Khanh Do
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Khanh Do has a total of 26 co-authors. Here are the top co-authors by number of shared publications:
- Cesunica E. Ivey
Civil & Environmental Engineering, University of California, Berkeley, CA, United States of America
7 shared publications - Yang Zhang
Department of Civil and Environmental Engineering, Northeastern University, Boston, MA 02115, USA
3 shared publications Arash Kashfi Yeganeh
Department of Chemical and Environmental Engineering, University of California Riverside, Riverside, CA, USA
2 shared publications- Duncan Quevedo
Department of Civil and Environmental Engineering
2 shared publications Ziqi Gao
Department of Civil and Environmental Engineering, Georgia Institute of Technology, Atlanta, GA, USA
2 shared publications- Ziqi Gao
University of Virginia
2 shared publications
Latest Publications
- Open Access
- 22 October 2026
- Published by Elsevier in SSRN Electronic Journal
Comparison of Machine Learning Algorithms in Reproducing Real-Time Air Quality Forecast of a 3-D Air Quality Model over Southeastern U.S
Three-dimensional air quality models, or chemical transport models (CTMs), are powerful tools in predicting criteria air pollutants (CAPs), but they tend to have systematic biases due to uncertainties in input data, deficient model configuration and insufficient representations of complex physio-chemical processes. Also, the requirement of costly computational resources is another major challenge that hinders the broad use of such model systems. However, with the emergence of the rapidly booming machine learning (ML) techniques, including deep learning, those long-lasting limitations can now be addressed. ML techniques do not require being explicitly programmed but enable computers to learn patterns from data and make predictions. With the aid of ML techniques, it becomes possible to develop fast and easy-to-apply real-time air quality forecasting (RT-AQF) systems. This work explores the use of a spatial interpolation based bias correction technique (the k-nearest neighbor inverse distance weighting method) to improve the RT-AQF results from a CTM, Weather Research and Forecasting model with the Model of Aerosol Dynamics, Reaction, Ionization, and Dissolution (WRF-Chem-MADRID), and then the use of three ML models to reproduce bias-corrected predictions. The main objective is to assess how well the different ML models (multi-layer perceptron, random forest, and U-Net) can reproduce the bias corrected WRF-Chem-MADRID results through a “leave-one-out” 5-fold cross validation method. The evaluation shows a better temporal representation by random forest than multi-layer perceptron, and the success of U-Net in reproducing both the raw and enhance CTM outputs. This work demonstrates a strong capability of different ML models in capturing the spatiotemporal characteristics of two important CAPs, O3 and PM2.5, over southeastern U.S. in warm and cold months in 2022. It would lay a foundation in the development of a physics-guided ML technique in RT-AQF applications with less computational resources, while retaining the stability and accuracy of CTMs.
- 22 October 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessComparison of Machine Learning Algorithms in Reproducing Real-Time Air Quality Forecast of a 3-D Air Quality Model over Southeastern U.S
Three-dimensional air quality models, or chemical transport models (CTMs), are powerful tools in predicting criteria air pollutants (CAPs), but they tend to have systematic biases due to uncertainties in input data, deficient model configuration and insufficient representations of complex physio-chemical processes. Also, the requirement of costly computational resources is another major challenge that hinders the broad use of such model systems. However, with the emergence of the rapidly booming machine learning (ML) techniques, including deep learning, those long-lasting limitations can now be addressed. ML techniques do not require being explicitly programmed but enable computers to learn patterns from data and make predictions. With the aid of ML techniques, it becomes possible to develop fast and easy-to-apply real-time air quality forecasting (RT-AQF) systems. This work explores the use of a spatial interpolation based bias correction technique (the k-nearest neighbor inverse distance weighting method) to improve the RT-AQF results from a CTM, Weather Research and Forecasting model with the Model of Aerosol Dynamics, Reaction, Ionization, and Dissolution (WRF-Chem-MADRID), and then the use of three ML models to reproduce bias-corrected predictions. The main objective is to assess how well the different ML models (multi-layer perceptron, random forest, and U-Net) can reproduce the bias corrected WRF-Chem-MADRID results through a “leave-one-out” 5-fold cross validation method. The evaluation shows a better temporal representation by random forest than multi-layer perceptron, and the success of U-Net in reproducing both the raw and enhance CTM outputs. This work demonstrates a strong capability of different ML models in capturing the spatiotemporal characteristics of two important CAPs, O3 and PM2.5, over southeastern U.S. in warm and cold months in 2022. It would lay a foundation in the development of a physics-guided ML technique in RT-AQF applications with less computational resources, while retaining the stability and accuracy of CTMs.
- Open Access
- 26 April 2025
- Published by Elsevier in Environmental Pollution
- Vol. 375, 126311
Assessing health risks and socioeconomic disparities associated with ambient air pollution and point sources across the United States
- 26 April 2025
- Published by Elsevier in Environmental Pollution
- Vol. 375, 126311
Open AccessAssessing health risks and socioeconomic disparities associated with ambient air pollution and point sources across the United States
- Open Access
- 14 February 2025
- Published by IOP Publishing in Environmental Research Letters
- Vol. 20 (3) , 034018
Assessing air pollution exposure disparities in disadvantaged communities of Greater Boston: a new cumulative environmental justice score system
Disadvantaged communities in the U.S. experience higher levels of air pollution, leading to increased environmental justice (EJ) challenges. In this work, we combine socioeconomic metrics with spatially interpolated fine particulate matter and ozone concentrations from fifty-nine low-cost sensors and nine reference monitors to investigate inequality in air pollution exposure among different population groups in Greater Boston. The results indicate that the disadvantaged groups are exposed to higher levels of air pollution. Air pollution is concentrated in the center of Boston, Cambridge, Chelsea, and Everett, corresponding to high population density and anthropogenic activities. We further assess the environmental equality by reconstructing the EJ map for Massachusetts using median household income, percentage of minority, and percentage of limited English speaking as indicators. The EJ blocks are clustered in large cities across Massachusetts. Many communities in Greater Boston are classified as EJ communities due to the high percentage of minorities. Chelsea and Everett face more severe EJ issues, with most of their communities exceeding multiple EJ indicators. Additionally, we introduce and demonstrate a new method for determining EJ communities by utilizing the multi-criterion framework to calculate the cumulative EJ score for every census block, integrating air quality with socioeconomic indicators, and each indicator can be viewed as a policy action. Chelsea, Everett, and Lynn consistently exhibit the highest EJ scores with no improvements over the years. The EJ score system provides the time evolution to assist authorities and the residents in tracking the progress in improving the EJ scores of their communities. The method indicates the strength of the efforts needed to maintain or improve the EJ scores for achieving better EJ in the communities, allowing authorities to take immediate actions to improve the environmental well-being of vulnerable communities.
- 14 February 2025
- Published by IOP Publishing in Environmental Research Letters
- Vol. 20 (3) , 034018
Open AccessAssessing air pollution exposure disparities in disadvantaged communities of Greater Boston: a new cumulative environmental justice score system
Disadvantaged communities in the U.S. experience higher levels of air pollution, leading to increased environmental justice (EJ) challenges. In this work, we combine socioeconomic metrics with spatially interpolated fine particulate matter and ozone concentrations from fifty-nine low-cost sensors and nine reference monitors to investigate inequality in air pollution exposure among different population groups in Greater Boston. The results indicate that the disadvantaged groups are exposed to higher levels of air pollution. Air pollution is concentrated in the center of Boston, Cambridge, Chelsea, and Everett, corresponding to high population density and anthropogenic activities. We further assess the environmental equality by reconstructing the EJ map for Massachusetts using median household income, percentage of minority, and percentage of limited English speaking as indicators. The EJ blocks are clustered in large cities across Massachusetts. Many communities in Greater Boston are classified as EJ communities due to the high percentage of minorities. Chelsea and Everett face more severe EJ issues, with most of their communities exceeding multiple EJ indicators. Additionally, we introduce and demonstrate a new method for determining EJ communities by utilizing the multi-criterion framework to calculate the cumulative EJ score for every census block, integrating air quality with socioeconomic indicators, and each indicator can be viewed as a policy action. Chelsea, Everett, and Lynn consistently exhibit the highest EJ scores with no improvements over the years. The EJ score system provides the time evolution to assist authorities and the residents in tracking the progress in improving the EJ scores of their communities. The method indicates the strength of the efforts needed to maintain or improve the EJ scores for achieving better EJ in the communities, allowing authorities to take immediate actions to improve the environmental well-being of vulnerable communities.
- Open Access
- 6 January 2025
- Published by American Chemical Society (ACS) in ACS ES&T Air
- Vol. 2 (2) , 226-235
GPU Implementation of a Gas-Phase Chemistry Solver in the CMAQ Chemical Transport Model
The Community Multiscale Air Quality (CMAQ) model simulates atmospheric phenomena, including advection, diffusion, gas-phase chemistry, aerosol physics and chemistry, and cloud processes. Gas-phase chemistry is often a major computational bottleneck due to its representation as large systems of coupled nonlinear stiff differential equations. We leverage the parallel computational performance of graphics processing unit (GPU) hardware to accelerate the numerical integration of these systems in CMAQ's CHEM module. Our implementation, dubbed CMAQ-CUDA, in reference to its use in the Compute Unified Device Architecture (CUDA) general purpose GPU (GPGPU) computing solution, migrates CMAQ's Rosenbrock solver from Fortran to CUDA Fortran. CMAQ-CUDA accelerates the Rosenbrock solver such that simulations using the chemical mechanisms RACM2, CB6R5, and SAPRC07 require only 51%, 50%, or 35% as much time, respectively, as CMAQv5.4 to complete a chemistry time step. Our results demonstrate that CMAQ is amenable to GPU acceleration and highlight a novel Rosenbrock solver implementation for reducing the computational burden imposed by the CHEM module.
- 6 January 2025
- Published by American Chemical Society (ACS) in ACS ES&T Air
- Vol. 2 (2) , 226-235
Open AccessGPU Implementation of a Gas-Phase Chemistry Solver in the CMAQ Chemical Transport Model
The Community Multiscale Air Quality (CMAQ) model simulates atmospheric phenomena, including advection, diffusion, gas-phase chemistry, aerosol physics and chemistry, and cloud processes. Gas-phase chemistry is often a major computational bottleneck due to its representation as large systems of coupled nonlinear stiff differential equations. We leverage the parallel computational performance of graphics processing unit (GPU) hardware to accelerate the numerical integration of these systems in CMAQ's CHEM module. Our implementation, dubbed CMAQ-CUDA, in reference to its use in the Compute Unified Device Architecture (CUDA) general purpose GPU (GPGPU) computing solution, migrates CMAQ's Rosenbrock solver from Fortran to CUDA Fortran. CMAQ-CUDA accelerates the Rosenbrock solver such that simulations using the chemical mechanisms RACM2, CB6R5, and SAPRC07 require only 51%, 50%, or 35% as much time, respectively, as CMAQv5.4 to complete a chemistry time step. Our results demonstrate that CMAQ is amenable to GPU acceleration and highlight a novel Rosenbrock solver implementation for reducing the computational burden imposed by the CHEM module.
- Open Access
- 7 August 2024
- Published by Royal Society of Chemistry (RSC) in Environmental Science: Atmospheres
- Vol. 4 (9) , 1051-1063
Assessing CMAQ model discrepancies in a heavily polluted air basin using UAV vertical profiles and sensitivity analyses
An unmanned aerial vehicle was deployed daily in Riverside, CA from August to November 2020, capturing vertical ozone and particulate matter measurements. Flights took place in the early morning and late afternoon, resulting in 321 vertical profiles from the surface to 500 m above ground level. The measured ozone mixing ratio is statistically compared with ground-based measurements at the Riverside–Rubidoux regulatory air monitoring site in Jurupa Valley, CA and with CMAQ simulated concentrations to assess consistency with the nearest reference monitor and model skill at reproducing the observed vertical structure, respectively. The default model configuration overestimates ground-level ozone by 17.7 ppb in the morning and underestimates it by an average of 2.9 ppb in the afternoon. The sensitivity of the model to factors such as planetary boundary layer (PBL), eddy diffusivity, NOx emissions, and VOC emissions is investigated by modifying key physics and emissions settings in a series of simulations. We found that our default PBL scheme used in the default CMAQ simulation negatively biases the PBL height in the nighttime and positively biases it in the daytime compared to the observations retrieved from a ceilometer. For the observational region of interest, NOx emissions are concluded to be largely underestimated, leading to biases in modeled ozone concentration. We conclude with recommendations for achieving model parity with localized measurements.
- 7 August 2024
- Published by Royal Society of Chemistry (RSC) in Environmental Science: Atmospheres
- Vol. 4 (9) , 1051-1063
Open AccessAssessing CMAQ model discrepancies in a heavily polluted air basin using UAV vertical profiles and sensitivity analyses
An unmanned aerial vehicle was deployed daily in Riverside, CA from August to November 2020, capturing vertical ozone and particulate matter measurements. Flights took place in the early morning and late afternoon, resulting in 321 vertical profiles from the surface to 500 m above ground level. The measured ozone mixing ratio is statistically compared with ground-based measurements at the Riverside–Rubidoux regulatory air monitoring site in Jurupa Valley, CA and with CMAQ simulated concentrations to assess consistency with the nearest reference monitor and model skill at reproducing the observed vertical structure, respectively. The default model configuration overestimates ground-level ozone by 17.7 ppb in the morning and underestimates it by an average of 2.9 ppb in the afternoon. The sensitivity of the model to factors such as planetary boundary layer (PBL), eddy diffusivity, NOx emissions, and VOC emissions is investigated by modifying key physics and emissions settings in a series of simulations. We found that our default PBL scheme used in the default CMAQ simulation negatively biases the PBL height in the nighttime and positively biases it in the daytime compared to the observations retrieved from a ceilometer. For the observational region of interest, NOx emissions are concluded to be largely underestimated, leading to biases in modeled ozone concentration. We conclude with recommendations for achieving model parity with localized measurements.
- Open Access
- 29 March 2024
- Published by Royal Society of Chemistry (RSC) in Environmental Science: Atmospheres
- Vol. 4 (4) , 488-500
Performance of machine learning for ozone modeling in Southern California during the COVID-19 shutdown
We combine machine learning (ML) and geospatial interpolations to create two-dimensional high-resolution ozone concentration fields over the South Coast Air Basin (SoCAB) for the entire year of 2020. The interpolated ozone concentration fields were constructed using 15 building sites whose daily trends were predicted by random forest regression. Spatially interpolated ozone concentrations were evaluated at 12 sites that were independent from the machine learning sites and historical data to find the most suitable prediction method for SoCAB. Ordinary kriging interpolation had the best performance overall for 2020. The model is 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. All interpolation methods poorly predicted and underestimated ozone concentrations for Crestline during summer, indicating that the site has a distribution of ozone concentrations that is 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. The study demonstrates the utility of ML and geospatial techniques for evaluating air pollution levels during anomalous periods. Both ML and the Community Multiscale Air Quality model do not fully capture the irregularities caused by emission reductions during the COVID-19 lockdown period (March–May) in the SoCAB. Including 2020 training data in the ML model training improves the model's performance and its potential to predict future abnormalities in air quality.
- 29 March 2024
- Published by Royal Society of Chemistry (RSC) in Environmental Science: Atmospheres
- Vol. 4 (4) , 488-500
Open AccessPerformance of machine learning for ozone modeling in Southern California during the COVID-19 shutdown
We combine machine learning (ML) and geospatial interpolations to create two-dimensional high-resolution ozone concentration fields over the South Coast Air Basin (SoCAB) for the entire year of 2020. The interpolated ozone concentration fields were constructed using 15 building sites whose daily trends were predicted by random forest regression. Spatially interpolated ozone concentrations were evaluated at 12 sites that were independent from the machine learning sites and historical data to find the most suitable prediction method for SoCAB. Ordinary kriging interpolation had the best performance overall for 2020. The model is 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. All interpolation methods poorly predicted and underestimated ozone concentrations for Crestline during summer, indicating that the site has a distribution of ozone concentrations that is 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. The study demonstrates the utility of ML and geospatial techniques for evaluating air pollution levels during anomalous periods. Both ML and the Community Multiscale Air Quality model do not fully capture the irregularities caused by emission reductions during the COVID-19 lockdown period (March–May) in the SoCAB. Including 2020 training data in the ML model training improves the model's performance and its potential to predict future abnormalities in air quality.
- 10 February 2024
- Published by Elsevier in Atmospheric Environment
- Vol. 323
Predicting PM2.5 levels and exceedance days using machine learning methods
- 10 February 2024
- Published by Elsevier in Atmospheric Environment
- Vol. 323
Predicting PM2.5 levels and exceedance days using machine learning methods
- Open Access
- 29 November 2023
- Published by American Chemical Society (ACS) in ACS ES&T Air
- Vol. 1 (1) , 33-42
Multidecadal Analysis of Meteorological and Emissions Regimes for PM2.5 Across California
- 29 November 2023
- Published by American Chemical Society (ACS) in ACS ES&T Air
- Vol. 1 (1) , 33-42
Open AccessMultidecadal Analysis of Meteorological and Emissions Regimes for PM2.5 Across California
- Open Access
- 15 June 2023
- Published by Royal Society of Chemistry (RSC) in Environmental Science: Atmospheres
- Vol. 3 (8) , 1159-1173
Emerging investigator series: a machine learning approach to quantify the impact of meteorology on tropospheric ozone in the inland southern California
The role of meteorology in facilitating the formation and accumulation of ground-level ozone is of great theoretical and practical interest, especially due to changing global climate. In this study, with appropriate machine learning algorithms, we analyzed large meteorology and air quality datasets to train machine learning models to (1) enhance the prediction of ozone levels in the South Coast Air Basin of California, (2) investigate the impact of recent meteorological shifts on ozone formation, and (3) determine the most critical factors influencing ozone exceedance hours. Random forest regression was used to predict historical and future trends of ozone levels, and k-nearest neighbor was used as a binary classifier for ozone exceedance prediction. The models were trained on meteorology data from Ontario and Los Angeles International Airport stations and air quality data from the Fontana, California air monitoring station, and data were collected for the 1994 to 2018 time period. Upon model evaluation, the correlation of the RFR model was 0.92, and the probability of detection for ozone exceedances using k-nearest neighbors was 0.81 for the most recent years of the analysis (2014–2018). We also ran a 4 km Community Multiscale Air Quality model simulation to generate air pollution estimates over Southern California. As expected, ozone in Fontana was positively correlated with temperature. The ozone exceedance hours usually occurred when the temperature was above 25 °C, and the wind direction was from 270° (westerly). Ozone sensitivity as a function of temperature and NOx was also examined. Observed troughs in hourly NOx concentrations during midday under high temperatures suggests that most of the ambient NOx reacted, also as expected. The results indicate that machine learning can support state implementation planning by complementing traditional air quality modeling, reducing simulation time, and exploiting large datasets for historical simulations and future air quality predictions.
- 15 June 2023
- Published by Royal Society of Chemistry (RSC) in Environmental Science: Atmospheres
- Vol. 3 (8) , 1159-1173
Open AccessEmerging investigator series: a machine learning approach to quantify the impact of meteorology on tropospheric ozone in the inland southern California
The role of meteorology in facilitating the formation and accumulation of ground-level ozone is of great theoretical and practical interest, especially due to changing global climate. In this study, with appropriate machine learning algorithms, we analyzed large meteorology and air quality datasets to train machine learning models to (1) enhance the prediction of ozone levels in the South Coast Air Basin of California, (2) investigate the impact of recent meteorological shifts on ozone formation, and (3) determine the most critical factors influencing ozone exceedance hours. Random forest regression was used to predict historical and future trends of ozone levels, and k-nearest neighbor was used as a binary classifier for ozone exceedance prediction. The models were trained on meteorology data from Ontario and Los Angeles International Airport stations and air quality data from the Fontana, California air monitoring station, and data were collected for the 1994 to 2018 time period. Upon model evaluation, the correlation of the RFR model was 0.92, and the probability of detection for ozone exceedances using k-nearest neighbors was 0.81 for the most recent years of the analysis (2014–2018). We also ran a 4 km Community Multiscale Air Quality model simulation to generate air pollution estimates over Southern California. As expected, ozone in Fontana was positively correlated with temperature. The ozone exceedance hours usually occurred when the temperature was above 25 °C, and the wind direction was from 270° (westerly). Ozone sensitivity as a function of temperature and NOx was also examined. Observed troughs in hourly NOx concentrations during midday under high temperatures suggests that most of the ambient NOx reacted, also as expected. The results indicate that machine learning can support state implementation planning by complementing traditional air quality modeling, reducing simulation time, and exploiting large datasets for historical simulations and future air quality predictions.
- 1 February 2021
- Published by Elsevier in Journal of Aerosol Science
- Vol. 152
A data-driven approach for characterizing community scale air pollution exposure disparities in inland Southern California
- 1 February 2021
- Published by Elsevier in Journal of Aerosol Science
- Vol. 152
A data-driven approach for characterizing community scale air pollution exposure disparities in inland Southern California