Beginning in 1985, a continuous aerosol sampling program has been undertaken at a coastal location in northwestern Corsica. This site is 300 m above sea‐level and at least 20 km from local pollution sources. It is exposed during 80% or more of the time to maritime air masses which have travelled over the western Mediterranean from southern, western and northern directions. Daily 24‐h aerosol samples were collected on 0.4 μm pore size nuclepore filters from a 10 m high tower and analysed for Al, Si, P, S, K, Ca, Ti, Mn, Fe and Zn by X‐ray fluorescence, and for Na, Cu and Pb by flameless atomic absorption. The data obtained up to now, from April 1985 to April 1986, show that the temporal variability of the concentrations displays 2 distinctive patterns. First, a seasonal pattern is observed for the elements of continental origin, either natural (e.g., Al, Si) or anthropogenic (e.g., S, Pb). This pattern is inversely related to the frequency and amount of rainfall such that the highest concentrations are observed between May and October. Estimates based on these data indicate an average time of 2 days to reload that atmospheric environment with aerosol particles from continental sources. This seasonal pattern is not observed for locally produced sea‐salt aerosol particles (index Na), whose concentration is related to local wind speed. Elements associated with mineral aerosol particles exhibit sporadic but intense concentration peaks that are superimposed on this general pattern. 20 of these events were recorded for the sampling period considered with their frequency being maximal in spring and summer. 3‐D air‐mass trajectories show that all these events are associated with transport of soil dust from Africa.

International audienceA neural network combined to a neural classifier is used in a real time forecasting of hourly maximum ozone in the centre of France, in an urban atmosphere. This neural model is based on the MultiLayer Perceptron (MLP) structure. The inputs of the statistical network are model output statistics of the weather predictions from the French National Weather Service. With this neural classifier, the Success Index of forecasting is 78% whereas it is from 65% to 72% with the classical MLPs. During the validation phase, in the Summer of 2003, six ozone peaks above the threshold were detected. They actually were seven