A Data-Driven and Cost-Effective Framework for Urban Air Quality Monitoring and NO2 Forecasting in Cheltenham

Abstract

Effective urban air quality management requires monitoring systems that are reliable, affordable, and fairly distributed across commercial and residential areas. This study examines nitrogen dioxide (NO2 ) concentrations in Cheltenham by comparing commercial sites, Boots Corner/High Street, with the residential site London Road across pre-COVID-19, COVID-19, and post-COVID-19 periods. Missing values were handled using mean imputation, and nonparametric Kruskal–Wallis H-tests followed by Dunn’s post hoc pairwise comparisons were used to compare monitoring locations. Forecasting performance was assessed using Autoregressive Integrated Moving Average (ARIMA), Holt-Winters exponential smoothing (Holt-Winters), Extreme Gradient Boosting (XGBoost), Gated Recurrent Unit, and Long Short-Term Memory (LSTM) models, with root mean square error (RMSE) as the evaluation metric and bootstrap simulations for validation. Results showed that NO2 levels at London Road were statistically indistinguishable from Cheltenham citywide levels (p = 0.74), suggesting it can act as a cost-effective proxy for average citywide trends, although it should not replace monitoring of short-term fluctuations or localized pollution episodes. In contrast, Boots Corner/High Street differed significantly from citywide levels (p < 0.05). For forecasting, Holt-Winters slightly outperformed ARIMA, while LSTM achieved the best overall accuracy, with an RMSE of 2.66 and about 16% lower error than competing models. Bootstrap simulations supported these findings, though performance varied with dataset size. Overall, London Road can support cost-efficient NO2 surveillance in Cheltenham, while supplementary monitoring remains important. LSTM models should be prioritized for high-accuracy forecasting, with bootstrap validation guiding model selection.

Received: 17 November 2025 | Revised: 28 May 2026 | Accepted: 24 July 2026

Conflicts of Interest

The authors declare that they have no conflicts of interest to this work.

Data Availability Statement

The data that support the findings of this study are openly available from the Cheltenham Borough Council at https://doi.org/10.5281/zenodo.21478965.

Author Contribution Statement

Johnson Ohakwe: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Data curation, Writing – original draft, Writing – review & editing, Visualization, Project administration. Derrick Ofori: Methodology, Data curation, Writing – review & editing. Bhupesh Kumar Mishra: Conceptualization, Methodology, Validation, Investigation, Resources, Data curation, Writing – original draft, Writing – review & editing, Supervision, Project administration. William Sayers: Conceptualization, Validation, Resources, Writing – review & editing, Supervision, Project administration. Timothy Olusakin: Formal analysis, Methodology, Writing – original draft. John Chisimkwuo: Validation, Formal analysis, Supervision.