Ayako Kawano , Makoto Kelp, Minghao Qiu , Kirat Singh, Eeshan Chaturvedi, Ines Azevedo, Marshall Burke

Poor ambient air quality represents a substantial threat to public health globally. However, accurate measurement of air quality remains challenging in many parts of the world, including in populous countries like India, where ground monitors are scarce yet exposure and health burdens are expected to be high. This lack of precise measurement impedes understanding of how pollution exposure changes over time and varies across different populations, and inhibits monitoring of interventions to improve air quality. Here we develop open-source daily fine particulate matter (PM2.5) datasets at a 10 km resolution for India from 2005 to 2023, using a region-specific two-stage machine learning model carefully validated on held-out monitor data that it was not trained on. Our model demonstrates robust out-of-sample performance, substantially outperforming existing publicly-available monthly PM2.5 datasets. We use model output to analyze long-term air quality trends, finding that PM2.5 increased across most of the country until around 2016 and then began to decline thereafter, partially driven by favorable meteorology in southern India. Importantly, recent PM2.5 reductions were substantially larger in wealthier areas, albeit from a higher initial level, but we find no evidence that the recently-adopted National Clean Air Program has improved air quality in targeted urban areas to date. Our results highlight the urgency of air quality control policies that effectively target both lower and higher socioeconomic groups. To further enhance air quality monitoring across populations in India and other countries, we use model output to propose locations where new ground monitors should be installed in India, and examine the adaptability of our method to other settings with scarce ground monitoring data.