### Abstract

Contamination by nitrates is an important cause of groundwater pollution and represents a potential risk to human health. Management decisions must be made using probability maps that assess the nitrate concentration potential of exceeding regulatory thresholds. However these maps are obtained with only a small number of sparse monitoring locations where the nitrate concentrations have been measured. It is therefore of great interest to have an efficient methodology for obtaining those probability maps. In this paper, we make use of the fact that the discrete probability density function is a compositional variable. The spatial discrete probability density function is estimated by compositional cokriging. There are several advantages in using this approach: (i) problems of classical indicator cokriging, like estimates outside the interval (0,1) and order relations, are avoided; (ii) secondary variables (e.g. aquifer parameters) can be included in the estimation of the probability maps; (iii) uncertainty maps of the probability maps can be obtained; (iv) finally there are modelling advantages because the variograms and cross-variograms of real variables that do not have the restrictions of indicator variograms and indicator cross-variograms. The methodology was applied to the Vega de Granada aquifer in Southern Spain and the advantages of the compositional cokriging approach were demonstrated.