An EMD and PCA hybrid approach for separating noise from signal, and signal in climate change detection

Abstract

One of the important issues in climate change detection is the selection of climate models for the background noise. The background noise is generally chosen in a somewhat subjective manner. In the current study, we propose an approach of detecting climate change signal in order to mitigate the effects of background noise and to improve climate change detection ability. At first, the high-frequency components of three climate datasets (climate signal, observation, background noise) induced from the random noise process are extracted from empirical mode decomposition (EMD) analysis. Then, statistical detection techniques are applied to the datasets from which the high-frequency random components are excluded. The proposed approach is tested with synthetically generated data and with a real-world case study represented by global surface temperature anomaly (GSTA) data. The case study reveals that each component of the observed GSTA data from EMD contains the information related to external and internal forcings such as solar activity and oceanic circulation. Among these components, the statistically significant low-frequency components are employed in climate change detection. Compared to one of the existing approaches, some improvements in the slope coefficient estimates and the signal-to-noise ratio (SNR) are observed in the synthetic application of the proposed model. The application to the GSTA data shows higher SNR in the proposed approach than in the existing approach. Copyright (C) 2011 Royal Meteorological Society