The statistical behavior of PM10 events over guadeloupean archipelago: Stationarity, modelling and extreme events

Abstract

Environmental pollution management is one of the most important features in pollution risk assessment. Several studies have shown that exposure to particulate matter with an aerodynamic diameter of 10 μm or less, i.e. PM10, were associated to adverse health effects. To our knowledge, no study has yet investigated the modelling of PM10 frequency distribution and extreme events in the Caribbean basin. Here, the descriptive statistics and four theoretical distributions (lognormal, Weibull, Burr and stable) were used to fit the parent distribution of PM10 daily average concentrations in Guadeloupe archipelago with a database of 11 years. In order to determine the best distribution, the Kolmogorov–Smirnov statistic test (KS test) was computed as performance indicator value. With an annual average of 26.4 ± 16.1 μg/m3, the descriptive statistics highlighted that PM10 concentrations in Guadeloupe are lower than those measured in cities of Europe, Asia or Africa. Contrary to other megacities, we found that high PM10 levels in Guadeloupe are mainly due to natural large-scale sources, i.e. African dust. From May to September, i.e. high dust season, PM10 concentrations are 1.5 times larger since dust outbreaks are more frequent. A statistical stationarity threshold of 66 months is estimated using the distribution analysis. This underlines the cycle stability of African dust over this last decade. Concerning the statistical modelling, our results showed that Burr & Weibull mixture model is the best distribution to represent PM10 daily average concentrations with a first statistical behavior corresponding to the low dust season and an another to the high dust season. By analysing the extreme events statistic with the classical power-law distribution, we observed that Burr & Weibull mixture model could also improve the modelling of these events. In summary, the Burr & Weibull mixture model is suitable to model both classical and extreme events.