quality_Model

[2] M. Jacobson and R.P. Turco (1994), Atmos. Environ. 28, 273-284

[3] The author gratefully acknowledges help from Kristen Foley (EPA), Ed Anderson (GDIT), and Elizabeth Adams (UNC) in providing model data and resolving implementation issues.

Yuzhi Jin^1,2, Jiandong Wang^1,2, David C. Wong³, Chao Liu^1,2

1. China Meteorological Administration Aerosol-Cloud and Precipitation Key Laboratory, School of Atmospheric Physics, Nanjing University of Information Science and Technology, Nanjing 210044, China

2. Collaborative Innovation Center on Forecast and Evaluation of Meteorological Disasters, Nanjing University of Information Science and Technology, Nanjing, 210044, China

3. US Environmental Protection Agency, Research Triangle Park, NC, USA

Black carbon(BC), as a dominant light-absorbing aerosol, has significant impacts on atmospheric environment and global climate. Currently, the uncertainty in numerical simulations of BC is mainly due to its coating induced by complex aging process. Moreover, the simulation of BC aging has not yet been fully considered in the Community Multiscale Air Quality Modeling System(CMAQ). In this study, we developed a BC aging module and incorporated it into two-way coupled WRF-CMAQ model. To begin with, two new species, “Bare BC” and “Coated BC”, were added in the model to distinguish the aging status of BC. Then the transformation of bare BC to coated BC has been considered. We proposed two aging schemes according to an existing methods summary and the model characteristics: Scheme 1 assumes a fixed aging timescale of 1.15 days, while Scheme 2 represents the aging rate by the rate of OH radical concentration change. We conducted a simulation by using WRF-CMAQ model with BC aging schemes in the United States in June 2010 and validated the results with observational data from the Carbonaceous Aerosols and Radiative Effects Study(CARES) campaign. We analyzed three aspects: aging-related variables, BC mixing state, and optical properties. The results indicate that: firstly, compared to the default model that does not consider BC aging, the new model incorporating the BC aging module with Scheme 2 demonstrates pronounced spatiotemporal variations in aging-related variables. Secondly, the new model results can capture the temporal distribution characteristics of significantly higher coated BC proportions during the afternoon period and the spatial distribution pattern with bare BC dominating near the sources and coated BC dominating away from the sources. The mean number fraction of coated BC is approximately 57%. Between the two aging schemes, the diurnal variation of coated BC number fraction simulated with Scheme 2 is closer to the observed values. Thirdly, the default model simulates a BC mass absorption cross section(MAC) value of 532nm wavelength is about 11.6 m²/g at the T0 site, while both aging schemes in the new model simulate MAC values around 8 m²/g, which are closer to other research results. In conclusion, considering BC aging process in the WRF-CMAQ model, particularly with Scheme 2, improves the simulation of dynamic variations of mixing states, which allows for a more accurate assessment of aerosol optical properties.

Khanh Do^1,2,5, Jose Rodriguez Borbon³, George Delic⁴, Bryan Wong³, Yang Zhang⁵, and Cesunica Ivey^1,2,6

¹Department of Chemical and Environmental Engineering, University of California, Riverside, CA

²Center for Environmental Research and Technology, Riverside, CA

³Department of Materials Science & Engineering Program, University of California, Riverside, CA

⁴Hiperism Consulting, LLC, Chapel Hill, NC

⁵Department of Civil and Environmental Engineering, Northeastern University, Boston, MA 6Now at Department of Civil and Environmental Engineering, University of California, Berkeley, CA

The Earth’s atmosphere is extremely complex because of the many physical and chemical processes, such as dispersion, diffusion, deposition, and chemical reactions. Researchers seek to improve the predictability of air quality models by integrating more scientific processes with an increasing number of chemical species to the reaction mechanisms. These enhancements degrade the computational efficiency for the most comprehensive modeling applications with the disadvantage of increased simulation time. Offline chemical transport models (CTM) require a significant amount of simulation time for large spatial domains. As an example, the majority of the simulation time is expended in solving for the gas-phase chemistry when a CTM is applied to simulate gaseous air pollutants. To reduce the simulation time while maintaining the integrity of the models, we utilize graphics processing units (GPU) to replace the central processing units (CPU) for computing the most intensive science processes with a successful port of the gas-phase chemistry solver onto a GPU. The actual kernel computing time for the solver is twice as fast as the CPU with a BLKSIZE of 8,000. However, the GPU solver comes at the cost of communication time incurred by moving data back and forth between the host system memory to the GPU memory. In this paper, we focus on the details of (1) the compilation of the Community Multiscale Air Quality (CMAQ) model with Compute Unified Device Architecture (CUDA) kernels, (2) porting the gas-phase CTM solver onto the GPU, and (3) optimizing the solver to improve GPU computational efficiency. The positive results from the ported solver show the promising future for intensive parallel computing applications on GPU devices, benefiting researchers in reducing the simulation time and accelerating the research.

Deanna Huff ADEC, Tom Carlson with Trinity Consultants, Chao-Jung Chien and Pradeepa Vennam with Ramboll Consultants, Kathleen Fahey USEPA-ORD, Rob Gilliam USEPA-ORD, Nicole Briggs USEPA R10, Nick Czarnecki ADEC

Fairbanks Alaska was designated as a non-attainment area for PM2.5 in 2009 for exceeding the 24-hr National Ambient Air Quality Standards (NAAQS) of 35 ug/m3. Design values, which are 3-year averages for comparison to the NAAQS, have varied from 43 ug/m3 to 135 ug/m3 due to localized hot spots, monitor location, meteorological variations, and control measure implementation. The current design value for 2020 – 2022 is 64ug/m3. Winter conditions in Fairbanks, AK include strong inversions trapping pollutants close to the ground leading to elevated concentrations of PM2.5 and its precursor gases. The two largest contributors to PM2.5 in Fairbanks are organic carbon and sulfate. Control strategies have focused on reducing organic carbon through wood stove measures and sulfate with fuel reductions. Attainment modeling is a regulatory requirement for non-attainment area State Implementation Plans (SIPs) and is used to demonstrate the most expeditious date an area could reach attainment with a given control strategy. In previous SIPs the Alaska Department of Environmental Conservation (ADEC) based attainment demonstrations on an outdated modeling platform, meteorological data, and episodes. ADEC has been working on resolving these deficiencies. Updates include the CMAQ Model version 5.33 and updated WRF meteorology in collaboration with USEPA-ORD RARE grant ALPACA studies, emissions inventory, and all other pre-processing models. The results are better model performance for representing stable boundary layers, model performance of secondary sulfate and these updates have allowed DEC to accurately represent modeling of control strategies that will bring the area into attainment for PM2.5.

Kiran Alapaty1 , Jesse Bash1 , Christian Hogrefe1, Daiwen Kang1, Rob Gilliam1, Barron Henderson1, Chris Nolte1, Alan Vette1, Bin Cheng2, and Sarav Arunachalam2

For about half-a-century, similarity profile functions have been the only option to model planetary boundary layer (PBL) processes in meteorological and air quality models. These functions represent boundary layer stability conditions so that PBL processes are realistically modeled. However, differences in these functions across regional and global meteorological and air quality models can contribute to sizeable model-to-model differences in simulation results. To address this, we present the development, implementation, and testing of a new paradigm for PBL modeling that is independent of similarity functions in the WRF and CMAQ models.

Previously, a 3-D turbulence velocity scale, e*, was proposed and validated with a decade of measurements from a 3-D sonic anemometer. Then, this turbulence velocity scale was successfully tested and validated in a box model to simulate aerosol deposition velocities where it replaced the default friction velocity and similarity functions.

In this study, the e* was used to develop a new surface layer parameterization in the WRF model to estimate surface fluxes without the use of similarity functions. The e* was also used to develop new formulations in a boundary layer parameterization, again without using similarity functions. These formulations were also implemented in the CMAQ model to maintain consistency in representing boundary layer processes. Further, several resistances in the ozone dry deposition estimations were revised by the incorporation of e*. Thus, the new paradigm of PBL modeling was implemented in the WRF and CMAQ models. Numerical simulations for a hemispheric domain using WRF and CMAQ models were performed for the year 2018.

We compare and evaluate model estimates of selected fields (e.g., simulated surface precipitation, 2 m air temperature(T2) and water vapor mixing ratios, 10 m winds, and surface HNO3 and ozone) from 2018 WRF and CMAQ simulations using the OLD (i.e., using similarity functions) and NEW (without using similarity functions) surface and mixed layer PBL schemes. Results for 2018 indicate that the new paradigm for PBL modeling works very well (e.g., T2 bias differences between NEW and OLD are well within measurements accuracy), providing the very first alternative to the 50-year old traditional method of PBL modeling.

Michael Barna, National Park Service, Air Resources Division, Fort Collins, CO

Ralph Morris, Ramboll, Novato, CA

Tom Moore, Regional Air Quality Council, Denver, CO

Gail Tonnesen, U.S. Environmental Protection Agency, Region 8, Denver, CO

Kevin Briggs, Colorado Department of Public Health and Environment, Denver, CO

The Western Regional Air Partnership (WRAP) has developed a modeling platform to simulate the formation of haze-causing particles that impact federally protected lands in the western United States. To assist state air quality planners in determining which emission sources are likely candidates for future mitigation, several source apportionment scenarios were evaluated, and two sets of results for the year 2028 are presented here: 1) a ‘high-level important regional sources’ version, with broad emission categories (i.e., U.S. anthropogenic, international anthropogenic, natural, and fires), and 2) a ‘low-level anthropogenic emission sources within individual states’ version, which refines the U.S. anthropogenic contribution to specific emission sectors within individual WRAP region states. Eight examples are discussed, which reflect the variation in source apportionment results at national parks, wilderness areas, and wildlife refuges in the western U.S., and suggest which emission sectors are candidates for mitigation to improve future visibility.

Thibaud Sarica^1,2, Youngseob Kim¹, Yelva Roustan¹, Karine Sartelet¹, and Yang Zhang²

¹ CEREA, École des Ponts ParisTech, EDF R&D, IPSL, Marne-la-Vallée, France

² Department of Civil and Environmental Engineering, Northeastern University, Boston, MA, USA

High concentrations of air pollutants, such as nitrogen dioxide (NO₂) and particulate matter (PM), are observed in urban areas due to unfavorable dispersion conditions and the proximity to traffic. Historically, street-network models such as the Model of Urban Network of Intersecting Canyons and Highways (MUNICH) represent these concentrations considering their homogeneity in each street segment. A new version of MUNICH discretizes the street volume to represent concentration heterogeneities (referred to as the heterogeneous version of MUNICH). In this new version of the model, three model layers in the vertical direction are introduced to limit the artificial dilution of both traffic emissions and concentrations. Based on the parameterization from the Operational Street Pollution Model, a recirculation zone and a ventilation zone are considered under specific conditions to represent horizontal heterogeneities.

In the homogeneous version of MUNICH, processes of deposition of gases and PM, and resuspension of previously deposited material have limited impacts (at ~1% for both gases and PM) on the concentration levels. These processes, however, are expected to have larger impacts on the pollutant concentrations, particularly in the first layer. In this work, the representation of these processes is adapted in the new heterogeneous version of MUNICH. In the first model layer at the bottom of the street, dry deposition is simulated at both road and wall surfaces, where it is represented only to wall surface in the two other layers. Scavenging by precipitation is computed separately for each layer. Finally, resuspension, occurring at the road surface, is only considered in the first layer. Simulations with the new heterogeneous version of MUNICH are performed over a district of Greater Paris, France. Resulting concentrations of NO₂ and PM with and without deposition and resuspension processes are compared with observational data in the street and to the concentrations modeled with the homogeneous version of MUNICH. This work is an important step toward developing a new operational version of MUNICH representing concentration heterogeneities while considering the main physicochemical processes taking place in the street.

Nancy Daher, Lexie Wilson, Rachel Edie and Mark Sghiatti

The Salt Lake Valley in Utah continues to exceed the National Ambient Air Quality Standard (NAAQS) for ozone (O3) during the summer, with O3 forming from both local and non-local anthropogenic and natural sources. In this work, a typical summertime O3 air pollution episode that occurred in the Salt Lake Valley in 2017 was simulated using the Comprehensive Air Quality Model with extensions (CAMxv7.1). Three 12/4/1.33 km nested grid domains were considered for this analysis. The Ozone Source Apportionment Technology (OSAT) CAMx probing tool was used to estimate contributions to ozone formation from 6 geographic regions and 26 source emission groups separately. Results showed that non-local natural and anthropogenic emissions contribute to most of the ozone measured at the controlling monitor in the Salt Lake Valley, accounting for about 66% (39.8 ppbv) of its modeled maximum daily 8-hour average O3 (MDA8) on average during the modeling episode. Local anthropogenic and biogenic sources had smaller contributions, accounting for nearly 16.7% (10.04 ppbv) and 6.1% (3.7 ppbv) of MDA8 at the same location, while international anthropogenic emissions source contribution averaged 6.5% (3.9 ppbv). Contributions from other sources, such as wildfires, agricultural fires, lightning NOx, were much smaller (< 5% and 3 ppbv). Among local sources within the non-attainment area, biogenic emissions and on-road emissions, followed by point source emissions, were predominant contributors to O3 measured at the controlling monitor. These findings have important policy implications for emissions control development.

Lexie Wilson, Nancy Daher, Mark Sghiatti

Several different versions of the Biogenic Emissions Inventory System (BEIS) and the Biogenic Emissions Landuse Database (BELD) were used to simulate emissions for a 1.33 km modeling domain covering the Northern Wasatch Front ozone nonattainment area in Utah. The Northern Wasatch Front nonattainment area includes a densely populated urban corridor - home to Salt Lake City and surrounding municipalities – alongside several lush national forests and other vegetation. This regime requires special consideration of biogenic volatile organic compounds (BVOCs), since emissions are proximate to Utah’s main population center. To assess differences among recent BEIS and BELD versions and their impact on ozone model performance along the Northern Wasatch Front, three model simulations in SMOKE-BEIS are compared: (1) BEIS 3.6.1 with BELD 4.1, (2) BEIS 3.7 with BELD 5, and (3) BEIS 4 with BELD 6. These emissions are further modeled in CAMx. Isoprene concentrations are directly and qualitatively compared at two Utah monitoring stations to assess model performance. Biogenic emissions have a noted impact on ozone model performance as well. We propose next steps for improved model performance related to BVOCs in the intermountain west.

Rich Mason, David Cooley, Hannah Derrick

Estimating the nonpoint component for Industrial and Commercial/Institutional (ICI) fuel combustion is challenging in the absence of available fuel consumption estimates from the point inventory. ICI fuel consumption is computed separately by sector and fuel type for each State and Local agency by subtracting estimated point inventory fuel consumption from total fuel consumption provided by the Energy Information Administration (EIA) State Energy Data System (SEDS). With varying levels of controls by facility and geographic area, direct point inventory emissions subtraction is not a technically defensible approach for computing nonpoint emissions. Therefore, for the 2020 National Emissions Inventory (NEI), we analyzed the relationship between State/Local-submitted point inventory fuel consumption estimates and their point inventory carbon monoxide (CO) emissions by sector and fuel type. Screening for outliers and using a median ratio of results by sector/fuel, we developed a default approach for estimating point inventory fuel consumption based on state-submitted CO emissions for the 2020 NEI. Limitations to this approach and suggestions for refinements for the 2023 NEI are also discussed.

Jia Xing^1,2, Bok H. Baek¹, Siwei Li³, Chi-Tsan Wang¹, Ge Song³, Siqi Ma¹, Daniel Tong¹

¹Center for Spatial Information Science and Systems, George Mason University, Fairfax, VA 22030, USA

²Department of Civil and Environmental Engineering, the University of Tennessee, Knoxville, TN 37996, USA

³Hubei Key Laboratory of Quantitative Remote Sensing of Land and Atmosphere, School of Remote Sensing and Information Engineering, Wuhan University, Hubei, 430000, China

The improvement of emission estimates at the sector-level is crucial for policymakers to identify pollution sources and design effective control strategies. However, previous top-down methods mostly focused on total emissions. A novel machine learning-based top-down method, Variational Autoencoder (VAE), exhibits strong ability in estimating emissions from observations in a highly efficient way. As a follow-up development of the VAE method, this study aims to enhance its ability to estimate sector-level emissions by utilizing multiple observations with spatial/temporal profiles for each sector. To achieve this, we carefully designed the VAE model structure by incorporating a deep Convolutional Long Short-Term Memory model (ConvLSTM) to better capture the spatiotemporal variations of atmospheric chemicals. Additionally, we selected relevant features related to emission and concentration of atmospheric chemicals, including surface upper-layer emissions and meteorological variables, to accurately represent the relationship between emissions and concentrations. The VAE was trained using a dataset generated from multiple simulations with a chemistry transport model (WRF-CMAQ), enabling it to successfully mimic the correlations between layer-specific emissions and surface/column concentrations. The sensitivity analysis of the VAE further suggests the importance of NOx emissions in driving the variation of NO2 concentration. We applied this method to the CONUS domain for the years of 2019 and 2020. The VAE method successfully captured the changes in NOx emissions due to the COVID-19 pandemic, demonstrating agreement with the estimation obtained from bottom-up emissions.

Karl Seltzer, Venkatesh Rao, Rich Mason, Jennifer Snyder, Brandy Albertson, Andrew Bollman, Susan McCusker

Cooking is a source of both gaseous and particulate pollutants. These emissions occur both commercially and in residential settings, and observations suggest cooking contributes ~16% of observed organic aerosol and ~8% of observed PM_2.5 in U.S. cities. Source testing also reports considerable carbonyl emissions from cooking, including air toxics such as formaldehyde and acetaldehyde. Currently, the U.S. EPA’s National Emissions Inventory strictly considers commercial cooking and backyard barbequing emissions sources. Here, we review the existing methods used to estimate cooking emissions in the NEI, propose alternative methods for comprehensively estimating cooking emissions nationwide, illustrate the uncertainty in emission estimates due to widely varying emission factors, and compare an updated inventory with estimates from the 2020 NEI. In addition, we cover the chemical characterization of direct emissions, including the speciation of primary particulate matter, volatility of primary organic aerosol, and the contributions of hazardous air pollutants.

Benjamin N. Murphy

Karl Seltzer

Amara Holder

Gabriel Isaacman-VanWertz

Havala O. T. Pye

Residential wood burning is currently one of the largest anthropogenic sources of organic carbon particles and vapors to the atmosphere in the United States, according to recent U.S. EPA National Emission Inventory estimates. The impact of these emissions on air quality is profound, especially in the wintertime when wood is used for heating, and it is expected to grow in relative importance in the future. Existing inventories and photochemical air quality models often use an outdated conceptual model of the phase partitioning of organic particulate and vapor mass. Specifically, regulatory test methods are used to quantify particulate matter emission factors from wood stoves with an operational definition of particulate matter (i.e. mass captured on a Teflon filter) that is susceptible to systematic biases corresponding to the temperature and dilution conditions of each individual test. Meanwhile, total hydrocarbons vapors are characterized using flame-ionization detection, which provides an uncertain measure of gas mixtures containing significant contribution from oxygenated molecules. Finally, the speciation of residential wood burning emissions needs to be revised with state-of-science understanding of key semivolatile and intermediate volatility compounds that are potent secondary organic aerosol precursors.

This presentation discusses the physical basis for and methodology we use to implement revisions to the PM and VOC emission factors and speciation in U.S. EPA emissions modeling tools and quantifies the impact these updates have on primary and secondary organic aerosol concentrations in the U.S. with the Community Multiscale Air Quality (CMAQ) model employing the new Community Regional Atmospheric Chemistry Multiphase Mechanism (CRACMM). The new approach provides the most rigorous and complete translation to date of organic compound mass from residential wood burning emissions to air pollutant concentrations. Identification of key uncertainties including volatility distributions and OM:OC metrics point to critical areas for future research.

Ty Hosler, Dr. Brad Adams

Dust emission and transport along the Wasatch Front of Northern Utah impacts various environmental processes (snow melt, eutrophication, alpine soil buildup, etc.). Coupled with the contemporary recession of the Great Salt Lake (GSL), increased emission of particulate matter poses serious health and economic risks to the local population as well. For these reasons it is important that dust emissions are accurately characterized to better quantify their influence on environmental and social processes. This work better characterizes dust emission predictions by using local soil properties and land cover conditions in the modeling process. By changing default CMAQ particle size distribution and soil composition to those of locally measured values, predicted vertical dust emission fluxes were observed to decrease by approximately 50%. Incorporation of Non-Photosynthetic Vegetation (NPV) or “brown vegetation” was also significant as it decreased the erodible land area in Utah dust sources by approximately 40% during fall, winter and spring seasons. Both changes suggest that CMAQ default values were significantly overestimating the dust emissions in the region.

Joey(Jiaoyan) Huang, Carey Jang, Jim Kelly, Shicheng Long, Yun Zhu, Jia Xia

Fine particulate matter (PM_2.5) and ozone (O₃) are criteria air pollutants that have been associated with adverse effects on human health and the environment. Predicting the effectiveness of policy options for improving PM_2.5 and O₃ air quality using chemical transport models (CTM) is computationally expensive. Response surface models (RSMs) are comprehensive tools for simulating the nonlinear response of O₃ and PM_2.5 across chemical regimes with less computing resources compared to the traditional CTM method. Previous studies have developed various response surface models. The required number of CTM simulations for RSM development decreased from hundreds in the early-generation statistical RSMs to dozens in the polynomial function-RSM (pf-RSM) method. Moreover, the deep-learning RSM (DeepRSM) requires only a few CTM simulations to capture the nonlinear response of O₃ and PM_2.5 to emission changes. We developed a DeepRSM for the US CONUS domain with 10 climate regions using a 2018 CMAQ modeling platform. We used five “out-of-sample” cases for cross validation. We compared O3 and PM_2.5 concentrations and responses due to the emission changes between DeepRSM and CMAQ outputs using these five case studies. Monthly O₃ DMA-8hour concentrations and responses (grid-by-grid and spatial variations) agree well between CMAQ and DeepRSM results (r = 0.98 – 0.99). Monthly average PM2.5 concentrations (grid-by-grid and spatial variations) show high agreement between CMAQ and DeepRSM results (r ~ 0.999). However, the monthly average PM_2.5 responses due to the emission changes show slight disagreement (r ~ 0.95). Overall, DeepRSM can capture the O₃ nonlinear response simulated by the CMAQ brute force method for a wide range of NOx and VOCs reductions (25, 50, 75, and 100%). We will present comparison of DeepRSM results and HDDM/ISAM results at the CMAS conference. The DeepRSM tool will benefit the policy analysis by minimizing CTM simulations and capturing nonlinear responses of O3 and PM2.5 to emission changes from potential future rules.

Ty Hosler, Dr. Brad Adams

Dust emission, transport, and deposition play an integral role in the hydrology of Utah. Dust emitted from the Great Basin is transported into the Wasatch and Uinta mountain ranges where it is deposited. The deposited dust brings nutrients both essential and detrimental to the local flora and fauna. The modern desiccation of the Great Salt Lake (GSL) has exposed hundreds of miles of shoreline previously covered by water. These increased emission sources potentially increase dust deposition to Utah watersheds. This work quantifies the amounts of important nutrients (N, K, Ca, Fe, etc.) deposited by dust in key Utah watersheds to help identify the role dust plays in water quality. This is accomplished using speciated element tracking in the WRF/CMAQ dust modeling framework. One advantage to this framework is that wet and dry dust deposition can be resolved both spatially and temporally throughout the domain of interest (e.g., state-wide, regional, watershed). The timeframe studied is the 2022 spring dust season in Utah (i.e., March-May). This timeframe is of interest because the dust deposited on snow during this period both accelerates snow melt and influences runoff water quality. Cumulative deposition of critical species are calculated and used to better estimate nutrient loading in specific watersheds.

Ahmed Khan Salman, Yunsoo Choi, Jincheol Park, Seyedali Mousavinezhad, Mahsa Payami, Mahmoudreza Momeni, Masoud Ghahremanloo

This study details the development and evaluation of a digital twin model of the Community Multiscale Air Quality (CMAQ) model, utilizing a U-Net deep learning architecture to accelerate the simulation of surface NO2 concentrations across the Contiguous United States (CONUS). The digital twin employs a subset of meteorological, land cover, and emission input variables identical to those in CMAQ. An initial assessment of the model based on 3-fold monthly cross-validation during the summer (JJA) demonstrates excellent accuracy, with a correlation coefficient (R) of 0.979 and an Index of Agreement (IOA) of 0.989. Subsequently, the model's long-term sustainability was examined by training it with NEI 2011 and 2014 data, and then evaluating it using NEI 2017 data. This yielded an R of 0.949 and an IOA of 0.974. We utilized the digital twin to investigate the semi-normalized sensitivity of NO2 concentrations to NOx emissions, which exhibited a satisfactory alignment with CMAQ Decoupled Direct Method (DDM) sensitivities, with an MAE of 0.271 ppb. Diurnal cycle analysis of NOx sensitivity coefficients in 15 major urban environments indicated slight over- and underestimations of the morning and evening peaks, respectively, with the MAE varying from 0.27 (Dallas) to 0.92 ppb (Los Angeles). Remarkably, the digital twin’s computational efficiency significantly surpasses CMAQ’s, providing more than 400 times the simulation speed on a single CPU and over 600 times when utilizing both CPU and GPU. As such, the digital twin represents a promising tool for efficient CMAQ modeling, with potential applications in health impact assessments, emission reduction strategies, and emission inventory optimization.

Zhen Qu, Daven K. Henze, Helen M. Worden, Zhe Jiang, Benjamin Gaubert, Nicolas Theys, Wei Wang

Accurate quantification of the magnitude, trend, and national contribution of air pollutant emissions from each human activity is critical for the planning and verification of emission reduction efforts. Top-down estimates with satellite data provide important information on the sources of air pollutants. We apply a newly developed sector-based inversion method to quantify NOx, SO2, and CO emissions over 2005–2012 from various activities, including transportation, industry, residential, aviation, shipping, energy, and biomass burning. We incorporate OMI NO2, OMI SO2, and MOPITT CO observations and leverage the co-emission of these gases to identify the source sectors. The framework improves emission estimates at the process level by optimizing emission factors and activity rates without relying on explicit knowledge of their values and resolves discrepancies with bottom-up inventories at the sector level. We first applied this approach to estimate sectoral emissions in China and India, where posterior evaluations with surface measurements show reduced normalized mean bias (NMB) by 7% (NO2)–15% (SO2) and normalized mean square error (NMSE) by 8% (SO2)–9% (NO2) compared to a species-based inversion. The posterior estimates capture the peak years of Chinese SO2 (2007) and NOx (2011) emissions and attributes their drivers to industry and energy activities. The CO peak in 2007 in China is driven by residential and industry emissions. In India, the inversion attributes NOx and SO2 trends mostly to energy and CO trend to residential emissions. We are extending this framework to estimate sectoral emissions in the US and evaluate the posterior estimates with the EPA NEI.

Khanh Do^1,2,4, Arash Kashfi Yeganeh^1,2, Ziqi Gao³, Yang Zhang⁴, and Cesunica E. Ivey^1,2,5

¹Department of Chemical and Environmental Engineering, University of California, Riverside, CA

²Center for Environmental Research and Technology, Riverside, CA

³Department of Civil and Environmental Engineering, Georgia Institute of Technology, Atlanta, GA, USA

⁴Department of Civil and Environmental Engineering, Northeastern University, Boston, MA

⁵Now at Department of Civil and Environmental Engineering, University of California, Berkeley, CA

Machine learning (ML) and chemical transport models (CTM) have a similar goal to accurately predict air pollution. ML heavily depends on the quality and quantity of data available. Conversely, CTMs are based on first principles equations and are initiated with interpolated observation data, hence avoiding most obstacles introduced by data missingness in observations. In contrast with CTMs, which produce larger scale, spatially resolved outputs, ML only provides predictions strictly at trained locations when used for ambient air quality applications. In this study, we combine machine learning and geospatial interpolations to create a two-dimensional high-resolution ozone concentration field over the South Coast Air Basin (SoCAB) for the entire year of 2020 to investigate the ozone trend under the rapid changes in emissions and meteorological conditions due to the COVID-19 shutdown. Three spatial interpolation methods (bicubic, inverse distance weighing, and ordinary kriging) are employed. The predicted ozone concentration fields were constructed using 15 building sites, and random forest regression was employed to test predictability of 2020 data based on input data from past years. Spatially interpolated ozone concentrations were evaluated at twelve sites that were independent of the actual spatial interpolations to find the most suitable method for SoCAB. Ordinary kriging interpolation had the best performance overall for 2020: concentrations were overestimated for Anaheim, Compton, LA North Main Street, LAX, Rubidoux, and San Gabriel sites (mostly coastal Basin) and underestimated for Banning, Glendora, Lake Elsinore, and Mira Loma sites (mostly inland Basin). The model performance improved from the West to the East, exhibiting better predictions for inland sites. The model performs the best at interpolating ozone concentrations inside the sampling region (bounded by the building sites), with R2 ranging from 0.56 to 0.85 for those sites, as prediction deficiencies occurred at the periphery of the sampling region, with the lowest R2 of 0.39 for Winchester. All interpolation methods poorly predicted and underestimated ozone concentrations in Crestline during summer by up to 19 ppb. Poor performance for Crestline indicates that the site has a distribution of air pollution levels independent from all other sites. Therefore, historical data from coastal and inland sites should not be used to predict ozone in Crestline using data-driven spatial interpolation approaches, despite Crestline being a design value site for the Basin in previous years. The study demonstrates the utility of machine learning and geospatial techniques for evaluating air pollution levels during anomalous periods.

Verica Savic-Jovcic¹, Craig Stroud¹, Junhua Zhang¹, Qiong Zheng¹, Elisa Boutzis¹, Jack Chen¹, Sylvain Ménard², Dragana Kornic^2
1Air Quality Research Division, Environment and Climate Change Canada, Toronto, Ontario, Canada
²National Predictions Development Division, Environment and Climate Change Canada, Montreal, Quebec, Canada

Environment and Climate Change Canada operates the Regional Air Quality Deterministic Prediction System (RAQDPS) to provide forecasts of O₃, NO₂ and PM_2.5 pollutant concentrations. Based on these forecasts, Air-Quality Health Index (AQHI) is calculated and provided to Canadian population. Additionally, a twin operational system, RAQDPS-FW known as FireWork, provides information about the concentration of PM2.5 that stems only from forest fires. These two systems run twice daily to produce 72-h long forecasts.

At the heart of each system is an on-line chemistry transport model, GEM-MACH, which is driven by the meteorological inputs available from the operational meteorological Regional Deterministic Prediction System (RDPS) and by pre-processed chemistry-related inputs. Chemical inputs in RAQDPS are in nature long-term averages or projections, except for the forest-fire emissions, which are produced twice-daily from the satellite observations of hotspots and RDPS meteorological conditions.

Updates to RAQDPS, which are currently under evaluation, include: a. Update to meteorological piloting from regional to global forecasts; b. Merging of twin systems into one that will provide above-mentioned forecasts based on the inputs that include forest-fire emissions, and that will advise on the location of the forest-fire induced plumes of higher PM2.5 concentrations; c. Updates to the chemical inputs, i.e. anthropogenic emissions, lateral boundary conditions, leaf-area index and biogenic-emissions rates; and d. Updates to dynamical core, physical and chemistry libraries of GEM-MACH.

Anthropogenic emissions were upgraded only for Canada, where both surrogates and inventories have undergone revision. In this presentation, we will focus on the details of the changes in the surrogates, and on the evolution of the inventories. We will conclude with the discussion of the impact of these improvements on the concentrations of the forecasted pollutants. 

A. Eyth, J. Vukovich, C. Farkas, J. Godfrey, K. Seltzer, R. Mason, A. Diem, S. Roberts, C. Allen, J. Beidler

The U.S. EPA has developed an emissions modeling platform for the year 2020 based on the 2020 National Emissions Inventory (NEI). Model-ready emissions were prepared to support modeling with the Community Multiscale Air Quality (CMAQ) modeling system and the AERMOD dispersion model. Special considerations related to preparing emissions data appropriate for modeling the pandemic year of 2020 will be discussed, along with other recent updates to the platform unrelated to the pandemic.

Shih Ying Chang, Nathan Pavlovic, Jiaoyan Huang, Kenneth Craig, Charles Scarborough, Charles Driscoll, Justin Coughlin, Jeffrey Herrick, Christopher Clark, and Kevin Horn

Critical loads (CL_D) of atmospheric deposition for nitrogen (N) and sulfur (S) and critical levels (CL_E) of ozone (O₃) are used to support decision-making related to air quality policies and resource management. Previously, CL_D of N and S were established using empirical methods to investigate the relationship between ecosystem health endpoints and deposition estimates from a data fusion process that combines CMAQ-modeled deposition and measurements. The certainty of the CL_D estimates depends on accurately representing the underlying environmental conditions that drive sensitivity to N and S. Conversely, O₃ CL_E were established based on seedling experiments with controlled O₃ exposure, which may fail to accurately reflect O₃ impacts on mature trees. In this study, we used bootstrap-ensemble machine-learning methods to develop CL_D and CL_E estimates and assess uncertainties for 108 tree species (CL_D) and 88 trees species (CL_E) in the U.S. Machine learning models were trained to predict tree growth and survival probability in relation to air pollutant deposition/concentration. This study reports the first results for empirically derived, species-specific O₃ CL_E for tree growth and tree survival using a database of ~1.5 million mature trees observed between 2001 and 2021 across the U.S. O₃ exposure levels across the U.S. have been below the growth CL_E for most tree species, while levels may have exceeded the survival CL_E for some species. For N, the deposition level exceeds the lower bound of the survival CL_D, detrimentally impacting 80% or more of tree species and over 50% of the species range (i.e., the area where the species can be found). At the upper bound, however, less than 20% of tree species are adversely impacted across more than 60% of the species range. Our results provide new evidence of the magnitude and uncertainty of O3 exposure impacts to mature trees across the U.S. The uncertainty of N and S CL_D is sufficiently large to warrant consideration in resource management and regulatory decision-making s with respect to atmospheric deposition.

George Pouliot, Kristen Foley, Alison Eyth, Jeff Vukovich, Venkatesh Rao, Caroline Farkas, Janice Godfrey 

The EPA has developed a set of modeled meteorology, emissions, air quality and pollutant deposition spanning the years 2002 through 2019, known as EPA’s Air Quality Time Series (EQUATES). The emission inventory data for this time series was a key component of this large project. The key goal was to develop North American emissions inventories using, to the extent possible, consistent year-specific input data and the most up to date methods across all years, including emissions from motor vehicles, fires, and oil and gas sources. Now that this dataset is available for use and has been incorporated into many modeling studies and into the latest release of the EPA air emission trends data in 2023, (https://www.epa.gov/air-emissions-inventories/air-pollutant-emissions-trends-data), we will provide an overall summary of the project, lessons learned, and ideas for the next steps for EPA’s time series project. When new and improved inventory methods are developed in the future for any sector, we would need to update the current EQUATES dataset for 2002-2019 to a new version. In addition, looking ahead to the current decade that started in 2020, we will likely at some point want to develop a new time series starting in 2020 that would incorporate the latest information and methods for inventory development, possibly including inventory improvements derived from NASA’s Tropospheric Emissions: Monitoring of Pollution (TEMPO) satellite that was launched in April 2023 and adding hazardous air pollutants (HAPs) trends to the suite of these criteria air pollutants (CAPs).

Colin Lee, Paul Makar, Kenjiro Toyota, Christian Hogrefe, Olivia Clifton, Mhairi Coyle, Erick Fredj, Orestis Gazetas, Ignacio Goded, László Horváth, Qian Li, Ivan Mammarella, Giovanni Manca, J. William Munger, Ralf Staebler, Eran Tas, Timo Vesala, Tamás Weidinger, Zhiyong Wu, Leiming Zhang

Modern machine learning approaches such as deep learning and random forests have shown remarkable progress and promising results in atmospheric modeling, but there remain valid concerns which have largely prevented their deployment in operational forecast settings. Physically-based machine learning is a recent approach where physical insights are coded into the machine learning model and taken into consideration during the training process. Providing these kinds of constraints has been shown to actually improve model representation of observational data compared to unconstrained data-based approaches in some situations. In this study, we extract the Wesely dry-deposition scheme written in Fortran from Environment and Climate Change Canada's online weather and air quality model, GEM-MACH, and translate it to TensorFlow, a modern deep learning framework which uses python as the programming language. From here we use physically-based machine learning, applying various different physical constraints to our machine learning model, in order to best reproduce ozone dry-deposition velocity data derived from field measurements at eight terrestrial sites with different land cover types (peat bog, grass, temperate mixed forest, deciduous broadleaf forest, Evergreen needleleaf forest and shrub) under different climate conditions. This dataset, compiled as part of the Activity 2 of the Air Quality Model Evaluation International Initiative Phase 4 (AQMEII-4) study, includes a wealth of ancillary data such as leaf area index, soil moisture conditions, CO2 concentrations, meteorological and micrometeorological quantities observed simultaneously at the sites. Using this training data, we are able to improve the model’s ability to represent the observations while still maintaining the overall physical structure of the model, which should make it easier to justify the use of such a model in an operational setting.

Anas Alhusban, ShunLiu Zhao, Amir Hakami (Carleton University) Petros Vasilakos, Ted Russell (Georgia Tech)

Grid resolution of an air quality model (AQM) is one of the parameters that affect estimates of marginal societal benefits of emissions abatement in metropolitan areas. The use of higher resolution is often constrained by the availability of sufficiently resolved inputs, and computational costs. This work evaluates the impacts of the horizontal grid resolution on the population health benefit due to reductions in primary or precursor PM2.5 emissions estimated using adjoint sensitivity analysis.

We use U.S. EPA’s CMAQv5.0 and its multiphase adjoint to examine the impacts of horizontal grid resolution on attributed source health impacts estimates. We chose two of the largest metropolitan areas in North America (New York City & Los Angeles), modeled at progressively increased resolutions of 36 km, 12 km, 4km, and 1km. The simulation covered a two-week episode during the summer of 2016. Emissions were produced using the 2016 emissions platform. Meteorological inputs were driven from the Weather Research and Forecasting (WRFv3.9.1). We use the Global Exposure Mortality Model (Burnett, et al., 2018) which provides a nonlinear concentration-response function for mortality due to chronic exposure. We find that the health burden estimates across resolutions for Primary PM2.5 range between 11 to 14 billion dollars for the New York area and 11 to 13 billion for Los Angeles. Our findings suggest that despite the increased sub-grid variability in benefit per ton (BPT) for PM2.5 and its inorganic precursors, estimates with a relatively coarser resolution such as 12km can be sufficient to estimate the total health burden in a metropolitan area within a regional study. However, higher resolution modeling is required for any policy action at the city level as it depicts the spatial features clearly.

Y. Burak Oztaner, ShunLiu Zhao, Amir Hakami (Carleton University), Rohit Mathur, Barron Henderson (EPA)

The impact of environmental stressor such as air pollution on public health raises concerns about inequality across socio-economic status (SES) groups on any spatial scale. The long-term PM2.5 exposure and mortality often result in higher health burden in low- and mid-income populations. Previous studies have often focused on the disparities of air pollution health burden at local (urban) or national scales. In this study, we evaluate the impact of primary PM2.5 and its precursor emissions (NOX, SO2, NH3) on inequality of PM2.5 mortality across income groups on a hemispheric scale.

We employ the multiphase CMAQ-Adjoint v5.0 to attribute environmental health inequality of PM2.5 to each emission source location using a health inequality index (the Concentration Index) over the Northern Hemisphere. The CMAQ and its adjoint model are driven by meteorological inputs from the Weather Research and Forecasting (WRFv3.8.1), and emissions are retrieved from the 2016 Hemispheric Emission modelling platform. The simulations are conducted over a 108-km resolution for the year 2016. We use the Global Exposure Mortality Model (Burnett, et al., 2018) to estimate PM2.5 exposure. The national-level household income populations (16 income bins) are applied to evaluate the health inequality of PM2.5 exposure. We monetize PM2.5 health inequality using generalized and country-specific estimates of Value of Statistical Life (VSL) and a transfer function between the utility of emission reductions and income enhancement for lowering hemispheric inequality. Our results reveal India as the country most affected by disparities in air pollution health burden. Primary PM2.5 emissions carry the largest impact in India, with valuation approaching and exceeding $1M in large areas of India. By contrast, emission reductions in North America and Europe are often associated with negative sensitivities, indicating that improved air quality in these regions would result in increased disparity in air pollution health burden across the hemisphere.

Kai Wu, Shupeng Zhu, Michael Mac Kinnon, Scott Samuelsen

Medium and heavy-duty vehicles (MHDV) operate along freight corridors serving major transportation hubs including large shipping ports and contribute substantially to poor air quality and associated health outcomes in exposed populations. Moreover, MHDV-related pollution disproportionately affects communities, particularly disadvantaged communities situated nearby major highways and hubs. Therefore, a transition to clean MHDV and clean fuels that reduce emissions including NOx and particulate matter (PM) will subsequently improve both local and regional air quality, and significantly enhance both the environmental quality and public health in disadvantaged communities. In this study, we investigate the co-benefits of various clean energy scenarios (based on The 2022 Scoping Plan for Achieving Carbon Neutrality) on air quality and health implications over the South Coast Air Basin of California using the Community Multiscale Air Quality (CMAQ) model. The comparison between different scenarios indicates that substantial health benefits could accrue from achieving clean fleets, which aligns with the requirement of California’s scoping plan. Furthermore, we highlight the considerable reduction of premature deaths attributed to scoping plan-driven anthropogenic emission reductions for the communities of Wilmington, Carson, and West Long Beach (WCWLB) adjacent to the Ports of Los Angeles and Long Beach, which has been identified as one of the most pollution impacted disadvantaged communities in the state by California Air Resources Board.

Shupeng Zhu¹, Michael Mac Kinnon¹, Xiangyu Jiang², Kai Wu¹, Amaya Hernandez¹, G.S. Samuelsen^1,3,4

1 Advanced Power and Energy Program, University of California, Irvine, CA 92697, USA

2 Georgia Environmental Protection Division, Atlanta, GA 30354, USA

3 Department of Mechanical and Aerospace Engineering, University of California, Irvine, CA 92697 USA

4 Department of Civil and Environmental Engineering, University of California, Irvine, CA 92697 USA

In the United States (U.S.), exposure to health-damaging air pollution has decreased significantly over the last two decades. However, the regulatory drivers responsible for these improvements have not achieved the same level of improving the inequity of exposure among different racial and social-economic groups. In this study, we analyze mortality risks related to PM2.5 and ozone in the contiguous U.S. from 2002 to 2019, using daily pollutant concentrations, race, and ethnicity-specific health impact functions, and incidence rates at the census tract level. This allows us to examine temporal trends and variations in mortality burdens and their spatial distribution differences. To assess these inequalities, we introduce the Environmental Justice Index (EJI), which measures distribution differences in relation to the Social Vulnerability Index (SVI) of various communities. Our findings reveal that air pollution-related mortality risk decreased significantly for both pollutants. However, minority populations saw smaller reductions with an increasing ozone-related risk for Asian and Hispanic groups. The EJI reveals a growing inequality of both pollutants, especially after 2013 and among elderly populations, with 3~5 times higher inequality levels found among the elderly population for PM2.5 compared to the general population throughout the study period. Moreover, vulnerable communities (i.e., communities with top 10% SVI) experienced disproportionate air pollution risks in most states, and growing inequalities in nearly half of all states. In 2019, communities with the highest 10% mortality risk were more likely to have larger minority populations, limited English proficiency, and a higher housing density than in 2002. The study illustrates the importance of demographic-specific assessments in identifying and addressing environmental injustices.

Daniel Schuch, Yang Zhang, Mariana Império, Roberto Schaeffer, Sergio Ibarra-Espinosa, Maria de Fatima Andrade, Mario Eduardo Gavidia Calderón, and Michelle L. Bell

Air pollution has a crucial effect on human health. It has been considered the number one environmental cause of death and is deeply uneven across different socioeconomic groups. Future air quality will be affected by many different factors such as climate change and changes in natural and anthropogenic emissions. The objective of this work is to explore how the current emission control programs, technologies, and policy measures impact future air quality in Brazil in 2050. In this work, the Weather Research and Forecasting model coupled with Chemistry (WRF-Chem) is used to simulate air quality over triple-nested domains for Brazil, Southwest region, Metropolitan Area of São Paulo (MASP) and Metropolitan Area of Rio de Janeiro (MARJ) considering the current (2012-2016) and future (2048-2052) emissions and climate conditions. The emissions for 2012-2016 are generated based on a combination of the Emissions Database for Global Atmospheric Research (EDGAR), the local emissions estimated using the Vehicular Emissions Inventories model (VEIN) applied to MASP and a local inventory for MARJ. The projected emissions for 2048-2052 are based on the present emissions and projected emission change factors (ECFs) calculated from the Brazilian Land Use and Energy Systems (BLUES) model outputs for each region, sector, and pollutant considering current policies. For the current polices scenario, the BLUES model considers the maintenance and expansion of the country’s existing installed power capacity, considers the control of deforestation, and the implementation of existing control measures for reducing emissions from stationary sources and fleet renovation, which imposes stricter emissions restrictions on new vehicles. The calculated ECFs show increased emissions of criteria pollutants from the energy and industrial sectors. The country’s PM2.5 emissions increase by 80% due to changes in industrial emissions over all regions and energy emissions in the North, Northeast, and Center-West of Brazil. VOCs concentrations are reduced by 25% due to reduced VOCs emissions from multiple sectors. NOx emissions increase 22% due to increased industrial emissions. Two sets of simulations are performed: (1) a baseline simulation considering present emissions and (2) a simulation with emission change scenario considering projected emissions under the business-as-usual (BAU) scenario. Results show an average increase of 13.7% for the maximum 8-hour O3 and an increase of 42.8% for PM2.5 over Brazil under the BAU scenario. These simulations are used to assess health impact of O3 and PM2.5 and demonstrate the impacts of current emission reduction polices on both air quality and human health in MASP and MARJ, Brazil.

Erfan Hajiparvaneh 

Hossein Alizadeh 

Charles Robert Koch 

Vahid Hosseini 

Air pollution in Canada was responsible for 15300 premature deaths in 2021. To motivate continuous actions for improving human health and emission reduction, Canadian Ambient Air Quality Standards (CAAQS) have been revised and will be more stringent starting 2025. The Province of Alberta, which has one of the largest oil reservoirs in the World, is the highest NOx emitter in Canada and has exceeded the CAAQS standard for NO2. In this study, an air quality model composed of WRF and CMAQ  was deployed to determine the sensitivity of ambient NO2 to the major anthropogenic NO2 sources including the upstream oil and gas (UOG) and transportation in Alberta. The modeling results show that Transportation and mobile sources which contribute to the 15% of the total NOx emissions in Alberta are responsible for 53% of the ambient NO2 concentration in the populated cities of Edmonton and Calgary. Analyzing emission abatement scenarios for UOG and transportation sources indicates that there is a linear correlation between the ambient NO2 concentration and emission reduction of the major sources.  The modeling results suggest that to achieve a new CAAQS 12 ppb threshold for NO2 concentrations, potential mitigation policies in urban areas should focus on mobile and transportation emissions. Considering the observed NO2 concentration in 2019, the results show that a minimum of 23% emission reduction of transportation sources is required for complying with the new standards.

Charles Baschnagel, P.h.D.,

EpiMaps is an innovative population health prediction approach developed by Booz Allen Hamilton that harnesses AI to identify on upstream drivers of poor community outcomes, including social determinants of health, and allows for targeted population intervention. Demonstrations of EpiMaps have identified several use cases to date due as its predictive capabilities allow users to identify factors that have the greatest impacts on public health. These include mental health outcomes, the impacts of exposure to high temperatures, tracking covid-19, and assisting in scenario planning. It has great potential for applications related to environmental justice, public health, air quality and climate change. Through predictive modeling and easy to comprehend visualization, EpiMaps allows leaders to implement targeted investigations and interventions which saves resources while improving their delivery.

Barron H. Henderson, Halil Cakir, Brett Gantt, Ben Wells, Janica Gordon, Phil Dickerson, Hai Zhang, Alqamah Sayeed, Shobha Kondragunta, and help from the HAQAST AirNow team

This team has been developing fusions of monitors, sensors, models, and satellites to make the best air quality estimates available for AirNow. We will present a multi-data-source data fusion method, characterize the performance with validation techniques, highlight the strengths and weaknesses, and characterize the feasibility of integration into AirNow.

It is well known that surface monitors represent the best estimate of the state of the atmosphere near the monitors, but that monitors are spatially and temporally sparse compared to other data sources like low-cost sensors, satellites, or numerical models. Surface monitors are often interpolated to cover the gaps, but interpolations have known issues, especially in spatially/functionally biased observations. The monitors that report to AirNow are biased towards urban locations, and biased towards areas of high pollution where the monitors are most needed. Even so, they may miss pollution hot spots – either anthropogenic or natural like wild fires. Low-cost sensors, satellites and forecast models provide better spatial or temporal coverage, but are well-known to have their own issues. So, the question is how do we bring them together and how do we evaluate the result?

Models and surface monitors have long been “fused” using techniques that boil down to bias correction. Biases can be known as specific monitor locations, which can be interpolated to correct the continuous model. The challenge is deciding how to develop a bias correction methodology from many sources of data and leverage their strengths and weaknesses.

This talk will cover a low-complexity solution. The model results have been fused with regulatory grade surface monitors, fused with calibrated and aggregated sensor measurements, and fused with satellite inferred PM25. The model serves as the synthesizing element, which allows each sparse observation product to be differentially applied before a final data fusion.

The system has been applied to a year of data (June 2021 to June 2022) including a separate training/testing cross-validation configurations for each fusion component and the multi-source fusions. This talk will cover the methodology, validation, and outline a plan for integration into AirNow.

Maria de Fatima Andrade*, Mario Gavidia Calderon*, Edmilson Dias de Freitas*, Taciana Toledo de Almeida Albuquerque#.

*Instituto de Astronomia, Geofísica e Ciências Atmosféricas, Universidade de São Paulo

# Departamento de Engenharia Sanitária e Ambiental, Universidade Federal de Minas Gerais

The recognition of the necessity to enhance the representation of the urban process in air quality models and UHI (urban heat island) intensities has led to a better description of the factors which influence meteorological fields, topography, and urban characteristics. These factors include radiative balance, turbulent transport, and resistance effects, as well as anthropogenic energy production resulting from the presence of the buildings and vegetation within the urban canopy. To be possible to incorporate in the Chemical Transport Model these heterogeneities and accurately quantify population exposure to air pollution, it is essential to establish a coupling between a mesoscale and a microscale model. This coupling allows for a comprehensive understanding of the complex interactions between different scales and provides a basis for accurate environmental modeling applications, such as air quality forecasts, and the health effect of air pollution. One prominent international project addressing these challenges is the World Urban Database and Access Port Tool (WUDAPT) (Ching et al., 2018). WUDAPT aims to provide the necessary means, framework, and infrastructure to support a wide range of grid-based environmental modeling applications that are tailored to specific purposes and spatial scales. Its comprehensive database enables customized utilization of the tool in any location worldwide. In Brazil, the inclusion of microscale processes in air quality models has evolved to incorporate various schemes for radiative balance within the urban canopy layer. Ferreira et al. (2017) demonstrated that WUDAPT offers a satisfactory combination of accuracy and time efficiency for Land Climate Zone (LCZ) classification. Similarly, Pellegatti et al. (2019) showed that utilizing LCZ parameters from the WUDAPT map for Sao Paulo improved the meteorological fields and simulated ozone concentrations within the WRF-Chem model. In a study conducted by Hoppe et al. (2022) in Santa Catarina, Brazil, the authors elaborated on the LCZ map using the WUDAPT database. This information, combined with temperature measurements, was employed to describe the formation of the urban heat island. In another investigation by Gavidia-Calderon et al. (2021) in Sao Paulo, a high-resolution model called MUNICH (Model of Urban Network of Intersecting Canyons and Highways) was utilized to simulate ozone and NOx in street urban canyons. To accurately represent building heights and the local climate zone map, the researchers incorporated data from the WUDAPT database. Overall, the integration of WUDAPT and its associated tools and databases provides valuable insights into the representation of urban processes in air quality models, facilitating improved accuracy and understanding of the complex interactions between meteorological factors, topography, and urban features.

References:

Ching, Jason, et al. "WUDAPT: An urban weather, climate, and environmental modeling infrastructure for the Anthropocene." Bulletin of the American Meteorological Society 99.9 (2018): 1907-1924.

Ferreira, Luciana Schwandner, et al. "Mapping Local Climate Zones for São Paulo Metropolitan Region: a comparison between the local climate zone map and two other local maps." PASSIVE LOW ENERGY ARCHITECTURE (2017): 255-262.

Franco, Dirce Maria Pellegatti, et al. "Effect of Local Climate Zone (LCZ) classification on ozone chemical transport model simulations in Sao Paulo, Brazil." Urban Climate 27 (2019): 293-313.

Gavidia-Calderón, Mario Eduardo, et al. "Simulation of O 3 and NO x in São Paulo street urban canyons with VEIN (v0. 2.2) and MUNICH (v1. 0)." Geoscientific Model Development 14.6 (2021): 3251-3268.

Hoppe, Ismael Luiz, et al. "Local climate zones, sky view factor and magnitude of daytime/nighttime urban heat islands in Balneário Camboriú, SC, Brazil." Climate 10.12 (2022): 197.

Ankur Prabhat Sati and Gerald Mills

Multi-scale atmospheric modelling allows the integration of drivers and processes that dominate at different spatial and temporal scales. In cities, this approach is needed to capture the effects of micro-scale (building and street) impacts on local (neighborhood) and urban scale atmosphere and vice-versa. This integration is needed to understand and manage urban pollutant emissions, which are highly heterogenous over time and space. This work focuses on Carbon emissions which are the subject of urban-scale policies to create low (and even zero) Carbon cities. The research presented here couples the Weather Research Forecasting (WRF) and Model of Urban Network of Intersecting Canyons and Highways (MUNICH) models to simulate Carbon emissions in Dublin (Ireland) with a view to developing neighborhood-scale mitigation policies. Initially, WRF-chem is applied to nested domains to capture regional, national, and urban effects: the island of Ireland is simulated at a 1 km resolution and Dublin city at 250 m resolution. The model setup includes a comprehensive treatment of meteorology, land use (including urban cover using WUDAPT-derived Local Climate Zones (LCZs)), and biogenic/anthropogenic emissions and sequestration. Following evaluation, the WRF simulations are used to simulate the mixing of transport-based Carbon emissions using MUNICH, which requires details on the characteristics of the street network (e.g., width and height). The Carbon emissions will be estimated based on modelled traffic flows. The results of the simulations will be compared with street-level measurements of Carbon concentrations. The objective of this work is to create a modelling infrastructure suited to the development of urban policies to mitigate carbon emissions.

Corey Smithson, Bradley Adams

The increasing urbanization of the greater Salt Lake City area (GSLA) has contributed to the development of an urban canopy over this area. This canopy refers to the effects of building profiles, varying land surface properties and anthropogenic heating on local meteorological conditions including temperature, humidity, and wind velocity. Urban Canopy Models (UCMs) can be used to represent these characteristics on a mesoscale without needing to develop models accounting for effects of individual buildings. An effective method used to classify urban areas is WUDAPT Local Climate Zones (LCZs), which assign properties to 10 different types of urban areas.

A baseline model that represents current GSLA conditions was developed using a series of sensitivity studies, which focused on the effects of nested mesh resolution, land surface models, UCMs, and LCZ urban classifications. Use of detailed local climate zones (LCZs) to classify urban properties resulted in greater temperature variations between regions across the GSLA. Use of LCZs produced 5-10 °C higher temperatures in more urban areas whereas use of LCZs produced 4-6 °C lower temperatures around the edges of the valley. This was consistent with the more refined heating characteristics available with LCZs. The baseline GSLA model was validated using measured meteorological data. Three urban growth scenarios were simulated using modified LCZs and compared to this baseline model to evaluate the effects of future growth on local 2-meter air temperatures, 2-meter relative humidity, and 10-meter wind speed. Results showed increased urban density did not affect daytime temperatures within the GSLA, but did increase local nighttime temperatures 3-11 °C, depending on location. The effects of anthropogenic heating rates were most noticeable during early nighttime hours. Predicted relative humidity generally was lower with the urban development, whereas 10-m wind speed was higher or lower depending on the valley location and type of LCZ changes.

Sarah Conner, Chris Round PhD, Prachi Suthantankar, Christine Capozzi

Booz Allen’s Cloud-native Climate Intelligence Ecosystem accelerates research-to-operations by enabling timely, scientifically grounded decisions in climate change-impacted areas of policy, mission, and business. The Ecosystem’s "as a Service" features can support a community of data providers and flex to meet the common data challenges across any environmental and earth science domain. The Ecosystem has the potential to transform the study of air quality issues, accelerating insights using AI and advanced analytics in the cloud. The Climate Intelligence Ecosystem’s best-in-class open-source and cloud-native services are built on open architecture and hosted on Google Cloud Platform, with the ability to integrate with multiple clouds. By leveraging reusable software components, data delivery, and machine learning patterns to furnish necessary cloud architecture, the Ecosystem provides a cloud-based data science workbench capable of analytics and modeling on demand with no environment or infrastructure setup required for users. The Ecosystem’s data catalog contains pre-loaded, high-quality environmental data, and users can add their own datasets to maximize impact in analytics and prediction. On ingestion, data is flattened and stored in a common schema, solving challenges with data conversion and enabling fusion of disparate datasets. The stored data is programmatically optimized for parallel computing allowing greater throughput processing, including for scientific prediction models, effectively scaling horizontally instead of vertically. Any output produced by the Climate Intelligence Ecosystem can easily be migrated to other platforms, with the option to produce data visualizations and dashboards. A secure API layer offers the ability to feed outside data products, such as Digital Twins or commercial visualization tools like ESRI, Tableau, and Qlik.

Jeffrey M Vukovich (USEPA), George Pouliot (USEPA), James Beidler (GDIT)

The National Emissions Inventory (NEI) published by the U.S. Environmental Protection Agency (EPA) every three years details the annual emissions of criteria and hazardous air pollutants from all sources, including wildland fires (wild and prescribed fires). For the last five cycles of the NEI (2008-2020), a Bluesky modeling framework has been used that includes a module to identify fuels burned, a second module to estimate the amount of fuel consumed, and a third module to apply emissions factors on the consumed fuels to get the final emissions estimate. The Fire Emission Production Simulator (FEPS) emissions factors have been used in the third module for these previous NEI cycles. This case study examines the impacts on emissions on year 2021 wildland fires using the Smoke Emissions Repository Application (SERA) emissions factor database instead of the FEPS emissions factors. The SERA database is a compilation of published field and laboratory emission factors of wildland fire across the United States and Canada. SERA is a searchable online database of existing emissions factors of 276 known air pollutants that includes factors for fuel type, study type (laboratory or field), measurement platform (ground, tower or air-based), geographic location and source reference. The database was created to enable analysis and summaries of existing emissions factors, and creation of average emissions factors to be used in generating fire inventory emissions and other decision support tools for smoke management. This case study will compare year 2021 wildland fire emissions using FEPS and SERA emissions factors by fire type, by combustion phase, by region and by month.

James Beidler, Jeff Vukovich, George Pouliot

Smoke originating from wildland fires in Canada has a substantial impact on the summer air quality in the United States. In previous emissions modeling platforms the U.S. Environmental Protection Agency (EPA) has used a combination of wildland fire inventory data from Environment and Climate Change Canada (ECCC) and the Fire Inventory from NCAR (FINN). For year 2021 a Canada wildland fire emissions estimate approach consistent with the National Emissions Inventory (NEI) United States wildland fire emissions methods was developed and applied. The 2021 Canada approach utilized the Hazard Mapping System (HMS) fire and smoke product along with Canada’s National Burned Area Composite (NBAC) as the primary sources of fire activity. Fuel beds were identified from Canada’s National Forest Inventory (NFI) raster. The resulting fire activity and emissions estimates were compared to other 2021 Canada fire inventories.

Jeremiah Johnson, Pradeepa Vennam, Chris Emery and Greg Yarwood

Fires are large emission sources affecting ozone and PM_2.5 concentrations over regional scales. Therefore, accurate Fire Emission Inventories (FEIs) are needed for exceptional event analyses and State Implementation Plan (SIP) modeling. Emission estimates from currently available FEIs can differ by an order of magnitude so the decision of which FEI to include in modeling may have a significant impact on modeled air quality. In 2022, we developed a Python-based tool to process three different global FEIs into model-ready inputs: 1) FINN2.5; 2) Global Fire Assimilation System version 1.2 (GFAS1.2); and 3) Quick Fire Emissions Dataset version 2.5 (QFED2.5). This year, we developed the capability to process two additional FEIs: 1) Fire Energetics and Emissions Research (FEER1.0): and 2) Regional ABI and VIIRS fire Emissions version 1.0 (RAVE1.0). Additionally, we implemented new options for temporal emission allocation and vertical plume rise schemes. This project also evaluated photochemical model performance using the different FEIs from which to make recommendations on the FEI(s) to include in SIP modeling. We evaluated CAMx model performance for four different FEIs (FINN2.5, GFAS1.2, QFED2.5 and FEER1.0) for an April-May 2019 episode in Texas. We found large positive ozone biases related to FINN2.5 fire emissions throughout the episode when transport of smoke from biomass burning in Mexico and Central America was frequent. FINN2.5 emissions are estimated from satellite-estimated burned area. However, the three FEIs (GFAS, QFED, FEER) based on satellite-derived Fire Radiative Power (FRP) all showed substantially smaller ozone biases and overall better statistical agreement with observations. We identified GFAS1.2 as the best representation of fires due to overall ozone model performance and its reporting of useful parameters such as FRP and vertical plume information. We then conducted additional testing using fire emission inputs based on all four combinations of vertical and temporal allocation schemes applied to the GFAS1.2 FEI. Ozone and PM_2.5 concentrations were nearly identical across the four tests given long range transport distances between the fires and Texas, which moderated effects from plume rise and temporal treatments. We plan to augment air quality model testing for additional years and seasons using all available FEIs and to evaluate ozone and PM_2.5 model performance against observations.

Havala Pye (US EPA), Lu Xu (NOAA/CIRES), Christine Allen (General Dynamics Information Technology), Eric Apel (NCAR), Donald Blake (University of California Irvine), Pedro Campuzano-Jost (CU Boulder), Matt Coggon (NOAA), Emma D’Ambro (US EPA), Jessica Gilman (NOAA), Thomas Hanisco (NASA), Barron Henderson (US EPA), Greg Huey (Georgia Tech), Jose Jimenez (CU Boulder), Faye McNeill (Columbia University), Ben Murphy (US EPA), Jeff Peischl (NOAA), T. Nash Skipper (ORISE at US EPA), Jason St. Clair (UMBC), Carsten Warneke (NOAA), Paul Wennberg (Caltech), Forwood Wiser (Columbia University), Glenn Wolfe (NASA/UMBC), and Caroline Womack (NOAA/CIRES)

Wildfires are an increasingly prominent source of emissions to air including particulate matter and hazardous air pollutants. Understanding the health implications of wildfire smoke is complicated by the fact that the composition of smoke emissions as well as their transformation products are incompletely characterized. In this work, we aim to build a relatively complete description of reactive organic carbon (ROC) emissions and their secondary products in wildland fires using a combination of observations and model predictions. Specifically, we gather observations from the DC-8 aircraft for western U.S. wildfires during the 2019 Fire Influence on Regional to Global Environments and Air Quality (FIREX-AQ) field campaign and compare to predictions from the Community Multiscale Air Quality (CMAQ) model. Within CMAQ, we use the Community Regional Atmospheric Chemistry Multiphase Mechanism (CRACMM) with AMORE isoprene chemistry which has expanded secondary organic aerosol precursors including phenols, cresols, furans, semivolatile organic compounds, and intermediate volatility organic compounds relevant for wildland fires. We find the base model captures 77 % by mole of measured gas-phase ROC emissions. However, underestimates in emissions result in organic aerosol being underestimated by a factor of 6 near source. After updating the emission inputs and chemical evolution of wildfire smoke, species concentrations will be extended to cancer and non-cancer estimates of toxicity.

Grace Vincent, Laura DeSantis, Ethan Patten, Sambit Bhattacharya

Real-time fire detection from image sequences is a highly sought-after capability in video surveillance applications, as it plays a crucial role in preventing environmental disasters and ensuring continuous monitoring of both urban and forest environments. To achieve this goal, there is a strong need to deploy "intelligent cameras" equipped with onboard video analytics algorithms capable of detecting fires (flames and/or smoke) in real time. These cameras are strategically placed throughout the territory, often in remote locations, and operate independently with minimal computational support. Their primary function is to process the image sequences using a fire detection algorithm and promptly generate notifications for relevant authorities' alarm centers. In this context, it is essential to strike a balance between fire detection accuracy, notification speed, and computational resources, as overly resource-intensive methods are impractical for this application. Therefore, the focus is on developing highly efficient techniques that can deliver reliable results without overwhelming the available processing capabilities. In this project we are developing and training machine learning (ML) models which can classify images based on presence or absence of image features corresponding to fire or no fire seen in the images. Of specific interest are ML models that are computationally efficient, low latency with regards to model inference, and do not require memory that can exceed what is typically available onboard for intelligent cameras.

Contact Information::Joseph Wilkins, PhD, Howard University, joseph.wilkins@howard.edu; Additional Presenters/Authors: Miriam Marlier, PhD, UCLA Department of Environmental Health Sciences, mmarlier@ucla.edu; Rachel Connolly, PhD, UCLA Department of Environmental Health Sciences, rachelconnolly@g.ucla.edu; Jihoon Jung, PhD, University of North Carolina at Chapel Hill, climategeo@gmail.com Claire Schoellart, PhD, University of Washington, Dept. of Environmental and Occupational Health Sciences, cscholla@uw.edu Eimy Bonilla, PhD, Howard University, eimy.bonilla@howard.edu Osinachi Ajoku, PhD, Howard University, osinachi.ajoku@howard.edu

In California, wildfire frequency and severity have grown substantially in the last several decades due to climate change and other anthropogenic factors such as increasing development at the wildland-urban interface (WUI). Currently, 30% of the population (2.5 million persons) live in “at-risk” areas with approximately 35% of the state’s housing structures in the WUI. Residents living in California face increased risks from wildfires that threaten health and safety compared to the rest of the United States. Such exposures include direct risks, such as active wildfire threats, as well as indirect risks, such as smoke exposure from nearby fires. While existing studies have evaluated direct and indirect risks separately, to our knowledge no research has explored the spatial variation in the two types of risk and explored associated population vulnerability. In this study, we use recently released California-specific WUI data developed using remote sensing estimates, Cal FIRE perimeters, and 11 years of Community Multiscale Air Quality (CMAQ) model wildland fire-specific PM2.5 estimates to map and analyze exposures in the WUI. We use the fire perimeters, alongside available structure damage data, to analyze the spatial and temporal distribution of direct threats. We then use the modeled air pollution data to estimate adverse health outcomes associated with smoke exposure in the WUI. Finally, we overlay these findings with vulnerability data from the California Office of Health Hazard Assessment’s CalEnviroScreen environmental health screening tool and the Healthy Places Index to explore population vulnerability trends. These findings will have implications for wildfire management in California, as well as the provision of resources to vulnerable communities living in the state’s WUI regions.

Matthew Alvarado, Jen Hegarty, Rick Pernak, Mike Iacono, John Henderson

Verisk Atmospheric and Environmental Research

Domestic fire emissions can have major impacts on surface ozone (O₃) concentrations both near the fires and hundreds of miles downwind. Understanding the impact of biomass burning smoke on urban O₃ air quality requires (i) understanding the chemistry of the smoke before it reaches the city and (ii) the changes in the urban production rate of O₃ caused by the smoke. The relative importance of these two pathways on the air quality impacts of domestic fire smoke is not well understood and it is unclear which processes should be targeted to reduce the overall uncertainty.

We present an examination of the impact of wildland fires on urban ozone (O₃) in Houston, Texas and El Paso, Texas using statistical modeling. Analyzing air from upwind background and urban surface sites separately allowed us to examine the change in O₃ due to the mixing of smoke with urban pollution separately from the impact of smoke before it mixes with urban pollution. We applied the same statistical methods to both the real-world surface observations and CAMx-simulated surface observations to determine if the impact of fires on urban air quality as simulated in CAMx is statistically equivalent to the impacts seen in the real-world data.

Our results suggest that on days when the NOAA Hazard Monitoring System (HMS) indicated smoke over Houston and El Paso, the background MDA8 O₃ was elevated by 6-8 ppbv on average. These impacts may be related to long-distance transport of smoke from the Yucatan (Houston and El Paso) and California (El Paso only). In Houston, the impact on the maximum MDA8 O₃ was much higher than the background (6 ppbv on average), suggesting urban area chemistry amplified the impact of the smoke on ozone. However, in El Paso we instead see a decrease of 1.5 ppbv in the average impact of smoke, suggesting the response of the chemistry in these urban areas to smoke transport is very different. For El Paso, our CAMx statistical analysis suggested that there were statistically significant differences between CAMx and the ambient data, but further analysis showed that the predicted impacts of fires in both cases were very similar. For Houston, however, the differences between CAMx and the ambient data fits were not statistically significant for maximum O₃, but the CAMx data strongly overestimates the background O₃ for Houston on both smoky and non-smoky days.

Kamal J. Maji¹, Zongrun Li¹, Jennifer D. Stowell², Chad Milando², Yongtao Hu¹, Gregory A. Wellenius², Patrick L. Kinney², Armistead G. Russell¹, Ambarish Vaidyanathan^1,3, and M. Talat Odman¹

¹Georgia Institute of Technology, Atlanta, Georgia, USA;

²Boston University, Boston, Massachusetts, USA;

³Centers for Disease Control and Prevention, Atlanta, Georgia, USA

Background. Prescribed fire (PF) is planned and intentionally set wildland fires, which are commonly used for reducing the risk of wildfires and managing wildland ecosystems. PFs are conducted under weather conditions different from those conducive to wildfires, resulting in different types of smoke with different constituents, concentrations, and proportions. To protect public health, there is a need for a better understanding of the impact of prescribed fires on air quality, especially in the southeastern US where many people live at the wildland-urban interface.

Methods. This study aimed to estimate the daily impact of PFs on air quality using satellite fire products, ground-based observations, and air quality modelling. (1) Daily PF information was estimated by discerning agricultural fire and wildfire from satellite-derived Fire INventory from NCAR (FINN). (2) A linear regression model was used to adjust burned area from FINN with burn permit records. The adjusted burned area was used in BlueSky Smoke Modeling Framework to estimate three-dimensional prescribed burning emissions. (3) The WRF and CMAQ models were used to simulate PF impacts on daily average PM_2.5 and daily maximum 8-hr average O₃ (MDA8-O₃) in 2020, at 4-km horizontal grid resolution over the southeastern US. The simulated PM_2.5 and MDA8-O₃ fields were then fused with daily observations at ambient surface monitors to generate “observation-adjusted burn impacts”.

Results. The CMAQ-fused model simulation captures the variability in PM2.5 (r²=0.68) and MDA8-O3 (r²=0.91) observations reasonably well. The fires simulation also captures the concentrations observed in the peaks and ambient conditions well for PM_2.5 (MB=-0.51 μg/m³, RMSE = 2.57 μg/m³, NMB = −6.3%) and MDA8-O3 (MB=-1.61 ppb, RMSE = 3.68 ppb, NMB = −4.45%). The impacts of PF on spatiotemporal variations of total-PM_2.5 and MDA8-O₃ are significant across the study region. In 2020, PFs contribution to annual average PM2.5 across the region was 0.81±1.18 µg/m³ (14.0±13.0% of atmospheric PM_2.5), whereas during active fire season (January-April) the average contribution was 0.95±1.18 µg/m³ (16.0±13.0% of atmospheric PM_2.5). However, the highest impact was observed in November, 1.72±1.78 µg/m³ (24.0±19% of atmospheric PM_2.5). Contribution to annual average MDA8-O₃ was relatively low, 0.32±0.43 ppb (0.6±0.9% of atmospheric MDA8-O₃) in 2020. However, PFs contribution increases to >1% of ambient MDA8-O₃ during March in large parts of the region, with an average contribution of 0.55± 0.54 ppb.

Conclusions. Prescribed fires contribute significantly to daily PM_2.5 and O₃ concentrations in the southeastern US. Exposure to air pollution from prescribed burns may be associated with a higher risk of ED visits for respiratory and cardiovascular problems.

Zongrun Li, M. Talat Odman, Yongtao Hu, Armistead G. Russell

Accurate simulation of prescribed burning is important for air quality management and fire management practices, especially in the southeastern United States. Fire behavior models can simulate fire propagation and resulting emissions in detail though typically do not consider atmospheric transformations. Chemical transport models simulate the transport and transformation of pollutants with simplified plume rise models. We designed and implemented a coupling algorithm to integrate a fire behavior model and a chemical transport model.

The fire behavior model simulates the spreading process of the fire and provides spatiotemporal information for the burned area, released heat, and dispersion of chemically inert smoke tracer. WRF-SFIRE is a fire behavior model which considers the interactions between fire and meteorology. WRF-SFIRE estimates fire propagation under real-time meteorological and fuel moisture conditions and has feedback to the local meteorology. The time profile of the smoke tracer in WRF-SFIRE is estimated by the burned area, fuel type, and consumed fuel load during the fire spreading process. The smoke is elevated by the buoyancy generated from the fire heat flux. Although the fire behavior model simulates fire processes and initial pollutant atmospheric transport, it is limited by the lack of chemical reactions, which hinders its ability to provide comprehensive air quality information, particularly secondary pollutant formation, e.g., ozone and secondary organic carbon (SOC).

On the other hand, while the chemical transport model considers the transport and chemical reactions of pollutants, it uses simple assumptions or parameterized plume rise models for simulating initial pollutant transport during fire events. CMAQ uses the Briggs plume rise model for point sources; WRF-Chem uses a 1D plume rise model proposed by Freitas et al. to estimate the injection height of fire emissions. GEOS-Chem assumes that all fire emissions are emitted into the boundary layer. Also, the incorporated plume rise model in chemical transport models cannot simulate plume structures specific to different ignition methods, which is important for fire management practices.

In this presentation, we present a coupling algorithm for integrating fire behavior and chemical transport models. We conducted the coupling algorithm for integrating WRF-SFIRE with CMAQ. We used the coupled model to simulate prescribed burning events in Fort Stewart and Fort Benning. The model performance was evaluated by smoke observations from instruments on the grounds. The method and findings will be useful for developing prescribed burning air quality models to assist fire management practices.