Hawaiian rainfall climatography

Abstract

Monthly and annual rainfall data from 591 stations on the island of Oahu, Hawaii, have been compiled as a computerized data set. Data from 76 stations are used in a presentation of the techniques and procedures for preparing a rainfall atlas. The relative merits of the median versus the mean as representative normal statistics are examined. Normal is defined here as the central point of a distribution about which the other values are evenly distributed. It is concluded that the median is the better choice both because it stabilizes more quickly and because it is less affected by the extreme values often present in hydrological data. A method of biased estimation, ridge regression, is introduced as a technique for extrapolating long period normal rainfalls from short records. Ridge regression is shown to be superior to ordinary least squares regression, double-mass analysis, and a robust estimator of location for extrapolating Hawaiian rainfall normals. Some aspects of the effects of topography on rainfall are discussed. It is shown that, at points equidistant from the crest, there is no substantial difference between rainfalls on leeward ridges and valleys on Oahu. It is suggested that aerodynamic, rather than topographic, effects might explain the results of previous study which found the valleys to be wetter than the ridges. It is concluded that maps of median rainfalls, constructed using the technique of ridge regression, will best represent the distributions of normal rainfall on the Hawaiian Islands but, because of the many errors inherent in the measurement of rainfall, exact rainfall maps cannot be produced.