A prediction method for nonlinear time series analysis of air temperature data by combining the false nearest neighbors and subspace identification methods

A climate system can be considered as a dynamical system with a chaotic behavior. This system is built onto different subsystems. One of them is the air temperature. This magnitude is of particular relevance on evapotranspiration models, general circulation models, and generally in any hydric budget model, becoming essential its determination as accurate as possible. Due to the fact that its value can vary significantly, and it is demonstrated that this quantity obeys a chaotic behavior, it is essential to be able to predict it. Thus, accurate temperature measurements or predictions are required. In this paper, we have considered a one-year time series of temperature data. The experimental data have been recorded by a meteorological station located on the top of the building of the Escuela Politécnica Superior de Albacete, which belongs to the University of Castilla-La Mancha. In order to make predictions, we propose a method which combine the minimal embedding dimension obtained by the method of false nearest neighbors and a model estimated by means of a subspace identification method. The results, in terms of predicted error, show the reliability of this new approach. Key–Words: Nonlinear Time Series Analysis, False Nearest Neighbors Method, Subspace Identification Methods, Prediction