The limitations of numerical simulations and single-site machine-learning approaches in air quality forecasting

The gap

Evidence profile

Sourced from the limitations and stated research gap and inline gaps and future work and recommendations and future-work section of the source papers, classified as general, spanning 7 journals. Those papers have been cited 1 times in total.

Supporting evidence — 7 representative gaps

• Ensemble Learning Approaches for Air Pollution Classification and Environmental Health Risk Assessment (2026) · JURNAL KESEHATAN LINGKUNGAN · doi

  Although the ensemble models achieved an impressive accuracy rate, studies have identified some major issues that need to be resolved. First, the studies were conducted using open-source datasets with very little time and space resolution; thus, these 166 Jurnal Kesehatan Lingkungan/10.20473/jkl.v18i2.2026.159-170 Vol. 18 No.2 April 2026 (159-170) datasets do not adequately represent the potential for variability in pollutant exposure for populations living in urban areas. Second, the primary focus of the study was on classification accuracy and did not provide an explicit representation of health outcomes (i.e., hospital admission and death rates). This study did not incorporate spatio-temporal dimensions, as the primary objective was to evaluate the performance of ensemble learning models in pollutant and health-risk classification based on historical exposure records. Future research should integrate spatial information (e.g., geographic pollutant distribution) and temporal variation (e.g., hourly, daily, or seasonal fluctuations) to enable spatial risk mapping, forecasting, and more comprehensive time-series environmental health surveillance. In future studies, researchers should incorporate spatio-temporal data and population vulnerability factors (age, socioeconomic status) to enable risk stratification. The potential to increase the interpretability of the model and to support the application of this type of model to real-world policy decisions will also be enhanced through the expansion of the model into a multi-source hybrid ensemble model. In addition to enhancing the potential of this type of model to be applied to real-world decision- making, collaborative efforts between environmental agencies and public health institutions will be needed to develop and implement the infrastructure necessary to support the implementation of these types of intelligent monitoring systems nationally (39,43–44).

  generallimitationsevidence 5/5
  Keywords: model health ensemble potential pollutant temporal risk models accuracy source datasets time exposure primary classification
• FuXi-Air: air quality forecasting based on emission-meteorology-pollutant multimodal machine learning (2026) · npj Clean Air · cited 1× · doi

  The limitations of numerical simulations and single-site machine-learning approaches in air quality forecasting. The need for a multimodal machine learning model that integrates meteorological, emission, and observational data. The lack of a scientific reference and a practical example for applying deep machine learning to support rapid air pollution risk warning.

  generalstated research gapevidence 5/5
  Keywords: limitations numerical simulations single-site machine-learning approaches air quality
• Residual Stacking with Temporal Convolutional Network and Light Gradient Boosting for Air Quality Index Prediction (2026) · International Journal of Intelligent Engineering and Systems · doi

  Future work will focus on incorporating spatial correlation modeling using Graph Neural Networks to capture inter-city dependencies, integrating real- time data from IoT-enabled sensors to enhance satellite-derived responsiveness, and aerosol and climate parameters for broader coverage. Future work will explore adaptive attention within the deviation module and integrate unified SHAP–LIME explanations to deepen causal interpretation of pollutant–meteorology interactions.

  generalinline gapsevidence 5/5
  Keywords: future focus incorporating spatial correlation modeling using graph neural networks capture inter city dependencies integrating
• A VMD–PSO hybrid framework with machine and deep learning for Aaccurate spatial–temporal daily rainfall occurrence prediction in Pakistan’s diverse climatic regions (2026) · Modeling Earth Systems and Environment · doi

  The current method can be improved in future studies by adding satellite observations and remote sensing, which will provide a more thorough spatial representation and enable the detection of exceptional rainfall occurrences. In order to better capture temporal correlations over large distances than recurrent networks, more complex deep learning archi- tectures, including Transformer models and attention, may be investigated. Additionally, the system can be tailored to incorporate real-time early warning in order to mitigate disaster risks in sensitive locations and encourage flood preparedness. Addi- tionally, multi-objective optimization would be useful in achieving a balance between interpretability of the model, computational efficiency, and forecast accuracy, making the framework more practical to use on a broad scale. Author contributions H.F. and S.H. designed and implemented the deep learning framework and performed experiments. M.A.H. contrib- uted to data preprocessing, model benchmarking, and statistical analy- sis. S.S. assisted with theoretical formulation and result interpretation. All authors reviewed and approved the final manuscript. Funding Open access funding provided by NTNU Norwegian Univer- sity of Science and Technology (incl St. Olavs Hospital - Trondheim University Hospital) Data availability No datasets were generated or analysed during the current study.

  generalfuture workevidence 5/5
  Keywords: current order deep learning model framework funding hospital improved future adding satellite observations remote sensing
• PM10 Concentration Forecasting: A Comparative Evaluation of Deep Learning and Time Series Methods (2026) · Academic Platform Journal of Engineering and Smart Systems · doi

  This study provides many opportunities for further study and application in air quality forecast models. First, since deep-learning-based methods (e.g., LSTM and GRU) have shown generally lower error rates than traditional statistical models in terms of predicting air pollution levels, but are limited in the accuracy of their predictions due to the lack of data and assumptions used in their development, they have the potential to be used as complementary tools to traditional statistical models in early warning systems provided that sufficient contextual information is available and proper validation techniques are employed. Second, researchers should develop spatiotemporal models (e.g., CNN-LSTM models, attention-based neural networks) that can learn both temporal patterns and spatial relationships among different air pollution monitoring sites in order to improve the robustness of predictive models in urban areas where emissions vary greatly over space and time. Third, incorporating additional variables that describe the context in which an area's air pollution level is being measured (e.g., intensity, industrial production indicators, land-use characteristics, population density, extreme weather events) has the potential to increase the explanatory power of models and better represent the various factors that influence variability in PM10 levels. traffic Lastly, providing the results of PM10 forecasts to the public through digital decision-making tools (e.g., web-based dashboards, mobile apps), could allow for timely access to air quality information and potentially support preventative action by vulnerable populations (i.e., children, older adults, individuals with chronic respiratory or cardiovascular disease). If uncertainty associated with the models is communicated evaluated continuously, the integration of PM10 forecasts into digital decision-making tools has the potential to provide significant contributions to evidence-based public health and environmental management. and models clearly are Author contributions: All authors have contributed equally to work. Conflict of Interest: No conflict of interest was declared by the authors. Financial Disclosure: The authors declared that this study has received no financial support. REFERENCES [1] Q. Liu, B. Cui, and Z. Liu, “Air quality class prediction using machine learning methods based on monitoring data and secondary modeling,” Atmosphere (Basel), vol. 15, no. 5, p. 553, 2024 [2] Who.int. [Online]. Available: https://www.who.int/publications/m/item/who- ambient-air-quality-database-(update-jan-2024. [Accessed: 29-Oct-2025]. [3] L. Zhang, D. Su, W. Guo, and S. Li, “Empirical study on urban sustainable development model based on identification of advantages and disadvantages,” Frontiers in Sustainable Cities, vol. 4, 2022. [4] P. Verma et al., “Assessment of human and meteorological influences on PM10 concentrations: Insights from machine learning algorithms,” Atmos. Pollut. Res., vol. 15, no. 6, p. 102123, 2024. [5] L. Mampitiya, N. Rathnayake, Y. Hoshino, and U. Rathnayake, “Forecasting PM10 levels in Sri Lanka: A comparative analysis of machine learning models for PM10,” J. Hazard. Mater. Adv., vol. 13, no. 100395, p. 100395, 2024. Academic Platform Journal of Engineering and Smart Systems (APJESS) 14(2), 114-124, 2026 123 PM10 Concentration Forecasting: A Comparative Evaluation of Deep Learning and Time Series Methods Sertaç CARAN, Adem KORKMAZ [21] D. Chicco, M. J. Warrens, and G. Jurman, “The coefficient of determination R-squared is more informative than SMAPE, MAE, MAPE, MSE, and RMSE in regression analysis evaluation,” PeerJ Comput. Sci., vol. 7, no. e623, p. e623, 2021. [22] R. J. Genetski and J. S. Armstrong, “Long-range forecasting: From crystal ball to computer,” J. Am. Stat. Assoc., vol. 82, no. 400, p. 1194, 1987. [23] C. D. Lewis, Industrial and business forecasting methods: A practical guide to exponential smoothing and curve fitting. London: Butterworths. 1982. [24] S. Makridakis, E. Spiliotis, and V. Assimakopoulos, “Statistical forecasting and Machine Learning methods: Concerns and ways forward,” PLoS One, vol. 13, no. 3, p. e0194889, 2018. [6] B. I. E. Heredia et al., “Air quality monitoring in Mendoza, Argentina: machine learning approaches for PM10 prediction,” Environ. Sci. Pollut. Res. Int., vol. 32, no. 29, pp. 17511–17527, 2025. [7] A. Mishra and Y. Gupta, “Comparative analysis of Air learning

  generalrecommendationsevidence 5/5
  Keywords: models learning based quality machine forecasting statistical pollution levels potential tools monitoring authors comparative deep
• Explainable PSO-optimised machine learning models for multi-pollutant air quality forecasting in major African cities with transfer learning (2026) · Frontiers in Environmental Science · doi

  investigate learning methodologies, incorporate supplementary machine including deep learning, spatiotemporal models, and hybrid frameworks, so as to enhance the understanding of complex pollutant interactions and urban heterogeneity. Adding secondary and trace pollutants to the list of pollutants and using high- resolution data on weather, transportation, and land use, could make predictions even more accurate. Research could also examine scalable domain adaptation and transfer learning methodologies inadequate monitoring infrastructure, encompassing real-time model updating and ensemble methods that integrate several source-city models. Finally, in operational air quality management systems would allow for an understanding of how strong, long-lasting, and useful predictive forecasting may be for policy and public health actions. long-term studies of how models are used cities with for

  generalfuture workevidence 5/5
  Keywords: learning models methodologies understanding pollutants long investigate incorporate supplementary machine including deep spatiotemporal hybrid frameworks
• Estimation of PM2.5 Concentration Based on PSO-Optimized Machine Learning Models and SHAP Analysis: A Case Study of Wuhan, Hubei Province (2026) · Applied Sciences · doi

  integrate multi-source environmental data and socioeconomic datasets, including satellite-derived aerosol optical depth (AOD) products, emission inventories, land-use information, and human activity indicators - explore the combination of temporal deep learning models, graph neural networks (GNNs), and explainable artificial intelligence methods - investigate transfer learning strategies to improve model generalization and adaptability across different regions

  generalfuture-work sectionevidence 5/5
  Keywords: integrate multi-source environmental data socioeconomic datasets including satellite-derived

Questions about this gap