The North Pacific marine ecosystems are primary sources of ecosystem services (e.g., fishing, shipping, and recreation) for its surrounding countries including Canada, U.S., China, Russia, Korea, and Japan. Coastlines of the North Pacific are seriously damaged by extreme sea level rise (Nicholls et al., 2007). In particular, the coastal zones are immediately affected by submergence and increased flooding of coastal land, as well as saltwater intrusion of surface waters (Nicholls and Cazenave, 2010).

In addition to the rising sea level trend associated with the global warming, extreme sea level events occur in association with natural climate variability such as the Pacific Decadal Oscillation (PDO), the Interdecadal Pacific Oscillation (IPO), the North Pacific Gyre Oscillation (NPGO), and the El Niño-Southern Oscillation (ENSO) in the North Pacific (Mantua et al., 1997; Zhang et al., 1997; McGowan et al., 1998; Lombard et al., 2005; Di Lorenzo et al., 2008; Hamlington et al., 2019; Han et al., 2019). To address relatively short-term risks, stakeholders desire a forecast of monthly/seasonal rising or falling sea levels caused by those climate variabilities. Hinkel et al. (2019) analyzed user needs for sea level rise information, and how they are able to be met given the state-of-the-art of sea level forecast science. Jacox et al. (2020) reviewed statistical and dynamical marine ecosystem forecasting methods and highlighted examples of their application along U.S. coastlines for seasonal-to-interannual prediction. Payne et al. (2017) also reviewed the state of the art marine ecological forecasts and suggested forecasts ranging from seasonal to decadal time scales are now a reality. Tommasi et al. (2017a) evaluated the multi-annual SST predictions over Large Marine Ecosystems (LMEs), a coastal scale relevant to managed fisheries stocks. Tommasi et al. (2017b) also highlighted advances in seasonal to decadal prediction of managing living marine resources in a dynamic environment. Those previous studies provide information relevant for supporting coastal adaptation decision making.

Although skillful predictions of SST have already proven useful for a number of marine resource applications (e.g., Hobday et al., 2014; Stock et al., 2015), further studies about sea level anomaly are necessary. Rebert et al. (1985) showed that the oceanic Kelvin and Rossby waves have a direct relation between thermocline depth and sea level, while they have only an indirect relation to SST. These ocean dynamics are responsible for the relatively high skill of sea level prediction relative to SST prediction (Miles et al., 2014). Zainuddin et al. (2017) found that SST was an important variable for detecting hotspots of skipjack tuna distribution, as they are sensitive to the changes in temperature. Sea level anomaly is related to the changes in the depth of the thermocline and mesoscale variability. They combined these variables to improve detection of potential pelagic hotspots for skipjack tuna.

To expand prediction of large-scale sea level anomalies into coastal areas and to further the understanding of its potential predictability, it is necessary to evaluate the lead-time and locations in which a dynamical, physics-based prediction system performs well. It might allow coastal communities to better adapt for the impacts of severe flooding and erosion driven by high sea levels.

Although decadal climate variation is more predictable than previously thought, it is still challenging (Meehl et al., 2014; Smith et al., 2019). Here, our focus is on seasonal-to-interannual prediction. Generally speaking, the most important potential source of seasonal-to-interannual predictability is often from ENSO events, which develop via air-sea coupled feedback. Therefore, application of an ocean-atmosphere coupled general circulation model (GCM) is naturally a proven approach to overcome shortcomings of stand-alone atmospheric/oceanic models. Miles et al. (2014) initially attempted to apply a coupled GCM to predict seasonal sea level anomalies, and assessed the skill globally for up to 7 months in advance. McIntosh et al. (2015) showed the prediction skill by dynamical GCMs is better relative to statistical approaches for coastal sea level. Polkova et al. (2015) used the decadal prediction system and found predictability in the subtropics. Roberts et al. (2016) assessed the predictability of large-scale dynamic sea level anomalies up to 15 months using a climate model and found that prediction of seasonal-to-interannual sea level variability in the extratropics will be governed by the predictability of surface wind stress and modes of atmospheric variability.

This study is a follow-up study of those pioneering studies. Here, we have revisited the predictability of sea level up to about 2 years in advance by analyzing results of a coupled ocean–atmosphere general circulation model “SINTEX-F.” Such a long lead time retrospective forecast is beyond most current operational capabilities, and hence a skill assessment of the model results is conducted here as a first attempt. We believe that the obtained result is useful to attempt systematic and synergistic prediction of marine ecosystem responses to regional and global climate variations.

Sea level anomalies in the region of (30°–40°N, 180°–150°W) may be related to the PDO. The PDO is now interpreted as an empirical mode, which includes teleconnection from ENSO and stochastic atmospheric/oceanic fluctuations (Schneider and Cornuelle, 2005; Newman et al., 2016). Decadal or longer timescale signals appear also to be important for the 2000 event. About 50% of the sea level anomaly averaged over 30°–40°N, 180°–150°W during March-April-May 2000 is due to the decadal signal in the prediction (Figure 2). Actually, 2000 is an extreme year for decadal variations in the IPO index and also basin-wide sea level in the Pacific (see Figure 3 in Lyu et al., 2017), which is closely related to the ENSO-like low-frequency variability. Although low-frequency sea level variations with periods longer than interannual time scales are interesting, it might be difficult to clearly separate the interannual variations from the decadal and longer timescale variations mainly due to limitation of the sample size and the lead time of the reforecast experiments. Although the focus of this study is on seasonal-to-interannual scale prediction, in the future, we may need to develop skillful seamless prediction abilities from seasonal-to-decadal (S2D) timescale.

Tropical and North Pacific processes are interlinked, which means that the North Pacific processes might also contribute to ENSO predictability (e.g., Ogata et al., 2019). The build-up of subsurface ocean heat content in the tropical western Pacific as well as the northeastern subtropical Pacific is identified as ENSO precursors (Capotondi et al., 2015; Yu and Paek, 2015). Chikamoto et al. (2015) also showed that the low-frequency trans-basin tropical climate variations between the Pacific and the other two adjacent ocean basins can be predicted up to 3 years ahead. Further studies are necessary to estimate the role of the inter-basin coupling on multi-year predictability of the tropical and North Pacific using partial assimilation reforecast experiments.

Since our results are based on a single-model system, we need to check them by a multi-model ensemble system (e.g., Kirtman et al., 2014; Tompkins et al., 2017; Widlansky et al., 2017).

A noble path to systematic and synergistic prediction of marine ecosystem variations may be to develop an earth system model, to incorporate biogeochemical processes into a climate model to represent the interacting physical, chemical, and biological processes. It can provide outlooks for marine-resource–relevant changes beyond physical variables. Along this line, Park et al. (2019) showed that an earth system model can skillfully predict seasonal to multiannual chlorophyll fluctuations in many regions.

Although predictability of open-ocean anomalies was the focus in this study, its connection to coastal sea level is also important. However, it is still challenging to resolve the complicated topography near the coastal regions for the resolution used in current climate models. Therefore, downscaling techniques are helpful to capture the open-ocean and coastal region connections (e.g., Jacox et al., 2020) in a manner similar to successful examples for atmospheric downscaling (e.g., Ratnam et al., 2016, 2017).

Enhancement of the relatively coarse ocean model grid will help to resolve more accurately some islands and narrow upwelling regions. Higher resolution in the atmospheric model may also help to improve winds that are an important component of the ENSO teleconnection. In addition, the accuracy should be improved by better initial conditions by explicit use of altimeter data and in situ subsurface ocean temperature and salinity observation from the expendable bathythermographs (XBTs), mooring buoys, sea stations, Argo floats, etc. Increasing the ensemble size may be beneficial for improving prediction of the extratropics, where the signal-to-noise ratio is relatively low. Actually, we have been developing the new version of the SINTEX-F prediction system called as SINTEX-F2 based on a high-resolution model by updating the initialization scheme and increasing the ensemble size (Doi et al., 2016, 2017, 2019, 2020a; Morioka et al., 2019). However, because the computational cost is expensive, the SINTEX-F2 mainly targets for prediction up to 11-month lead time at this stage. We are now extending the lead time up to 23 months because its benefit was shown in this paper.

We assessed the prediction skill of sea level anomaly up to 23 months in advance by the SINTEX-F system and found a skillful prediction region in the North Pacific (30–40 N, 180–150 W) at about 2-year lead. The successful prediction of the long-lasting ENSO/ENSO-Modoki in the tropical Pacific seems to contribute to that sea level predictability. The result may be useful to attempt systematic and synergistic prediction of marine ecosystem responses to regional and global climate variations.

The datasets generated for this study can be found in the online repositories. The names of the repository/repositories and accession number(s) can be found below: https://fbox.jamstec.go.jp/public/3vnQgARS2EDAjpEBzL10R3qP8jcy2W-xWLfXy5mE7u8Z/m/3x5AD2lA.

TD performed the seasonal prediction experiments, and analyzed the observation data and model prediction outputs. All authors contributed to designing the research, interpreting results, and writing the manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.