Non-parametric estimation of the influence of meteorology on indoor and outdoor air

Indoor air quality is known to be highly influenced by outdoor air quality through the exchange of indoor and outdoor air, whether by natural or mechanical means[1]. Local meteorology (wind speed, wind direction, temperature, precipitation and relative humidity) also play a role in modifying indoor and outdoor air quality[2]. Here, we present analysis of indoor/outdoor air pollution measurements and the influence of meteorology and temporal trends to predict the concentrations of measured indoor air quality parameters. Generalised Additive Models (GAMs) are applied to examine temporal trends in, and the effect of, meteorology on concentrations of PM10 and NO, NO2, NOx, O3, and SO2, based on four years (2004-2007) of simultaneous indoor and outdoor measurements in Haifa, Israel. Classification and Regression Trees (CARTs) are also applied to the data to determine which of the above parameters are most relevant to explaining differences in air quality and prediction from both GAMs and CARTs is performed for validation. Wind direction was found to have a significant influence on increasing indoor concentrations of PM10 and all the other gases. Temperature, wind speed and relative humidity were associated with changes of outdoor Ozone and PM10. The key drivers of seasonal variation in outdoors NO2, NOx and NO are related to atmospheric photochemistry. An advantage of using CARTs is that influential predictor variables can be identified for further modelling within the GAM framework. The resulting GAMs are then able to explain, with a great deal of flexibility in capturing non-linear effects, the influence of these meteorological parameters, as well as the smooth, periodic, annual trend