Tratamento de grandes volumes de dados hidrometeorológicos apoiados por validação cruzada em Workflows científicos

Abstract

Researchers face several challenges in order to handle large volumes of high-quality meteorological data, free of missing (or gaps). The detailed comprehension of rainfall temporal data is important in daily activities such as in the evaluation of the distribution of rain in a given area. The proper study of such dataset may aid administrators in planning cities, industries and farmlands. In Brazil, meteorological dataset is collect by meteorological rainfall stations that are spread in the geographic space. Part of such dataset are maintained Agência Nacional de Águas (ANA) by means of HidroWeb system. Activities like detect and fill temporal series gaps are crucial to control the quality of meteorological data. This dissertation presents a computational approach based on scientific workflows and crossvalidation and interpolation methods to face the above-mentioned challenges. The interpolation methods we have used in this dissertation are linear regression,regional weighting, inverse square distance, regional weighting with linear regression. The scientific workflows we have developed are able to use the four methods to generate large volumes of high meteorological data; they are also able to collect retrospective provenance from the in silico experiments. Our experiments evaluated dataset from 34 (from 77) meteorological rainfall stations which the temporal serial has at least ten years of data. Our experimental results showed that the best results were achieved with PR method