Distinguishing Regional And Local PM2.5 Influences Using Sensor Networks: Evidence From West Bengal, India

SATHISH SWAMINATHAN, Panagiotis Reskos, V. Faye McNeill, Columbia University

     Abstract Number: 333
     Working Group: Remote and Regional Atmospheric Aerosol

Abstract
The Indo-Gangetic Plain (IGP), situated along the foothills of the Himalayas, is among the most polluted regions globally, with air quality influenced by both anthropogenic and natural factors. Key anthropogenic sources include stubble burning, industrial emissions, and urban activities from densely populated cities. Meteorological phenomena such as long-range atmospheric transport, temperature inversions, and fluctuations in boundary layer height further modulate pollutant dispersion and concentration. West Bengal is uniquely positioned as the only state in the IGP that is bordered by both the Himalayan foothills to the north and the Bay of Bengal to the south. This geographical placement makes it particularly susceptible to the combined effects of cold air diversion from the mountains and monsoonal precipitation driven by the sea. West Bengal is also one of the first Indian states with an extensive state-owned, sensor-based air quality monitoring network comprising over 200 low-cost sensors. This infrastructure enables high-resolution analysis of spatiotemporal PM2.5 patterns at the district level. In this study, we apply machine learning techniques to distinguish between regional and local influences on PM2.5 concentrations using sensor data alone. Time series decomposition and clustering techniques are used to identify signatures at different temporal scales and group locations with similar signatures. Meteorological inputs are then introduced to assess the effectiveness of the method in describing the relative contribution of regional and local factors across a geographically diverse state.