VERS DES INDICATEURS GÉOSTATISTIQUES POUR LES COMPTES DE LA QUALITÉ DE L'EAU
Résumé
The recommendations stemming from the European "water framework directive" are to build quality indicators from elementary statistics on the annual mean and the 90 percentile. But the exploratory analysis of the nutrient concentrations measurements (nitrates etc.) per station shows that the implicit usual hypotheses are not verified. The concentrations present an annual periodic component as well as time correlation, revealed by the experimental variograms. How to define then the distribution on which mean and quantile are inferred, and what hypotheses on stationnarity are really necessary ? The annual mean value, defined as a time integral, is estimated by kriging. The corresponding weighting associated with a linear interpolation of the quantile function, allows reducing strongly the bias on the quantile. But its estimation remains often imprecise.