Generating robust rainfall intensity–duration–frequency estimates with short-record satellite data

Abstract

Summary In poorly instrumented areas, rainfall records are often absent or short, compounding the statistical challenge of estimating extreme rainfall characteristics and designing flood control structures. In this research two separate rainfall datasets in the country of Ghana were used with two different probability distribution frequency analysis methods to estimate intensity–duration–frequency (IDF) parameters. General extreme value type II (GEV-II) probability distributions were fit using two robust parameterization approaches, one called regional for combining regional and local data, the other called global for combining a global parameter and local data. Annual maximum rainfall data were from the Ghanaian Meteorological Service Department (GMSD) eight ground gages, containing N -min to 24 h durations at 5–24 years record lengths, and from 381 TRMM satellite precipitation bins covering the country, containing 3–24 h duration at 9 years record length in 2007. TRMM rainfall depths were statistically greater at 63% of the stations. Goodness of fit was computed for a maximum return interval of 43 years, and error analysis was computed for 24-h data at return intervals from 77 to 109 years, for all four separately parameterized GEV-II equations and corresponding IDF curves. The best fit and least error was obtained with the global method applied regionally, which used GMSD data to derive time-based parameters and TRMM to derive distribution parameters. The regional method could not incorporate GMSD data, and used only TRMM data, which led to substantial overestimation of rainfall intensity for short durations. This research shows it was essential to combine GMSD and TRMM data for IDF generation in Ghana. The short TRMM record sets with duplicate observed depths at increasing durations caused the regional method to generate internal consistency errors in the IDF plots, suggesting the global method with regional application is best for Ghana.