Por Clara Cordeiro (CEAUL & Departamento Matemática, FCT - Universidade do Algarve).
The search for the future is an appealing challenge in time series. The diversity of forecasting methodologies is inevitable and is a topic still in expansion. Exponential smoothing methods (EXPOS) are the launch platform for modelling and forecasting and through bootstrap, several sample paths for a time series are simulated. Since its beginning, Boot.EXPOS algorithm has shown promising results when combining exponential smoothing and bootstrap (Cordeiro & Neves, 2009, 2010, 2013, 2014).
Satellite Remote Sensing (RS) data are widely used for the environmental monitoring of the Earth. The study, analysis and interpretation of satellite-derived data are based on statistical techniques for dependent data. The decomposition of RS time series into seasonal, trend and irregular components is commonly used in this research field. In Cristina et al. (2016), st.fit() selects the best Seasonal-Trend decomposition by Loess (STL) for each combination of the seasonal and trend smoothing parameters, based on an error measure. In the case of an uncorrelated irregular component, the forecast of the STL will rely on the forecast of trend and seasonal components obtained through the Boot.EXPOS. This work presents an overview of the Boot.EXPOS and its new challenges.
(This is a joint work with Professor Manuela Neves - CEAUL and Instituto Superior de Agronomia, Universidade de Lisboa and Sónia Cristina - CIMA, Universidade Algarve and Sagremarisco Lda).