Evaluation of Two Different Methods of Data Quality Analysis Using Local Daily Precipitation Measurements at Santa Catarina State

Abstract Daily rainfall data from a meteorological network in Southern Brazil is used to assess the performance of two different outliers detection algorithms. Both methods use a statistical and spatial consistency approach based in distance and elevation difference between two rain gauge measurements. A variation of the Multiple Interval Gamma Distribution method of You, Hubbard, Nadarajah e Kunkel (2007) is considered in this study. Neighboring stations data is gathered to get the local average rainfall distribution. The precipitation range of values is partitioned so one makes the assumption that every interval can be modeled by a Gamma distribution. The second method assumes no prior distribution characteristic, and instead uses point spatial and cumulated temporal information from neighboring rain gauge stations to consist daily rainfall data. In order to assess the reliability of the detected outliers, as well the accuracy, seeded errors are introduced in the historical rainfall series. A two dimensional probability model of introduced/detected error (yes-no) is used to compute metrics related to the correct detection and false alarm probabilities made by the algorithm. We verify that the new proposed method overcomes the Multiple Interval Gamma Distribution method.