Forecasting the Air Quality Index (AQI) in Sub-Saharan Africa: Leveraging Seasonal Patterns and Deep Learning for Abuja, Nigeria
Author
Onyeche Vera Adikwu, Nkiruka Favour Ozoemena, Jophia Nicholas Kemi, Mariam Jibrin Usman, Laura Yusuf-Ufua
Abstract
Air pollution is one of the leading environmental health risks in Sub-Saharan Africa, yet the region remains among the most sparsely instrumented for continuous air quality monitoring. Abuja, Nigeria's capital, records Air Quality Index (AQI) levels that swing sharply between a comparatively clean wet season and a heavily polluted Harmattan-dominated dry season, driven by desert dust transport, biomass burning, and reduced atmospheric mixing. Existing Nigerian AQI forecasting studies have generally relied on single-model approaches evaluated over short study windows or leakage-prone random data splits, and none has yet targeted Abuja with a design that explicitly encodes seasonal structure. This paper presents a seasonally-aware deep learning methodology for forecasting AQI in Abuja, built on two years (January 2024–December 2025) of hourly data comprising 17,208 observations across AQI, PM2.5, PM10, NO2, O3, SO2, CO, NH3 and NO. Following a leakage-safe chronological 70/15/15 split, a gradient-boosted tree ensemble (XGBoost) and a Bidirectional LSTM were benchmarked against persistence and seasonal-naive baselines. XGBoost achieved the strongest one-hour-ahead forecasting performance (RMSE = 0.123, MAE = 0.070, R² = 0.988, classification accuracy = 98.9%), outperforming the Bi-LSTM (RMSE = 0.314, R² = 0.920) and both baselines, with the persistence baseline (RMSE = 0.347, R² = 0.902) itself proving difficult to beat -- underscoring the strong hour-to-hour autocorrelation in Abuja's AQI series. Both candidate models generalized across seasons, with lower error in the wet season than the Harmattan-dominated dry season. These results, together with the underlying seasonal feature engineering and evaluation framework, provide an empirically validated forecasting methodology for Abuja and a template for other data-scarce Sub-Saharan African cities.
Keywords
Air Quality Index; AQI forecasting; deep learning; ensemble learning; seasonal patterns; Sub-Saharan Africa; Abuja; Nigeria; time-series forecasting
Full Text:
References
Akintola, A., et al. (2026). Systematic review of AI in predicting air pollutant concentrations: Global versus Nigeria-specific approaches.
Al Rukabie, J. S. A., Al-Jiboori, M. H., & Chlaib, H. K. (2026). Evaluation of air pollution with NO2, SO2, and PM2.5 in Nasiriyah city center, south of Iraq. Iraqi Journal of Science, 67(4), 2298–2312. https://doi.org/10.24996/ijs.2026.67.4.32
Balogun, H., & Zakari, Y. (2025). When simpler wins: Facebook's Prophet vs LSTM for air pollution forecasting in data-constrained Northern Nigeria. arXiv:2508.16244. https://arxiv.org/abs/2508.16244
Bernacki, J., & Scherer, R. (2025). A comprehensive review of data-driven techniques for air pollution concentration forecasting. Sensors, 25(19), Article 6044. https://doi.org/10.3390/s25196044
IQAir. (n.d.). Abuja air quality index (AQI) and Nigeria air pollution. Retrieved August 2026, from https://www.iqair.com/nigeria/fct/abuja
Morapedi, T. D., & Obagbuwa, I. C. (2023). Air pollution particulate matter (PM2.5) prediction in South African cities using machine learning techniques. Frontiers in Artificial Intelligence, 6, Article 1230087. https://doi.org/10.3389/frai.2023.1230087
Ogunsanwo, G. O., Odulaja, P. T., Omotunde, A. A., & Solanke, O. O. (2025). Air quality index prediction using deep learning for Lagos State in Nigeria. Lafia Journal of Scientific & Industrial Research (LJSIR).
Okure, D. (2026). Towards improving urban air quality in Africa [Doctoral dissertation].
Omokpariola, D. O. (2025). Spatiotemporal analysis of atmospheric aerosols in African environments using MERRA-2 data.
Omokungbe, O. R., Olufemi, A. P., Ediagbonya, T. F., Ajileye, O. O., & Adepehin, D. S. (2025). Seasonal variability of particulate matter pollutants in Abuja, Nigeria. Geomatics, Natural Hazards and Risk. https://doi.org/10.1080/19475705.2025.2605114
Prastyo, D., Yudhana, A., & Murinto. (2026). Application of LSTM for SO2 emission prediction in incinerator monitoring using CEMS. Bit-Tech (Binary Digital-Technology), 9(1), 647–658. https://doi.org/10.32877/bt.v9i1.3955
Shao, S. (2026, February 5). Comparative analysis of machine learning models for multi-horizon PM2.5 forecasting. TechRxiv. https://doi.org/10.36227/techrxiv.177031335.58180594/v1
Singh, R., & Shukla, V. (2026). Operational air-quality forecasting with compact rolling trends: Lightweight boosted-tree approach. Procedia Computer Science, 282, 725–734. https://doi.org/10.1016/j.procs.2026.XX.XXX
Taylor, O. E., & Ezekiel, P. S. (2023). A model for forecasting air quality index in Port Harcourt Nigeria using Bi-LSTM algorithm. arXiv:2302.03930. https://doi.org/10.48550/arXiv.2302.03930
Wambebe, N. M., & Duan, X. (2020). Air quality levels and health risk assessment of particulate matters in Abuja municipal area, Nigeria. Atmosphere, 11(8), 817. https://doi.org/10.3390/atmos11080817