An alternative approach for estimating large-area indoor PM2.5 concentration – A case study of schools

For indoor air modelling, difficulties in collecting parameters including life activity patterns and building characteristics are dilemmas when conducting a large-area study. Land-use/land cover information which is easier to obtain could represent as surrogates of emission sources for assessing quality. Moreover, low-cost sensors machine learning provide better way enhance model accuracy. This study proposed an alternative estimation approach assess daily PM2.5 concentration environments schools large area by integrating sensors, land-use/land predictors, learning-based modelling approaches. Indoor data was collected from 145 AirBox Kaohsiung Pingtung Counties Taiwan. Geospatial predictors were extracted the circular buffers surrounding each sensor. Spearman correlation analysis stepwise variable selection procedures performed select variables land-use regression (LUR) integrated with XGBoost, Random Forest (RF), LGBM models. The results revealed that outdoor distance nearest thermal power plant main determinants variations, there no sources. When incorporating learning, R2 increased 0.59 LUR 0.85 LUR-XGBoost while RMSE decreased 8.63 5.27 ?g/m3, than both LUR-RF LUR-LGBM. demonstrates value sensor algorithm estimating spatiotemporal variability area.