Long-term meteorology-adjusted and unadjusted trends of PM2.5 using the AirGAM model over Delhi, 2007–2022

Abstract

This study investigates the impact of meteorological variations on the long-term patterns of PM_2.5 in Delhi from 2007 to 2022 using the AirGAM 2022r1 model. Generalized Additive Modeling was employed to analyze meteorology-adjusted (removing the influence of inter-annual variations in meteorology) and unadjusted trends (trends without considering meteorology) while addressing auto-correlation. PM_2.5 levels showed a modest decline of 14 μg m⁻³ unadjusted and 18 μg m⁻³ meteorology-adjusted over the study period. Meteorological conditions and time factors significantly influenced trends. Temperature, wind speed, wind direction, humidity, boundary layer height, medium-height cloud cover, precipitation, and time variables including day-of-week, day-of-year, and overall time, were used as GAM model inputs. The model accounted for 55% of PM_2.5 variability (adjusted R-squared = 0.55). Day-of-week and medium-height cloud cover were non-significant, while other covariates were significant (p < 0.05), except for precipitation (p < 0.1). Wind speed (F-value: 98) showed the strongest correlation, followed by day-of-year (61), years (41.8), planetary boundary layer height (13.7), and temperature (13). Meteorological parameters exhibited significant long-term trends, except for temperature. Inter-annual meteorological variations minimally affected PM_2.5 trends. The model had a Pearson correlation of 0.72 with observed PM_2.5, underestimating episodic peaks due to long-range transport. Partial dependencies revealed a non-linear PM_2.5 relationship with meteorology. Break-point detection identified two potential breakpoints in PM_2.5 time series. The first, on October 1, 2010, saw a significant increase from 103.4 to 162.6 μg m⁻³, potentially due to long-range transport. Comparing meteorology-adjusted and unadjusted trends can aid policymakers in understanding pollution change causes.