Systematic review of artificial intelligence in predicting air pollutant concentrations: comparing global and Nigeria-specific studies

Résumé

Abstract Accurate prediction of air pollutant concentrations is essential for environmental policy, public health protection, and early warning systems. In recent years, Artificial Intelligence (AI) techniques, including machine learning (ML), deep learning (DL), and hybrid models, have significantly improved the forecasting of particulate matter, gaseous pollutants, and real-time Air Quality Index (AQI). This study presents a systematic review of AI applications for predicting air pollutant concentrations, with emphasis on comparing global and Nigeria-specific studies. To ensure methodological consistency and avoid inappropriate comparisons, the analysis was structured according to modelling task: station-level particulate-matter forecasting, satellite-based particulate-matter mapping, and AQI prediction. Results were interpreted descriptively within each category because of differences in datasets, validation strategies, and prediction objectives. Findings show that data availability and monitoring infrastructure strongly influenced model selection and reported performance. Global studies commonly apply more complex ML, DL, and hybrid models using long-term multi-station and multi-source datasets, whereas Nigeria-specific studies rely more on simpler ML approaches because of limited data availability and sparse monitoring networks. Differences in reported performances are therefore linked mainly to data conditions and validation methodology rather than the inherent superiority of any specific algorithm. The study highlights the importance of aligning modelling approaches with available data, improving validation practices, and expanding monitoring infrastructure in data-limited regions. It emphasizes the value of context-specific modelling frameworks and multi-source data integration for evidence-based environmental decision-making. Beyond synthesis of existing literature, the study further contributes to a context-aware analytical framework for interpreting and applying AI models for air-quality prediction.

Citer ce document

Accès au document

Texte intégral en lecture en ligne, réservé aux abonnés SPHAERO et aux membres de l'institution. Se connecter

Auteur(s)

Statistiques

Consultations : 1

Téléchargements : 0