Quantifying local and background contributions to PM10 concentrations in Haute‐Normandie, using random forests

Abstract

AbstractThe problem is to quantify local and background contributions to PM10 concentrations in Haute‐Normandie region. We use measures of pollution variables on a network of 11 monitoring sites, completed by meteorological variables, during 2004–2009. Random forests (RFs), a recent statistical method, are used to put in evidence the marginal effects of explanatory variables, and to classify parameters of influence on PM10 pollution in different situations: roadside, urban background, industrial, and rural. The local pollution is the most important source and is marked by the classic tracers NO and NO2 as urban activity pollution, and SO2 as industrial one. The entire process of statistical quantification of local and background contributions, without neither direct information nor measurements about sources, can be divided in three main steps. The first one is to classify the explanatory variables (pollutants and meteorological parameters) into five groups: pollutants from urban activity, pollutants from industrial activity, and three groups of meteorological variables. The second step is to handle the specificity of a rural and coastal station for which there is a priori no local pollution sources, to use it as a marker of the background pollution. The final step consists of quantifying the local and background contributions to the PM10 pollution, using only markers of urban and industrial activities and PM10 background. As a synthetic conclusion, it appears as often seen in publications about other geographical locations, that the local part contributes half of the total pollution. Copyright © 2011 John Wiley & Sons, Ltd.