Reconstruction of Meteorological Records With Pca-Based Analog Ensembles Methods

The Analogue Ensembles (AnEn) method has been used to reconstruct missing data in time series with base on other correlated time series with full data. As the AnEn method benefits from the use of large volumes of data, there is a great interest in improving its efficiency. In this paper, the Principal Component Analysis (PCA) technique is combined with the classical AnEn method and a K-means cluster-based variant, within the context of reconstructing missing meteorological data at a particular station using information from neighbouring stations. This combination allows to reduce the dimension of the number of predictor time series, while ensuring better accuracy and higher computational performance than the AnEn methods: it reduces prediction errors by up to 30% and achieves a computational speedup of up to 2x.