Long-term sea-level variability along the coast of Japan during the 20th century revealed by a 1/10$^{\circ }$ OGCM

Abstract

We explore long-term sea-level variability along the coast of Japan during the 20th century, using a 1/10$^{\circ }$ ocean general circulation model driven by two 20th century atmospheric reanalysis data. The modeled sea level anomalies along the coast of Japan (JPN-SLAs) show a consistent upward trend throughout the 20th century, which is comparable to global-mean sea-level rise, whereas no trend is obvious for the observed JPN-SLAs based on tide gauge data carefully selected by the Japan Meteorological Agency (JMA). We point out that the major difference between the model results and the tide gauge data may be due to the vertical land movements (VLMs) at the tide gauge stations, despite the JMA’s assumption that the VLMs are relatively small there. If this is correct, the estimates from our model combined with the barystatic component by a recent study would yield a linear trend of 1.79 [0.89$\sim$2.28] mm $\hbox {yr}^{-1}$ for JPN-SLAs without VLMs from 1900 to 2010, which is close to the global average SLAs estimated in recent studies. The empirical orthogonal function (EOF) analysis shows that the first mode of the modeled JPN-SLAs is almost spatially uniform with a peak in the 1950s. The peak is created by coastal trapped waves (CTWs), which are excited when positive sea level anomalies produced by change in wind stress, propagate westward as baroclinic Rossby waves and reach Japan. From idealized experiments, we find that the first EOF mode is well reproduced by the CTWs excited east of Hokkaido.

Appendix: Trend estimation

We compute noise parameters and the resulting trend and standard deviation ($\sigma$) using the Hector software (Bos et al. 2013) for each annual-mean time series of the SLAs components in exps 20CRv3 and ERA20C. The uncertainty of the trend is then defined as 1.64 times the standard deviation (corresponding to the 90% confidence interval). We assume that the spectrum of all time series can be approximated by a generalized Gauss Markov spectrum following Frederikse et al. (2020). Trend is estimated as the mean of these two experiments. We count for their trend errors by adding the half of difference in trends and uncertainties calculated in the Hector software in quadrature. The trend of the barystatic component estimated by Frederikse et al. (2020) and its uncertainty is calculated using their python scripts. Total trend is the sum of the steric, dynamic, and barystatic trends. The uncertainties are added in quadrature.

We also estimate the linear trend of the tide gauge observation using the Hector software for the four tide gauges stations assuming the AR(1) (first-order autoregressive) noise model. We count for their trend errors by adding the standard deviation of the estimated trends for the four stations and their uncertainties calculated in Hector software in quadrature.