Cascade-Machine-learning quantifies hourly wildfire smoke exposure and acute health risks in California

The rising frequency and severity of wildfires have intensified concerns regarding their adverse impacts on public health. However, accurately quantifying acute wildfire smoke exposure and its associated health impacts remains challenging due to limitations in existing high-resolution exposure data. Here, we develop a novel Cascade-Machine-Learning framework to generate unprecedented hourly wildfire-specific PM_2.5 concentrations at a 1 km × 1 km resolution across California, achieving substantial accuracy improvements over traditional chemical transport models and satellite-derived datasets. Leveraging this high-resolution dataset with health records from the University of California, we identify critical relationships between short-term wildfire smoke exposure and acute pneumonia-related health risks. Notably, we introduce a new exposure metric, P_max, capturing the intensity of hourly peak exposures relative to daily accumulated exposure, and reveal that short-lived, pulse-type wildfire smoke events are associated with nearly tenfold higher pneumonia-related medical risks compared to sustained exposure at equivalent daily average concentrations. Our results further highlight heightened vulnerability among individuals younger than 18 years and the African American populations. This work underscores the urgent need for temporally detailed exposure assessment in wildfire health studies and provides a robust scientific foundation for targeted public health interventions and emergency preparedness in an era of intensifying wildfire risks.