High-Resolution Map** of Air Pollution in Delhi Using Detrended Kriging Model

Abstract

Air quality information is not captured adequately due to limited numbers of air quality monitoring stations across many cities worldwide. Limited studies apply advanced spatial map** techniques to predict pollutant concentrations in highly polluted regions with high spatial variability. This paper demonstrates an advanced detrending spatial map** technique to assess the variations of particulate matter concentrations across different land use categories in a highly polluted city—Delhi—and estimate population-weighted average concentrations in the city. The “Detrended Kriging” method uses the city’s monitored datasets and land use information to predict pollutant concentrations. Concentrations are detrended based on high-resolution local land use characteristics and then interpolated using ordinary kriging before retrending again. The model estimates population-weighted concentrations (more important for health exposures) of PM_2.5 (113 µg/m³) and PM₁₀ (248 µg/m³) for Delhi and finds them to be 21–36% higher than the monitored values in the crucial winter season of 2018. The model demonstrates satisfactory performance on both spatial and temporal scales in Delhi and shows high index of agreement (d = 0.86 for PM₁₀ and 0.81 for PM_2.5), low RMSE (27.3 µg/m³ for PM₁₀ and 11.8 µg/m³ for PM_2.5), and low bias (− 1.6 µg/m³ for PM₁₀ and − 0.5 µg/m³ for PM_2.5) for the detrended kriging model, in comparison to ordinary kriging (PM_2.5 (d = 0.54, RMSE = 13.81, bias =  − 0.86) and PM₁₀ (d = 0.33, RMSE = 41.73, bias =  − 4.7)) and inverse distance weighting method (PM_2.5 (d = 0.65, RMSE = 16.08, bias = 2.93) and PM₁₀ (d = 0.55, RMSE = 46.10, bias = 7.8)). Statistical measure “d” varies between 0 (no agreement) and 1 (perfect match).