Continental Shelf Research 26 (2006) 1448–1468 The Red Sea outflow regulated by the Indian monsoon Hidenori Aiki a, , Keiko Takahashi b , Toshio Yamagata a,c a Frontier Research Center for Global Change, Japan Agency for Marine-Earth Science and Technology, Yokohama 236-0001, Japan b Earth Simulator Center, Japan Agency for Marine-Earth Science and Technology, Yokohama 236-0001, Japan c Department of Earth and Planetary Science, University of Tokyo, Tokyo 113-0033, Japan Available online 30 June 2006 Abstract To investigate why the Red Sea water overflows less in summer and more in winter, we have developed a locally high- resolution global OGCM with transposed poles in the Arabian peninsula and India. Based on a series of sensitivity experiments with different sets of idealized atmospheric forcing, the present study shows that the summer cessation of the strait outflow is remotely induced by the monsoonal wind over the Indian Ocean, in particular that over the western Arabian Sea. During the southwest monsoon (May–September), thermocline in the Gulf of Aden shoals as a result of coastal Ekman upwelling induced by the predominantly northeastward wind in the Gulf of Aden and the Arabian Sea. Because this shoaling is maximum during the southwest summer monsoon, the Red Sea water is blocked at the Bab el Mandeb Strait by upwelling of the intermediate water of the Gulf of Aden in late summer. The simulation also shows the three-dimensional evolution of the Red Sea water tongue at the mid-depths in the Gulf of Aden. While the tongue meanders, the discharged Red Sea outflow water (RSOW) (incoming Indian Ocean intermediate water (IOIW)) is always characterized by anticyclonic (cyclonic) vorticity, as suggested from the potential vorticity difference. r 2006 Elsevier Ltd. All rights reserved. Keywords: Stretched-coordinate OGCM; Monsoon winds; Coastal upwelling; Subsurface tongue 1. Introduction The Red Sea is a semi-enclosed mediterranean sea surrounded by the African and Eurasian continents, which is linked to the Indian Ocean by a very shallow sill. The excess evaporation over the Red Sea produces extremely salty and dense water which intermittently spills from the sill and cascades down to the intermediate depths of the Indian Ocean. Passing through the Gulf of Aden (Fig. 1), the Red Sea outflow water (RSOW) becomes a part of the intermediate circulation in the Indian Ocean, which has been observed as a mid-depth salinity maximum in the Arabian Sea and even in the southern hemisphere (Wyrtki, 1971; Quadfasel and Schott, 1982; Gordon et al., 1987; Beal and Bryden, 1997; Mecking and Warner, 1999). Compared with the other marginal overflows in the world oceans (e.g. Mediterranean water outflow in the North Atlan- tic), the Red Sea outflow system has several unique features: (i) the outflow from the Red Sea ceases in late summer, (ii) the strait connecting the Red Sea to the Gulf of Aden is very steep and narrow, and (iii) the Gulf of Aden is dominated by a chain of mesoscale eddies interacting with the tongue of RSOW underneath. ARTICLE IN PRESS www.elsevier.com/locate/csr 0278-4343/$ - see front matter r 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.csr.2006.02.017 Corresponding author. E-mail addresses: [email protected] (H. Aiki), [email protected] (T. Yamagata).
Transcript
ARTICLE IN PRESS
0278-4343/$ - se
doi:10.1016/j.csr
�CorrespondiE-mail addre
yamagata@eps.
Continental Shelf Research 26 (2006) 1448–1468
www.elsevier.com/locate/csr
The Red Sea outflow regulated by the Indian monsoon
aFrontier Research Center for Global Change, Japan Agency for Marine-Earth Science and Technology, Yokohama 236-0001, JapanbEarth Simulator Center, Japan Agency for Marine-Earth Science and Technology, Yokohama 236-0001, Japan
cDepartment of Earth and Planetary Science, University of Tokyo, Tokyo 113-0033, Japan
Available online 30 June 2006
Abstract
To investigate why the Red Sea water overflows less in summer and more in winter, we have developed a locally high-
resolution global OGCM with transposed poles in the Arabian peninsula and India. Based on a series of sensitivity
experiments with different sets of idealized atmospheric forcing, the present study shows that the summer cessation of the
strait outflow is remotely induced by the monsoonal wind over the Indian Ocean, in particular that over the western
Arabian Sea. During the southwest monsoon (May–September), thermocline in the Gulf of Aden shoals as a result of
coastal Ekman upwelling induced by the predominantly northeastward wind in the Gulf of Aden and the Arabian Sea.
Because this shoaling is maximum during the southwest summer monsoon, the Red Sea water is blocked at the Bab el
Mandeb Strait by upwelling of the intermediate water of the Gulf of Aden in late summer. The simulation also shows the
three-dimensional evolution of the Red Sea water tongue at the mid-depths in the Gulf of Aden. While the tongue
meanders, the discharged Red Sea outflow water (RSOW) (incoming Indian Ocean intermediate water (IOIW)) is always
characterized by anticyclonic (cyclonic) vorticity, as suggested from the potential vorticity difference.
The Red Sea is a semi-enclosed mediterranean seasurrounded by the African and Eurasian continents,which is linked to the Indian Ocean by a veryshallow sill. The excess evaporation over the RedSea produces extremely salty and dense water whichintermittently spills from the sill and cascades downto the intermediate depths of the Indian Ocean.Passing through the Gulf of Aden (Fig. 1), the RedSea outflow water (RSOW) becomes a part of the
e front matter r 2006 Elsevier Ltd. All rights reserved
intermediate circulation in the Indian Ocean, whichhas been observed as a mid-depth salinity maximumin the Arabian Sea and even in the southernhemisphere (Wyrtki, 1971; Quadfasel and Schott,1982; Gordon et al., 1987; Beal and Bryden, 1997;Mecking and Warner, 1999). Compared with theother marginal overflows in the world oceans (e.g.Mediterranean water outflow in the North Atlan-tic), the Red Sea outflow system has several uniquefeatures: (i) the outflow from the Red Sea ceases inlate summer, (ii) the strait connecting the Red Sea tothe Gulf of Aden is very steep and narrow, and(iii) the Gulf of Aden is dominated by a chain ofmesoscale eddies interacting with the tongue ofRSOW underneath.
Fig. 1. (a) Bathymetry of the region from the Red Sea to the Indian Ocean. Contours show bottom depths of 100, 500, and 1000m. (b)
Curvilinear orthogonal coordinates used in the present numerical model. (c) The global map of the transformed coordinates in Mercator
projection, with the upper (lower) end corresponding to the transposed pole in the Arabian peninsula (India). (d) The close-up topography
of the numerical model around the Bab el Mandeb Strait. Contours show bottom depths of 150, 200, 300, 600, and 800m.
H. Aiki et al. / Continental Shelf Research 26 (2006) 1448–1468 1449
The Bab el Mandeb Strait is located at the exit ofthe Red Sea and consists of several complextopographic features. From the Red Sea to theIndian Ocean along the 200-km-long Strait, thereare the Hanish Sill, the Perim Narrows, andtwo bottom outlets connected to the Tadjura Rift
(Fig. 1). The Hanish Sill is only 160m deep and5 km wide at the bottom, so that it is crucial indetermining the nature of RSOW. At the tip of thesouthwestern corner of the Arabian peninsula is thePerim Narrows of about 20 km wide, in which thestrait deepens to 300m. As the strait further slopes
ARTICLE IN PRESS
Table 1
The annual mean values, together with the winter (November–
May) values over the summer (June–October) values in the
parentheses, of observed transport Qobs (Sv) salinity Sobs (psu),
temperature Tobs (1C), and potential density robs ðsyÞ through the
Bab el Mandeb Strait
Qobs (Sv) Sobs (psu) Tobs (1C) robs (sy)
Surface layer �0.31 �0:540:01
� �37.0 36:6
37:3
� �28.3 26:7
30:5
� �23.8 24:0
23:7
� �Middle layer �0.07 �
�0:18
� �36.7 �
36:7
� �24.4 �
24:4
� �24.8 �
24:8
� �Bottom layer 0.36 0:52
0:14
� �39.7 39:6
39:8
� �22.8 23:3
22:1
� �27.6 27:3
27:8
� �
These are calculated from the monthly mean values given in
Table 7 of Sofianos et al. (2002). A positive (negative) Q denotes
an outflow (inflow). The surface, middle, and bottom layers are
defined by the zero crossings of the observed velocity profile.The
middle layer does not exist in the winter season, which is denoted
by the ‘�’.
H. Aiki et al. / Continental Shelf Research 26 (2006) 1448–14681450
down to 600–800m in the southern subsection ofthe strait, it separates into two bottom branchescalled the Northern and Southern Channels. Thesetiny trenches (not visible in Fig. 1a) are curved onthe flank of the Tadjura Rift of 1600m deep. Themain outlet of RSOW is the 120-km-long NorthernChannel (Peters et al., 2005; Peters and Johns,2005), which is a narrow bottomed valley of only5 km wide. As the descending RSOW is sheltered bythe confined topography of the Northern Channel,the bottom salty water does not experience sig-nificant mixing with the ambient water until it isejected into the mid-depths of the Tadjura Rift. It isvery challenging for a numerical model to resolvethe above described topography of the Bab elMandeb Strait. Ozgokmen et al. (2003) used atwo-dimensional nonhydrostatic model to partlyinvestigate the sinking of RSOW through theNorthern Channel. The present study uses a three-dimensional stretched-coordinate model that incor-porates the narrow and steep topography of the Babel Mandeb.
At the western end of the Gulf of Aden, thedischarged RSOW reaches a neutral-buoyancydepth of 400–800m (isopycnals of 27.0–27:4sy) andleaves the coast to propagate eastward (cf. Boweret al., 2000). The excursion of RSOW is traceable asa mid-depth salinity maximum in the Gulf of Adenand further downstream in the Indian Ocean(Warren et al., 1966; Quadfasel and Schott, 1982;Meschanov and Shapiro, 1998; Mecking andWarner, 1999; Han and McCreary, 2001). Recentobservations by Bower et al. (2002) have revealedthe meandering pathways of RSOW produced byinteraction with a chain of deep-reaching mesoscaleeddies in the Gulf of Aden. Due to the lateral stirringby the mesoscale eddies, the RSOW undergoesconsiderable (isopycnal) mixing in the Gulf of Aden.The well-developed core of the RSOW enters theArabian Sea and sheds lenses of salty water, whichare often called Reddies (Red Sea water eddies). Insummarizing the distribution of the observed salinityanomalies in the Arabian Sea, Shapiro and Mescha-nov (1991) suggest that the most probable mechan-ism of Reddy generation is instability of the RSOWmain tongue. The numerical simulations in thepresent study complement sparse measurements inthe northwestern Indian Ocean, and allow us tounderstand the three-dimensional structure of theRSOW tongue (Section 3).
At the Bab el Mandeb Strait, the salty Red Seawater is discharged mainly in winter and it ceases in
summer. The winter exchange (about 0.5 Sv inTable 1, 1 Sv ¼ 106m3/s) occurs in the surface andbottom layers, whereas the summer exchange(about 0.16 Sv in Table 1) occurs in the middleand bottom layers (Thompson, 1939; Murray andJohns, 1997). During the summer monsoon (May–September), some water mass from the intermediatelayer of the Gulf of Aden (i.e. the Indian Ocean)intrudes into the Red Sea (Jones and Browning,1971; Patzert, 1974; Bethoux, 1987; Maillard andSoliman, 1986; Murray and Johns, 1997; Saafaniand Shenoi, 2004). It has been generally believed(Thompson, 1939) that the above-mentioned seaso-nal cycle of the water exchange is driven primarilyby the local wind stress which is constrained by theorography of the Bab el Mandeb. The predominantwinds over the strait blow towards the Red Seaduring the winter monsoon (October–April) toenhance the surface layer inflow (and bottomoutflow), whereas the wind blows towards theIndian Ocean during the summer monsoon (May–September) to stop the surface layer inflow (andbottom outflow). In contrast, the water exchangethrough the Bab el Mandeb appears to be hydrau-lically controlled (Pratt et al., 1999; Smeed, 2000,2004; Siddall et al., 2002). This highlights thestratification in the upstream and downstream ofthe strait: upwelling in the Gulf of Aden duringsummer may play an important role for the seasonaloutflow and inflow (Jones and Browning, 1971;Patzert, 1974; Smeed, 2000). The relative impor-tance of these two wind-forced effects, one directand one indirect, is not clear. To examine this issue,we will present in Section 4 a series of sensitivity
ARTICLE IN PRESSH. Aiki et al. / Continental Shelf Research 26 (2006) 1448–1468 1451
experiments with different sets of idealized atmo-spheric forcing.
This paper is organized as follows. Section 2introduces a locally high-resolution OGCM for theRed Sea and Gulf of Aden region, which isintegrated for 15 years in our reference experiment.We investigate in Section 3 the three-dimensionalstructure and time evolution of the RSOW tonguein the Gulf of Aden. Section 4 presents a series ofsensitivity experiments demonstrating that theseasonal overflow through the Bab el Mandeb isremotely constrained by the monsoonal currents inthe Indian Ocean. This paper ends with a briefsummary in Section 5.
2. Numerical model
We have developed a stretched-coordinateOGCM focusing on the Red Sea and the Gulf ofAden region (Fig. 1). The present configuration isdesigned to incorporate the interactions between thesmall-scale processes around the Bab el MandebStrait and the larger-scale currents in the IndianOcean. The computational domain covers the entireglobe, and hence there are no open boundaries.
2.1. Stretched coordinates
The present model uses the curvilinear orthogo-nal coordinates given by the pole-transpositionmethod of Bentsen et al. (1999). The region coveringthe Red Sea and the Gulf of Aden is focused on bylocating one pole at the southwestern corner of theArabian peninsula (10 km inside the coastline,closest to the Perim Narrows) and another pole inIndia (Fig. 1b). As shown in Fig. 1c (with theMercator projection), the transformed coordinatesdeform the world oceans to enlarge the Arabianmarginal seas; about half of the computationaldomain is occupied by the analysis region of thepresent study.
This transformed map is discretized on a 240�256 grid in the present model, giving a smallesthorizontal grid spacing of about 2 km around thenarrowest region of the Bab el Mandeb. The PerimNarrows (about 20 km wide at the sea surface) arewell resolved, but the resolution is insufficient tocapture realistically the steep and narrow topogra-phy at the bottom of the strait, in particular theNorthern Channel that is 5 km wide at the outletto the Gulf of Aden (see Section 1). The gridspacing increases from 4 to 20 km in the Gulf of
Aden (Fig. 7a), and is capable of capturing themesoscale eddies (scales �100 km) reported byBower et al. (2002). As the resolution around thehead of the Somali Peninsula is about 20 km, it issufficient for simulating the Great Whirl and theSocotra Gyre, along with the seasonal evolutionof the Somali Current along the African coast(Swallow and Bruce, 1966; Warren et al., 1966;Schott, 1983; Schott et al., 1997; Esenkov andOlson, 2002; Wirth et al., 2002). In the interior ofthe Indian Ocean, the large-scale currents driven bymonsoonal winds can be resolved with the averagegrid spacing of about 80 km (cf. Schott andMcCreary, 2001; Shankar et al., 2002). The Pacific,Atlantic and Southern Oceans are included at lowresolutions of 200–400 km and represent almostpassive absorbing boundaries. The CFL condition isvery severe at the finest mesh around the Bab elMandeb, giving a time step of 1.5min in thenumerical integration. Therefore the computationalcost of integrating the present model with 61 440ð¼ 240� 256Þ horizontal grid points for 15 years iscomparable to that of integrating a 11 globalOGCM with 64 800 ð¼ 360� 180Þ grid points for150 years with a time step of 15min.
The vertical grid of the present model is given bya hybrid terrain-following and z-level coordinate,called the ‘stepped sigma’ coordinate (Hanney,1991; Beckmann and Haidvogel, 1993; Ezer andMellor, 2004). This coordinate has the advantage ofpacking the numerical mesh near the sloped bottomwithout causing a significant error in the pressuregradient. The grid spacing is the smallest (8m) atthe sea surface, which monotonically increases tothe bottom with 30 elements. As a result, the HanishSill (160m deep) in the Bab el Mandeb is grideduniformly with the minimum vertical spacing (of8m), whereas at a mid-depth of 800m in the Gulf ofAden (where parcels of RSOW drift) the verticalresolution is about 80m; the vertical grid distribu-tion is shown in Figs. 6, 9b, 10b. The modeltopography of the sea floor is constructed from the1-min global bathymetric dataset of the GeneralBathymetric Chart of the Oceans. Over the regionsof the Southern and Northern Channel, we neededto modify the topographic data to represent thesteep outlet to the Gulf of Aden as shown in Fig. 1d.The model Northern Channel (between point A andB in Fig. 1d) lowers the bottom depth from 250 to900m. In contrast to the real bathymetry (e.g.Figure 1 of Peters et al., 2005 and Figure 1c ofBower et al., 2005), the route of the model Northern
ARTICLE IN PRESSH. Aiki et al. / Continental Shelf Research 26 (2006) 1448–14681452
Channel is almost straight. The model Channel hassomewhat taller side walls than that in the realtopography, as we discuss later in Section 3.2. TheSouthern Channel is not included in the presentmodel, so that the preset model is unable to simulatethe multi-layer structure of RSOW above theTadjura Rift (cf. Bower et al., 2005). The presentstudy is more concerned with the main body ofRSOW at intermediate depths 500–900m in theGulf of Aden, and its interaction with currents inthe northwestern Indian Ocean.
2.2. Formulation and forcing
The strong tidal effects near the Perim Narrowsand the Hanish Sill may change the characteristicsof the strait water (Jarosz et al., 2005a,b). Thus thepresent model includes the tide-generating potentialof its major four constituents (M2; S2; K1, and O1)and the nonhydrostatic pressure formulation ofMarshall et al. (1997). A prototype of the presentmodel was used in Aiki and Yamagata (2004) tosimulate the shedding of Meddy-like lenses from theMediterranean Sea. The bottom friction is calcu-lated from a standard quadratic equation. Thehorizontal eddy viscosity is given by Smagorinsky(1963) with a nondimensional coefficient of 0.1(after squaring). The vertical eddy viscosity is set100 cm2=s ð1 cm2=sÞ when the Richardson number issmaller (larger) than 0.5 to parameterize eitherKelvin–Helmholtz or Holmboe’s instability. ThePrandtl number is unity near the surface (shallowerthan 100m) as the upper and surface mixed layersare somewhat diabatic, whereas the Prandtl numberis set to 10 in the subsurface layer below 100mdepth to avoid excessive diffusion of RSOW.Ozgokmen et al. (2003) adopted a large Prandtlnumber of 20 in their nonhydrostatic numericalexperiments. The present model adopts no specialmeasure for the surface mixed layer and the bottomboundary layer. The flux-corrected transportscheme (Zalesak, 1979) is used for the advectionof tracers. The temperature and salinity fields of theWorld Ocean Atlas 1998 are used for the initialcondition with no motion: the velocity field is laterdetermined by geostrophic adjustment during theinitial steps of time integration. The model isintegrated for 15 years with the atmospheric forcingfrom monthly ECMWF (European Centre forMedium Range Weather Forecast) reanalysis windstress data and COADS (Coastal ObservationalAnd Data Ship) water and heat fluxes dataset.
There is no restoration to the temperature andsalinity climatologies at the sea surface.
As in Sofianos and Johns’s (2002, 2003) numer-ical study on the Red Sea circulation, we correct theCOADS flux data over the Red Sea by adding0.5m/year for evaporation–precipitation and60W/m2 for surface heat flux, respectively. Thesevalues are based on Sofianos et al.’s (2002) estimatesof the annual mean freshwater and heat fluxes overthe Red Sea: these are 2m/year and 11W/m2
whereas the COADS values are 1.5m/year and�40W/m2, respectively. The Knudsen formulaeused in Sofianos et al. (2002) dictate the freshwaterand heat budgets inside the Red Sea:
r1Q1S1 þ r2Q2S2 þ r3Q3S3 ¼ 0, (1)
ðE � PÞA ¼ Q1 þQ2 þQ3, (2)
FT ¼1
Ar0cpðQ1T1 þQ2T2 þQ3T3 � AðE � PÞTSÞ,
(3)
where ri, Qi Si, and Ti are the density, layertransport, salinity, and temperature for the surfacewater ði ¼ 1Þ, the intermediate water ði ¼ 2Þ, and thebottom water ði ¼ 3Þ through the Bab el Mandeb;E � P is the fresh water flux over the Red Sea; A isthe surface area of the Red Sea; FT is the heat lossover the Red Sea; r0 is the reference densityð1026 kg/m3
Þ; cp is the heat capacity of waterð3986 J �C=kgÞ; TS is the Red Sea surface tempera-ture. The overbar denotes the annual mean. Tocalculate E � P or FT , Sofianos et al. (2002) usedthe values of Qi, Ti, and Si given by Murray andJohns (1997) except for the surface layer transportQ1 (due to the large error in the observationalestimates). Note that the above correction isonly for the annual mean budget. More detailedcorrection including seasonal and spatial variationsis necessary but it is left for a future study(cf. Cromwell and Smeed, 1998).
3. Results of the reference experiment
Throughout the 15 years of simulation for thereference experiment, we released a passive tracerrepresenting the concentration of RSOW from theRed Sea. This tracer is in a statistically equilibriumstate during the last 5 years of the referenceexperiment (Fig. 2), which also suggests a meanresident time of about 2.3 years for RSOW in theGulf of Aden. Model results from the final year are
ARTICLE IN PRESS
RSOW inside Gulf
RSOW over the globe
tota
l am
ount
tota
l am
ount
year
(a)
(b)
Fig. 2. Total amount of the passive tracer (a) over the Gulf of
Aden and (b) over the entire region outside the Red Sea, during
the last 5 years of the reference experiment. Both ordinates are
nondimensionalized by 6:3072� 1015 m3 ð¼ 0:22Sv� 86400 s�
365dayÞ: an annual discharge amount of the RSOW tracer,
calculated from the bottom figure (i.e. the slope of the line in the
bottom figure is made unity).
H. Aiki et al. / Continental Shelf Research 26 (2006) 1448–1468 1453
shown in the followings to discuss the developmentof the subsurface and surface flows in the north-western Indian Ocean. Sections 3.1 and 3.2 willserve to validate the model results with observa-tions. Sections 3.3 and 3.4 focus on the three-dimensional distribution of the RSOW tracer in theGulf of Aden.
3.1. Currents at the sea surface
Upper layer currents around the Gulf of Aden aredisplayed with the vertical component of relativevorticity at the sea surface in Fig. 3. The simulatedlife-cycle of the Somali Current system is consistentwith previous reports (Swallow and Bruce, 1966;Warren et al., 1966; Schott et al., 1997; Schott andFischer, 2000; Esenkov and Olson, 2002; Wirthet al., 2002). The southwest monsoon (May–September) in the Indian Ocean drives the SomaliCurrent to flow northward along the African coast(Figs. 3a,b), which develops into the Great Whirland Socotra Gyre in late summer (Fig. 3c). TheRossby number of these intense currents and eddiesis large (above 0.5), attributable in part to the smallCoriolis parameter at low latitudes. The onset of the
northeast monsoon (October–April) results in thereversal of the Somali Current (Fig. 3d). It isinteresting that the Gulf of Aden is filled with anarray of mesoscale eddies throughout the year; theseare the Gulf of Aden eddies reported in Bower et al.(2002). These Gulf of Aden eddies are highlybaroclinic with respective to the main thermoclineat a depth of 300m and the lower part extends tothe bottom, as we explain in Section 3.4 (Fig. 10).Over the Bab el Mandeb Strait in Fig. 3, anenergetic inflow to the Red Sea is found in allmonths except September when the surface showsan outflow to the Gulf of Aden. The simulatedreversal of the surface current is consistent with theobservation of the surface layer outflow duringJuly–September (Sofianos et al., 2002). The seasonalcontrast of the water exchange through the strait isbetter exhibited in the sea surface salinity. In March(Fig. 4a) there is an inflow of fresher water from theIndian Ocean that extends to 171N inside the RedSea. In August (Fig. 4b) there is a counter current ofsalty Red Sea water along the African coast, whoseoutgoing edge forms an anticyclonic eddy in thewestern gulf. Although the local wind stress overthe Bab el Mandeb Strait is underestimated in thepresent study, the surface current through the Bab elMandeb Strait is shown to reverse seasonally. Thebottom outflow at the Bab el Mandeb will bedetailed in the next subsection.
3.2. The Bab el Mandeb Strait
The reference experiment is successful in reprodu-cing the seasonal transition of the water exchangethrough the Bab el Mandeb Strait. The along-straitvelocity at the Perim Narrows (Fig. 5a) shows atwo-layer exchange from October to June and athree-layer exchange during late summer. Thebottom outflow has a maximum speed of 120 cm/sduring March–April, which is comparable to theobserved speed in Figure 3 of Murray and Johns(1997) and Figure 3 of Sofianos et al. (2002). Inorder to estimate the volume transport of thebottom outflow, one needs to define the upperinterface of the bottom water. We have firstcalculated a cross-strait integral of the along-straitvelocity at each depth in the cross-section. Then, asin Sofianos et al. (2002), zero-crossing in the verticalprofile of the each-depth transport ðm2=sÞ is used toestimate the overflow rate. The reference experiment(see solid line in Fig. 11) gives a winter maximum of0.51 Sv and a summer minimum of 0.11 Sv with an
ARTICLE IN PRESS
Fig. 3. Seasonal evolution of relative vorticity (RV; the vertical component) at the sea surface during the 15th year of the reference
experiment. Positive (negative) values indicate cyclonic (anticyclonic) rotation, with units of the Coriolis parameter.
H. Aiki et al. / Continental Shelf Research 26 (2006) 1448–14681454
annual average of 0.36 Sv. Thus, even using thelarge-scale wind forcing as in the present study, weare able to reproduce a seasonal difference of 0.4 Svin the overflow rate of RSOW. The simulatedseasonal difference of 0.4 Sv is greater than boththeoretical and model estimates in Siddall et al.(2002) and Sofianos and Johns (2002). The overflowrate of the present study compares qualitatively withthe observations (Table 1), although the directobservations of Murray and Johns (1997) showeda short-term peak of 0.7 Sv in February and acomplete cessation of bottom overflow in August.
The discrepancies are on short time scalesð�1monthÞ and are possibly due to the limitedwind and buoyancy forcing applied in the presentmodel (Section 2.2). The lack of high-frequency(and also locally intensified) wind in the presentsimulation may be one of the reasons why themiddle layer velocity (during July–September) doesnot reach the bottom in Fig. 5a.
Temperature profile at the Perim Narrows high-lights intrusion of the Gulf of Aden intermediatewater in late summer. In Fig. 5b, a minimumtemperature of 18 1C is found at 160m depth during
ARTICLE IN PRESS
Fig. 4. Sea surface salinity (SSS, [psu]) around the Bab el Mandeb Strait in (a) March and (b) September, simulated in the 15th year of the
reference experiment.
H. Aiki et al. / Continental Shelf Research 26 (2006) 1448–1468 1455
August–September. Indeed such cold intrusion hasbeen observed by Saafani and Shenoi (2004). Thecold signal of the Gulf of Aden intermediate watercontinues northward across the Hanish Sill where thecore is warmed to about 22 1C (not shown). Alsocomparison between the temperature in Fig. 5b andthat in Table 1 is meaningful. For example in Augustin Fig. 5b, if one focuses on values of temperature atthe depths of the velocity maximum, one finds 31 1Cat the sea surface, 23–24 1C at 60m depth, and 22 1Cat 190m depth. These temperature values are withinan accuracy of �1 �C if compared with the values ofthe surface, middle, and bottom layer in Table 1 forsummer season; see Table 2 for a summary of thesimulated temperature, salinity, and density profilesat the Perim Narrows. In Figs. 5b–d, all oftemperature, salinity and density profiles are verti-cally homogenized in the vicinity of the upperinterface of the bottom water, which results fromenhanced diabatic mixing inside the Strait. While thediabatic mixing allows detrainment of RSOW tooverlying fluids, bottom salinity remains greater than39.5 psu throughout the year (Fig. 5c). The intrusionof the Gulf of Aden water is also exhibited byuplifted isopycnals during July–September in Fig. 5d.Again, if salinity and density values in Fig. 5 arepicked up at the depth of velocity maximum at anytime in (see Table 2), the simulation result provides agood comparison to the observed values in Table 1.
Once being ejected from the Perim Narrows, mostparcels of RSOW enter the model NorthernChannel and cascade downslope. As has been
observed by Peters et al. (2005), the bottom salinityis greater than 39.5 psu down to 500–700m depth inMarch when the overflow of RSOW is strongest(Fig. 6a). However, the overlying fluids are un-realistically simulated in salinity range between 39.0and 39.5 psu, distributing up to 250m depth andover the along-channel distances between 40 and100 km. This thick layer of the detrained RSOW isshaped by the sidewalls of the model NorthernChannel which is somewhat taller than the real one:the wall height is 400m at an along-channel distanceof 100 km in Fig. 6, whereas the height in the realtopography is less than 300m at Station 37 inFigure 1a of Peters et al. (2005). In August (Fig. 6b),the bottom core of RSOW retreats from theChannel leaving behind water parcels of decreasedsalinity. In summary, the present simulation is ableto reproduce bottom salinity and seasonal changesof RSOW in the Northern Channel. The model biasat 300–500m depth will not be serious for theanalysis of the RSOW tongue distributing below500m depth in the offshore region. Hopefully theexcessive dilution at the upper depths compensatesthe lack of the Southern Channel in the presentmodel.
3.3. Spreading of RSOW
The Gulf of Aden is filled with random energeticeddies (Fig. 3) and hence the distribution of RSOWincorporates year-to-year variability. Results ex-plained in the following are an example for one
ARTICLE IN PRESS
(a)
(b)
(c)
(d)
dept
h [m
]de
pth
[m]
dept
h [m
]de
pth
[m]
Perim velocity
Perim temperature
Perim salinity
Perim density
Fig. 5. Vertical profiles at the Perim Narrows for (a) along-strait
velocity (cm/s), (b) temperature (1C) , (c) salinity (psu), and
potential density ðkg=m3Þ, averaged over the last 3 years of the
reference experiment. These four quantities are taken at a deepest
point in the cross-section (point A in Fig. 1d), corresponding to
the measurements of Murray and Johns (1997) and Sofianos et al.
(2002). Positive (negative) values in Fig. 5a indicate an outflow
(inflow). Shade in each figure shows depths where the each-depth
transport ðm2=sÞ is negative (directed to the Red Sea).
Table 2
Same as Table 1 except for the result of the reference experiment
Smodel (psu) Tmodel (1C) rmodel ðsyÞ
Surface layer 36.6 36:437:1
� �28.1 27:2
30:9
� �23.5 23:7
23:0
� �Middle layer 36.8 �
36:8
� �25.5 �
25:5
� �24.5 �
24:5
� �Bottom layer 39.0 39:0
39:0
� �23.2 23:5
22:1
� �26.9 26:8
27:2
� �
Salinity, temperature, and density values in Fig. 5 are picked up
by using the transport-weighted-mean as in Sofianos et al. (2002).
The model’s winter (summer) mean values are from October to
June (July–September).
H. Aiki et al. / Continental Shelf Research 26 (2006) 1448–14681456
particular year. Fig. 7 shows the time evolution ofthe subsurface distribution of RSOW: the verticalintegral of the RSOW concentration (the passivetracer) is interpreted as an effective thickness ofRSOW. After being discharged from the Bab elMandeb, RSOW initially piles up in the TadjuraRift (the western edge of the Gulf) with a thicknessof 700m. In January (Fig. 7a), RSOW forms twolarge blobs (about 120 m thick) in the western Gulfof Aden. The main blob centered at (461E, 121N)spans the meridional extent of the gulf, and isassociated with anticyclonic rotation exhibited inFig. 7g. This large anticyclone is coupled with asmall cyclone to the west at (44.71E, 121N), whichconsists of a patch of low concentration RSOW: themixed product of RSOW and Indian Ocean Inter-mediate Water (IOIW). The second blob of RSOWis located along the African coast at (44.51E, 111N)in Fig. 7a, which is part of the southward-flowingbranch of newly ejected RSOW wrapping aroundthe cyclonic vortex mentioned above. This bound-ary stream of RSOW appears to leave the Africancoast at 451E to continue to the above-mentionedanticyclonic blob of RSOW. Meanwhile, the easternhalf of the Gulf of Aden contains a thin and large(about 400 km long and 150 km wide) tongue ofRSOW that is associated with the negative (antic-yclonic) relative vorticity (Fig. 7g). This indicatesthat the RSOW tongue is here on the south of anoutgoing current to the Indian Ocean as is a frontcomprising RSOW and IOIW.
In March (Fig. 7b), the RSOW tongue (100mthick) in the eastern gulf slightly rotates its axis tothe south while exhibiting a clear pair of relativevorticity (Fig. 7h), so that the intermediate layercurrent is now directed to Socotra Island. In thewestern gulf, the aforementioned main anticyclonicblob of RSOW is pushed northward to the Arabiancoast. At the western edge of the Gulf, the newly
ARTICLE IN PRESS
dept
h [m
]de
pth
[m]
distance [km]
March
August
A B
(a)
(b)
Fig. 6. Vertical distribution of salinity (psu) along the model Northern Channel (between point A and B in Fig. 1d) in (a) March and (b)
August, from the 15th year of the reference experiment. Contours show the vertical grid distribution with every five elements.
H. Aiki et al. / Continental Shelf Research 26 (2006) 1448–1468 1457
ejected stream of RSOW from the TadjuraRift remains following the African coast. In May(Fig. 7c), the body of RSOW exhibits a gulf-longstream, which meanders but extends over the wholerange of the main axis of the Gulf of Aden. Thismeandering suggests frontal instability at the sub-surface layers (Fig. 7i), which may lead to theisopycnal mixing. The intense anticyclonic rotationfound at (461E, 12.51N) is associated with the thickRSOW blob evident in Fig. 7c at the same position.Interestingly in July (Fig. 7d), as a result of thebreaking of the RSOW stream, the eastern gulfcontains a well-mixed product of RSOW and IOIWof about 100m thick. On the other hand, thewestern gulf contains two thick blobs of RSOW ofabout 200m thick, with the offshore one at (46.51E,121N) consisting of the anticyclonic part of adipolar vortex in Fig. 7j.
The end of the summer monsoon brings a drasticchange in the Gulf of Aden as most of RSOWparcels shift southward to the African side of theGulf of Aden (Fig. 7e). Interestingly, a large blob ofIOIW is seen on the Arabian side at (481E, 131N) in
Fig. 7e, indicating the intrusion of IOIW from theeast. Indeed, the time sequence during July–November (Figs. 7c–e) shows several patches ofIOIW translating westward along the Arabian coast.This westward progression replaces the western gulfwith IOIW (Fig. 7f), which in turn pushes the mainbody of the RSOW tongue eastward, towards theIndian Ocean. These movements of intermediatewater is part of a large cyclonic circulation at themid-depth (Fig. 7j). Results from the other yearsshow that the above-mentioned scenario of thereplacement of IOIW and RSOW does not apply toall years. Occasionally no major patch of IOIWintrudes into the Gulf of Aden in a year: in someyears a small patch of IOIW intrude from theSomalian Coast. Regardless of these year-to-yeardifferences, the total amount of RSOW is in astatistically equilibrium state (Fig. 2a). Detailedinvestigation is devoted to a future study.
The above results have demonstrated that, whilethe tongue meanders, the discharged RSOW (theincoming IOIW) is always characterized by antic-yclonic (cyclonic) vorticity. This may result from the
ARTICLE IN PRESS
Fig. 7. Time series of the reference experiment during the 15th year. (a)–(f) Vertical integral (below 26:5sy) of the passive tracer releasedfrom the Red Sea (which is initially unity), displaying an effective thickness (m) of the Red Sea outflow water (RSOW). (g)–(l) Relative
vorticity (RV; the vertical component) averaged between isopycnals of 27.0–27:5sy, around a depth of 800m. Positive (negative) values
indicate cyclonic (anticyclonic) rotation, with units of the Coriolis parameter. The contours in panel (a) show the horizontal resolution
(km) of the present model.
H. Aiki et al. / Continental Shelf Research 26 (2006) 1448–14681458
ARTICLE IN PRESSH. Aiki et al. / Continental Shelf Research 26 (2006) 1448–1468 1459
difference of potential vorticity, as has beensuggested for Meddies in the North Atlantic (Moreland McWilliams, 1997; Aiki and Yamagata, 2004).
3.4. Vertical profiles of RSOW
To investigate how RSOW leaves the coast andpenetrates into the offshore stratification, we haveprepared Fig. 8 which shows a vertical section alongthe main axis of the Gulf of Aden. The body of theRSOW tongue is distributed at depths between 600and 1000m, which is similar to the observation ofBower et al. (2002). The RSOW parcels mainly driftalong the 27.3–27:4sy isopycnal in Fig. 8, which issomewhat deeper than the observed isopycnal(27.15–27:35sy) of the salinity maximum (Mescha-nov and Shapiro, 1998). The density and depth ofthe simulated RSOW tongue are highly sensitive toboth the model topography of the Bab el Mandeband the parameterization of diapycnal mixing nearthe bottom (Ezer, 2005; Legg et al., 2006). Fig. 8shows two blobs of RSOW in the downstream at491E and 511E. These are associated with themeandering of the stream (Fig. 7c). The RSOW issubject to enhanced isopycnal mixing by the frontalinstability in the Gulf of Aden (where the concen-tration reduced to 0.2), whereas further upstreaminside the strait the RSOW is sheltered by theconfined topography and thus experiences lessmixing (the concentration of RSOW was 0.6 afterbeing discharged from the strait).
Fig. 8. Vertical distribution of the concentration of the Red Sea water (
oblique line in Figs. 7c and i. The dotted line shows isopycnal contour
North–south sections along 44.41E (Fig. 9) and501E (Fig. 10) are shown for salinity and (thevertical component of) relative vorticity. In thewestern side of the gulf (Fig. 9a), the dischargedhigh salinity water (over 37 psu) is well above thebottom and is here neutrally buoyant (cf. Boweret al., 2005). The relative vorticity in Fig. 9b exhibitsa bottom reaching cyclone associated with thedepression of RSOW in Fig. 9a near the northernboundary (around 12.11N). The thermocline islocated at 300m depth above which there is a pairof positive and negative relative vorticity. Fig. 10a,which is another vertical section crossing the tip ofthe RSOW tongue, displays a small core (about50 km wide) of salty water drifting at 800m depthwhich is surrounded by much diluted RSOWparcels distributed in parallel to the isopycnalsurfaces. This confirms the enhanced isopycnalmixing due to the swinging of RSOW tongue. Therelative vorticity distribution in Fig. 10b is quiteinteresting because it is dominated below thethermocline at 300m depth by a tall vorticity pairwhose zero-crossing (at 13.21N) is exactly aligned atthe aforementioned RSOW core drifting at the mid-depth. This tall vorticity pair is associated with theaforementioned eastward (outgoing) current to theIndian Ocean (Figs. 7g–i).
4. Sensitivity experiments on the seasonal overflow
The following subsections present the results of aseries of sensitivity experiments on the last 5 years of
shaded) and the potential density (dotted line) ðkg=m3Þ, along the
s.
ARTICLE IN PRESS
Fig. 9. Vertical distribution of (a) salinity (psu) and (b) relative vorticity (normalized by the Coriolis parameter) along 44.41E of Figs. 7c
and i. The dotted line in panel (a) shows isopycnal contours. The solid line in panel (b) shows the vertical grid distribution with every five
elements.
Fig. 10. Same as Fig. 9 except for the section along 501E.
H. Aiki et al. / Continental Shelf Research 26 (2006) 1448–14681460
the reference experiment, which we performed toelucidate how and why the Red Sea water overflowsseasonally through the Bab el Mandeb.
4.1. Wind or buoyancy forcing?
In order to verify whether the seasonal variationin the water exchange through the strait is due towind stress or atmospheric buoyancy (water andheat) fluxes, we have performed two sensitivityexperiments using different forcing. The first run(‘wind-run’ in Fig. 11) is driven by the monthlymean wind stress (as in the reference experiment)but with steady (i.e. annual mean) climatology heatand water fluxes. The result (Fig. 11a) is almostidentical to the reference experiment (abbreviated‘ref-run’), showing the seasonal variation in theoverflow rate with a winter maximum of 0.46 Sv anda summer minimum of 0.07 Sv. Another run (‘buoy-run’ in Fig. 11) has used the monthly mean heat andwater fluxes (as in the reference experiment) but
with a steady (i.e. annual mean) wind stress. Thisresults in little seasonal variability (Fig. 11a): theoverflow rate is about 0:38� 0:5 Sv throughout theyear. From these two results it can be concludedthat the summer stopping and the winter culmina-tion of RSOW transport is primarily due to theseasonal evolution of wind stress rather than toatmospheric buoyancy forcing. The dominant roleof wind stress is consistent with the previous studiesthat have attributed the seasonal water exchangeto wind stress applied above the Bab el Mandeb(cf. Thompson, 1939).
4.2. Where is the wind stress important?
We have next conducted three sensitivity experi-ments to identify which region of the wind stress isessential to the seasonal variability in the straitwater exchange. In the first experiment (‘red-run’),the monthly wind is applied only over the Red Sea(north of 141N), with other regions receiving the
ARTICLE IN PRESS
3 sensitivity experiments
2 + 1 sensitivity experiments
2 sensitivity experiments
over
flow
rat
e [S
v]ov
erfl
ow r
ate
[Sv]
over
flow
rat
e [S
v]
(a)
(b)
(c)
Fig. 11. Overflow transport (Sv) ð1 Sv ¼ 106 m3=sÞ of the Red
Sea water through the Bab el Mandeb Strait, through a cross-
section at point A in Fig. 1d. Details of the estimation are
provided in the first paragraph of Section 3.2. The solid line in
each panel shows the result of the reference experiment (‘ref-
run’). (a) Two sensitivity experiments (‘buoy-run’ and ‘wind-run’)
in Section 4.1. (b) Three sensitivity experiments (‘indian-run’,
‘aden-run’ and ‘red-run’) in Section 4.2. (c) Three sensitivity
experiments (‘arab-run’, ‘equa-run’ and ‘fpln-run’) in Section 4.3.
H. Aiki et al. / Continental Shelf Research 26 (2006) 1448–1468 1461
annual mean wind. The heat and water fluxes aremonthly over the whole computational domain (asin the reference experiment). For the second andthird experiments, the monthly wind is limited tothe Gulf of Aden (‘aden-run’) and the Indian Ocean(‘indian-run’), respectively (the two regions areseparated at 501E). Fig. 11b shows that ‘red-run’and ‘aden-run’ produce almost identical results,with less significant seasonal evolution in theoverflow rate. Interestingly, however, the monthlywind over the Indian Ocean (‘indian-run’) producesa maximum transport of 0.48 Sv in March–May(end of the northeast monsoon) and a minimumtransport of 0.24 Sv in August (end of the southwestmonsoon), which resembles the reference run. Theseresults suggest that the monsoonal wind in theIndian Ocean can remotely control the waterexchange through the Bab el Mandeb.
To elucidate the dynamic link between the IndianOcean and the Bab el Mandeb, we have nextinvestigated the depth of the thermocline in theArabian Sea and the Gulf of Aden. Fig. 12 showsthe depth of a 26:0sy isopycnal during the final yearof the reference experiment. Its depth is about 200min all months except September when the Gulf ofAden shoals in response to the southwest monsoon(cf. Luther and O’Brien, 1985; Bauer et al., 1991;Manghnani et al., 1998; Smith et al., 1998; Vecchiet al., 2004). The lifted isopycnal (about 120m deep)in the Gulf of Aden clearly continues to the Red Seathrough the Bab el Mandeb, which leads to blockingof the Red Sea outflow during this period. Thehydraulic theory of Siddall et al. (2002) has alreadyexplained that the Red Sea outflow minimizes whenan isopycnal in the Gulf of Aden is uplifted. Theshoaling is consistent with the intrusion of the Gulfof Aden intermediate water to the Red Sea (Section3.2). The late summer shoaling of the Gulf of Adenis also evident in the World Ocean Atlas. In themonthly climatology, the depth of a 26:0sy iso-pycnal is about 180m in February–March andabout 70m in August–September, which extendsover the whole region of the Gulf of Aden.
We have produced Fig. 13 which shows timeevolution of the depth of upper thermocline,averaged over the western Gulf of Aden, for allexperiments. The reference experiment (‘ref-run’,solid line), ‘wind-run’, and ‘indian-run’ demonstratea distinct seasonal cycle of 200m deep in March and130m deep in August–September. In contrast,without seasonal wind forcing in the Indian Ocean,‘buoy-run’, ‘red-run’, and ‘aden-run’ show little
ARTICLE IN PRESS
Fig. 12. Seasonal evolution of the upper thermocline depth (shaded, [m]), defined by a 26:0sy isopycnal, during the 15th year of the
reference experiment. Climatological wind stress vectors ðN=m2Þ are superimposed.
H. Aiki et al. / Continental Shelf Research 26 (2006) 1448–14681462
seasonal variability in thermocline depth. Theshoaling of the western Gulf of Aden is evidentlycaused by the monsoonal winds over the IndianOcean, instead of the local wind forcing. Eachtiming of the deepest depth (March) and shallowestdepth (August–September) matches that of max-imum and minimum outflow in Fig. 11. This resultis consistent with Siddall et al. (2002) in that theseasonal cycle of thermocline depth in the Gulf ofAden is in phase with that of the bottom wateroutflow at the Bab el Mandeb.
4.3. Which part of the Indian monsoon is important?
We have conducted two further sensitivity experi-ments to explore which part of the Indian monsoonis important for the seasonal water exchangethrough the strait (cf. Bruce et al., 1994; Senguptaet al., 2001; Brandt et al., 2002; Prasad et al., 2005;Rao and Behera, 2005). We have limited themonthly wind stress to the Arabian sea (north of101N) and then to the equatorial Indian Ocean(south of 101N), with a steady wind stress elsewhere
ARTICLE IN PRESS
2 sensitivity experiments
3 sensitivity experiments
2 + 1 sensitivity experiments
dept
h [m
]de
pth
[m]
dept
h [m
]
(a)
(b)
(c)
Fig. 13. Same as Fig. 11 except for the depth (m) of a 26.0syisopycnal, averaged over the western Gulf of Aden, west of 451E.
H. Aiki et al. / Continental Shelf Research 26 (2006) 1448–1468 1463
of each. In Fig. 11c, for the case of the seasonalwind over the Arabian sea (‘arab-run’), the overflowrate shows a maximum (0.47 Sv) in March and aminimum (0.28 Sv) in August, whereas the secondcase with variability in the equatorial Indian Ocean
(‘equa-run’) does not show such enhanced season-ality. Also Fig. 13c shows that the western Gulf ofAden shoals when wind stress changes seasonallyover the Arabian sea. These indicate that themonsoonal wind in the northern side of the IndianOcean is essential for the seasonal water exchange atthe Bab el Mandeb, as has been presumed by Jonesand Browning (1971).
In an attempt to confirm the above conclusion,we present an analytical theory for the coastalupwelling in the Gulf of Aden. We adopt a 1.5-layerplanetary geostrophic model on an f-plane:
�fhv ¼ �g0hhx þ tx=r0, (4)
fhu ¼ �g0hhy þ ty=r0, (5)
ht þ ðhuÞx þ ðhvÞy ¼ 0, (6)
where f is the Coriolis parameter, h and g0 are layerthickness and reduced gravity appropriate for thedefinition of a 1.5-layer model, ðu; vÞ and ðtx; tyÞ arethe eastward and northward component of velocityand wind stress, respectively. Taking curl of Eqs. (4)and (5) to obtain a form of velocity divergence, werewrite the continuity Eq. (6):
ht ¼ �ty
x � txy
f r0. (7)
An areal integral of Eq. (7) over the Gulf of Adenyields
ht ¼ �1
Bf r0
Ztyjx¼51�E dy, (8)
where B is the surface area of the Gulf, west of 511Ewhere the tip of the Somalian Peninsula locates. Theright-hand side of Eq. (8) is derived by assumingthat the line integral of wind stress along thenorthern, western, southern boundaries of the Gulf(Fig. 14a) is negligible. It follows from Eq. (8) thatthe seasonal shoaling of the thermocline in the Gulfof Aden is controlled by the seasonal variability inthe meridional wind in between the Gulf of Adenand the Arabian Sea. The theoretical rate ofshoaling and deepening is estimated in Fig. 14busing the wind stress of the reference experiment. Ifthis rate is integrated for a year, the thermoclinedepth can change seasonally �40m, which iscomparable to the simulated variability in Fig. 13.Also in Fig. 14b, the sign reverses in April–May andin September–October. Each timing lags behind theperiod of deepest and shallowest depth in thewestern Gulf in the reference experiment (Fig. 13)
ARTICLE IN PRESS
GULF OF ADEN
51E
(a)
(b)
Fig. 14. (a) Schematic of the route of line integral on the right-hand side of Eq. (8). (b) Theoretical rate of coastal Ekman down/upwelling
(positive/negative values) (m/month) over the Gulf of Aden, given by Eq. (8) for the wind stress of the reference experiment with
B ¼ 2:35� 1011 m2, f ¼ 3:0� 10�5 s�1, and r0 ¼ 1026:0kg=m3. Anomaly from the annual mean is shown.
H. Aiki et al. / Continental Shelf Research 26 (2006) 1448–14681464
by more than 1 month. This discrepancy suggestsimportance of other processes neglected in theabove simplified theory, such as impacts of subsur-face layers, fluxes through the Bab el Mandeb, andnongeostrophic dynamics: these examinations aredevoted a later study.
We have conducted a final sensitivity experiment(‘fpln-run’) by fixing the Coriolis parameter over theArabian marginal seas (the Arabian sea, the Gulf ofAden, and the Red Sea). This intends to support thef-plane assumption used in the above theory andalso to investigate the impact of Rossby waves on
the seasonal exchange at the Bab el Mandeb. Theresult (see Fig. 11c) turns out to be almost the sameas the reference run in that there is a wintermaximum and a summer minimum, even withoutthe effect of Rossby waves over the northwesternIndian Ocean. We have also checked if any patternof coastal Kelvin waves is found in the thermoclinedepth during the period of shoaling in the Gulfof Aden (cf. Shankar and Shetye, 1997). It turnsout that the thermocline shoals almost simulta-neously over both the Gulf of Aden and the coastalregions of the Arabian sea (not shown). This is
ARTICLE IN PRESSH. Aiki et al. / Continental Shelf Research 26 (2006) 1448–1468 1465
contrasted with the dynamics of subsurface layers:in Section 3.3 we show that patches of IOIWpropagate westward along the Arabian coast at themid-depth. The upper layer and intermediate layerseem to exert different time scales in the Gulf ofAden.
The analyses in Sections 4.1–4.3 have presentedboth numerical and theoretical evidences for ex-plaining how the Indian monsoon regulates remo-tely the late summer stopping of the Red Seaoutflow. The Red Sea outflow is regulated by theseasonal shoaling of the Gulf of Aden as a part ofthe Indian monsoon system. It may be possible in afuture study to deduce the interannual variability ofthe water exchange at the Bab el Mandeb fromreanalysis data of monsoonal wind over the IndianOcean.
4.4. Discussion
Some readers may find that the experimentaldesign of ‘ref-run’, ‘wind-run’, and ‘buoy-run’ in thepresent study is analogous to that in the E1, E4, andE5 experiments, respectively, in Sofianos and Johns(2002). A careful examination is needed for thedefinition of layer interface, in discussing the two-and three-layer exchange at the Bab el Mandeb. InSofianos and Johns (2002) layer interfaces aredefined by fixed density values, rather than thezero-crossing in velocity profile. Hence both thesurface and middle layer transports in their Figs. 7and 8 are directed to the Red Sea, and have littlerelevance to the three-layer exchange in observa-tional studies. Moreover, the surface layer inSofianos and Johns (2002) has densities less than25:5sy, which include the observed middle layerdensity of 24:8sy in Table 1. With insufficientvertical resolution for upper layers, isopycnal-coordinate models can produce misleading dy-namics of water exchange at the Bab el Mandeb.
Nonetheless, comparison with the results ofSofianos and Johns (2002) follows. The seasonaldifference in the bottom outflow is 0.4 Sv in ‘ref-run’and about 0.2 Sv in the E1 experiment, which maybe compared with observed estimates of 0.7 Sv( ¼ 0.7–0 Sv) in Murray and Johns (1997) and0.56 Sv ( ¼ 0.61–0.05 Sv) in Sofianos et al. (2002).While the ‘wind-run’ is successful in reproducing themain characteristics of ‘ref-run’, neither of the E4and E5 experiments reproduces the seasonal bottomoutflow of the E1 experiment. The E4 experimentreports reduction in the annual mean bottom
transport due to absence of seasonal cycle in thethermohaline forcing, which does not contradict thepresent result from ‘wind-run’. The above-men-tioned improvements in the present study over theresult of Sofianos and Johns (2002) are brought byfiner model resolution (both vertical and horizontal)and improvements in outer boundary condition.
The conclusion derived from the present studyconcerns only impacts of the large-scale wind. Thepresent model used the ECMWF wind whichsignificantly underestimates the local wind over theBab el Mandeb. We show that, even using the large-scale wind, one is able to reproduce a seasonaldifference of 0.4 Sv (‘ref-run’) in the overflow rateof RSOW, of which at least a seasonal differenceof 0.24 Sv (‘indian-run’) is caused by the monsoonalwinds over the Indian Ocean. Satellite windswill present an important comparison with theresult of the present study. Also devoted to a futurestudy is a simulation with increased horizontalresolution up to 0.5–1 km, which will enable a morerealistic simulation of the Northern and SouthernChannels, especially the nonhydrostatic formula-tion. Although integration time is limited to lessthan a few 10 days, several recent studies attemptsuch high resolution simulation (cf. Legg, 2004;Nakamura and Awaji, 2004).
5. Summary
The mechanism of seasonal cycle in the Red Seawater outflow has been investigated in the presentstudy using a locally high-resolution global OGCM,in which the horizontal resolution varied from 2kmaround the Bab el Mandeb Strait (the narrow exit ofthe Red Sea), to less than 20 km in the Gulf of Adenand about 80 km in the Indian Ocean. Although thenarrow topography of the Bab el Mandeb leavesseveral challenges (such as the insufficient resolutionof the present model to fully resolve the bottomchannels, and the excessive computational load dueto the CFL condition), the 15 years referenceexperiment has provided good simulations of thebasic features of the Red Sea outflow system,including the evolution of the RSOW tongue inthe Gulf of Aden and the summer minimum andwinter maxima in the overflow transport throughthe Bab el Mandeb. Moreover, we have exampledthe three-dimensional structure of the RSOWtongue (Section 3). Interestingly, while the tonguemeanders, the discharged RSOW (incoming IOIW)is always characterized by anticyclonic (cyclonic)
ARTICLE IN PRESSH. Aiki et al. / Continental Shelf Research 26 (2006) 1448–14681466
vorticity, as is suggested from the potential vorticitydifference. The RSOW is subject to enhancedisopycnal mixing by the frontal instability in theGulf of Aden (where the concentration reduced to0.2), whereas further upstream inside the strait theRSOW is sheltered by the confined topography andthus experiences less mixing (the concentration ofRSOW was 0.6 after being discharged from thestrait).
We have also examined why the overflow occursseasonally at the Bab el Mandeb (Section 4). Aseries of sensitivity experiments with different sets ofidealized atmospheric forcing have revealed that thesummer reduction of the Red Sea water overflow issensitive to the monsoonal wind over the ArabianSea, which we attribute to the following mechanism.The southwest (summer) monsoon blows north-eastward inside the Indian Ocean. The positive curlof the wind stress on the left of the wind’s axisinduces coastal Ekman upwelling, particularly overthe Gulf of Aden. The thermocline in the Gulf ofAden keeps shoaling until the end of the summermonsoon (May–September). The elevated thermo-cline in the Gulf of Aden can block overflow of theRed Sea water at the Bab el Mandeb, as is explainedby the hydraulic analysis of Siddall et al. (2002). Theabove mechanism has not been investigated byprevious OGCM studies. The present researchsuggests future field experiments focusing on theinteraction between the Red Sea water outflow andthe Indian monsoon system. These experimentsshould contribute to a deeper understanding of thehydrological cycle in the Arabian Sea.
Acknowledgments
The authors thank Drs. Yukio Masumoto,Swadhin Behera and Suryachandra Rao for valuablediscussions. All experiments have been performedusing the Earth Simulator. This paper is improvedby comments from five anonymous reviewers.
References
Aiki, H., Yamagata, T., 2004. A numerical study on the
successive formation of Meddy-like lenses. Journal of
Geophysical Research 109, C06020.
Bauer, S., Hitchcock, G.L., Olson, D.B., 1991. Influence of
monsoonally-forced Ekman dynamics upon surface layer
depth and plankton biomass distribution in the Arabian Sea.
Deep-Sea Research 38, 531–553.
Beal, L.M., Bryden, H.L., 1997. Observations of an Agulhas
undercurrent. Deep-Sea Research 44, 1715–1724.
Beckmann, A., Haidvogel, D., 1993. Numerical simulation of
flow around a tall isolated seamount. Journal of Physical
Oceanography 23, 1736–1753.
Bentsen, M., Evensen, G., Drange, H., Jenkins, A.D., 1999.
Coordinate transformation on a sphere using conformal
mapping. Monthly Weather Review 127, 2733–2740.
Bethoux, J.P., 1987. Variabilie climatique des echanges entre la
Mer Rouge et l’Ocean Indien. Oceanologica Acta 10,
285–291.
Bower, A.S., Hunt, H.D., Price, J.F., 2000. Character and
dynamics of the Red Sea and Persian Gulf outflows. Journal
