Descripción de dos métodos de rellenado de datos ausentes en series de tiempo meteorológicas

Abstract

Two methods for filling missing data gaps in geophysical time series are presented. The first one is based on the principal component decomposition of the correlation matrix built for close spatial stations with common time series records of the same variable. This multivariate method allows the incorporation in the estimated values of large scale features based on the information shared by the stations. The second method could be used when there are no close station and the missing data must be calculated from the same station historical information. This method adjusts an auto-regressive model to the time series which is then used to estimate the missing data. Two algorithms were used to calculate the auto-regressive coefficients: the Burg estimator and the one proposed by Ulrych and Clayton. The first one is appropriate for stochastic processes and the second for deterministic series. The two methodologies described in this work are recursive: a first estimation of the missing data is done running the algorithms but ignoring or using a crude approximation of them. Then, the algorithm runs again with the new estimated data, replacing the previous run missing data estimations. The run stops when the maximum difference between two successive estimations is smaller than the value fixed by the user. Filled data conserves the mean and standard deviation of the original time series. These algorithms have been adapted and modified for its use in SCILAB using also Graphic User Interfaces. Scilab is an open source platform, similar to MATLAB, and runs indistinctively in Windows and Linux. They were elaborated as an extension activity of the University of Costa Rica.