Preenchimento de falhas de séries micrometeorológicas utilizando técnicas estatísticas combinadas

Abstract

With the progress of information technology, regardless of context, data is the most precious asset an organization can have. However, they are subject to failures, which can occur due to several factors, such as measurement errors, equipment failures, anthropic actions, among others. This fact makes it difficult to analyze and affects its applicability. This thesis intends to develop a computational program that automates the filling of these flaws and creates an inventory of time series of micrometeorological data, supported by statistical techniques like frequency, Monte Carlo, Bootstrap, Moving Average, interpolation and linear regression, seeking to preserve the characteristics of the series, seasonality, trend, variance and amplitude. This program also has routines to treat situations in which the technique leads to divergences of results (exceptions). To validate the model, artificially produced flaws were filled in a time series of air temperatures and relative air humidity collected in the micrometeorological tower of Sinop. Statistical correlation tests, regression coefficient of the order 0.95, and f test for variance between the original series and the estimated series were applied. In the range of 1% to 70% of failures, being recorded every 5% of this range, the results validated the model.