Estimating background PM_2.5 concentrations using a wavelet-based convolution framework

Quantifying regional background pollution, driven by long-range transport and diffuse emissions from surrounding regions, is crucial for isolating the impact of local sources on air quality. The existing methods to estimate background such as statistical clustering and spectral filtering have the limitation to not account for the spatial resolution that is essential to define background pollution. In this study, we adopted a signal processing method to obtain a spatiotemporal distribution of PM_2.5 background concentrations over India using MERRA2 reanalysis dataset. This method utilizes wavelet transforms to capture the coherence of low-frequency components in time series from neighborhood regions. The derived PM_2.5 background map captures the major seasonal and regional pollution regimes across India, including winter buildup over the Indo-Gangetic Plain (IGP), pre-monsoon dust-driven enhancements in the northwest. The method reproduces expected transitions between polluted and clean periods, yielding a consistent background pattern. A comparison of derived PM_2.5 background with those obtained from the Gaussian Mixture Model-Hidden Markov Model (GMM-HMM) at four major Indian cities also showed good agreement with MAD ranging from 2.2 (Kolkata) to 10.8 (Delhi) µg/m³ across cities. Analysis of high- and low-emission grids showed that the method correctly yields higher background contributions at low-emission grids, reflecting dominant regional transport and higher increments at high-emission grids linked to local sources. Finally, evaluation against 5th percentile PCHIP-derived background concentrations shows moderate agreement across the Delhi monitoring network (MAE = 21.95 µg/m³, RMSE = 29.45 µg/m³), improving substantially at the background site in Delhi (MAE = 14.03 µg/m³, RMSE = 17.65 µg/m³), consistent with reduced local-source influence at this location and supporting the reliability of the proposed method.