A Transformer-based agent model of GEOS-Chem v14.2.2 for informative prediction of PM2.5 and O3 levels to future emission scenarios: TGEOS v1.0
Dehao Li (Nanjing University of Information Science and Technology)
Jianbing Jin (Nanjing University of Information Science and Technology)
Guoqiang Wang (Shanghai University of Engineering Science)
Mijie Pang (TU Delft - Mathematical Physics)
Weihong Zhang (Ecological Environment Monitoring Center of Ningxia Hui Autonomous Region)
Hong Liao (Nanjing University, Nanjing University of Information Science and Technology)
More Info
expand_more
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Abstract
Efficient and informative air quality modeling in future emission scenarios is vital for effective formulation of emission reduction policies. Traditional chemical transport models (CTMs) struggle with the computational demands required for timely predictions. While advanced emulator techniques greatly accelerate CTM simulating process, they fall short in providing comprehensive estimates of future air quality due to their limited model structure. Additionally, these emulators often have difficulty simultaneously accounting for varying emission variables and the effects of regional transport, which limits their applicability and undermines prediction accuracy. In this study, an informative future air quality prediction model “TGEOS v1.0” based on the Transformer framework is developed as an efficient agent model of GEOS-Chem v14.2.2. TGEOS is able to efficiently estimate key statistical indicators of PM2.5 and O3 concentrations under future emission scenarios and capture potential extreme pollution events, with approximately 2.51 s to execute one-year estimation. The model incorporates sectoral emissions of up to 26 distinct species as well as the impacts of regional emissions and meteorology on pollutant concentrations, enhancing its versatility and predictive accuracy. The spatial and probability distributions predicted by TGEOS are in good agreement with GEOS-Chem, with the correlation coefficients for PM2.5 and O3 exceed 0.98 in high-pollution months. Compared with other machine learning models, TGEOS based on Transformer framework showcases superior performance, underscoring the potential of the Transformer framework in air quality modeling.
help_outline