Assessing Rainfall Data Homogeneity and Estimating Missing Records in Mākaha Valley, O‘ahu, Hawai‘i

Abstract

The objectives of this study were to examine records from long-term rain gauges in Mākaha Valley for data homogeneity and to compare methods of estimating missing data. Double mass analysis was used to investigate data homogeneity. Results show that tree growth near one gauge has reduced rainfall catch by 21–25% since 1974. Four methods for estimating missing daily rainfall data were then tested using index gauges selected from a network of 21 active rain gauges. The number of index gauges and their order of selection were varied according to proximity and correlation. Selection by correlation significantly improved the performance of the station average and inverse distance methods for most cases, as well as the normal ratio method for the case when only one gauge is used. The normal ratio method produces the lowest error when two to five index gauges are used; the inverse distance method yields the lowest error when six or seven index gauges are used. Direct substitution produces better accuracy than the normal ratio method when using only one index gauge. Problems related to multicollinearity, heteroscedasticity, and assumptions of data normality preclude the use of multiple linear regression.