Answers are grounded in this paper and cite their sources.
High Resolution Image Download MS PowerPoint Slide We introduce SmartAQ 2+, a data fusion model combining a chemical transport model-based system (SmartAQ), low-cost PM 2.5 sensors, and machine learning (ML) to improve 3-day high-resolution PM 2.5 forecasts in urban areas. The ML model was trained using 2021–2023 data from Patras, Greece and evaluated in the same city with data from 2024. SmartAQ 2+ had better seasonal, monthly, and hourly performance than SmartAQ, reducing its error by a factor of 2. Across the 31 Patras monitoring sites, SmartAQ 2+ had good-to-excellent (fractional error ≤50%) monthly average performance in 21 sites in winter and spring, 17 in summer, and 22 in autumn. For exceedances of the EU daily mean threshold of 25 μg m –3, SmartAQ 2+ correctly predicted 23 of 54 exceedance days (43%). Its performance was similar for forecasts issued one and 2 days in advance and deteriorated by 30% after 48 h. Including cooking emissions in SmartAQ and as a separate input feature in SmartAQ 2+ reduced daily mean error by 30% at areas with high restaurant density. Sensitivity analysis showed highest reliance on sensors in the urban core and increased reliance on SmartAQ and static spatial features toward the suburbs. Without retraining, the model showed preliminary evidence of spatial transferability to Athens.
No takes yet. Share an insight, caveat, or question.
Apostolopoulos et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: