1. Introduction
Global sea level rise over the past century, a major indicator of anthropogenic climate change, is projected to continue in the future (Church et al. 2013a). However, because of various physical processes driven by surface buoyancy and momentum fluxes, sea level rise is nonuniform around the globe (e.g., Zhang et al. 2014; Meyssignac et al. 2017). Sea level changes affect densely populated coastal regions, thus understanding future sea level rise is of great socioeconomic interest.
For decades, global climate models (GCMs) have provided scientific insights about large-scale climate change and variability, including sea level rise (e.g., Church et al. 2013b; Slangen et al. 2017), with increasing complexity and advanced algorithms (e.g., Rasp et al. 2018; Horowitz et al. 2020). However, GCMs, including those in the Climate Model Intercomparison Project phase 5 (CMIP5; Taylor et al. 2012) and phase 6 (CMIP6; Eyring et al. 2016), are sometimes not adequate for addressing regional climate issues, because the coarse resolution (typically 100 km in ocean components; e.g., Grose et al. 2020), mean state biases (e.g., Lyu et al. 2020), and reduced vertical layers in shallow regions (e.g., Seo et al. 2014) are common barriers for direct regional applications. Although the High Resolution Model Intercomparison Project (HighResMIP; Haarsma et al. 2016) partly fills this gap, the increasing computational cost means it is not practical for high-resolution GCMs to carry out perturbation experiments for specific regional issues. Moreover, the vertical resolution improvement is not always adequate in HighResMIP to determine the structure in shallow region. Consequently, in parallel with the development of high-resolution global models, downscaling methods have also been developed.
Downscaling methods can be categorized into two broad groups: statistical downscaling and dynamical downscaling (e.g., Hewitson and Crane 1996; Fowler et al. 2007; Pielke and Wilby 2012). Statistical downscaling refers to the methodology of utilizing statistical relationship between large-scale climate effects and observed local climate responses to transform the GCM output to regional climate information (e.g., Chen et al. 2010). Since the success of statistical downscaling relies heavily on long-term, high-quality observations, which are often unavailable in oceanography, here we employ dynamical downscaling using a high-resolution model to dynamically translate the large-scale climate processes to finer regional information (e.g., Meier 2006; Ådlandsvik 2008; Chamberlain et al. 2012; Sun et al. 2012; Liu et al. 2012, 2013; Li et al. 2014; Seo et al. 2014; Liu et al. 2015; Oliver et al. 2015; Liu et al. 2016; Zhang et al. 2017; Toste et al. 2018; Hermans et al. 2020; Shin and Alexander 2020). Specifically, for the applications of sea level projection, Liu et al. (2016) compared the future sea level changes in the western North Pacific using the Regional Ocean Modeling System (ROMS) driven by three GCMs respectively; Zhang et al. (2017) carried out dynamical downscaling for sea level projections based on a near-global high-resolution (1/10°) ocean model driven by an ensemble average of 17 CMIP5 models; Toste et al. (2018) discussed projected sea level changes on the east coast of Brazil using the ROMS driven by one single GCM; and Hermans et al. (2020) investigated the dynamic sea level (DSL) projections for the northwestern European shelf downscaled from two CMIP5 models respectively. However, previous downscaling studies have seldom differentiated the sea level change resulting from oceanic versus atmospheric changes.
Here, we apply dynamical downscaling to understand changes in the East China marginal seas and the South China Sea (SCS). The East China marginal seas are the Bohai Sea, the Yellow Sea, and the East China Sea from north to south, with average depths of 18, 44, and 350 m, respectively (Fig. 1). In contrast, the semi-enclosed SCS has an average depth of 1212 m, resulting in more complex dynamical processes. As these marginal seas are highly influenced by the East Asian monsoon, ocean variables exhibit obvious seasonal signals (e.g., Su 2004; Isobe 2008). The strength and location of the Kuroshio also affect these marginal seas significantly (e.g., Xue et al. 2004; Guo et al. 2006).
The model domain of all the experiments carried out in this paper. The bathymetry is intentionally shown with a nonlinear color bar to provide more information for shallow regions. The red dots denote the locations of tide gauges along the China mainland coast labeled with the tide gauge numbers.
Citation: Journal of Climate 34, 17; 10.1175/JCLI-D-20-0796.1
The rate of relative sea level rise along the China coast (3.4 mm yr−1 on average from 1980 to 2019 according to the Ministry of Natural Resources of the People’s Republic of China) is higher than the global mean sea level (GMSL) rise of 2.7 ± 0.3 mm yr−1 during 1980–2019 (Church and White 2011) or 2.5 ± 0.2 mm yr−1 during 1980–2018 (Frederikse et al. 2020). During the altimetry period (1993–2020), the mean rate of geocentric sea level rise of the marginal seas near China and the neighboring ocean (region in Fig. 1) is 3.4 ± 0.5 mm yr−1, higher than the concurrent GMSL rise of 3.1 ± 0.4 mm yr−1 without glacial isostatic adjustment correction (Ablain et al. 2019) based on the altimetry product from the Archiving, Validation and Interpretation of Satellite Oceanographic (AVISO). In addition to sea level rise, sea levels in the marginal seas near China are also modulated by large-scale climate modes, such as El Niño–Southern Oscillation (ENSO; Rong et al. 2007; Chang et al. 2008; Liu et al. 2010), the Pacific decadal oscillation (Han and Huang 2008), and the North Pacific Gyre Oscillation (Moon and Song 2017). However, there are limited studies investigating future sea level projections for this region (Qu et al. 2019).
In this paper, we aim to apply a dynamical downscaling technique to derive high-resolution mean DSL changes in the marginal seas near China by the end of the twenty-first century, and examine the contributions from changes in local atmosphere versus remote Pacific basin. The DSL in this study is defined as the local height of the sea surface above the regional mean sea level (RMSL) over the model domain, which is closely related to ocean density and circulation (e.g., Zhang et al. 2014), like the way DSL is commonly derived in CMIP models (i.e., regional deviations from the global mean; e.g., Yin 2012). The remainder of this manuscript is organized as follows. In section 2, we introduce the regional ocean model configuration and dynamical downscaling methodology. Next, the sea level simulations for the historical period are evaluated against available observations and reanalyses at different spatiotemporal scales (section 3). Then the dynamical downscaling results are present in section 4. In section 5, the experiment demonstrates that the downscaled results driven by the ensemble mean forcing are almost identical with the ensemble mean of individually downscaled cases. Two perturbation experiments are implemented to separate the contributions to sea level changes from the open ocean and the local atmosphere (section 6). Finally, we summarize this study and discuss the innovations and limitations of this dynamical downscaling approach in section 7.
2. Model and method
All the model simulations in this study are based on the ROMS with the model design shown in Table 1. The ROMS, a free-surface, primitive equation, terrain-following sigma vertical coordinate ocean model (Shchepetkin and McWilliams 2003, 2005), has been widely applied for modeling coastal regions (e.g., Ådlandsvik 2008; Liu et al. 2016). In this study, the model domain covers the Bohai Sea, the Yellow Sea, the East China Sea, most of the SCS, and a part of the west Pacific (99.2°–140.3°E, 1.5°–41.3°N; Fig. 1). The horizontal resolution is 1/12° (~8 km) with 30 (s coordinate) vertical levels. The bathymetry is obtained from the U.S. Navy Digital Bathymetric Data Base Variable (DBDBV) 5-min resolution dataset, and a linear programming procedure (Sikirić et al. 2009) is used to smooth the bathymetry to avoid horizontal pressure gradient errors.
The initial condition, forcing, and simulation period for each experiment; ΔF is the monthly climatology climate change signal derived from individual global climate models (GCMs) [Eq. (1)], while
a. Model configuration of the Historical experiment
1) Open boundary conditions
The Bluelink Reanalysis (BRAN) version 2016, an eddy-resolving near-global ocean reanalysis (Oke et al. 2013), provides open boundary conditions for the Historical experiment (Table 1) every 5 days. Our regional model ROMS adopts the Boussinesq approximation, which conserves ocean volume rather than mass. For all experiments, the net water volume flux (i.e., sum of volume transports across all open boundaries, and river runoff) is balanced at each model time step to keep the RMSL over the model domain zero. Through this approach, we focus on the sea level deviations from the RMSL over the model domain, which can be derived and added back in postprocessing (section 5b).
For the open boundary conditions, clamped conditions are used for the depth-integrated volume flux and tracers (temperature and salinity). Radiation conditions are applied to depth-dependent momentum (Raymond and Kuo 1984). Sea level is not provided on the boundaries, since it can be calculated by the model based on other variables. As a common practice in regional ocean modeling, the sponge layer within 2° of open boundaries is used to stabilize the model by damping unphysical signals near open boundaries and to nudge temperature and salinity to the prescribed values. Moreover, viscosity and diffusivity coefficients are linearly increased from the interior sponger layer edge to the open boundaries by multiplying a factor from 1 to 10 and the nudging time scale is decreased from 200 days at the interior sponge layer edge to 5 days on the boundaries. The monthly climatological sea surface salinity (SSS) obtained from BRAN, which contains data assimilation of observed salinity from a range of sources (Oke et al. 2013), is used to constrain long-term drift in the model with a relaxation time scale of 30 days. Although the tide is an important driver of coastal sea level variations, it contributes little to the long-term mean sea level spatial pattern so the tide is omitted in our study as done by other modeling studies focusing on mean sea level spatial pattern changes (e.g., Liu et al. 2016).
2) Surface forcing and river runoff
For the surface forcing of the Historical experiment, the ERA-Interim reanalysis (ERAI herein; Dee et al. 2011), also adopted in BRAN, is used to force the regional model every 6 h with horizontal resolution of 0.75°. The surface forcing variables include 2-m air temperature, mean sea level air pressure, relative humidity, precipitation, 10-m wind, net longwave radiation, and net shortwave radiation. Bulk formulas are adopted to calculate the surface fluxes (Fairall et al. 2003).
The monthly climatological river discharge from the Yangtze River is included (Lv et al. 2006), which has the largest annual mean river transport into the model domain (Wohl 2007). The salinity of runoff is set at 5 psu, and the temperature is the same as the surrounding water. Considering that there is no sea ice module in this model configuration, the lowest temperature is set as −1.5°C, which is the typical lowest winter temperature in the model domain (mainly in the Bohai Sea) based on 4-μm nighttime sea surface temperature (SST) product from the Terra Moderate Resolution Imaging Spectroradiometer (MODIS; Werdell et al. 2013).
3) Spinup experiment
To ensure a realistic initial condition for the historical simulation, the model was initialized with temperature, salinity, and velocity fields from the BRAN on 1 January 1994, and then was driven by 6-h ERAI forcing and 5-day averaged open boundary conditions from BRAN. Year 1994 was repeated until the model reached a quasi-equilibrium. Because it is a regional model and the initial temperature and salinity fields and open boundary forcing are extracted from the same ocean reanalysis product, it only takes about 3 years for the model to stabilize. After a 5-yr spinup, the first day ocean state of the sixth year was used as the initial condition for the Historical experiment over 1994–2015, which is the available period for BRAN.
b. Dynamical downscaling
The dynamical downscaling is carried out in the following steps:
- 1) The climate change signals (ΔF) of CMIP5 models were obtained as the differences between the historical monthly climatology (1980–2005) and the future monthly climatology (2079–2100, the same length as the Historical experiment). The climate changes under representative concentration pathways (RCPs) 4.5 and 8.5 are both considered. Specifically, the climate change signals of ocean variables (temperature, salinity, and velocity) and atmosphere variables (2-m air temperature, mean sea level air pressure, relative humidity, precipitation, 10-m wind, net longwave radiation, and net shortwave radiation) are further labeled as ΔFO and ΔFA, respectively.Note that the climatology is not sensitive to the exact base period length as long as it is longer than 20 years. Thus the derived climate change signal ΔF is robust.
2) The climate change signal ΔF is added to the forcing of the Historical experiment to produce forcings for the Future experiments, including the open ocean boundary forcing (with volume transport balanced), surface atmospheric forcing, climatological SSS for relaxation, and the climatological water temperature of river discharge.
3) The first month of ΔFO is added to the initial condition of the Historical experiment to complete a spinup experiment with repeated year 2079 conditions, in the same manner as the historical spinup. It takes about 6–7 years for the model to stabilize. After a 9-yr spinup, the first day ocean state of the 10th year was used as the initial condition for the Future experiments.
4) After spinup, the regional model is integrated for the future period 2079–2100 driven by the merged future forcing derived from step 2.
Rather than just use the long-term mean differences between the future and historical simulations from CMIP5, we use the monthly climatological change ΔF, because the marginal seas near China display strong seasonal cycles mainly driven by seasonal variations of solar radiation and the East Asian monsoon (e.g., Li et al. 2016), as shown in the Historical experiment and observations (section 3 and Figs. S2–S5 in the online supplemental material).
To reduce model physics uncertainty and internal climate variability in individual CMIP5 models, a multimodel ensemble based on available CMIP5 models is often used to provide a more robust assessment of the climate change signal (e.g., Fowler et al. 2007). However, due to computation resource limitation, a subset of CMIP5 ensemble is downscaled here rather than the whole ensemble. With the underlying assumption that good-performance models of historical observations tend to give better future projections (e.g., Tedeschi and Collins 2017), eight CMIP5 models (ACCESS1.0, CanESM2, CMCC-CM, CMCC-CMS, GISS-E2-R, HadGRM2-CC, MRI-CGCM3, and MPI-ESM-MR; Table 2; Flato et al. 2013) are selected. They have better representation of historical long-term mean sea level spatial pattern and interannual variability of sea level in the Pacific (Landerer et al. 2014; Lyu et al. 2016). These eight selected CMIP5 models (CMIP5 ensemble in this study) are first downscaled individually, and the downscaled ensemble mean is compared with the corresponding CMIP5 ensemble mean.
The eight selected GCMs from phase 5 of the Climate Model Intercomparison Project (CMIP5) (Flato et al. 2013).
3. Model evaluation of the historical simulation
To demonstrate that the high-resolution regional model can simulate ocean conditions realistically, the historical simulation is evaluated against available historical oceanic observations and reanalysis products for the period 1994–2015 at different spatiotemporal scales (such as daily to interannual variabilities, small- to large-scale spatial patterns). Only the sea level evaluation is present here, while more evaluations (like ocean temperature and current) can be found in the online supplemental material (Figs. S2–S5).
The sea level simulation is compared with the absolute dynamic topography of the Data Unification and Altimeter Combination System (DUACS DT2014; Pujol et al. 2016) from AVISO, BRAN, and coastal tide gauge records available from the University of Hawaii Sea Level Center (UHSLC; Caldwell et al. 2015) and Ding et al. (2018). By design (section 2), the RMSL over the model domain is zero at every time step, so sea level data from both AVISO and BRAN are adjusted by removing the spatial mean over the model domain before comparing with model sea level simulations.
a. Sea level daily variability
A total of 23 daily records from 20 tide gauges (locations shown in Fig. 1) along the coastline of China from two separate periods (1994–96/97, 2006–07) are used for model evaluation (Fig. 2). The RMSL obtained from BRAN and the inverse barometer effect for each grid box based on ERAI sea level pressure are added to the simulated sea level to make it comparable with the tide gauge observations. For 20 of the 23 records, the temporal correlations between tide gauge records (after de-tiding) and model simulations are above 0.80, and the root-mean-square difference (RMSD) is approximately half of the standard deviation (STD) of the corresponding tide-gauge record (Table 3). The detailed comparisons of the daily time series show that the regional ocean model captures the synoptic sea level fluctuations well which are most obvious in winter (Fig. 2). Based on Ding et al. (2018) and Ding et al. (2019), these synoptic sea level fluctuations along Bohai Sea and Yellow Sea in winter (Fig. 2b) are closely related to winter storm events, coastal-trapped waves propagating along the coast, and adjustment to the strength of the Yellow Sea Warm Current. In contrast, the sea level fluctuations are weaker under relatively milder summer wind. Generally, the seasonal signal combined with synoptic sea level fluctuations during winter contributes to the high correlation of model sea level with tide-gauge observations. The lowest correlation coefficient (0.75) exists at Station 12 (Lusi), located near the estuary of the Yangtze River, and may be possibly induced by the monthly climatological river transport applied in the model rather than real-time monthly river transport.
Comparison between daily sea level time series (m) of tide gauges records after de-tiding (red line) and the nearest grid point from the Historical experiment (blue line) during (a) 1994–97 (or 1994–96) and (b) 2006–07. The three numbers in the parentheses behind the tide gauge number are the correlation coefficient, root-mean-square difference (m), and regression coefficient, respectively.
Citation: Journal of Climate 34, 17; 10.1175/JCLI-D-20-0796.1
The evaluation of daily sea level from the Historical experiment against tide gauges after removing the tide signal during 2006–07 and 1994–97 (or 1994–96). ID indicates the tide gauge number. STD indicates the standard deviation of tide gauge records. Cor and RMSD show the correlation coefficient and root-mean-square deviation between the Historical experiment simulations and tide gauge records. Reg presents the regression coefficient of the Historical experiment results with respect to tide gauge records.
b. Sea level seasonal variability
The amplitude of seasonal variations of sea level from the Historical experiment (Fig. 3c) compares well with that from both AVISO (Fig. 3a) and BRAN (Fig. 3b), as evidenced by the high spatial correlation and low RMSD between AVISO and model (Figs. 3a,c; correlation 0.84 and RMSD 0.04 m), and between BRAN and model (Figs. 3b,c; correlation 0.76 and RMSD 0.05 m). Compared with BRAN, the Historical experiment displays closer seasonal cycle agreement with the altimeter observation in the East China marginal seas and coastal region of the SCS, presenting a more obvious seasonal reversal of coastal currents (Figs. S5c,f) driven by the East Asian monsoon. The model also realistically simulates the sea level seasonal cycle phase, as shown in the maps of the months when sea levels reach maximum and minimum values (Figs. 3f,i). The seasonal phase of sea level shows a north–south dipole (roughly divided by 22°–24°N) that may be related to solar radiation varying with latitude and seasonal reversal of surface wind.
The seasonal amplitude of sea level (m) from (a) the Archiving, Validation and Interpretation of Satellite Oceanographic (AVISO), (b) BRAN, and (c) the Historical experiment, which is obtained from the difference between monthly climatological maximum and minimum values. The maximum month of sea level from (d) AVISO, (e) BRAN, and (f) the Historical experiment, when the sea level reaches the maximum values. The minimum month of sea level from (g) AVISO, (h) BRAN, and (i) the Historical experiment, when the sea level reaches the minimum values.
Citation: Journal of Climate 34, 17; 10.1175/JCLI-D-20-0796.1
c. Sea level interannual variability
To evaluate the interannual variability, the monthly sea level anomalies are obtained from AVISO, BRAN, and the Historical experiment, and smoothed with the 5-month running-mean filter. The spatial distribution of the STD of interannual variability from the Historical experiment compares well with that derived from AVISO and BRAN, indicated by high spatial correlation and low RMSD between AVISO and model (Figs. 4a,c; correlation 0.81 and RMSD 0.02 m), and between BRAN and model (Figs. 4b,c; correlation 0.80 and RMSD 0.02 m). The interannual sea level STD from the model in the East China marginal seas is slightly larger (<0.01 m) than AVISO and BRAN, but smaller (<0.02 m) on the east side of the Kuroshio. A high STD center is found near the Luzon Strait (Fig. 4c), which suggests the model has a stronger Kuroshio intrusion into the SCS than the observation and reanalysis data. The Kuroshio intrusion process and its controlling mechanisms are not well understood (Nan et al. 2015), but are potentially controlled by surface winds (e.g., Wu and Hsin 2012), the interbasin pressure gradient (e.g., Song et al. 2006), potential vorticity (e.g., Xue et al. 2004), and eddy activity (e.g., Lu and Liu 2013). Considering that the stronger intrusion is only limited to a small region immediately west of the Luzon Strait, we did not explore this complex issue further.
Comparison of interannual sea level variability. The standard deviation (STD) (m) of (a) AVISO, (b) BRAN, and (c) the Historical experiment during 1994–2015. The first empirical orthogonal functions based on (d) AVISO, (e) BRAN, (f) the Historical experiment over 1994–2015, and the percentages shown in the spatial patterns are the explained variances. Also shown is (g) a comparison of the first principal components and the 5-month running-mean filtered multivariate El Niño–Southern Oscillation index.
Citation: Journal of Climate 34, 17; 10.1175/JCLI-D-20-0796.1
Empirical orthogonal functions (EOF) analysis is also applied to the interannual sea level variability. The first EOF (EOF1) from the Historical experiment (Fig. 4f) is consistent with that from AVISO (Fig. 4d; correlation 0.95) and BRAN (Fig. 4e; correlation 0.96), with a relatively stronger signal in the East China marginal seas, the Japan Sea, and the southeast corner of the model domain. The principal components (PCs) of AVISO, BRAN, and the Historical experiment correspond well with each other with correlation coefficients larger than 0.95 (Fig. 4g). The high consistency revealed by the EOF analysis indicates that interannual variations of sea level are reliably simulated. The 5-month running-mean filtered multivariate El Niño–Southern Oscillation index (MEI; Wolter and Timlin 2011) is also shown (Fig. 4g) as a reference. The correlation between the first PC (PC1) of AVISO (the Historical experiment) and MEI is 0.80 (0.70), indicating that interannual sea level variability in the model region is strongly modulated by ENSO events (Rong et al. 2007; Chang et al. 2008; Liu et al. 2010; Zhang and Church 2012).
Generally, various evaluations done in this section (and in Figs. S2–S5) indicate that this regional model reproduces the historical ocean states reasonably well, giving us confidence to examine the dynamically downscaled results from our model experiments.
4. Dynamical downscaling results
In this section, we compare the high-resolution downscaling ensemble and the corresponding coarse-resolution CMIP5 ensemble, focusing on SST, DSL, and the upper 50-m mean current under RCP8.5. The results of RCP4.5 are present in the online supplemental material (Fig. S1).
a. SST
For the summer SST simulation, the pattern from the Historical experiment (Fig. 5f) is close to the satellite observations (Fig. 5a) and BRAN (Fig. 5e), with a clear Kuroshio warm path and relatively low temperature (<27°C) in the East China marginal seas, the Japan Sea, and south of the Kuroshio Extension. The spatial correlation coefficient and RMSD between MODIS and the Historical experiment (Figs. 5a and 9f) are 0.96° and 0.67°C, respectively, and between BRAN and the Historical experiment (Figs. 5e,f) are 0.98° and 0.49°C, respectively. In contrast to the model results, BRAN, and satellite observation, the Kuroshio path in the CMIP5 ensemble (Fig. 5b) is not well defined and the SST in the marginal seas displays cold biases. Although the differences between the CMIP5 ensemble and the ROMS simulations in both historical (Figs. 5b,f) and future periods (Figs. 5c,g) are larger than 4.0°C, the differences in projected temperature rise (e.g., 3.0°–5.0°C in the East China marginal seas and the Japan Sea with a smaller uniform rise 2.5°–3.0°C elsewhere) between the CMIP5 ensemble (Fig. 5d) and the downscaling ensemble (Fig. 5h) are small, within 1.0°C. This is not surprising considering the climate change signals extracted from CMIP5 ensemble (reflected in Fig. 5d) are used to drive the regional ocean model. The finer resolution provides regional details, indicating the value of dynamical downscaling. For example, the curved isotherms near the west coast of East China marginal seas (Fig. 5h) are absent in the CMIP5 ensemble (Fig. 5d), which are related to the northward coastal current in summer.
The mean sea surface temperature (SST) (°C) in summer (JJA) from (a) the Terra Moderate Resolution Imaging Spectroradiometer (MODIS) observation (2001–16) and (e) BRAN (1994–2015). The mean SST (°C) in summer (JJA) during (b) the historical period (1980–2005) and (c) the future period (2079–2100) under RCP8.5, and (d) the difference between (b) and (c) based on the CMIP5 ensemble. The mean SST (°C) in summer (JJA) from (f) the Historical experiment with ROMS and (g) the Future experiments ensemble, and (h) the difference between (f) and (g) from the downscaling ensemble. (i)–(p) As in (a)–(h), but in winter (DJF).
Citation: Journal of Climate 34, 17; 10.1175/JCLI-D-20-0796.1
During winter, high spatial correlation and low RMSD are revealed between MODIS and the Historical experiment (Figs. 5i,n; correlation 0.99 and RMSD 0.68°C), and between BRAN and the Historical experiment (Figs. 5m,n; correlation 0.99 and RMSD 0.50°C). The path of the Kuroshio Current, the Yellow Sea Warm Current intrusion in the East China marginal seas, and the temperature gradient normal to the coastline are clear in MODIS (Fig. 5i), BRAN (Fig. 5m), and the Historical experiment (Fig. 5n), but are all nearly absent in the coarse-resolution CMIP5 ensemble (Fig. 5j). Similar to the situation during the summer, although differences in winter SST between CMIP5 ensemble and ROMS simulations in both historical (Figs. 5j,n) and future periods (Figs. 5k,o) are roughly 4.0°C along the coast, the projected SST increase of 3.5°–5.0°C in the East China marginal seas from the downscaling and CMIP5 ensembles (Figs. 5p,i) are not very different (within 1.0°C). There is a smaller warming (2.0°–2.5°C) along the path of the Kuroshio, particularly in the downscaled results. In the East China marginal seas and the Japan Sea, the SST change pattern presented in the high-resolution downscaling ensemble (Fig. 5p) has more significant warming along the coast, especially north of 24°N.
b. DSL
Compared with the mean DSL of CMIP5 ensemble (Fig. 6b), the ROMS DSL gradient along the Kuroshio path in the Historical experiment (Fig. 6f) is sharper and the curvature of isolines is closer to the observations of mean dynamic topography (MDT; Maximenko et al. 2009; Fig. 6a) and BRAN (Fig. 6e), as indicated by the high spatial correlation and low RMSD between the observed MDT and the Historical experiment (Figs. 6a,f; correlation 0.98 and RMSD 0.05 m) and between BRAN and the Historical experiment (Figs. 6e,f; correlation 0.97 and RMSD 0.06 m). The simulated Kuroshio intrusions into the SCS in CMIP5 ensemble (Fig. 6b), BRAN (Fig. 6e), and ROMS (Fig. 6f) are all stronger than the observations (Fig. 6a), indicating the challenging nature of representing the intrusion realistically in models.
(a) The observed mean dynamic topography (1992–2012) and (e) the mean dynamic sea level (DSL) of BRAN (1994–2015). The mean DSL (m) during (b) the historical period (1980–2005) and (c) the future period (2079–2100) under RCP8.5, and (d) the difference between (b) and (c) based on the CMIP5 ensemble. The mean DSL (m) from (f) the Historical experiment with ROMS and (g) the Future experiments ensemble, and (h) the difference between (f) and (g) from the downscaling ensemble. In (d) and (h) the intermodel STD of the CMIP5 ensemble and downscaling ensemble is shown in contours. Stippling indicates intermodel agreement over the regions where the mean DSL change is larger than the intermodel STD.
Citation: Journal of Climate 34, 17; 10.1175/JCLI-D-20-0796.1
For the DSL change, generally the CMIP5 (Fig. 6d) and the downscaling ensembles (Fig. 6h) have similar large-scale patterns, but more regional details are revealed in the high-resolution downscaling ensemble (see the difference map in Fig. S6), and also in the comparisons between individual pairs of downscaling experiment and its parent CMIP5 model (Fig. S7). In particular, there is a low DSL center of −0.15 m in the middle of the SCS and a high center of 0.1 m around Luzon Strait in the dynamical downscaling ensemble results (Fig. 6h), which are nearly absent in the CMIP5 ensemble. However, both centers correspond to large intermodel uncertainty (contours in Fig. 6h), which reflects the intermodel spread among the downscaling ensemble caused by the ΔF derived from different parent GCMs.
The intermodel agreement (indicated by stippling in Figs. 6d,h) is defined as where the magnitude of multimodel mean change is larger than the intermodel STD, which means the changes of different models agree well with each other. This agreement appears only in limited regions (e.g., south of Japan) in the CMIP5 ensemble, and it is generally absent along the coastline (Fig. 6d). After downscaling, the ratio of intermodel agreement area over the total area (model domain) increases from 21.1% to 54.1%, especially in almost all of the East China marginal seas, and along the west coast of SCS (Fig. 6h). The improvement can be partially related to the reduced mean state biases, because Lyu et al. (2020) found there are good connections between mean state biases and future projections in both CMIP5 and CMIP6 ensembles. In this study, the same Historical experiment is used as the benchmark to compare with the Future experiment driven by individual climate model, which provides a mean ocean state without significant biases as demonstrated in the validation analysis (see section 3). This comparison between the CMIP5 and the downscaling ensembles indicates the added values of dynamical downscaling in providing more reliable future projections of DSL, especially along the coastline.
c. Upper ocean current
The upper ocean current averaged over 0–50 m during the historical period displays a more sharply defined Kuroshio and more obvious cyclonic circulation in the SCS in the Historical experiment with ROMS (Fig. 7d) than in the CMIP5 ensemble (Fig. 7a). Under the future climate, the direction of ocean circulation is almost unchanged in both the CMIP5 (Figs. 7a,b) and the downscaling (Figs. 7d,e) ensembles. In contrast, the 0–50-m westward transport from the eastern boundary between 14° and 31°N in the downscaling ensemble and CMIP5 ensemble are enhanced by 1.14 Sv (from 3.24 to 4.38 Sv; Fig. 7f) and 1.23 Sv (from 2.38 to 3.61 Sv; Fig. 7c) respectively (1 Sv ≡ 106 m3 s−1). The most significant changes in the downscaling ensemble are the enhanced Kuroshio intrusion into the SCS and the strengthened cyclonic circulation in the SCS, which mainly result from the remote Pacific influence combined with local atmospheric surface forcing (section 6).
The mean current (m s−1) over upper 50 m during (a) the historical period (1980–2005) and (b) the future period (2079–2100) under RCP8.5, and (c) the difference between (a) and (b) based on the CMIP5 ensemble shown as vectors with the mean current speed overlaid in color shading. The mean current (m s−1) over upper 50 m from (d) the Historical experiment with ROMS and (e) the Future experiments ensemble, and (f) the difference between (d) and (e) shown as vectors with the mean current speed overlaid in color shading.
Citation: Journal of Climate 34, 17; 10.1175/JCLI-D-20-0796.1
The projected changes of DSL and upper ocean current match each other quite well (cf. Figs. 6h and 7f) as they are closely related through the geostrophic balance. For example, the projected westward current strengthening between 16° and 24°N is consistent with the meridional contrast of DSL, with high (low) DSL in the north (south), and the strengthened cyclonic circulation in the SCS corresponds to the low DSL value center.
5. Ensemble-forcing experiment
a. Ensemble-forcing downscaling versus downscaling ensemble
The results reveal that the future projections of DSL (Fig. 8a), summer SST (Fig. 8e), winter SST (Fig. 8i), and mean current over the upper 50 m (Fig. 9a) of this Ensemble-forcing experiment are almost identical with those from the ensemble mean of eight individual downscaling experiments results (Figs. 6h, 5h,p, and 7f), indicated by almost perfect spatial correlations and low RMSDs (0.01 m for DSL, 0.05°C for summer SST, and 0.06°C for winter SST; the difference maps in Fig. S8). This means that ensemble averaging can be done either before downscaling by averaging forcings or after downscaling by averaging individually downscaled ocean states. It is often not practical to downscale many GCMs individually. However, if only ensemble-mean downscaled projections are desired, an alternative and more efficient approach is to average the forcing changes first and then apply the ensemble-mean forcing change to the ocean model just once, as is done with RCP4.5 in our study (Fig. S1).
The mean DSL (m) change based on (a) the Ensemble-forcing experiment, (b) the Atm-only perturbation experiment, and (c) the Ocean-only perturbation experiment. (d) The sum of these two perturbation experiments. (e)–(h),(i)–(l) As in (a)–(d), but for the summer (JJA) SST (°C) and winter (DJF) SST (°C), respectively.
Citation: Journal of Climate 34, 17; 10.1175/JCLI-D-20-0796.1
The mean current (m s−1) change over upper 50 m based on (a) the Ensemble-forcing experiment, (b) the Atm-only perturbation experiment; and (c) the Ocean-only perturbation experiment shown as vectors with the mean current speed change overlaid in color shading. (d) The upper 50-m volume transport of the east boundary from the Historical experiment (red line) and the Ensemble-forcing experiment (blue line) with negative value for flowing in, and (e) their difference (future period minus historical period; green line). The bounds in (d) and (e) indicate the STD of monthly climatology.
Citation: Journal of Climate 34, 17; 10.1175/JCLI-D-20-0796.1
b. Derivation of sterodynamic sea level based on ensemble-forcing experiment
The DSL change (Fig. 8a) expresses the relative changes within the model domain; however, it is not sufficient for local stakeholders and coastal adaptation planning without considering the regional mean change. To get more realistic sea level changes, the sterodynamic sea level (SDSL; Gregory et al. 2019) change is derived through the method introduced in the online supplemental material. SDSL is what the tide gauges would measure if the mass exchange between ocean and other components of the Earth system (e.g., added water into ocean due to glacier or polar ice sheet mass loss) and vertical land motion (e.g., glacial isostatic adjustment; Farrell and Clark 1976; Gregory et al. 2019) are ignored.
SDSL can be decomposed into the mass sea level component (defined as MassSL) and the steric sea level (SSL; e.g., Wu et al. 2017). The MassSL change over the ROMS regional domain (black box in Fig. 10a) is 0.05 m, which implies that more water flows into this regional domain. This is the result of water redistribution to shallow regions under global warming (e.g., Landerer et al. 2007; Yin et al. 2010). The regional mean of SSL change of dynamical downscaling results (Fig. 10c) is 0.23 m, mainly induced by thermal expansion as climate change. The SDSL change in the South China Sea is mainly caused by the SSL change, with a weaker SSL rise in the center of the South China Sea than the surrounding region closely connected with the cyclonic circulation change.
The change of (a) mass sea level (MassSL) over the global domain and (b) sterodynamic sea level (SDSL) within the ROMS domain based on the CMIP5 ensemble. The black box in (a) indicates the ROMS domain. (c) Steric sea level (SSL) change, (d) MassSL change, and (e) SDSL change based on the downscaling ensemble. The black contour in (d) is the 200-m isobath.
Citation: Journal of Climate 34, 17; 10.1175/JCLI-D-20-0796.1
By adding these two spatially uniform terms (0.05 m of regional mean MassSL and 0.23 m of regional mean SSL) to the mean DSL change pattern (Fig. 8a), the projected SDSL change (Fig. 10e) has a regional mean of 0.28 m. The MassSL change pattern of downscaling ensemble (Fig. 10d) indicates that shallow (deep) region gains (losses) mass (indicated by the change of sign across the 200-m isobath in Fig. 10d), similar to the CMIP5 result (Fig. 10a). The regional mean SDSL change of CMIP5 ensemble (Fig. 10b) is also 0.28 m [Eq. (S3)], consistent with that of downscaling ensemble (0.05 + 0.23 m; Fig. 10e), indicating that our dynamical downscaling protocol can provide regional sea level projections by the combination of direct simulation and postprocessing. However, greater spatial details are present in the downscaled SDSL (Fig. 10e; spatial STD 0.07 m) compared with that from CMIP5 (Fig. 10b; spatial STD 0.04 m). The downscaled SDSL change in most of the East China marginal seas (continental shelf of the SCS) is 0.35–0.40 m (0.30–0.35 m), roughly 0.05 m (0.05 m) higher than that in the CMIP5 ensemble. The projected larger sea level rise of 0.05 m should be considered in the future adaptation planning. The downscaled SDSL rises in the East China marginal seas and the shelf of SCS are between 0.3 and 0.4 m, which are mainly caused by the MassSL change (Figs. 10d,e). In contrast, the low center in the SCS and the high center near the Luzon Strait revealed in the downscaled SDSL are mainly induced by the SSL change (Figs. 10c,e). The SDSL component, improved by the downscaling in this study, is the dominant component for regional distribution of total sea level (e.g., Zhang et al. 2017; Frederikse et al. 2020). The other components such as the sea level changes related to land-ice mass changes are relatively spatially uniform, especially in the far fields away from melting sources (like the marginal seas near China) and could be merged with the SDSL change following Zhang et al. (2017).
6. Contributions from open boundary versus atmospheric changes
As an extension of the Ensemble-forcing experiment, two perturbation experiments are designed to further explain the changes in the downscaled ocean states. As shown in Table 1, using the Ensemble-forcing experiment as the control run, an Ocean-only experiment is designed, in which only
The DSL change pattern in the Ocean-only experiment (Fig. 8c) is close to that from the Ensemble-forcing experiment with all forcing applied (Fig. 8a) with spatial correlation of 0.93 and RMSD of 0.03 m, implying that ocean boundary condition changes (i.e., influences from the Pacific) explain most of the changes of DSL in the model domain. In comparison, the surface forcing change results in smaller changes on DSL (Fig. 8b) except in the East China marginal seas where the atmospheric changes play a more important role. The DSL distribution in the Atm-only experiment, having the higher values (>2.5 cm) in the East China marginal seas (Fig. 8b), is consistent with the northward wind change there (Fig. 11a). This conclusion is consistent with that from Li et al. (2016), which reveals that wind-induced redistributions of water plays a significant role in the sea level variability of the East China marginal seas.
The change of (a) mean surface wind (m s−1), (b) DSL (m) with global mean removed, (c) mean ocean current (m s−1) over upper 400 m, and (e) mean ocean current (m s−1) over 400–2000 m derived from CMIP5 ensemble. Also shown are the change of (d) mean ocean current (m s−1) over upper 400 m and (f) mean ocean current (m s−1) over 400–2000 m from downscaling ensemble. Except for (b), the vectors in the other figures show the mean wind (or current) change with the mean speed change overlaid in color shading. The black boxes in (a), (b), (c), and (e) are the ROMS domains.
Citation: Journal of Climate 34, 17; 10.1175/JCLI-D-20-0796.1
In contrast to the findings for sea level, the ocean boundary condition changes contribute little to the SST changes and the effect is limited to a narrow region near the open boundaries (Figs. 8g,k), while the surface forcing change dominates the SST changes (Figs. 8f,j) in both summer and winter. It is reasonable that the effects of surface forcing to ocean are limited in the upper ocean, and SST is closely constrained by the prescribed air temperature. However, the temperature and salinity changes in the deeper depths in the interior ocean are linked to the open boundary condition changes. For example, the SSL change caused by open boundary condition derived from the Ocean-only experiment (Fig. S9b) are close to the total SSL change (Fig. 10c) with spatial correlation of 0.97 and RMSD of 0.05 m. The mean SST increase over the model domain (excluding sponge layers) induced by the surface forcing change and the ocean boundary condition change are 2.75°C (Fig. 8f) and 0.51°C (Fig. 8g) in summer and 2.17°C (Fig. 8j) and 0.78°C (Fig. 8k) in winter, respectively. The SST warming induced by open boundary condition change is stronger during winter than during summer, while the reverse is true for the surface forcing. The sum of these two perturbation experiments results (Figs. 8d,h,l) are close to the Ensemble-forcing experiment (Figs. 8a,e,i) with small differences (DSL difference in Fig. S10), indicating these two changes can be added almost linearly.
The open boundary condition (Fig. 9c) is primarily responsible for the changes in upper ocean circulation pattern, with the atmospheric forcing change resulting in smaller changes (Fig. 9b). For the Ocean-only experiment, the eastern open boundary transport changes (Figs. 9d,e) indicate that more water flows into the model domain within the upper layer between 14° and 31°N (westward transport change in Fig. 9e) in the future climate. The larger volume inflow also results in enhanced surface Kuroshio Current, Mindanao Current, and Kuroshio intrusion into the SCS (Fig. 9c). Since more water flows into the SCS through the Luzon Strait, the cyclonic circulation in the upper SCS is strengthened, which corresponds to the negative DSL change (Fig. 8c). The atmosphere forcing change provides relatively weak effect on upper ocean circulation change (Fig. 9b), which is consistent with the weak DSL change from the Atm-only experiment (Fig. 8b). The upper Kuroshio flow to the south of Japan is enhanced under the influence of boundary condition change (Fig. 9c), but weakened due to the atmospheric change (Fig. 9b).
To further connect the regional change to larger-scale climate, the projected changes over the North Pacific basin from the CMIP5 ensemble are also presented. The wind change over the North Pacific roughly between 15° and 32°N (Fig. 11a) has increased cyclonic wind stress curl, leading to spindown of the subtropical gyre and a weakened western boundary current (Kuroshio) based on Sverdrup balance (Sverdrup 1947). This change is mainly reflected in subsurface circulation change (from 400 to 2000 m; Fig. 11e). However, a different circulation change is found for the upper ocean (from surface to 400 m; Fig. 11c) with spinup of the gyre between 16° and 40°N, and the DSL change (Fig. 11b) is mainly connected with the upper ocean gyre circulation with the positive DSL change corresponding to the anticyclonic ocean circulation change. This baroclinic response (i.e., spinup in the upper ocean and spindown at the subsurface) was first found by Zhang et al. (2014) and further discussed in other studies (e.g., Wang et al. 2015; Chen et al. 2019; Li et al. 2019). This baroclinic response in the North Pacific is caused by the projected larger baroclinicity with stronger (weaker) warming than the global mean in the upper (deeper) depths, which leads to stronger stratification and hinders heat transfer from the upper layers to the lower layers. Furthermore, the acceleration in the upper ocean is suggested to be mainly induced by surface warming rather than wind stress (Wang et al. 2015; Chen et al. 2019), which results in intensified Kuroshio and larger volume inflow on the eastern open boundary between 14° and 31°N. Through dynamical downscaling, the climate change signals coming from the Pacific basin (e.g., the gyre circulation changes discussed above) are passed into the regional model and reflected in the downscaling ensemble (cf. Figs. 11c–f).
7. Summary and discussion
In this study, a high-resolution regional ocean model of ROMS is set up for the marginal seas near China, with the aim to improve ocean climate projections by dynamical downscaling of eight selected CMIP5 climate models. The method of running models separately for historical and future periods, often called the time-slice method, has been widely applied in previous dynamical downscaling studies (e.g., Meier 2006; Ådlandsvik 2008; Chamberlain et al. 2012; Sun et al. 2012; Oliver et al. 2015; Liu et al. 2016; Toste et al. 2018), with the advantage of reducing the required computational effort of future projection. More spatial details are present in the downscaling ensemble than in their parent climate ensemble, including the better presentation of western boundary and coastal currents reflected in DSL, SST, and upper ocean circulation patterns. The comparison also reveals that the projected coastal DSL rise is 2.5–5.0 cm higher in the downscaling ensemble than that in the CMIP5 ensemble in the East China marginal seas and the continental shelf of the SCS (Fig. S6). These added values are the products of a high-resolution regional ocean model and the bias correction to GCMs based on robust historical simulation, which make the downscaling ensemble a useful extension to the corresponding CMIP5 ensemble.
As one of the advantages compared with the GCM, the regional model has the flexibility to be set up to run various perturbation experiments to investigate the specific regional issues. In this paper, the effects from ocean boundary conditions and local atmosphere forcing are discussed, indicating that the DSL change in the SCS is mainly caused by the climate change signal coming from the Pacific Ocean, while the DSL change in the East China marginal seas is induced by the climate change signal from both local atmosphere and remote Pacific Ocean.
As found by Fowler et al. (2007), the multimodel ensemble tends to give a more robust assessment of climate change than individual models. Therefore, it would be ideal to downscale as many climate models as possible. However, we note that the numbers of selected GCM in most of the dynamical downscaling work are limited within 1–3, most likely due to limited computation resources. Our downscaling experiments reveal that the downscaled results driven by the ensemble mean forcing are almost identical with the ensemble mean results from individual downscaling experiments, which means climate change signals from multiple GCMs can be combined linearly. This solution provides an efficient approach to downscaling GCMs, as done for RCP4.5 in this study results.
However, the disadvantage of this approach is that multimodel uncertainty among the downscaling ensemble cannot be derived directly. Nevertheless, the spatial pattern of multimodel uncertainty of the downscaling ensemble (solid contours in Fig. 6h) displays some common features with that of the CMIP5 ensemble (solid contours in Fig. 6d) with spatial correlation of 0.50. So it may be possible to derive the multimodel uncertainty of the downscaling ensemble by considering that from the corresponding CMIP5 ensemble, with other relevant information (e.g., ratio of DSL variability between ROMS and CMIP5). This approach is worth investigating in the future.
As a critical exercise of any regional modeling works, the boundary condition of dynamical downscaling should be tested carefully. In this work, the temperature and salinity are clamped so that temperature and salinity from GCMs were prescribed at the boundaries and nudged within the sponge layers, providing a robust mechanism to pass global climate change signals into the regional ocean domain. In addition, the radiation boundary condition for depth-independent momentum provides a mechanism for the regional ocean model to adjust and to radiate out signals across open boundaries. Despite the above configuration, there is a high center of DSL change in the Japan Sea with large intermodel uncertainty (Fig. 6h) which is nearly absent in the CMIP5 ensemble (Fig. 6d), as well as obvious SST rise in the Japan Sea (Figs. 5h,p). These differences are induced by the open boundary conditions (Figs. 8c,g,k). Since the Japan Sea is close to the northern open boundary and within the sponge layer, these significant changes should be interpreted with caution, and this uncertainty might be ameliorated by extending the model domain. However, considering the Japan Sea is a quasi-closed sea and the open boundaries are far from our key study area (i.e., the marginal seas near China), our main findings regarding the DSL change will likely not be affected.
In this study, we use sea level projections in marginal seas near China as an example to discuss the dynamical downscaling application. The dynamical downscaling protocol could be widely applied to any other regional domains by adding climate change signals from GCMs to the historical forcings. It is a useful modeling protocol to produce ocean climate projections information, which is critically needed for adaptation and mitigation planning for future climate changes.
Acknowledgments
This study was supported by the Centre for Southern Hemisphere Oceans Research (CSHOR), jointly funded by the Qingdao National Laboratory for Marine Science and Technology (QNLM, China) and the Commonwealth Scientific and Industrial Research Organisation (CSIRO, Australia), and the Australian Research Council’s Discovery Project funding scheme (Project DP190101173). We acknowledge the World Climate Research Programme, which, through its Working Group on Coupled Modelling, coordinated and promoted CMIP. We thank the climate modelling groups (listed in Table 2) for producing and making available their model output, the Earth System Grid Federation (ESGF) for archiving the data and providing access, and the multiple funding agencies who support CMIP and ESGF. This research was undertaken with the assistance of resources from the National Computational Infrastructure (NCI Australia), an NCRIS enabled capability supported by the Australian government.
Data availability statement
The altimetry products of the Archiving, Validation and Interpretation of Satellite Oceanographic (AVISO) are from https://resources.marine.copernicus.eu/?option=com_csw&view=details&product_id=SEALEVEL_GLO_PHY_L4_REP_OBSERVATIONS_008_047. The bathymetric data (DBDBV) are available from https://pubs.usgs.gov/of/2011/1127/dbdbv.html. The eight CMIP5 model monthly outputs of r1i1p1 (the used variables are zos, thetao, so, uo, vo, tas, hurs, psl, pr, uas, vas, rsds, rlds and rlus) used in this study can be accessed from https://esgf-node.llnl.gov/projects/esgf-llnl/. The Bluelink Reanalysis (BRAN) version 2016 can be downloaded from http://dapds00.nci.org.au/thredds/catalog/gb6/BRAN/catalog.html, while the ERA-Interim data can be downloaded from https://www.ecmwf.int/en/forecasts/datasets/reanalysis-datasets/era-interim.
The Regional Ocean Modelling System (ROMS) based on which our regional ocean model was set up can be downloaded from https://www.myroms.org. The ROMS-based regional ocean experiments upon which this study is based are too large to share via public domain. However, we provided all the information needed to replicate our experiments in section 2. The data, models, and codes produced out of this study are available from the corresponding author by request.
REFERENCES
Ablain, M., and Coauthors, 2019: Uncertainty in satellite estimates of global mean sea-level changes, trend and acceleration. Earth Syst. Sci. Data, 11, 1189–1202, https://doi.org/10.5194/essd-11-1189-2019.
Ådlandsvik, B., 2008: Marine downscaling of a future climate scenario for the North Sea. Tellus, 60A, 451–458, https://doi.org/10.1111/j.1600-0870.2007.00311.x.
Caldwell, P. C., M. A. Merrifield, and P. R. Thompson, 2015: Sea level measured by tide gauges from global oceans as part of the Joint Archive for Sea Level (JASL) since 1846 (NCEI Accession 0019568), version 5.5, NOAA National Centers for Environmental Information, accessed November 2018, https://doi.org/10.7289/V5V40S7W.
Chamberlain, M. A., C. Sun, R. J. Matear, M. Feng, and S. J. Phipps, 2012: Downscaling the climate change for oceans around Australia. Geosci. Model Dev., 5, 1177–1194, https://doi.org/10.5194/gmd-5-1177-2012.
Chang, C. W. J., H. H. Hsu, C. R. Wu, and W. J. Sheu, 2008: Interannual mode of sea level in the South China Sea and the roles of El Niño and El Niño Modoki. Geophys. Res. Lett., 35, L03601, https://doi.org/10.1029/2007GL032562.
Chen, C., G. Wang, S. P. Xie, and W. Liu, 2019: Why does global warming weaken the Gulf Stream but intensify the Kuroshio? J. Climate, 32, 7437–7451, https://doi.org/10.1175/JCLI-D-18-0895.1.
Chen, S. T., P. S. Yu, and Y. H. Tang, 2010: Statistical downscaling of daily precipitation using support vector machines and multivariate analysis. J. Hydrol., 385, 13–22, https://doi.org/10.1016/j.jhydrol.2010.01.021.
Church, J. A., and N. J. White, 2011: Sea-level rise from the late 19th to the early 21st century. Surv. Geophys., 32, 585–602, https://doi.org/10.1007/s10712-011-9119-1.
Church, J. A., and Coauthors, 2013a: Sea level change. Climate Change 2013: The Physical Science Basis, T. F. Stocker et al. Eds., Cambridge University Press, 1137–1216.
Church, J. A., D. Monselesan, J. M. Gregory, and B. Marzeion, 2013b: Evaluating the ability of process based models to project sea-level change. Environ. Res. Lett., 8, 014051, https://doi.org/10.1088/1748-9326/8/1/014051.
Dee, D., and Coauthors, 2011: The ERA-Interim reanalysis: Configuration and performance of the data assimilation system. Quart. J. Roy. Meteor. Soc., 137, 553–597, https://doi.org/10.1002/qj.828.
Ding, Y., X. Bao, Z. Yao, D. Song, J. Song, J. Gao, and J. Li, 2018: Effect of coastal-trapped waves on the synoptic variations of the Yellow Sea Warm Current during winter. Cont. Shelf Res., 167, 14–31, https://doi.org/10.1016/j.csr.2018.08.003.
Ding, Y., and Coauthors, 2019: Observational and model studies of synoptic current fluctuations in the Bohai Strait on the Chinese continental shelf. Ocean Dyn., 69, 323–351, https://doi.org/10.1007/s10236-019-01247-5.
Eyring, V., S. Bony, G. A. Meehl, C. A. Senior, B. Stevens, R. J. Stoufer, and K. E. Taylor, 2016: Overview of the Coupled Model Intercomparison Project phase 6 (CMIP6) experimental design and organization. Geosci. Model Dev., 9, 1937–1958, https://doi.org/10.5194/gmd-9-1937-2016.
Fairall, C. W., E. F. Bradley, J. E. Hare, A. A. Grachev, and J. B. Edson, 2003: Bulk parameterization of air–sea fluxes: Updates and verification for the COARE algorithm. J. Climate, 16, 571–591, https://doi.org/10.1175/1520-0442(2003)016<0571:BPOASF>2.0.CO;2.
Farrell, W. E., and J. A. Clark, 1976: On postglacial sea level. Geophys. J. Int., 46, 647–667, https://doi.org/10.1111/j.1365-246X.1976.tb01252.x.
Flato, G., and Coauthors, 2013: Evaluation of climate models. Climate Change 2013: The Physical Science Basis, T. F. Stocker et al., Eds., Cambridge University Press, 741–866.
Fowler, H. J., S. Blenkinsop, and C. Tebaldi, 2007: Linking climate change modelling to impacts studies: Recent advances in downscaling techniques for hydrological modelling. Int. J. Climatol., 27, 1547–1578, https://doi.org/10.1002/joc.1556.
Frederikse, T., and Coauthors, 2020: The causes of sea-level rise since 1900. Nature, 584, 393–397, https://doi.org/10.1038/s41586-020-2591-3.
Gregory, J. M., and Coauthors, 2019: Concepts and terminology for sea level: Mean, variability and change, both local and global. Surv. Geophys., 40, 1251–1289, https://doi.org/10.1007/s10712-019-09525-z.
Grose, M. R., and Coauthors, 2020: Insights from CMIP6 for Australia’s future climate. Earth’s Future, 8, https://doi.org/10.1029/2019EF001469.
Guo, X., Y. Miyazawa, and T. Yamagata, 2006: The Kuroshio onshore intrusion along the shelf break of the East China Sea: The origin of the Tsushima Warm Current. J. Phys. Oceanogr., 36, 2205–2231, https://doi.org/10.1175/JPO2976.1.
Haarsma, R. J., and Coauthors, 2016: High Resolution Model Intercomparison Project (HighResMIP v1.0) for CMIP6. Geosci. Model Dev., 9, 4185–4208, https://doi.org/10.5194/gmd-9-4185-2016.
Han, G., and W. Huang, 2008: Pacific decadal oscillation and sea level variability in the Bohai, Yellow, and East China Seas. J. Phys. Oceanogr., 38, 2772–2783, https://doi.org/10.1175/2008JPO3885.1.
Hermans, T. H. J., J. Tinker, M. D. Palmer, C. A. Katsman, B. L. A. Vermeersen, and A. B. A. Slangen, 2020: Improving sea-level projections on the northwestern European shelf using dynamical downscaling. Climate Dyn., 54, 1987–2011, https://doi.org/10.1007/s00382-019-05104-5.
Hewitson, B. C., and R. G. Crane, 1996: Climate downscaling: Techniques and application. Climate Res., 7, 85–95, https://doi.org/10.3354/cr007085.
Horowitz, L. W., and Coauthors, 2020: The GFDL global atmospheric chemistry-climate model AM4.1: Model description and simulation characteristics. J. Adv. Model. Earth Syst., 12, e2019MS002032, https://doi.org/10.1029/2019MS002032.
Isobe, A., 2008: Recent advances in ocean-circulation research on the Yellow Sea and East China Sea shelves. J. Oceanogr., 64, 569–584, https://doi.org/10.1007/s10872-008-0048-7.
Landerer, F. W., J. H. Jungclaus, and J. Marotzke, 2007: Regional dynamic and steric sea level change in response to the IPCC-A1B scenario. J. Phys. Oceanogr., 37, 296–312, https://doi.org/10.1175/JPO3013.1.
Landerer, F. W., P. J. Gleckler, and T. Lee, 2014: Evaluation of CMIP5 dynamic sea surface height multi-model simulations against satellite observations. Climate Dyn., 43, 1271–1283, https://doi.org/10.1007/s00382-013-1939-x.
Li, H., M. Kanamitsu, S. Y. Hong, K. Yoshimura, D. R. Cayan, and V. Misra, 2014: A high-resolution ocean–atmosphere coupled downscaling of the present climate over California. Climate Dyn., 42, 701–714, https://doi.org/10.1007/s00382-013-1670-7.
Li, Q., Y. Luo, and F. Liu, 2019: Response of the subtropical gyre circulation in the North Pacific Ocean to CO2 quadrupling. Atmos.–Ocean, 57, 307–317, https://doi.org/10.1080/07055900.2019.1666701.
Li, Y., J. Zuo, Q. Lu, H. Zhang, and M. Chen, 2016: Impacts of wind forcing on sea level variations in the East China Sea: Local and remote effects. J. Mar. Syst., 154, 172–180, https://doi.org/10.1016/j.jmarsys.2015.10.009.
Liu, X., Y. Liu, L. Guo, Z. Rong, Y. Gu, and Y. Liu, 2010: Interannual changes of sea level in the two regions of East China Sea and different responses to ENSO. Global Planet. Change, 72, 215–226, https://doi.org/10.1016/j.gloplacha.2010.04.009.
Liu, Y., S. K. Lee, B. A. Muhling, J. T. Lamkin, and D. B. Enfield, 2012: Significant reduction of the Loop Current in the 21st century and its impact on the Gulf of Mexico. J. Geophys. Res. Oceans, 117, C05039, https://doi.org/10.1029/2011JC007555.
Liu, Y., L. Xie, J. M. Morrison, and D. Kamykowski, 2013: Dynamic downscaling of the impact of climate change on the ocean circulation in the Galápagos Archipelago. Adv. Meteor., 2013, 837432, https://doi.org/10.1155/2013/837432.
Liu, Y., S. K. Lee, D. B. Enfield, B. A. Muhling, J. T. Lamkin, F. E. Muller-Karger, and A. R. Mitchell, 2015: Potential impact of climate change on the Intra-Americas Sea: Part 1. A dynamic downscaling of the CMIP5 model projections. J. Mar. Syst., 148, 56–69, https://doi.org/10.1016/j.jmarsys.2015.01.007.
Liu, Z. J., S. Minobe, Y. N. Sasaki, and M. Terada, 2016: Dynamical downscaling of future sea level change in the western North Pacific using ROMS. J. Oceanogr., 72, 905–922, https://doi.org/10.1007/s10872-016-0390-0.
Lu, J., and Q. Liu, 2013: Gap-leaping Kuroshio and blocking westward-propagating Rossby wave and eddy in the Luzon Strait. J. Geophys. Res. Oceans, 118, 1170–1181, https://doi.org/10.1002/jgrc.20116.
Lv, X., F. Qiao, C. Xia, J. Zhu, and Y. Yuan, 2006: Upwelling off Yangtze River estuary in summer. J. Geophys. Res. Oceans, 111, C11S08, https://doi.org/10.1029/2005JC003250.
Lyu, K., X. Zhang, J. A. Church, and J. Hu, 2016: Evaluation of the interdecadal variability of sea surface temperature and sea level in the Pacific in CMIP3 and CMIP5 models. Int. J. Climatol., 36, 3723–3740, https://doi.org/10.1002/joc.4587.
Lyu, K., X. Zhang, and J. A. Church, 2020: Regional dynamic sea level simulated in the CMIP5 and CMIP6 models: Mean biases, future projections, and their linkages. J. Climate, 33, 6377–6398, https://doi.org/10.1175/JCLI-D-19-1029.1.
Maximenko, N., P. Niiler, L. Centurioni, M. Rio, O. Melnichenko, D. Chambers, V. Zlotnicki, and B. Galperin, 2009: Mean dynamic topography of the ocean derived from satellite and drifting buoy data using three different techniques. J. Atmos. Oceanic Technol., 26, 1910–1919, https://doi.org/10.1175/2009JTECHO672.1.
Meier, H. E. M., 2006: Baltic Sea climate in the late twenty-first century: A dynamical downscaling approach using two global models and two emission scenarios. Climate Dyn., 27, 39–68, https://doi.org/10.1007/s00382-006-0124-x.
Meyssignac, B., and Coauthors, 2017: Evaluating model simulations of twentieth-century sea-level rise. Part II: Regional sea-level changes. J. Climate, 30, 8565–8593, https://doi.org/10.1175/JCLI-D-17-0112.1.
Moon, J. H., and Y. T. Song, 2017: Decadal sea level variability in the East China Sea linked to the North Pacific Gyre Oscillation. Cont. Shelf Res., 143, 278–285, https://doi.org/10.1016/j.csr.2016.05.003.
Nan, F., H. Xue, and F. Yu, 2015: Kuroshio intrusion into the South China Sea: A review. Prog. Oceanogr., 137, 314–333, https://doi.org/10.1016/j.pocean.2014.05.012.
Oke, P. R., and Coauthors, 2013: Towards a dynamically balanced eddy-resolving ocean reanalysis: BRAN3. Ocean Modell., 67, 52–70, https://doi.org/10.1016/j.ocemod.2013.03.008.
Oliver, E. C., T. J. O’Kane, and N. J. Holbrook, 2015: Projected changes to Tasman Sea eddies in a future climate. J. Geophys. Res. Oceans, 120, 7150–7165, https://doi.org/10.1002/2015JC010993.
Pielke, R. A., Sr., and R. L. Wilby, 2012: Regional climate downscaling: What’s the point? Eos, Trans. Amer. Geophys. Union, 93, 52–53, https://doi.org/10.1029/2012EO050008.
Pujol, M. I., Y. Faugère, G. Taburet, S. Dupuy, C. Pelloquin, M. Ablain, and N. Picot, 2016: DUACS DT2014: The new multi-mission altimeter data set reprocessed over 20 years. Ocean Sci., 12, 1067–1090, https://doi.org/10.5194/os-12-1067-2016.
Qu, Y., S. Jevrejeva, L. P. Jackson, and J. C. Moore, 2019: Coastal sea level rise around the China seas. Global Planet. Change, 172, 454–463, https://doi.org/10.1016/j.gloplacha.2018.11.005.
Rasp, S., M. S. Pritchard, and P. Gentine, 2018: Deep learning to represent subgrid processes in climate models. Proc. Natl. Acad. Sci. USA, 115, 9684–9689, https://doi.org/10.1073/pnas.1810286115.
Raymond, W. H., and H. L. Kuo, 1984: A radiation boundary condition for multi-dimensional flows. Quart. J. Roy. Meteor. Soc., 110, 535–551, https://doi.org/10.1002/qj.49711046414.
Rong, Z., Y. Liu, H. Zong, and Y. Chen, 2007: Interannual sea level variability in the South China Sea and its response to ENSO. Global Planet. Change, 55, 257–272, https://doi.org/10.1016/j.gloplacha.2006.08.001.
Seo, G.-H., Y.-K. Cho, B.-J. Choi, K.-Y. Kim, B. Kim, and Y. Tak, 2014: Climate change projection in the northwest Pacific marginal seas through dynamic downscaling. J. Geophys. Res. Oceans, 119, 3497–3516, https://doi.org/10.1002/2013JC009646.
Shchepetkin, A. F., and J. C. McWilliams, 2003: A method for computing horizontal pressure-gradient force in an ocean model with a non-aligned vertical coordinate. J. Geophys. Res., 108, 3090, https://doi.org/10.1029/2001JC001047.
Shchepetkin, A. F., and J. C. McWilliams, 2005: The Regional Oceanic Modeling System (ROMS): A split-explicit, free-surface, topography-following-coordinate oceanic model. Ocean Modell., 9, 347–404, https://doi.org/10.1016/j.ocemod.2004.08.002.
Shin, S. I., and M. A. Alexander, 2020: Dynamical downscaling of future hydrographic changes over the northwest Atlantic Ocean. J. Climate, 33, 2871–2890, https://doi.org/10.1175/JCLI-D-19-0483.1.
Sikirić, M. D., I. Janeković, and M. Kuzmić, 2009: A new approach to bathymetry smoothing in sigma-coordinate ocean models. Ocean Modell., 29, 128–136, https://doi.org/10.1016/j.ocemod.2009.03.009.
Slangen, A. B., and Coauthors, 2017: Evaluating model simulations of 20th century sea-level rise. Part 1: Global mean sea-level change. J. Climate, 30, 8539–8563, https://doi.org/10.1175/JCLI-D-17-0110.1.
Song, Y. T., 2006: Estimation of interbasin transport using ocean bottom pressure: Theory and model for Asian marginal seas. J. Geophys. Res. Oceans, 111, C11S19, https://doi.org/10.1029/2005JC003189.
Su, J., 2004: Overview of the South China Sea circulation and its influence on the coastal physical oceanography outside the Pearl River Estuary. Cont. Shelf Res., 24, 1745–1760, https://doi.org/10.1016/j.csr.2004.06.005.
Sun, C., M. Feng, R. J. Matear, M. A. Chamberlain, P. Craig, K. R. Ridgway, and A. Schiller, 2012: Marine downscaling of a future climate scenario for Australian boundary currents. J. Climate, 25, 2947–2962, https://doi.org/10.1175/JCLI-D-11-00159.1.
Sverdrup, H. U., 1947: Wind-driven currents in a baroclinic ocean: With application to the equatorial currents in the eastern Pacific. Proc. Natl. Acad. Sci. USA, 33, 318–326, https://doi.org/10.1073/pnas.33.11.318.
Taylor, K., R. J. Stoufer, and G. A. Meehl, 2012: An overview of CMIP5 and the experiment design. Bull. Amer. Meteor. Soc., 93, 485–498, https://doi.org/10.1175/BAMS-D-11-00094.1.
Tedeschi, R. G., and M. Collins, 2017: The influence of ENSO on South American precipitation: Simulation and projection in CMIP5 models. Int. J. Climatol., 37, 3319–3339, https://doi.org/10.1002/joc.4919.
Tinker, J., M. D. Palmer, D. Copsey, T. Howard, J. A. Lowe, and T. H. J. Hermans, 2020: Dynamical downscaling of unforced interannual sea-level variability in the North-West European shelf seas. Climate Dyn., 55, 2207–2236, https://doi.org/10.1007/s00382-020-05378-0.
Toste, R., L. P. de Freitas Assad, and L. Landau, 2018: Downscaling of the global HadGEM2-ES results to model the future and present-day ocean conditions of the southeastern Brazilian continental shelf. Climate Dyn., 51, 143–159, https://doi.org/10.1007/s00382-017-3911-7.
Wang, G., S. P. Xie, R. X. Huang, and C. Chen, 2015: Robust warming pattern of global subtropical oceans and its mechanism. J. Climate, 28, 8574–8584, https://doi.org/10.1175/JCLI-D-14-00809.1.
Werdell, P. J., and Coauthors, 2013: Generalized ocean color inversion model for retrieving marine inherent optical properties. Appl. Opt., 52, 2019–2037, https://doi.org/10.1364/AO.52.002019.
Wohl, E. E., 2007: Hydrology and discharge. Large Rivers: Geomorphology and Management, A. Gupta, Ed., Wiley, 29–44.
Wolter, K., and M. S. Timlin, 2011: El Niño/Southern Oscillation behaviour since 1871 as diagnosed in an extended multivariate ENSO index (MEI.ext). Intl. J. Climatol., 31, 1074–1087, https://doi.org/10.1002/joc.2336.
Wu, C. R., and Y. C. Hsin, 2012: The forcing mechanism leading to the Kuroshio intrusion into the South China Sea. J. Geophys. Res. Oceans, 117, C07015, https://doi.org/10.1029/2012JC007968.
Wu, Q., X. Zhang, J. A. Church, and J. Hu, 2017: Variability and change of sea level and its components in the Indo-Pacific region during the altimetry era. J. Geophys. Res. Oceans, 122, 1862–1881, https://doi.org/10.1002/2016JC012345.
Xue, H., F. Chai, N. Pettigrew, D. Xu, M. Shi, and J. Xu, 2004: Kuroshio intrusion and the circulation in the South China Sea. J. Geophys. Res. Oceans, 109, C02017, https://doi.org/10.1029/2002JC001724.
Yin, J., 2012: Century to multi-century sea level rise projections from CMIP5 models. Geophys. Res. Lett., 39, 17709, https://doi.org/10.1029/2012GL052947.
Yin, J., S. M. Griffies, and R. J. Stouffer, 2010: Spatial variability of sea level rise in twenty-first century projections. J. Climate, 23, 4585–4607, https://doi.org/10.1175/2010JCLI3533.1.
Zhang, X., and J. A. Church, 2012: Sea level trends, interannual and decadal variability in the Pacific Ocean. Geophys. Res. Lett., 39, L21701, https://doi.org/10.1029/2012GL053240.
Zhang, X., J. A. Church, S. M. Platten, and D. Monselesan, 2014: Projection of subtropical gyre circulation and associated sea level changes in the Pacific based on CMIP3 climate models. Climate Dyn., 43, 131–144, https://doi.org/10.1007/s00382-013-1902-x.
Zhang, X., J. A. Church, D. Monselesan, and L. M. Kathleen, 2017: Sea level projections for the Australian region in the 21st century. Geophys. Res. Lett., 44, 8481–8491, https://doi.org/10.1002/2017GL074176.
