Improved Rainfall Estimation by Integration of Radar Data : A Geostatistical Approach

Abstract

The integration of radar data has been shown a powerful tool for the improvement of rainfall spatio-temporal estimation with respect to the estimation using only raingage data. However, current techniques are limited to standard cokriging algorithms and consider the samples as time independent. We propose the use of the co-located cokriging and kriging with an external drift algorithms, originally developed for the inclusion of geophysical data for the estimation of petrophysical attributes in the petroleum industry, to improve rainfall estimation by integration of radar data. We also propose to include time in the estimation process as a third coordinate, therefore accounting for temporal correlations in the data. These algorithms are demonstrated using part of the information gathered in the “Cévennes 86–88” experiment, consisting of radar data, exhaustively covering the area under study, and point data from 39 raingages, for a rainfall event lasting 5 hours. For this particular data set, it is concluded that accounting for the temporal correlation is not worth the extra modeling effort, in part due to the high point to point correlation between rainfall and radar data. The differences in estimation between including and not the radar data are important even for this case in which the number of rainfall gages is large.