Missing data imputation of climate datasets: implications to modeling extreme drought events

Abstract

Time series from weather stations in Brazil have several missing data, outliers and spurious zeroes. In order to use this dataset in risk and meteorological studies, one should take into account alternative methodologies to deal with these problems. This article describes the statistical imputation and quality control procedures applied to a database of daily precipitation from meteorological stations located in the State of Parana, Brazil. After imputation, the data went through a process of quality control to identify possible errors, such as: identical precipitation over seven consecutive days and precipitation values that differ significantly from the values in neighboring weather stations. Next, we used the extreme value theory to model agricultural drought, considering the maximum number of consecutive days with precipitation below 7 mm for the period between January and February, in the main soybean agricultural regions in the State of Parana.