Non-linear Rainfall Trend Extraction Using Hybrid Empirical Mode Decomposition And Singular Spectrum Analysis

Identification and precise quantification of rainfall trend are crucial for researchers to understand the variations in rainfall over a longer period of time. On a global scale, climate change is influencing the intensity and frequency of rainfall, leading to extreme rainfall events. The expeditious and systematic consideration of rainfall changes is important in this context. In the given study, EMD-SSA, a hybrid data-adaptive multivariate multiscale method based on empirical mode decomposition (EMD) and singular spectrum analysis (SSA), is proposed to extract the non-linear trend present in the rainfall series. At the initial stage, EMD is applied to decompose the observed rainfall series into several intrinsic mode functions (IMFs) of different frequencies depicting trends and oscillatory patterns. Periodogram analysis of each IMF is performed by Lomb-Scargle spectral analysis to identify the important periodic signals and their period. These periods are considered suitable input (embedding dimension) to the SSA. The rainfall data is collected on a daily scale for the region of Mumbai, India, from NASA’s Prediction of Worldwide Energy Resource (POWER) archive from 1981–2020. The non-linear trend present in the rainfall is estimated by the EMD, SSA, and EMD-SSA methods. From the analysis, an increasing rainfall trend is observed in the Mumbai city, indicating more rainfall events in the future. Finally, the study suggests that a hybrid EMD-SSA is better than standalone EMD and SSA approach. In the future, the proposed EMD-SSA can also be applied to understand the variabilities in rainfall pattern with respect to the climate indices.