Testing Methods of Pattern Extraction for Climate Data Using Synthetic Modes

Abstract In this paper we develop a method to quantify the accuracy of different pattern extraction techniques for additive space–time modes often assumed be present in climate data. It has previously been shown that standard technique principal component analysis (PCA; also known as empirical orthogonal functions) may extract patterns are not physically meaningful. Here analyze two modern methods, namely dynamical mode decomposition (DMD) and slow feature (SFA), comparison with PCA. We Monte Carlo generate synthetic mimic properties described literature. The datasets composed these generated do satisfy assumptions any presented. find both alternative methods significantly outperform PCA extracting local global These had higher mean across 60 out mixed climates, SFA slightly outperforming DMD. show majority simple cases extracts better than random guess. Finally, when applied real data more coherent less noisy warming signal, well an El Niño signal clearer spectral peak time series, plausible spatial pattern.