A comparison of statistical and machine learning models for spatio-temporal prediction of ambient air pollutant concentrations in Scotland

Aug 6, 2026

Spatio-temporal prediction of air pollutant concentrations is essential for regulatory compliance assessment and for generating exposure estimates used in epidemiological studies. The work surveys land use regression, additive and spatio-temporal smoothing approaches alongside machine learning algorithms, noting that few studies compare predictive performance in a thorough, model-to-model way. For Scotland’s monthly averages of NO2, PM10 and PM2.5, random forests generally outperform or match traditional statistical methods. The best model is then used to produce a new 1 km grid data resource covering 2016–2020 with uncertainty quantification.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages24

Reading Time1 h 0 min

AccessFree to download

Abstract

Introduction

Air pollution monitoring and modeling needs

Goal and fusion approach

Existing methods for predicting air pollution

A comparison of statistical and machine learning models for spatio‑temporal prediction of ambient air pollutant concentrations in Scotland Qiangqiang Zhu1 · Duncan Lee1 · Oliver Stoner1 Received: 23 April 2024 / Revised: 4 October 2024 / Accepted: 15 October 2024 / Published online: 13 November 2024 © Crown 2024 Abstract The spatio-temporal prediction of air pollutant concentrations is vital for assessing regulatory compliance and for producing exposure estimates in epidemiological studies. Numerous approaches have been utilised for making such predictions, including land use regression models, additive models, spatio-temporal smoothing models and machine learning prediction algorithms. However, relatively few studies have compared the predictive performance of these models thoroughly, which isone of the novel contributions of this paper. For the specific challenge of predicting monthly average concentrations of NO , PM 􀀞 and 􀀞􀀝􀀜􀀛 in Scotland, we find that random forests typically outperform (or are as good as) more traditional statistical prediction approaches. Additionally, we utilise the best performing model to provide a new data resource, namely, predictions of monthly average concentrations (with uncertainty quantification) of the above pollutants on a regular 1 km grid for all of Scotland between 2016 and 2020. Keywords Ambient air pollutants · Data product · Model comparison · Spatiotemporal prediction Handling Editor: Luiz Duczmal. Duncan Lee and Oliver Stoner contributed equally to this work. * Qiangqiang Zhu [q.zhu.1@research.gla.ac.uk](q.zhu.1@research.gla.ac.uk) Duncan Lee [Duncan.Lee@glasgow.ac.uk](Duncan.Lee@glasgow.ac.uk) Oliver Stoner [Oliver.Stoner@glasgow.ac.uk](Oliver.Stoner@glasgow.ac.uk) 1 School of Mathematics and Statistics, University of Glasgow, University Place, Glasgow G12 8QQ, UK 1 Introduction Air pollution is a complex mixture of different components including nitrogen dioxide (NO ), ozone (O ), and particulate matter (PM), the latter being measured by particles ≤ 10 m (PM 􀀞), and ≤ 2.5 m ( 􀀞􀀝􀀜􀀛) in aerodynamic diameter. Longterm exposure to these pollutants has been associated with a range of adverse health outcomes, including respiratory diseases (Bălă et al., 2021), cardiovascular diseases (Rajagopalan et al., 2018) and mental ill health (Gu et al., 2020) . A recent summary of the evidence is given by Chief Medical Officer (2022) . Globally, it is estimated that 99% of the population are exposed to concentrations that exceed the World Health Organisation’s guideline limits (World Health Organization, 2021), with the burden disproportionately affecting low-and middle-income countries (https://www. [who.int/health-topics/air-pollution](who.int/health-topics/air-pollution)). In Scotland, the focus of this study, air pollution management is supported by a robust legislative framework, including the UK Air Quality Strategy in 2007, the Air Quality Standards (Scotland) Regulations in 2010 that enacted the European Union 2008 Ambient Air Quality Directive (2008/50/ EC), and the UK Environment Act in 2021. These regulations establish a set of air quality standards, objectives and targets, with a summary being available at [https://](https://)[ ](https://)[www.scottishairquality.scot/air-quality/standards](www.scottishairquality.scot/air-quality/standards). Comprehensive monitoring of air pollutant concentrations is thus essential for a number of reasons, including the assessment of whether the above targets are being met, as well for producing exposure estimates for epidemiological studies (e.g., Dibben and Clemens, 2015) . Ideally, high-resolution air pollution maps should be produced based on data from a dense network of air pollution monitors, but as these monitors are expensive to install and run they are spatially sparse, with only about 100 currently active in Scotland ([https://www.scottishairquality.scot/latest/](https://www.scottishairquality.scot/latest/)[ ]

FAQs

Why is spatio-temporal prediction of air pollution concentrations important?

It supports regulatory compliance checks and enables exposure estimates for epidemiological studies.

Which model type performs best for monthly averages of pollutants in Scotland?

Random forests typically outperform traditional statistical approaches or perform as well as them.

What new data product is produced in this study?

Predictions of monthly average pollutant concentrations on a regular 1 km grid for all of Scotland from 2016 to 2020, including uncertainty quantification.

Aug 6, 2026

Upload by Lucas Martin