Mémoire

Abstract

The ENSO (El Ni˜no Southern Oscillation) phenomenon is a complex climate process governed by various cycles, that are yet only partially understood. To better understand these cycles and their impact on the global climate, we employed a modal extraction method based on Continuous Wavelet Transform (CWT). Most particularly, we used the Wavelet-Induced Mode Extraction (WIME) algorithm, initially developed to decompose signals into cosinusoidal components varying in frequency, amplitude, and phase. In this thesis, the WIME algorithm was adapted to handle real-world signals by incorporating adjustable parameters to manage noise, peak concavity, and frequency variation range. ENSO was analyzed using two indices: the Southern Oscillation Index (SOI) for the atmospheric component and the Oceanic Ni˜no Index (ONI) for the oceanic component. The method was validated by reconstructing signals for temperatures, ONI, and SOI, with correlation coefficients of 0.86, 0.94, and 0.76, respectively, demonstrating the robustness of the approach. The identified periods are consistent with each other and with existing literature. Moreover, the indicators remain strong when considering El Ni˜no and La Ni˜na events exclusively. Furthermore, short- and medium-term predictions (2-3 years) were made by truncating the time series and testing the method’s ability to forecast subsequent data. This validation procedure shows a correlation of 0.92 with the ONI signal and satisfactory difference indicators. Understanding ENSO is vital due to its significant impacts on global weather patterns, economies, and ecosystems. This study introduces a novel approach for analyzing and forecasting the ENSO phenomenon, offering potential improvements in prediction accuracy. Additionally, this method has potential applications to other climate phenomena, such as the Arctic Oscillation and the North Atlantic Oscillation, as well as non-climatic events like seismic activities.