Daily high-resolution PM2.5 mapping using spatiotemporal CNN-transformer-KAN model

作者：Zhifei Liu, Kang Zheng, Yongze Song, Jianing Zhang · 发表于：International Journal of Applied Earth Observation and Geoinformation · 年份：2025 · DOI：10.1016/j.jag.2025.104900 · 被引用次数：2 · 研究领域：Air Quality Monitoring and Forecasting、Atmospheric and Environmental Gas Dynamics、Air Quality and Health Impacts

• A novel CTKNet was proposed for PM 2.5 estimation. • First application of the KAN in air pollution modeling. • Spatiotemporally continuous AOD data is utilized for high-resolution PM 2.5 estimation. • Spatial-temporal PM 2.5 trends are revealed to support air quality policy and planning. Daily high-resolution mapping of fine particulate matter (PM 2.5 ) is critical for air quality monitoring and public health. However, current methods struggle to achieve high accuracy over large spatial and temporal scales due to limitations in modeling complex spatiotemporal dependencies. This study proposed a novel hybrid deep learning model—CNN-Transformer-KAN Network (CTKNet)—which utilizes Convolutional Neural Networks (CNN) to capture spatial features, Transformers for capturing long-range dependencies, and the Kolmogorov–Arnold Network (KAN) for nonlinear representation learning. Utilizing spatially continuous satellite aerosol optical depth (AOD) data and other multi-source spatiotemporal inputs, CTKNet estimated daily PM 2.5 at a spatial resolution of 1 km across China for the period 2015–2020, marking the first application of KAN in PM 2.5 estimation. It outperformed existing models, achieving a high cross-validation coefficient of determination (R 2 ) of 0.95 (sample-based), 0.90 (station-based), and 0.78 (time-based), and corresponding RMSEs of 8.13, 11.03, and 17.76 µg/m 3 . Yearly sample-based cross-validation R 2 values ranged from 0.91 to 0.96 with RMSEs below 11.74 µg/m 3 ,...