Application of land-use regression model with regularization algorithm to assess PM2.5 and PM10 concentration and health risk in Kolkata Metropolitan

Abstract

Present research aimed to develop monthly and annual land-use regression (LUR) model for simulation of particulate matter (PM) concentration using conventional linear regression and shrinkage or regularization algorithms taken 124 predictor variables from 44 monitoring stations in Kolkata Metropolitan Area. The present study introduced shrinkage algorithms for LUR with six step approaches to understand the variation in results between conventional LUR model and shrinkage algorithms. The adjusted R2 of all monthly LUR models ranges from 0.54 to 0.82 for PM2.5 and 0.23 to 0.65 for PM10, respectively. In annual LUR model with shrinkage algorithms, R2 value and RMSE of leave-one-out cross-validation (LOOCV) are 0.58, 12.7 for PM2.5 and 0.547, 11.49 for PM10. In annual conventional linear model, R2 and RMSE of LOOCV are 0.62, 14.70 for PM2.5, and 0.549, 11.99 for PM10, respectively. The result shows regularization algorithm models performed better than conventional linear regression model. Furthermore, location based ambient PM induced three health risks at public space i.e. all-cause mortality, cardiopulmonary and lung cancer are determined using attributable fraction and excess risk. The results demonstrated that highest mean excess risk for lung cancer is high at city center where people exposed are more vulnerable due to high PM2.5 concentration.