A geostatistical approach to areal rainfall estimation using raingage and radar measurements is described. The so-called cokriging method is used to obtain a linear estimator of ground-level rainfall depths by combining gage and radar data under unbiasedness and optimality constraints. The statistical inference of the spatial structure of these two kinds of measurements (required to determine the cokriging system is discussed in the multi- and single-realization context. A simplified version of the cokriging method is then proposed to obtain a more tractable system for practical applications.

A validation procedure based on (i) the estimation of reference rainfall depths and (ii) the selection of a set of likeness criteria is defined. The reference values are computed in an original way by integrating raingage measurements over radar pixels containing a test raingage using the classical kriging method.

The test case deals with a set of 11 daily rainfall events observed in the Paris region by the 10 cm “Melodi” weather radar system. The available raingage network includes 98 stations spread over 20 000 km ² : 69 stations have been used for validation purposes and the remaining 29 For the simplified cokriging operating method.

The available radar dataset presents severe limitations for hydrological applications mainly in relation to ground echo effects within a 52 km radius of the radar site. In spite of these unfavorable conditions, the proposed combination method appears to improve slightly the performance of the raw radar data and to exceed that of the classical uniform calibration method.

Further application of this method using a more appropriate dataset is necessary to confirm these initial results.