<p>Regional sea level rise (SLR) analysis requires robust methods capable of capturing temporal variations beyond simple linear trends. The Korean Peninsula, surrounded by three distinct seas with diverse oceanographic characteristics, provides an ideal testbed for evaluating different methodological approaches to SLR trend estimation. This study compares three analytical methods—Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), Ensemble Empirical Mode Decomposition (EEMD), and linear regression—using 61 years of tide gauge data (1960–2020; coverage varies by station) from 21 stations around the Korean Peninsula. We investigate how methodological choice affects SLR rate estimates and temporal pattern detection, with particular focus on CEEMDAN’s potential advantages in handling high-quality preprocessed data. The three methods produce systematically different results: linear regression generally yields the highest estimates, CEEMDAN provides intermediate values, and EEMD produces the lowest rates. Pairwise method differences reach 2.22 mm/year at Seogwipo, while the median (max-min) spread across methods is 0.69 mm/year (mean 0.82). Depending on definition, relative differences reach 126% (CEEMDAN vs EEMD at Seogwipo) or 211% when expressed as (max-min)/min across the three methods. CEEMDAN decomposition reveals time-varying SLR patterns with distinct acceleration periods at multiple stations (e.g., Ulleungdo and Jeju), though the physical significance of these patterns requires further validation. We introduce amplitude-based spatial clustering as a consistency check for decomposed components and demonstrate parameter optimization strategies for applying CEEMDAN to preprocessed oceanographic data. Our comparative analysis highlights substantial methodological uncertainty in regional SLR estimation, even within a relatively small study area (<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(5^\circ\)</EquationSource> </InlineEquation> latitude <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(\times\)</EquationSource> </InlineEquation> <InlineEquation ID="IEq3"> <EquationSource Format="TEX">\(5^\circ\)</EquationSource> </InlineEquation> longitude). While CEEMDAN shows promise for capturing temporal variations that linear methods cannot detect, distinguishing genuine geophysical signals from methodological artifacts remains challenging. These findings emphasize the need for methodological standardization in regional SLR studies and provide a framework for evaluating decomposition-based approaches in coastal time series analysis. The results contribute to ongoing efforts to reduce uncertainty in localized SLR assessments critical for coastal adaptation planning.</p>

Comparative analysis of signal decomposition methods for regional sea level trend estimation: a case study of the Korean peninsula

期刊论文
出版模式：

开放获取
发表日期：

2026年1月24日

查看全文

Young Jin Kim,
Hyuk Kang,
Hark-Soo Song,
Jong-Ho Kim,
Okyu Kwon

摘要

Regional sea level rise (SLR) analysis requires robust methods capable of capturing temporal variations beyond simple linear trends. The Korean Peninsula, surrounded by three distinct seas with diverse oceanographic characteristics, provides an ideal testbed for evaluating different methodological approaches to SLR trend estimation. This study compares three analytical methods—Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), Ensemble Empirical Mode Decomposition (EEMD), and linear regression—using 61 years of tide gauge data (1960–2020; coverage varies by station) from 21 stations around the Korean Peninsula. We investigate how methodological choice affects SLR rate estimates and temporal pattern detection, with particular focus on CEEMDAN’s potential advantages in handling high-quality preprocessed data. The three methods produce systematically different results: linear regression generally yields the highest estimates, CEEMDAN provides intermediate values, and EEMD produces the lowest rates. Pairwise method differences reach 2.22 mm/year at Seogwipo, while the median (max-min) spread across methods is 0.69 mm/year (mean 0.82). Depending on definition, relative differences reach 126% (CEEMDAN vs EEMD at Seogwipo) or 211% when expressed as (max-min)/min across the three methods. CEEMDAN decomposition reveals time-varying SLR patterns with distinct acceleration periods at multiple stations (e.g., Ulleungdo and Jeju), though the physical significance of these patterns requires further validation. We introduce amplitude-based spatial clustering as a consistency check for decomposed components and demonstrate parameter optimization strategies for applying CEEMDAN to preprocessed oceanographic data. Our comparative analysis highlights substantial methodological uncertainty in regional SLR estimation, even within a relatively small study area ( \(5^\circ\) latitude \(\times\) \(5^\circ\) longitude). While CEEMDAN shows promise for capturing temporal variations that linear methods cannot detect, distinguishing genuine geophysical signals from methodological artifacts remains challenging. These findings emphasize the need for methodological standardization in regional SLR studies and provide a framework for evaluating decomposition-based approaches in coastal time series analysis. The results contribute to ongoing efforts to reduce uncertainty in localized SLR assessments critical for coastal adaptation planning.