Integrating WRF-CHEM simulation with Random Forest algorithm to evaluate the spatio-temporal variation of PM10 in Western Himalayan region of India

This present study develops an integrated framework of WRF-CHEM simulation and Random Forest algorithm to improve PM10 predictions over Himalayan region India. The WRF-CHEM simulations showed a clear seasonal fluctuation in PM10 levels, with wintertime concentrations being higher due to increased biomass burning, decreased boundary layer height, and atmospheric stability. Large-scale pollution patterns were well represented by WRF-CHEM. To compare the observed surface PM10 concentrations with WRF-CHEM simulations, the random forest algorithm is used by including WRF-CHEM simulated meteorological/air pollutants values and considering the spatiotemporal characteristics as input parameters for the random forest algorithm. The obtained bias-corrected predictions for Raipur (30.29°N, 78.09°E; ASL ∼483 m), Rudrapur (28.97°N, 79.40°E; ASL ∼217 m), Kashipur (29.21°N, 78.96°E; ASL ∼218 m), and Clock Tower (30.32°N, 78.04°E; ASL ∼447 m) achieved a correlation of 0.76, 0.71, 0.85 and 0.72 respectively while index of agreement were 0.86, 0.82, 0.6 and 0.77 respectively. The integrated WRF-CHEM and Random Forest framework demonstrated substantial improvements in PM10 estimations across diverse topography. These findings signify the necessity of incorporating topographic variability into post-processing of chemical transport model outputs to refine air quality predictions.