Evaluation of spectral methods for high-frequency multiannual time series in coastal transitional waters: advantages of combined analyses

Abstract

High-frequency monitoring is currently a major component in the management and research of the coastal system responses to ongoing global changes. This monitoring is essential in tidal systems to address the multiscale variability of physico-chemical parameters. The analysis of the resulting multiscale, nonlinear, non-stationary and noisy time series requires adequate techniques; however, to date, there are no standardized methods. Spectral methods might be useful tools to reveal the main variability time scales, and thus their associated forcings. The most widely used methods in coastal systems are Lomb-Scargle Periodogram (LSP), Singular Spectral Analysis (SSA), Continuous Wavelet Transform (CWT), and Empirical Mode Decomposition (EMD), but their relevance for high-frequency, long-term records is still largely unexplored. In this article, these spectral methods are tested and compared using a high-frequency 10-yr turbidity dataset in the Gironde estuary. Advantages and limitations of each method are evaluated on the basis of five criteria: (1) efficiency for incomplete time series, (2) appropriateness for time-varying analysis, (3) ability to recognize processes without complementary environmental variables, (4) capacity to calculate the relative importance of forcings, and (5) capacity to identify long-term trends. SSA is the only analysis method to satisfy all the criteria, even with 70% missing data. Combining methods is also a promising strategy; i.e., SSA + LSP for better recognition of processes; CWT + SSA and EMD + CWT for short-term (seasonal) and long-term (>1 yr) analysis, respectively. The purpose of this methodological framework is to serve as a reference for future post-processing of data from monitoring programs in coastal waters.