Strengthening our knowledge on regional sea level rise through proxy data and machine learning.

Abstract:

Successful prediction of sea level variability and change still fail, to a large extent, on the short-term (1 to 10 years time scale), especially on a regional scale, where there can be large deviations from the global mean change due to climate variability. This study examines the validity of some ocean parameters (e.g., temperature-based estimates over open-ocean regions) to model regional coastal sea level variability from interannual to decadal timescales. It considers a non-linear machine learning regression method between the relevant observed proxy-variables and sea level data. It has also been possible to obtain significance levels across the different depth layers in the ocean (down to 2000 m) to assess their relative importance in capturing sea level variability. Our method can explain up to 33% more of the variance in sea level than commonly used linear regression procedures. The best fit occurs in the case of the coastal regions of the Indo-Pacific and East Pacific Ocean (with correlation values close to 1), followed by the Atlantic Ocean (correlations always higher than 0.76) on time scales of several years. The relationships that have been established acquires a crucial role in the understanding of internal variability and the development of regional sea level predictions for the short time frame. In fact, it has been observed that, in some regions, the proxy can even anticipate changes in sea level. This model could also potentially be used for gap-filling or data reconstruction of the regional mean sea level anomalies time series.