Assessing Uncertainty and Heterogeneity in Machine Learning-Based Spatiotemporal Ozone Prediction in the Beijing-Tianjin-Hebei Region in China

Aug 6, 2026

Upload by Logic

Accurate prediction of spatiotemporal ozone concentration supports early warning systems and air pollution regulation. The work addresses a gap in evaluating uncertainty and heterogeneity in machine learning spatiotemporal ozone prediction. Hourly and daily predictive performance is analyzed across the Beijing-Tianjin-Hebei region from 2013–2018 using ConvLSTM and DCGAN. Results indicate improved performance under multiple meteorological conditions, and ConvLSTM feasibly reproduces high-ozone distributions and captures spatiotemporal variation at 15 km × 15 km. Comparison with NAQPMS and monitoring observations further validates practical applicability.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages13

Reading Time33 min

AccessFree to download

Introduction

Background on ozone formation and air pollution impacts

Limitations of physical-informed air quality models

Emergence of machine learning for ozone prediction

Methods and experimental setup

ML models for spatiotemporal ozone prediction

Study region and time period

Meteorological conditions and performance scenarios

Results and evaluation

Hourly and daily predictive performance

Uncertainty and heterogeneity assessment

Comparison with NAQPMS and monitoring observations

Conclusion

Practical feasibility and spatial resolution considerations

Science of the Total Environment 881 (2023) 163146 Contents lists available at ScienceDirect Science of the Total Environment journal [homepage:](homepage: www. elsevier. com/locate/scitoten v)[ www. elsevier. com/locate/scitoten v](homepage: www. elsevier. com/locate/scitoten v) Assessing uncertainty and heterogeneity in machine learning-based spatiotemporal ozone prediction in Beijing-Tianjin-Hebei region in China Meiling Cheng a, Fangxin Fang a,⁎, Ionel Michael Navon b, Jie Zheng c, Jiang Zhu d, Christopher Pain a a Applied Modelling and Computation Group, Department of Earth Science and Engineering, Imperial College London, SW7 2BP, UK b Department of Scientiﬁc Computing, Florida State University, Tallahassee, FL 32306-4120, USA c Center for Excellence in Regional Atmospheric Environment, Institute of Urban Environment, Chinese Academy of Sciences, Xiamen 361021, China d International Center for Climate and Environment Sciences, Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing 100029, China H I G H L I G H T S Keywords: Ozone concentration prediction Machine learning Uncertainty Spatiotemporal Generative adversarial network G R A P H I C A L A B S T R A C T |
| | --- | | A B S T R A C T | Accurate prediction of spatiotemporal ozone concentration is of great signiﬁcance to effectively establish advanced early warning systems and regulate air pollution control. However, the comprehensive assessment of uncertainty and heterogeneity in spatiotemporal ozone prediction remains unknown. Here, we systematically analyze the hourly and daily spatiotemporal predictive performances using convolutional long short term memory (ConvLSTM) and deep convolutional generative adversarial network (DCGAN) models over the Beijing-Tianjin-Hebei region in China from 2013 to 2018. In extensive scenarios, our results show that the machine learning-based (ML-based) models achieve better spatiotemporal ozone concentration prediction performance with multiple meteorological conditions. A further comparison to the air pollution model-Nested Air Quality Prediction Modelling System (NAQPMS) and monitoring observations, the ConvLSTM model demonstrates the practical feasibility of identifying high ozone concentration distribution and capturing spatiotemporal ozone variation patterns at a high spatial resolution (here 15 km × 15 km). 1. Introduction Air pollution has detrimental effects on human health, ecological environment, and climate change (Kuerban et al., 2020). Ground-level ozone as a major secondary pollutant in China has been the subject of intense studies in recent years (Zhao et al., 2021; Maji et al., 2020; Ou et al., ⁎ Corresponding author. E-mail address: [f.fang@imperial.ac.uk](f.fang@imperial.ac.uk) (F. Fang). 2020; Wang et al., 2017b). Ozone is generated through chemical reactions of nitrogen oxides and volatile organic compounds in the presence of favorable meteorological conditions, e.g., intense solar radiation, low humidity, high temperature, and low winds (Wang et al., 2017a). Ozone concentration exhibits certain spatiotemporal changes and involves nonlinear, strong coupling and multivariate problems (Ezimand and Kakroodi, 2019; Su et al., 2020). Traditional air quality models (Han et al., 2018; Grell et al., 2005), which have the potential capability of representing the pollutant transformation, diffusion, and migration mechanisms through [http://dx.doi.org/10.1016/j.scitotenv.2023.163146](http://dx.doi.org/10.1016/j.scitotenv.2023.163146) Received 17 December 2022; Received in revised form 3 March 2023; Accepted 25 March 2023 Available online 1 April 2023 0048-9697/© 2023 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license ([http://creativecommons.org/licenses/by/4.0/](http://creativecommons.org/licenses/by/4.0/)). the governing equations, have been widely used in regional air quality prediction (Yafouz et al., 20

FAQs

What challenge does the study focus on regarding ML-based ozone prediction?

It focuses on systematically assessing uncertainty and heterogeneity in spatiotemporal ozone prediction, which had not been comprehensively evaluated before.

Which models are used for spatiotemporal ozone concentration prediction?

The study uses convolutional long short-term memory (ConvLSTM) and a deep convolutional generative adversarial network (DCGAN) across the Beijing-Tianjin-Hebei region.

How does the ConvLSTM model perform compared with other references?

ConvLSTM shows practical feasibility for identifying high ozone concentration patterns and capturing spatiotemporal variation at high spatial resolution (15 km × 15 km), and it is compared against NAQPMS and monitoring observations.

Aug 6, 2026

Upload by Logic