Predicción de la temperatura ambiental mediante modelos estadísticos funcionales de las estaciones monitoriadas por el GEAA en la Provincia de Chimborazo (2014-2019)

Abstract

The objective of the study was to forecast the ambient temperature, through functional statistical models of the stations of Alao, Atillo, Tunshi, Tixán, Matus, Quimiag, Urbina, San Juan, Espoch, Multitud, Cumandá, using stored ambient temperature data. from 2014 to 2019 by the Alternative Energy and Environment Group (GEAA). The databases were analyzed, ordered and spaced to make imputations using Random Forest; functional data techniques such as: smoothing B-plines, detection of atypical curves, functional exploratory analysis, functional means, functional standard deviation and functional variance were used. Data modeling was performed by means of functional principal component regression, to decompose a time series of functions into a set of functional principal components and their associated scores, in addition univariate (ARIMA) and multivariate (VAR) time series were used for forecasting coefficients. For the validation of the models, errors were analyzed by means of scale-dependent measures and percentage-based measures, while the precision of the model was calculated using the Theil’s U coefficient and the Diebold-Mariano test. It is concluded that there are fewer errors, when using the regression model of the functional principal component with coefficients predicted by multivariate time series. It is recommended to perform functional data analysis continuously at the ambient temperature variable or other meteorological variables to know its behavior over time.