Mapping climate-conditioned PM2.5 response under climate change in Taiwan using a high-resolution Geo-AI model
Journal
Journal of Environmental Management
Journal Volume
412
Start Page
130232
ISSN
03014797
Date Issued
2026-07-15
Author(s)
Wong, Pei-Yi
Hsu, Wei-Lun
Chen, Chieh-Ying
Chen, Chien-Fei
Hsu, Chia-Wei
Lung, Shih-Chun Candice
Liu, Wan-Yu
Yu, Chia-Pin
Chen, Pau-Chung
Seow, Wei Jie
Sung, Chien-Hao
Wu, Chih-Da
Abstract
This study develops a high-resolution Geo-AI framework to quantify the impact of future climate change on PM2.5 concentrations using Taiwan as a subtropical, monsoon-influenced island case. The model integrates long-term ground-based monitoring data (1994–2019), multi-scale geo-environmental predictors, and statistically downscaled CMIP6 meteorology, implemented using a Gradient Boosting Machine. The resulting model demonstrates strong predictive performance (R2 = 0.81 and RMSE = 8.69 μg/m3) and effectively captures PM2.5 dynamics within complex islands and coastal environments. By explicitly coupling a Geo-AI model with Intergovernmental Panel on Climate Change (IPCC) Sixth Assessment Report (AR6) climate scenarios, this study extends data-driven PM2.5 modeling from historical estimation to climate-conditioned future projection, addressing a key methodological gap in existing air-quality research. SHAP-based interpretability analysis identifies temperature and precipitation as dominant predictors, underscoring their central role in shaping future aerosol variability. The SHAP results further indicate that both temperature and precipitation exhibit nonlinear relationships across different temporal and regional scales and overall inverse associations with PM2.5 concentrations, clarifying the climate-driven effects of warming and hydrological change on PM2.5 dynamics under humid subtropical conditions. Across four Shared Socioeconomic Pathway scenarios, projected PM2.5 concentrations consistently decline in the near and midterm (between −1.25 and −1.5 μg/m3), followed by increasing spatial heterogeneity in the long term, with localized PM2.5 hotspots emerging under severe warming conditions. These findings suggest that climate change may generate uneven air-quality responses across space, highlighting the limitations of regional mean assessments and the need for high-resolution, climate-informed mitigation and adaptation strategies. The proposed framework provides a transferable tool for climate-responsive air-quality planning in humid subtropical, monsoon-influenced, and densely populated regions worldwide.
Subjects
Climate-conditioned PM2.5 projection
CMIP6 downscaled climate data
Geospatial artificial intelligence (geo-AI)
High-resolution air quality modeling
Humid subtropical and monsoon-influenced regions
IPCC AR6 climate scenarios
Publisher
Academic Press
Type
journal article