Project goal: we aim to integrate deepctm, a cutting-edge machine-learning model designed to emulate chemical transport models (ctms), with noaas operational national air quality forecast capability…

The University of Tennessee received a $1,049,997 cooperative agreement from the National Oceanic and Atmospheric Administration (NOAA) under the Weather and Air Quality Research program (CFDA 11.459), awarded August 1, 2025, with completion targeted for July 31, 2028. The primary deliverable is UFS-DeepCTM, an integrated air quality forecasting system that embeds a machine learning model (DeepCTM) into NOAA's Unified Forecast System (UFS) to enhance the National Air Quality Forecast Capability (NAQFC). This system will deliver faster computational performance and improved atmospheric chemical concentration predictions at finer spatial resolutions and larger geographic domains compared to traditional chemical transport models, while enabling efficient data fusion with diverse observational datasets.

To accomplish this objective, the University of Tennessee will enhance DeepCTM's architecture using a ConvLSTM-ResNet design that incorporates near-real-time emission adjustments through a variational auto-encoder and applies super-resolution techniques for expanded global chemistry applications. The project includes pretraining on multi-scale UFS-AQM data and ensemble optimization strategies. George Mason University serves as a critical sub-grantee, providing essential training data from NOAA's operational systems and meteorological data from the Global Forecast System, while also leading research-to-operations efforts by facilitating coordination between the University of Tennessee team and NOAA/National Centers for Environmental Prediction partners. The integrated UFS-DeepCTM system is designed to support higher-resolution forecasting and future chemistry-to-weather feedback research through its online coupling architecture.

Project goal: we aim to integrate deepctm, a cutting-edge machine-learning model designed to emulate chemical transport models (ctms), with noaas operational national air quality forecast capability (naqfc) to enhance the forecasting of atmospheric chemical concentrations. This integration promises improved efficiency, accuracy, and resolution in air quality predictions on national and global scales. Problem/opportunity statement: noaas naqfc, part of the unified forecast system (UFS), faces performance challenges. Machine learning offers a promising alternative. Our prior work on deepctm demonstrates consistent accuracy with significantly faster computation compared to traditional ctms. Deepctm generates sensitivity matrices enabling scientifically informed tuning of inputs (emissions, ICS) to enhance forecasting accuracy. Methodology/activities to be performed: we will enhance deepctm for seamless online integration with UFS simulations. Leveraging a convlstm-resnet architecture, IT incorporates near-real-time emission adjustments via a variational auto-encoder, applies super-resolution techniques, and extends chemistry for global applications. Pretrained on multi-scale UFS-AQM data, IT will execute ensemble runs to optimize the fusion strategy with observations. Primary project products/outputs: the projects primary outcome will be UFS-deepctm, a deepctm-embedded UFS air quality forecast system that delivers faster performance and efficient data fusion with diverse observations. IT enables full-chemistry forecasts at finer resolutions across larger domains or global scales and supports future studies on chemistry-toweather feedback via its online coupling design. Expected results, outcomes, and benefits: UFS-deepctm provides high computational efficiency, supporting aqrf-7, with potential for online coupling with meteorology at higher resolutions (e.g., 3 KM) and global coverage to address aqrf-1. An optimized ensemble strategy, along with emission and ICS tuning, is expected to improve air quality forecasting accuracy, meeting aqrf-3, aqrf-4, and aqrf-5 priorities. Intended beneficiaries and recipients: the primary recipient is NOAA/ncep, which will benefit from daily weather and air quality forecasts as the outcome transitions from offline to online, ultimately providing high-resolution global air quality forecast system. The study will also generate scientific publications to advance global atmospheric chemistry forecasting and contribute to global scientific community. 1235-WPO_