We use a network of GPS receivers on the Big Island of Hawaii to examine in detail the precipitable water field of the Ka'$overline {rm u}$ storm, which generated record rainfall and flash floods over much of the southern and eastern portions of the island. With GPS stations distributed from sea level to the highest points on the island, the spatial and temporal variations in precipitable water as the storm passed over the network are investigated. The results highlight the role of the topography in controlling the location of convection, revealing relatively static zones of high precipitable water where onshore winds and terrain anchor the storm. Using the observed correlation between the precipitable water and rainfall, estimates of rainfall based on the GPS data are generated. These estimates compare well with most of the high-rainfall sites but tend to overestimate rainfall for the low-rainfall areas. Rainfall predictions from a mesoscale spectral model of the storm consistently underestimate the observed heavy values. Comparison with the GPS, however, suggests that this underestimate is probably due to a correlated underestimate of precipitable water, indicating that significant improvements in predictions could be expected if the precipitable water field can be better constrained.