| Linear co-kriging was used to merge raingage measurements and radar rainfall data. To evaluate various co-kriging estimators, three experiments were performed. The first two are simulation experiments. Assuming high-quality radar rainfall fields and the rainfall fields generated from a stochastic space-time rainfall model as the ground-truth rainfall fields, raingage measurements and radar rainfall fields were generated with varying raingage network density and the error characteristics of radar rainfall. Ordinary and universal co-kriging estimators were then used to merge the raingage measurements and the radar rainfall data. Since raingages are usually sparse, the second-order statistics required for co-kriging can only be known with large uncertainty. The effect of the uncertainty and the sampling error of raingage measurements were also evaluated by co-kriging ground-truth rainfall data and radar rainfall data. The third is a real-world experiment. A physically based rainfall model was used to estimate the mean field, or the trend, of ground-level rainfall field. Modified simple co-kriging was then performed to merge the residual raingage measurements and the residual radar rainfall data. Ordinary and universal co-kriging estimators were also included for comparison purposes. The results show that linear co-kriging is potentially useful for merging raingage measurements and radar rainfall data. Under a range of raingage network densities and error characteristics of radar rainfall data, the gage-radar estimation using co-kriging is shown to consistently provide rainfall estimates that are better, in the minimum-error-variance sense, to the gage-only or the radar-only estimates alone over the life cycle of an oceanic convective storm. The magnitude of improvement is only minor. However, the consistency of the improvement by the gage-radar estimation, under the various conditions of the error characteristics of radar rainfall makes co-kriging an attractive tool in rainfall estimation. |
