Aplicação de técnicas de preenchimento de falhas de dados de pluviosidade mensal e anual para o noroeste do estado de Minas Gerais - Brasil

Abstract

The use of historical series of meteorological data requires the application of statistical techniques to correct faults (absence, error of registration, etc.) and may compromise the analysis if not implemented. From this perspective, the purpose of this paper was to analyze different statistical methods of correcting rainfall data (Arithmetic Average, Regional Weighting, Multiple Linear Regression and Regional Weighting Based on Linear Regressions) and discuss their application. Daily rainfall data were obtained from seven rainfall stations of the National Water Agency (located in the northwestern portion of the state of Minas Gerais) between 1995/1996 and 2014/2015, whose validation was performed by the double technique consistency process. It was used a test station for correction (Fazenda Conceição) and six support, which totaled 1040 estimated rainfall records (240 months and another 20 years for each). Multiple Linear Regression was the technique that obtained the best estimates on the annual and monthly scales for the period from October to May (rainy months), but there was no prevalence of a technique from June to September (dry months).