Wei Li

Operational chemical transport models provide physically consistent air-quality forecasts but remain computationally expensive, constraining forecast throughput and broader ensemble and scenario-based applications. We developed an expert-guided cross-model learning framework that jointly uses historical WRF-CMAQ simulations and target-domain simulations from NOAA’s operational Unified Forecast System Air Quality Model (UFS-AQM) to accelerate surface PM _2.5 forecasting. Scenario-diverse CMAQ simulations provide transferable training information, while separate total, nonfire, and fire simulations support expert specialization. Source- and target-domain samples are optimized jointly, and the resulting surrogates are evaluated over an independent operational period using fixed forecast initializations and target-defined masks. Training-data diversity produces the largest and most spatially consistent improvements. Relative to a surrogate trained only on the UFS-AQM baseline simulation, incorporating source-domain CMAQ simulations reduces pooled RMSE by approximately 6%, while combining source-domain information with limited target-domain perturbation data reduces RMSE by approximately 7%. Expert fusion provides a smaller but generally consistent refinement, reducing overall, high-pollution, extreme-regime, and spatial-gradient errors while modestly improving P95 exceedance detection and plume-footprint skill. The largest fusion benefit occurs for the baseline-only parent, where the specialized experts retain the greatest complementarity; the fire expert becomes increasingly useful in the upper tail of the PM _2.5 distribution. These results demonstrate that historical CTM simulations can serve as reusable cross-model training resources for adapting AI surrogates to new operational systems, providing a pathway toward computationally efficient atmospheric foundation models that complement process-based models.

Accurate near-real-time (NRT) air quality forecasting is critical for protecting public health but remains challenged by uncertainties in emissions, meteorology, and initial conditions within traditional chemical transport models (CTMs). We present an observation-driven machine learning framework that integrates multisource satellite products, including VIIRS aerosol optical depth and TEMPO NO2 with U.S. EPA AirNow measurements into a machine-learning surrogate of the NOAA Unified Forecast System (DeepAQM). A ConvLSTM-ResNet fusion module reconstructs spatially continuous surface PM2.5 and O3 fields from sparse observations, providing updated initial conditions for 72 h forecasts across the continental United States. DeepAQM reproduces UFS-AQM spatial patterns with strong agreement (R2 > 0.9 for O3 and up to 0.80 for PM2.5) while maintaining normalized mean bias within ±10%. Observation-updated initial conditions improve early forecast skill, increasing site-level R2 during the first 10 forecast hours from approximately 0.10–0.15 to 0.40–0.50 for PM2.5 and from approximately 0.40 to 0.50 for O3. Observation-based fine-tuning provides additional improvements throughout the 72 h forecast period, increasing overall R2 to 0.41 for PM2.5 and 0.82 for O3 during the independent evaluation period. These results demonstrate that integrating satellite and ground observations within DeepAQM can enhance NRT air quality forecasting and provide a scalable pathway toward globally deployable observation-driven prediction systems.

Biogenic isoprene emissions play an important role in air quality so it is important to quantify their response to extreme events such as drought. This study aims at constraining drought stress on biogenic isoprene emissions in South Korea using satellite formaldehyde (HCHO) column, the key product of isoprene oxidation, and a chemistry transport model (GEOS-Chem) with Model of Emissions of Gases and Aerosols from Nature version 2.1 (MEGAN2.1). It was found that the HCHO column from the Ozone Monitoring Instrument (OMI) increased by 5.4 % under the drought condition compared to the normal condition, but GEOS-Chem simulated a 20.23 % increase indicating an overestimation of isoprene emissions under drought. We implemented two existing drought stress algorithms in MEGAN2.1 and found they were not effective to reduce HCHO column biases in South Korea because those algorithms were proposed and developed for the Southeast United States (SE US). To improve this, we applied an iterative finite difference mass balance (IFDMB) method to estimate isoprene emissions using the OMI HCHO column. With this method, isoprene emissions were reduced by 60 % under the drought condition compared to those simulated by the standard MEGAN2.1 implemented in GEOS-Chem. The increase of HCHO column under the drought condition compared to the normal condition was also reduced to 10.71 %, which is comparable to that in the satellite retrievals. Based on isoprene emission difference between MEGAN2.1 and IFDMB, we developed the empirical equations to adjust isoprene emissions in South Korea that also improved model prediction of the secondary pollutant such as ozone.

Biogenic isoprene emissions play an important role in air quality so it is important to quantify their response to extreme events such as drought. While there have been some efforts to reduce uncertainties in isoprene emissions under the drought conditions, the effort has not been made in the South Korean region. Here, we aimed to constrain drought stress on biogenic isoprene emissions in South Korea using satellite formaldehyde (HCHO) column, the key product of isoprene oxidation, and a chemistry transport model (GEOS-Chem) with Model of Emissions of Gases and Aerosols from Nature version 2.1 (MEGAN2.1). It was found that the HCHO column from the Ozone Monitoring Instrument (OMI) increased by 5.4 % under the drought condition compared to the normal condition, but GEOS-Chem simulated a 20.23 % increase indicating an overestimation of isoprene emissions under drought. We implemented two existing drought stress algorithms in MEGAN2.1 and found they were not effective to reduce HCHO column biases in South Korea because those algorithms were proposed and developed for the Southeast United States (SE US). To improve this, we applied an iterative finite difference mass balance (IFDMB) method to estimate isoprene emissions using the OMI HCHO column. With this method, isoprene emissions were reduced by 60 % under the drought conditions compared to those simulated by the standard MEGAN2.1 implemented in GEOS-Chem. The increase of HCHO column under the drought conditions compared to the normal condition was also reduced to 10.71 %, which is comparable to that in the satellite retrievals. Based on isoprene emission difference between MEGAN2.1 and IFDMB, we developed the empirical equations to adjust isoprene emissions in South Korea that also improved model prediction of the secondary pollutant such as ozone.

Biogenic isoprene emissions play an important role in air quality so it is important to quantify their response to extreme events such as drought. While there have been some efforts to reduce uncertainties in isoprene emissions under the drought conditions, the effort has not been made in the South Korean region. Here, we aimed to constrain drought stress on biogenic isoprene emissions in South Korea using satellite formaldehyde (HCHO) column, the key product of isoprene oxidation, and a chemistry transport model (GEOS-Chem) with Model of Emissions of Gases and Aerosols from Nature version 2.1 (MEGAN2.1). It was found that the HCHO column from the Ozone Monitoring Instrument (OMI) increased by 5.4 % under the drought condition compared to the normal condition, but GEOS-Chem simulated a 20.23 % increase indicating an overestimation of isoprene emissions under drought. We implemented two existing drought stress algorithms in MEGAN2.1 and found they were not effective to reduce HCHO column biases in South Korea because those algorithms were proposed and developed for the Southeast United States (SE US). To improve this, we applied an iterative finite difference mass balance (IFDMB) method to estimate isoprene emissions using the OMI HCHO column. With this method, isoprene emissions were reduced by 60 % under the drought conditions compared to those simulated by the standard MEGAN2.1 implemented in GEOS-Chem. The increase of HCHO column under the drought conditions compared to the normal condition was also reduced to 10.71 %, which is comparable to that in the satellite retrievals. Based on isoprene emission difference between MEGAN2.1 and IFDMB, we developed the empirical equations to adjust isoprene emissions in South Korea that also improved model prediction of the secondary pollutant such as ozone.

Biogenic isoprene emissions play an important role in air quality so it is important to quantify their response to extreme events such as drought. While there have been some efforts to reduce uncertainties in isoprene emissions under the drought conditions, the effort has not been made in the South Korean region. Here, we aimed to constrain drought stress on biogenic isoprene emissions in South Korea using satellite formaldehyde (HCHO) column, the key product of isoprene oxidation, and a chemistry transport model (GEOS-Chem) with Model of Emissions of Gases and Aerosols from Nature version 2.1 (MEGAN2.1). It was found that the HCHO column from the Ozone Monitoring Instrument (OMI) increased by 5.4 % under the drought condition compared to the normal condition, but GEOS-Chem simulated a 20.23 % increase indicating an overestimation of isoprene emissions under drought. We implemented two existing drought stress algorithms in MEGAN2.1 and found they were not effective to reduce HCHO column biases in South Korea because those algorithms were proposed and developed for the Southeast United States (SE US). To improve this, we applied an iterative finite difference mass balance (IFDMB) method to estimate isoprene emissions using the OMI HCHO column. With this method, isoprene emissions were reduced by 60 % under the drought conditions compared to those simulated by the standard MEGAN2.1 implemented in GEOS-Chem. The increase of HCHO column under the drought conditions compared to the normal condition was also reduced to 10.71 %, which is comparable to that in the satellite retrievals. Based on isoprene emission difference between MEGAN2.1 and IFDMB, we developed the empirical equations to adjust isoprene emissions in South Korea that also improved model prediction of the secondary pollutant such as ozone.

The U.S. National Oceanic and Atmospheric Administration (NOAA) provides operational air quality (AQ) predictions over the United States and global aerosol forecasts. The current operational model, the National Air Quality Forecast Capability (NAQFC) at NOAA, has undergone a fundamental paradigm shift through its integration into the Earth system modeling Unified Forecast System (UFS) as a coupled component, the Air Quality Modeling component (AQMv7). AQMv7 embeds the U.S. EPA Community Multiscale Air Quality Model (CMAQ) and it has been operational at the National Weather Service (NWS) since May 2024. The model was also updated with a larger domain size and new emissions, including the development of the NOAA Emission and eXchange Unified System (NEXUS) along with dynamic processes such as using Model of Emissions of Gases and Aerosols from Nature (MEGAN) v2.1 and the FENGSHA dust scheme, and the Regional Advanced Baseline Imager (ABI)-Visible Infrared Imaging Radiometer Suite (VIIRS) Emissions (RAVE) biomass burning algorithm. Operational global aerosol modeling is performed through the NOAA Global Ensemble Forecast System with Aerosols and has been operational in this manner since 2020. Planned future advances include upgrading to a fully coupled atmosphere/land/ocean/sea-ice/wave/aerosols systems developed within the Unified Forecast System (UFS) framework. In its final configuration, the coupled UFS system will consist of: (1) FV3 dynamical core and CCPP atmospheric physics package using the Noah-MP land model, (2) MOM6 ocean model, (3) CICE6 sea ice model, (4) WAVEWATCH III wave model, and (5) the UFS-Aerosol component, based on NASA’s 2nd generation GOCART aerosol model. GOCART is a simplified chemistry and aerosol component that predicts the major aerosol species including dust, organic and black carbon, sea salt, and sulfate aerosols. In this presentation we will discuss NOAA’s current and future AQ and air composition forecasting capabilities and the performance of each system.

The air quality forecasting system is an essential tool widely used by environmental managers to mitigate adverse health effects of air pollutants. This work presents the latest development of the next-generation regional air quality model (AQM) forecast system within the Unified Forecast System (UFS) framework in the National Oceanic and Atmospheric Administration (NOAA). The UFS air quality model incorporates the US Environmental Protection Agency (EPA) Community Multiscale Air Quality (CMAQ) model as its main chemistry component. In this system, CMAQ is integrated as a column model to solve gas and aerosol chemistry, while the transport of chemical species is processed by UFS. The current AQM version 7 (AQMv7) is coupled with an earlier version of CMAQ (version 5.2.1). Here we describe the development of the updated AQMv7 by coupling to a “state-of-the-science” CMAQ version 5.4. The updates include improvements in gas and aerosol chemistry, dry deposition processes, and structural changes to the input/output (I/O) interface, enhancing both computational efficiency and representation of air–surface exchange processes. A simulation was conducted for the period of June–August 2023 to assess the effects of these updates on the forecast performance of ozone (O₃) and fine particulate matter (PM_2.5), two major air pollutants over the continental United States (CONUS). The results show that the updated model demonstrates an enhanced capability in simulating O₃ over the CONUS by reducing the positive bias, leading to a reduction in the mean bias by 3 %–5 % and 8 %–12 % for hourly and the maximum daily 8 h average O₃, respectively. Spatially, the updated model lowers the positive bias of hourly O₃ in most of the 10 EPA regions, particularly within the central and northwest areas, while amplifying the O₃ underestimation over the sites with negative bias. Similarly, the updates induce uniformly lower fine particulate matter (PM_2.5) concentrations across the CONUS domain, reducing the positive bias at some sites over the northeast in August and central Great Plains. The updated model does not improve model performance for PM_2.5 in the vicinity and downwind of fire emission sources, where AQMv7 shows the highest negative bias, thus indicating a focal point of model uncertainty and needed improvement. Despite these challenges, the study highlights the importance of the ongoing refinements for reliable air quality predictions from the UFS-AQM model, which is a planned future update to NOAA's current operational air quality forecast system.

Biogenic isoprene emissions play an important role in air quality so it is important to quantify their response to extreme events such as drought. While there have been some efforts to reduce uncertainties in isoprene emissions under the drought conditions, the effort has not been made in the South Korean region. Here, we aimed to constrain drought stress on biogenic isoprene emissions in South Korea using satellite formaldehyde (HCHO) column, the key product of isoprene oxidation, and a chemistry transport model (GEOS-Chem) with Model of Emissions of Gases and Aerosols from Nature version 2.1 (MEGAN2.1). It was found that the HCHO column from the Ozone Monitoring Instrument (OMI) increased by 5.4 % under the drought condition compared to the normal condition, but GEOS-Chem simulated a 20.23 % increase indicating an overestimation of isoprene emissions under drought. We implemented two existing drought stress algorithms in MEGAN2.1 and found they were not effective to reduce HCHO column biases in South Korea because those algorithms were proposed and developed for the Southeast United States (SE US). To improve this, we applied an iterative finite difference mass balance (IFDMB) method to estimate isoprene emissions using the OMI HCHO column. With this method, isoprene emissions were reduced by 60 % under the drought conditions compared to those simulated by the standard MEGAN2.1 implemented in GEOS-Chem. The increase of HCHO column under the drought conditions compared to the normal condition was also reduced to 10.71 %, which is comparable to that in the satellite retrievals. Based on isoprene emission difference between MEGAN2.1 and IFDMB, we developed the empirical equations to adjust isoprene emissions in South Korea that also improved model prediction of the secondary pollutant such as ozone.