Comparison of missing value estimation techniques in rainfall data of Bangladesh

The presence of missing values in daily rainfall data may hamper the analyses to determine effective results for solving problemsof hydrological, agricultural, and climatological issues. The study attempts to select an appropriate method for estimating themissing value of daily rainfall data of Bangladesh. For this purpose, eight methods and seven comparison techniques areemployed. For imputation of missing values employing these methods, three sets of daily rainfall data (1, 5, and 10% missingvalues) with 1000 repetitions are considered randomly for five regions of the country. These samples are artificially created asmissing and then imputation for these missing values is made applying the selected methods. The relative performance of themethods are examined using some comparison criteria. The following observations can be made from the study regarding thechoice of the appropriate missing value estimation technique: for imputation of the missing values of daily rainfall data, thearithmetic average method for rainfall stations Chittagong and Rajshahi in the south-east region and the north-west region,respectively, is found as the best methods. Further, the single best estimator method for rainfall stations Sylhet and Dhaka inthe north-east region and the mid-region, respectively, and the EM-MCMC method for rainfall station Khulna of the south-eastregion are also identified as the best methods in respect of Kolmogorov-Smirnov test, the lowest bias of estimate, the value of Sindex, et