We defined a grid--scale time--phase relationship between precipitation (PR) and underlying sea surface temperature (SST) as the local PR--SST relationship, and compared the spatial distribution of the relationship in the observation with that in a reanalysis dataset. An analysis using observed pentad mean data shows precipitation lagging SST for two pentads over large areas of tropical oceans. A reanalysis dataset, on the other hand, has a shorter time--phase difference between precipitation and SST than that in the observation and does not match the spatial distribution of the observed local PR--SST relationship over the tropical Indian Ocean and the tropical western Pacific. A reanalysis dataset may not capture the observed local PR--SST relationship in terms of either time--phase difference or spatial distribution because a numerical model in a data assimilation system uses SST as a lower boundary condition as well as the inhomogeneity of observations the system used.