Persistent WHO PM2.5 non-attainment and exposure risk under sparse monitoring: QC-audited spatial exposure assessment in a tropical mixed-source region

Background Reliable assessment of long-term fine particulate matter (PM 2.5 ) and World Health Organization (WHO) guideline non-attainment remains challenging in sparse, evolving monitoring networks. Urban emissions and seasonal biomass burning may obscure persistent spatial patterns. We developed a reproducible framework for uncertainty-qualified annual screening surfaces using routine regulatory data. Methods Hourly PM 2.5 data (2017-2023) from Northeastern (NE) Thailand were standardized into quality-controlled (QC) station-level time series. Short gaps (≤24 h) were interpolated; longer gaps were preserved; extreme hourly values were winsorized at the 5 th and 95 th percentiles (P05-P95). Annual means were computed under two completeness criteria (≥270 valid days/year, strict; ≥120 valid days/year, mapping support). Annual concentration surfaces (5-km grid) were generated using inverse distance weighting. Reliability was evaluated via leave-one-out cross-validation (LOOCV) for p = 1-3; p = 2 residuals were assessed using exploratory Global Moran’s I. Non-attainment was classified using the WHO 2021 interim target 3 (WHO IT3; 15 µg/m³). Results Substantial spatial heterogeneity was observed, with station medians ranging from 13.70 to 28.09 µg/m³ and a maximum daily mean of 65.25 µg/m³. Network-level annual means declined from 29.97 µg/m³ in 2018 to 21.42 µg/m³ in 2022, before increasing to 25.67 µg/m³ in 2023. Harmonized raster classification showed non-attainment across 100.00% of the classified domain in 2019-2020, 92.68% in 2021, 89.52% in 2022, and 98.88% in 2023. Under p = 2, pooled LOOCV root mean square error (RMSE) and mean absolute error (MAE) were 6.85 and 5.70 µg/m³, respectively. Negative R² values indicated limited predictive performance, while p = 1 performed best and no significant residual clustering was detected during 2020-2023. Conclusions Persistent, near-domain-wide WHO IT3 PM 2.5 non-attainment was observed amid sparse, evolving monitoring. By integrating QC transparency, completeness regimes, and spatial uncertainty diagnostics, this framework provides reproducible screening-level spatial products for regional PM 2.5 assessment and monitoring network evaluation in data-limited tropical regions.