On the Applicability of Bartlett Lewis Model: With Reference to Missing Data

The availability of a complete hourly rainfall dataset is required for hydrological applications, statistical modeling and forecasting of precipitation. The issues of missing data are of serious concern in rainfall modelling due to the problem in computing the autocorrelation when gaps of missing data are encountered in the pooled data. The present paper discusses the applicability of three methods of handling missing hourly data when the Bartlett Lewis rectangular pulses model is utilized: zero substitution, single imputation and multiple imputations. The three methods are applied to the hourly rainfall data from the Bukit Bendera rain gauge station, which consists of a complete nine year rainfall series. The methods are tested with different percentages of randomly generated missing rainfall values. The performance of the methods is studied in terms of the mean absolute deviation errors that are found during different monsoon periods. The findings indicate that the best method to address missing data when applying the Bartlett Lewis rectangular pulses model is the single imputation method.