Climate change projection in the Northwest ... - Wiley Online Library

PUBLICATIONS Journal of Geophysical Research: Oceans RESEARCH ARTICLE 10.1002/2013JC009646

Special Section: Pacific-Asian Marginal Seas

Climate change projection in the Northwest Pacific marginal seas through dynamic downscaling Gwang-Ho Seo1, Yang-Ki Cho1, Byoung-Ju Choi2, Kwang-Yul Kim1, Bong-guk Kim1, and Yong-jin Tak1 1

Key Points: Future climate change projections in the Northwest Pacific marginal seas The Yellow Sea Bottom Cold Water will disappear in 2100 Most volume transports in major straits in the marginal seas increase in future

Correspondence to: Y.-K. Cho, [email protected] Citation: Seo, G.-H., Y.-K. Cho, B.-J. Choi, K.-Y. Kim, B.-g. Kim, and Y.-j. Tak (2014), Climate change projection in the Northwest Pacific marginal seas through dynamic downscaling, J. Geophys. Res. Oceans, 119, 3497–3516, doi:10.1002/2013JC009646. Received 27 NOV 2013 Accepted 15 MAY 2014 Accepted article online 21 MAY 2014 Published online 6 JUN 2014

2

School of Earth and Environmental Sciences, Research Institute of Oceanography, Seoul National University, Seoul, Korea, Department of Ocean Science and Engineering, Kunsan National University, Gunsan, Korea

Abstract This study presents future climate change projections in the Northwest Pacific (NWP) marginal seas using dynamic downscaling from global climate models (GCMs). A regional climate model (RCM) for the Northwest Pacific Ocean was setup and integrated over the period from 2001 to 2100. The model used forcing fields from three different GCM simulations to downscale the effect of global climate change. MIROC, ECHAM, and HADCM were selected to provide climate change signals for the RCM. These signals were calculated from the GCMs using Cyclostationary Empirical Orthogonal Function analysis and added to the present lateral open boundary and the surface forcing. The RCM was validated by comparing hindcast result with the observation. It was able to project detailed regional climate change processes that GCMs were not able to resolve. A relatively large increases of water temperature were found in the marginal seas. However, only a marginal change was found along the Kuroshio path. Heat supply to the atmosphere decreases in most study areas due to a slower warming of the sea surface compared to the atmosphere. The RCM projection suggests that the temperature of the Yellow Sea Bottom Cold Water will gradually increase by 2100. Volume transports through major straits except the Taiwan Strait in the marginal seas are projected to increase slightly in future. Increased northeasterly wind stress in the East China Sea may also result in the transport change.

1. Introduction Sea surface temperature increase in the Northwest Pacific (NWP) Ocean and its marginal seas has been significantly higher than the global mean [Belkin, 2009]. Temperature is considered to be one of the most important variables in quantifying climate change. A rapid temperature change is expected to have a large impact on regional ecosystems [Kim and Kang, 2000; Lehodey and Maury, 2010]. The NWP marginal seas are characterized by complex regional circulation and large variability, yet they are small enough for multiple numerical simulations to be economically feasible. The surface temperatures of the NWP marginal seas are largely influenced by major ocean currents such as the Kuroshio, the Tsushima Current (TC), the East Korean Warm Current (EKWC), and the Yellow Sea Warm Current (YSWC). These poleward currents provide a continuous supply of heat to the marginal seas from the subtropical ocean (Figure 1). Each marginal sea has unique physical and biological characteristics and their complicated bottom topography and tidal currents play an important role in shelf processes. Global Climate Models (GCMs) have provided future projections of large-scale distribution of heat and its impact on oceanic processes [Stock et al., 2011]. Due to the GCMs’ coarse grid resolution, however, changes in coastal and shallow water regions cannot be effectively investigated. A simple spatial interpolation of GCMs cannot resolve eddy-scale variability in coastal and shallow regions [Jones et al., 1995]. Furthermore, GCMs do not typically include relevant local shelf processes such as those due to tidal mixing or buoyancy input from rivers. A GCM with a relatively coarse resolution results in erroneously large or small transports through the straits. As a result, exchange of mass and energy between the open ocean and the semiclosed marginal seas tends to be inaccurate. An accurate rendition of mass and energy transports through the straits, of course, is an important factor for an accurate simulation of regional climate change. An important focus of this study is a detailed investigation of the climate change in temperature, volume transport and heat flux in the NWP marginal seas for the next 100 years. Dynamical downscaling technique adapted for a regional ocean circulation model is a useful strategy to capture local climate change in the marginal seas.

SEO ET AL.

C 2014. American Geophysical Union. All Rights Reserved. V

3497

Journal of Geophysical Research: Oceans

10.1002/2013JC009646

Detailed simulations using a regional ocean circulation model may increase our understanding and predictability of climate change in the NWP marginal seas. Several studies have investigated future change in temperature, transport, and circulation in this region using numerical models [Guo et al., 2003, 2006; Isobe and Beardsley, 2006; Lee and Takeshi, 2007; Yang et al., 2012; Cho et al., 2009, 2013; Kim et al., 2013; Seo et al., 2010]. There are also a number of downscaling studies for other regions [Meier, 2006; Ådlandsvik and Bentsen, 2007; Sun et al., 2012]. This study utilizes ocean dynamical downscaling to investigate the impact of global climate change, as estimated from GCMs, on the water temperature, mass, and transport in the NWP marginal seas. The Figure 1. Current system in the study area. KC, TWC, TC, EKWC, and YSWC represent the Kuroshio Current, Taiwan Warm Current, regional ocean climate model (RCM) is heavily influTsushima Current, East Korean Warm Current, and Yellow Sea enced by local forcing, lateral boundary conditions, Warm Current, respectively. and complex coastlines, which, in turn, depend on global climate simulations. RCMs can improve the temporal and spatial patterns of climate change on the local scales [Jones et al., 1995; Walsh and McGregor, 1995; Wang et al., 2000, 2003]. In order to incorporate the effect of global climate change to regional ocean model, global warming signal together with its temporal and spatial patterns of change is extracted from each of the three GCM simulations: MIROC, ECHAM, and HADCM. Based on the extracted global warming signal, a dynamical downscaling strategy is developed. Before applying the resulting downscaling procedure for a projection of climate in the NWP marginal seas, performance of the regional ocean circulation model is evaluated. The employed RCM is verified against the observed volume transport through the Korea Strait as well as satellite-observed sea surface temperature (SST). The spread and uncertainty is also addressed for the future climate projections based on the downscaled RCM.

2. Models 2.1. Global Climate Models We selected three GCMs for regional downscaling after evaluating the performance of available GCMs against the observations in the NWP and the marginal seas. The GCMs selected for our experiment are MIROC-3.2 (hire) of Centre for Climate System Research in Tokyo [K-1 Developers, 2004], the ECHAM5 of the Max Planck Institute for Meteorology in Hamburg [Roeckner et al., 2003], and the HadCM3 of the Hadley Centre in Reading [Johns et al., 1997]. The MIROC has the highest horizontal grid resolution among available GCMs. Reichler and Kim [2008] evaluated coupled GCMs and suggested that the ECHAM and the HADCM simulate better the present-day conditions. The SRES A1B in the 4th assessment report scenario of the 4th assessment report of the International Panel on Climate Change (IPCC) was selected for this experiment. The details of selected GCMs are shown in Table 1. MIROC has a relatively fine horizontal grid resolution of 0.2 3 0.3 , whereas ECHAM and HADCM have resolutions of 1.0 3 1.0 and 1.25 3 1.25 , respectively. The coarse resolution of the GCMs results in an unrealistic transport through Korea Strait, which is an important factor in determining heat transport in the NWP marginal seas. The mean seasonal variation of the

Table 1. The Details of the Global Climate Models Used for This Study Model

Institute, Country

MIROC3.2 (hires) ECHAM5/MPI-OM HadCM3

SEO ET AL.

Center for Climate System Research (University of Tokyo), JAMSTEC, Japan Max Planck Institute for Meteorology, Germany Hadley Centre/Met Office, U. K.


Resolution

References

0.2 3 0.3 47 layers

K-1 Developer [2004]

1.0 3 1.0 40 layers 1.25 3 1.25 20 layers

Marsland et al. [2003] Gordon et al. [2000]

3498


10.1002/2013JC009646

transport though the Korea Strait, as calculated by GCMs from 2001 to 2010, was compared with that of the observed transport from 2001 to 2006 in Figure 2 [Fukudome et al., 2010]. A comparison shows that MIROC overestimates the transport of the TC through the Korea Strait, whereas ECHAM and HADCM underestimate it. Due to their coarse resolution, GCMs do not effectively resolve the Kuroshio and the currents in the Figure 2. Monthly mean transports through the Korea Strait from the GCMs and NWP marginal seas, although they the observation. Transport from (red line) MIROC, (blue line) HADCM, and (black are able to clearly show the seasonal line) ECHAM are averaged over 2001–2010. The observation (dashed line) is avervariation of temperature (Figure 3). aged from 2001 to 2006 [Fukudome et al., 2010]. The SST of HADCM in the East/Japan Sea (EJS) is colder than the SST in the other two GCMs. This may be due to the fact that the flow through the Korea Strait in HADCM is in the opposite direction relative to the observation; the lack of warm water supply results in cooling in the EJS. 2.2. Regional Model We used Regional Ocean Modelling System (ROMS) for the projection of climate change in the NWP marginal seas [Shchepetkin and McWilliams, 2005]. ROMS is a free-surface, terrain-following, primitive equation

Figure 3. Mean surface temperature and current of GCMs: (left) MIROC, (middle) ECHAM, and (right) HADCM. (top) SST and surface currents in February and (bottom) those in August averaged from 2001 to 2010. Color indicates sea surface temperature and vectors surface currents.

SEO ET AL.


3499


10.1002/2013JC009646

ocean model discretized on the Arakawa-C staggered grid in the horizontal direction. The regional model domain includes the Northwest Pacific and its marginal seas (18 N to 49 N and 118 E to 155 E) (Figure 4). The horizontal grid spacing is approximately 10 km. In the vertical direction, 20r (s-coordinate) levels are implemented to the scheme of Song and Haidvogel [1994], which enhances resolution toward the surface and the bottom. We used daily mean surface ERAInterim daily mean surface variables with 1.5 resolution to force RCM; the ERA-Interim is the global atmospheric reanalysis produced by European Center for Medium-Range Weather ForeFigure 4. Bottom topography of model domain for the Northwest Pacific. The casts (ECMWF; http://www.ecmwf.int/ thick solid lines across the major straits represent the sections through which volresearch/era). The daily mean forcing— ume transports are estimated. Contour lines are water depths in meter. mean sea level pressure, winds, air temperatures, specific humidity, downward longwave, and shortwave radiation—represents the atmospheric condition in 2001. Ocean lateral boundary conditions derive from the monthly mean Simple Ocean Data Assimilation (SODA) global ocean reanalysis in 2001 [Carton et al., 2000a, 2000b]. Bulk flux algorithm is employed for the surface heat flux [Fairall et al., 2003]. Tides are included along the open boundaries using eight major tidal components (M2, S2, N2, K2, K1, O1, P1, and Q1) from TPXO7 in order to provide tidal mixing effect [Egbert and Erofeeva, 2002]. Tides play an important role in mixing, which affects SST and the heat flux between the ocean and the atmosphere in the Yellow Sea (YS) and the East China Sea (ECS). Long-term mean monthly freshwater discharges from Changjiang (Yangtze) River and Huang He (Yellow) River were included in this study, which would affect surface mixing and salinity distribution. 2.3. Model Validation The spatial distribution of mean SST simulated by the regional model is compared with that of the satellite observation in winter and summer from 2001 to 2009 (Figure 5). The satellite observation data were collected using the NOAA/AVHRR Pathfinder Version 5 and distributed by Jet Propulsion Laboratory (http://poet.jpl.nasa.gov/) with a grid spacing of 4 km. The paths of the Kuroshio, the TC, and the EKWC are characterized by warm SST signatures in winter. The Kuroshio can be clearly identified as warm water southwest of Japan in the simulation. The TC and the EKWC in the EJS are also relatively well defined as warmer water than the surrounding in the simulation. In general, the distribution of isotherms, as well as the maximum and minimum of the simulated mean SST are similar to those obtained from satellite observations. Relatively sharp distinction of SST along the Kuroshio may be contributed to strong horizontal advection or small surface mixing. The Kuroshio transport from the regional model (26–27 Sv) is similar to the observation (24–25 Sv) by Kawabe [1995] across PN line in the ECS, This implies that horizontal advection is about 8% higher in the regional model. When the simulated surface currents were compared with the observed surface velocity of the Kuroshio using the Argos satellite-tracking Lagrangian drifter data [Tseng et al., 2011], the simulated mean velocity is similar to the satellite drifter observation. The sensitivity of surface temperature distribution to the horizontal and vertical mixing parameters was examined by increasing horizontal mixing coefficient from 20 to 100 m2/s and vertical mixing coefficient from 1 3 1026 m2/s to 1 3 1025 m2/s, respectively. When the vertical mixing coefficient was raised, smoother SST distribution was produced along the Kuroshio path. Linear trend in the model SST over a 29 year period (1982–2010) was compared with that of satellite observations in order to examine the regional ocean model’s performance in simulating long-term temperature

SEO ET AL.


3500


10.1002/2013JC009646

Figure 5. Mean sea surface temperature from (upper) satellite observations and (lower) the regional ocean model in (left) February and (right) August from 2001 to 2010.

changes (Figure 6). Both the model and the satellite observations indicate that SST has increased in most parts of the marginal seas from 1982 to 2010 (Figure 6). The increased SST over majority of the area appears to be the result of the global warming. Both model and observations consistently show a larger SST increase in the EJS and east of China. However, smaller SST change is seen along the Kuroshio path and over the coastal area of the YS. The volume transport through the Korea Strait is important, since it is a crucial factor for the distribution of heat and salinity in the EJS as well as in the ECS. An accurate simulation of the transport through the Korea Strait is a key element in accurately reproducing the circulation in the marginal seas of the NWP. Fukudome

SEO ET AL.


3501


10.1002/2013JC009646

Figure 6. Linear trend of sea surface temperature from (left) the regional ocean model and (right) the satellite observation during 29 years from 1982 to 2010.

et al. [2010] estimated the transport based on currents measured from an ADCP mounted on a ferry boat. The observed transport as well as the regional model transport show remarkable seasonal variation during 2001– 2006 (Figure 7). The mean transport in the entire transport record is roughly Figure 7. Comparison of volume transports through the Korea Strait among (filled 2.65 Sv. However, the transport of the circle) the regional model, (filled diamond), the ADCP observation [Fukudome SODA is about 1 Sv smaller than the et al., 2010], and (cross) the SODA reanalysis. observation and the regional model result. The unrealistic transport of the SODA might result from the coarse grid resolution which is not sufficient in representing complicate topography and coastline in the study area. Tidal mixing, which the SODA does not include, should also improve the three-dimensional structure of water column and transport in the regional model. Transport from the Ocean General Circulation Model (OGCM) did not show significant seasonal variation (Figure 2). While the regional model follows the observations closely (Figure 7), the MIROC transport is about 1 Sv larger than the observations for the whole period. The ECHAM and HADCM transports also deviated from observations, with either close to zero or negative transports.

Figure 8. PC time series of the second mode derived from air temperature in each GCM using CSEOF analysis. The second mode represents global warming component from 2001 to 2100. Black, blue, and red lines represent air temperature from MIROC, ECHAM, and HADCM, respectively.

SEO ET AL.


2.4. Warming Trend From GCM The regional climate model (RCM) is heavily influenced by local forcing induced by climate change. In order to incorporate the effect of climate change in an accurate manner, we calculated the long-term warming trend and its spatial distribution from the GCMs using CycloStationary Empirical Orthogonal Function analysis (CSEOF). The CSEOF analysis enables us to separate the anthropogenic warming trend form

3502


10.1002/2013JC009646

Figure 9. Second mode loading vector of air temperature derived from GCM using CSEOF analysis in (top) February and (bottom) August. (left) MIROC, (middle) ECHAM, and (right) HADCM.

natural variability with diverse temporal and spatial scales. The CSEOF technique is useful for extracting physically evolving spatial patterns [Kim et al., 1996; Kim and North, 1997]. Time-space data X(r,t) are decomposed into a set of Eigen modes Vn(r,t), also called loading vector, and their principal component (PC) time series, Pn(t): Xðr; tÞ5

M X

Vn ðr; tÞ Pn ðtÞ;

(1)

n51

where n, r, and t represent the mode number, space and time, respectively. Each loading vector is periodic: Vn ðr; tÞ5Vn ðr; t1dÞ;

(2)

where the nested period d is set to 12 months in this study. As a result, each loading vector consists of 12 monthly patterns.

SEO ET AL.


3503


10.1002/2013JC009646

Figure 10. Comparison of sea surface temperature between (top) GCMs and (bottom) RCMs in February 2100. (left) MIROC, (middle) ECHAM, and (right) HADCM. Color indicates the sea surface temperature and vector surface currents.

The first CSEOF mode from GCM projection represents the seasonal cycle and the second mode depicts trend of global climate change with a conspicuous trend. The PC time series of the second mode derived from air temperature from each GCM are compared in Figure 8. This figure together with Figure 9 implies about 4 C increase in air temperature averaged over the domain by 2100. Thus, the second mode is used to generate the lateral boundary conditions and the surface forcing during the period of 2001–2100. It should be noted that the three GCMs show similar temperature increases, although the model physics and grid resolutions are not identical. Loading vectors indicate spatially uneven warming (Figure 9). More rapid warming in the northern part of the domain compared to the southern part is a common feature to all three GCMs. The Figure 10 shows the SST monthly pattern associated with the second mode. As can be seen, all three GCMs simulate a significant warming along the southern lateral boundary of the RCM domain. The warming persists down to 300 m depth, below which temperature does not significantly warms up. Climate change signals,V2(r,t), P2(t), extracted from each GCM projection using CSEOF analysis was added to the surface forcing values of 2001: Fðr; tÞ5F2001 ðr; tÞ1V2 ðr; tÞ P2 ðtÞ:

(3)

Here F2001(r,t) is the daily surface forcing in 2001 and F2001(r, t) 5 F2002(r,t) 5 . . . 5 F2100(r,t). The added forcing, V2(r,t), P2(t), obtained from GCM output using CSEOF analysis includes the climate change signals in

SEO ET AL.


3504


10.1002/2013JC009646

Figure 11. Comparison of sea surface temperature between (top) GCMs and (bottom) RCMs in August 2100. (left) MIROC, (middle) ECHAM, and (right) HADCM. Color indicates the sea surface temperature and vector surface currents.

mean sea level pressure, winds, air temperatures, specific humidity, downward long wave radiation, and short wave radiation. Open boundary conditions are implemented as: Bðr; tÞ5B2001 ðr; tÞ1V2 ðr; tÞ P2 ðtÞ;

(4)

Here B2001(r,t) denotes the monthly averaged lateral boundary condition in 2001 and B2001(r,t) 5 B2002(r,t) 5 . . . 5B2100(r,t). The added boundary conditions, V2(r,t), P2(t), supplied from GCM using CSEOF analysis, include the climate change signals in temperature, salinity, velocity, and surface elevation along each lateral boundary.

3. Results and Discussion 3.1. Surface Current and Temperature Regional climate projection experiments were performed for the period from 2001 to 2100 by adding the warming signal to the present atmospheric forcing and lateral boundary condition. We compare the similarities and differences of the surface current and temperature obtained from the GCMs and RCMs in the NWP marginal seas. The Kuroshio and the TC, which are prominent ocean currents in the NWP and its marginal seas, were properly resolved in the projected surface current and temperature in February 2100 by the regional model. On

SEO ET AL.


3505


10.1002/2013JC009646

Figure 12. Yearly mean sea surface warming trend from (top) GCMs and (bottom) RCMs from 2010 to 2100. (left) MIROC, (middle) ECHAM, and (right) HADCM.

Figure 13. Yearly mean warming trend of (left) the sea surface and (right) the air calculated by Ensemble mean of RCMs from 2010 to 2100.

SEO ET AL.


3506


10.1002/2013JC009646

Figure 14. Annual mean net surface heat flux estimated from the regional models in (top) 2010 and (bottom) 2100. (left) MIROC, (middle) ECHAM, and (right) HADCM.

Figure 15. Long-term trend of annual mean net surface heat flux from the regional model result from 2010to 2100. (left) MIROC, (middle) ECHAM, and (right) HADCM. Unit: W/m2/yr.

SEO ET AL.


3507


10.1002/2013JC009646

the other hand, GCMs are not able to adequately simulate these currents due to their coarse resolution (Figure 10). The RCMs reproduce similar meanders of the Kuroshio despite the difference GCM forcing. While GCM projections showed large differences in SST distribution in the marginal seas, RCMs showed consistent SST distribution over the marginal seas. In general, HADCM produces the coldest SST, and MIROC generates the warmest SST; RCMs do not result in large differences in SST distribution compared to GCMs. In summer, SST projections by the GCMs are colder than those by the RCMs (Figure 11). The GCMs exhibit a large area of cold water in the north of the EJS and in the center of the YS than the RCMs. In the southern EJS and in the Pacific, the distribution of SST from RCMs is similar to those of GCMs. A close examination of Figure 16. Horizontal distribution of the bottom water temperature in August GCM results shows that the Kuroshio and from the oceanographic atlas of the East Asian Seas (http://kodc.nfrdi.re.kr/ atlas/). Blue represents the YSBCW colder than 10 C. the TC is very weak. The Kuroshio together with its strong meander, however, is clearly expressed in the RCM results. In RCM simulations, SST in certain coastal regions of the YS reaches a maximum temperature of 30–32 C in August 2100. To conduct detailed quantitative analysis, we calculated the spatial distribution of SST warming rates from GCMs and RCMs for the period 2010 to 2100 (Figure 12). The spatial patterns of warming trend are not quit consistent among the GCMs. On the other hand, RCMs produce similar patterns of SST warming rate despite having different GCM forcing. Relatively rapid warming is seen in the YS, while warming is sluggish along the Kuroshio path. The shallow and semienclosed YS might be sensitive to surface heating. The surface warming rate is higher in the northern area of the EJS than in the southern region. The Kuroshio, which originates from the subtropical ocean, experiences relatively slow warming. The ensemble mean of the surface warming trend was calculated from the RCMs, and it was compared with the air temperature warming trend over the same period. The sea surface warming trend is weaker than that of air temperature over most of the study area (Figure 13). SST exhibits a larger spatial differences (0.02–0.05 C/yr) in warming rate than the air temperature (0.04–0.05 C/yr). In the YS and in the Polar front region in EJS, sea surface warming rate is >0.04 C/yr. The warming speed of the surface air is about two times larger than that of the sea surface in the Kuroshio area. However, the warming speed of the sea surface in the YS is almost the same as the air. Figure 17. Zonal section of temperature along the 36.05 N in August from the oceanographic atlas of the East Asian Seas (http://kodc.nfrdi. re.kr/atlas/). Blue represents the YSBCW.

SEO ET AL.


3508


10.1002/2013JC009646

Figure 18. Bottom temperature of GCMs in August (top) 2010 and (bottom) 2100. (left) MIROC, (middle) ECHAM, and (right) HADCM.

3.2. Net Surface Heat Flux Warm currents have been known to transport huge amounts of heat from low to high latitude and supply heat to the atmosphere from the sea surface in the NWP marginal seas [Hirose et al., 1999]. All three RCMs show consistently that the annual mean heat supply to the atmosphere is larger than 100 W/m 2 along the Kuroshio and the TC paths as well as in the northeastern part of the EJS in both the present simulations and the future projections (Figure 14). The net heat flux in the northeastern EJS is larger than that estimated by Hirose et al. [1996]. This seems to result from the fact that our model does not have sea ice. The overshooting of the TC in the absence of sea ice in the northeastern EJS results in a warmer sea surface, which induces a larger net heat flux. Another plausible cause is SST bias due to inaccurate forcing from the atmospheric model. However, heat supply to the atmosphere is nearly zero or negative in the YS, where bottom depth is low and the SST warming rate is similar to that of the air temperature (see Figure 13). Regional net heat flux from the ocean to the atmosphere in 2100 is projected to be smaller than that in 2010. The long-term trend of the net surface heat flux was examined during the warming period during 2010– 2100 (Figure 15). Change in the heat loss is large in the Kuroshio and the TC area, while it is small in the YS. The net surface heat supply to the air decreases in most of the model domain except over the shallow region of the YS and some frontal regions, where SST increases faster than air temperature due to the variation of the current path. This change of net surface heat flux might affect the atmospheric circulation.

SEO ET AL.


3509


10.1002/2013JC009646

Figure 19. Bottom temperature of RCMs in August (top) 2010 and (bottom) 2100. (left) RCM-MIROC, (middle) RCM-ECHAM, and (right) RCM-HADCM.

3.3. The Yellow Sea Bottom Cold Water There is a unique cold water mass below the thermocline in the YS in summer. It is called the Yellow Sea Bottom Cold Water (YSBCW), which is defined as cold water with temperature below10 C [Hur et al., 1999; Zhang et al., 2008; Park et al., 2011]. The YSBCW is formed in winter and stored in the lower layer below the strong thermocline during summer. Its spatial distribution can be identified in the bottom temperature distribution from the Oceanographic Atlas of the East Asian Seas (http://kodc.nfrdi.re.kr/atlas/) in the August (Figure 16). The blue colored region in the middle of the YS represents the YSBCW. The dimension is 100– 200 km in longitudinal direction and is about 500 km in latitudinal direction. The vertical section of temperature along 36.05 N shows that the maximum thickness of the YSBCW is about 40 m (Figure 17). In August, the water column is highly stratified, and the minimum temperature is about 8 C at the bottom and the maximum temperature is about 27 C at the surface. GCMs cannot adequately simulate the characteristics of the YSBCW due to their coarse resolution and unrealistic topography. The bottom water temperature in August 2010 from MIROC and ECHAM show cold water in the north and southwest, respectively (Figure 18). HADCM shows bottom cold water lower than 10 C in most of the YS. These results deviate from the observations shown in Figure 16. The YSBCW with temperature lower than 10 C will disappear in the future according to the MIROC and ECHAM projections. Meanwhile, the bottom temperature of the YS as projected by HADCM will increase by a relatively small amount in future. The bottom depth of the YS in HADCM is >200 m, whereas the maximum depth of the YS is

SEO ET AL.


3510


10.1002/2013JC009646

Figure 20. Evolution of bottom temperature from RCM-MIROC in August 2010, 2030, 2050, 2070, 2090, and 2100.