Was the GreatExample Salinity image Anomaly slide with smallerin the title1970s font Induced by an Extreme Fram Strait Sea Ice GHCNDExport? 1

0.75 Who Kim, Steve Yeager, & Gokhan Danabasoglu 0.5

0.25

2020 CESM Workshop 0 CVCWG Session Cumulative fraction of precipitation 10 20June 17 302020 40 50 Days Example figure with text TRMM 3B42 Thanks: Alper Altuntas (NCAR) 1

0.75

0.5

0.25

0 Cumulative fraction of precipitation 10 20 30 40 50 Days Great Salinity Anomaly of the 1970s

4 I.M. Belkin et al. /Progress in Oceanography 41 (1998) 1–68 (1) • There has been decadal-scale low salinity events in the subpolar North Atlantic (SPNA), first emerging in the sub-Arctic seas, entering the North Atlantic, and moving along the subpolar gyre • The most pronounced event: during the late 1960s and 1970s, called Great Salinity Anomaly (GSA) (2) • Conventional view of GSA (Dickson et al. 1988): 1) Enhanced Fram Strait sea-ice export (FSSIE) in the late 1960s 2) Freshwater anomaly advected to the Labrador Sea, (3) shutting down deep convection during 1969-1971 3) Continued to move following the subpolar gyre and returned back to sub-Arctic seas a decade later Belkin et al. (1998)

2020 CESM Workshop, CVCWG, W. M. Kim ([email protected]) 2

Fig. 1. A scheme by Ellett and Blindheim (1992, Fig. 6) showing transit dates for the minimal values of the “Great Salinity Anomaly” of the 1970s (by Dickson et al., 1988). The Krauss (1986) scheme of circulation has been modified in the Northeast Atlantic by Ellett and Blindheim (1992) after Meincke (1986) and others. T. Koenigk et al.: Variability of Fram Strait sea ice export: causes, impacts and feedbacks in a coupled climate model 25

JUNE 1999 HA¨ KKINEN 1785 Table 2 Same as Table 1 but positive salinity changes using years in the Labrador Sea, there would be a large 23 years with negative ice export anomalies exceeding 1.5 standard deviations influence on the entire thermohaline circulation (Jung- claus et al. 2004, in press). This might happen after large 3). Altogether, the excess ice export through the Den- Great Salinity Anomaly of the 1970sS>1 0.5< S<1 0< S<0.5 S<0 ice export anomalies that were shown by Mikolajewicz D D D D mark Strait was about 900 km3 during the 6-month pe- et al. (submitted) in model simulations with a regional riod, October 1968–March 1969. Timing of this fresh- Number years 13 6 1 3 coupled atmosphere–ocean–sea ice model. % 56.5 26.1 4.3 13.1 water pulse has a strong effect on convection down- The anomalous sea ice export through Fram Strait stream because in late fall–early winter, the new deep affects sea ice concentration in the Greenland (especially waters are formed through densification driven by ther- in the first year after the anomalies, not shown) and the • GSA has received attention because of a potential role of Arcticmeridional-origin overturning couldfreshwater be found in our model. in shuttingLabrador Sea 1–2 years later (Fig. 7 e)). The increase of mal fluxes with freshwater fluxes opposing the process. This agrees well with results of Power et al. (1994) but is sea ice in the Greenland Sea after high ice export events However, besides timing, the intensity of the pulse,down 900 deep convection contrary to findings of Ha¨ kkinen (1999). However, for is due to the direct effect of enhanced ice transport into km3 of freshwater in a few months, is equally important. longer timescales the surface salinity of the Labrador the Greenland Sea (Walsh and Chapman 1990; Wang Also noteworthy is that the Fram Strait ice export was Sea and the meridional overturning are highly positively and Ikeda 2000). In addition, increased ice export is much lower than average for several years before the correlated. It takes several years to consume the reser- associated with the anomalous advection of cold air • Several modeling studies support the conventional view voir of deep water in the Labrador Sea, which is formed from the north into the Greenland Sea. This leads to an 1968 event (H1993), which would make the 1968 export during such 1 year with strong deep convection. Thus, a intensified sea ice formation in winter and less melting in event nearly three times the preceding annual exports. single year with large or without deep convection has no summer. Sea ice cover increases in the Labrador Sea, With this guidance from the limited-area model, the significant influence on the meridional overturning. If because anomalous cold and fresh water is transported simulated ice export volume can be modulated in an salinity anomalies could persist for more than a few into the Labrador Sea after large ice export events. In Arctic–North Atlantic model by changing the ice shear Hakkinen (1999) 24 T. Koenigk etKoenigk al.: Variability of Framet Straital. sea(2006) ice export: causes, - Composite impacts and feedbacks in analysis a coupled climate using model CGCM viscosity. Thus we can arrive at an idealized simulation Fig. 8 Composite analysis for Fig. 9 Composite analysis for of the GSA and its effects downstream without requiring annual mean 10 m salinity late winter/early spring afull-blownlong-termsimulationwithinterannually anomalies in psu 1 and 2 years (February, March, April) mixed after high (left) and low (rightSalinity) layer depth anomalies in metres MLD varying forcing. The two simulations, GSA1 and GSA2, Artificially enhanced ice exports through Fram Strait 1 and 2 years after high (left) have an anomalous ice mass accumulation in the Green- and low (right) ice exports through Fram Strait land Sea (Fig. 4a) shown against the limited-area model FSSIE

ice mass anomaly (shifted in y axis). In H1993 the total ; Wang ice export from the Arctic was 1700 km3 above the 1990 equilibrium. In GSA1 and GSA2, the total ice export 3 from the Arctic is 3355 and 3207 km above the equi- e)). The increase of 7 3 summarises that low librium (of 2664 and 1907 km ,respectively)forthe 2 first year and is close to zero for the following years. , left). In the Greenland Sea, These exports are much larger than in the limited-area 9 ect of enhanced ice transport into

model, but most of this ice export does not exit to the ff subpolarJUNE 1999 gyre, as the excess ice export to the LabradorHA¨ KKINEN 1787 ). In addition, increased ice export is Sea is 1170 and 1240 km3 in GSA1 and GSA2, re- , in press). This might happen after large 2000 spectively during the 6-month period covering October– 2004 December of the first year and January–March of the Control The anomalous sea ice export through Fram Strait ects sea ice concentration in the Greenland (especially The fresh water input in the Labrador Sea after large

second year. These ice melt anomalies constitute about ff years in theinfluence Labrador on the Sea, entireclaus there thermohaline et circulation al. would (Jung- beice export a anomalies large thatet were al. shown (submitted) by in Mikolajewicz coupled model atmosphere–ocean–sea simulations ice with model. a regional a in the first year afterLabrador the Sea anomalies, 1–2 not shown) years and later the (Fig. Perturbed sea ice in the Greenlandis Sea after due high to ice the export directthe events e Greenland Sea (Walshand and Ikeda Chapman associated with thefrom anomalous the north advection into the ofintensified Greenland sea cold Sea. ice This formation air leads in to wintersummer. an and less Sea melting in ice coverbecause increases anomalous cold in and the freshinto Labrador water the Sea, is Labrador transported Sea after large ice export events. In 50% of the climatological freshwater (P 2 E)fluxto of surface watersdriven with by deeper atmospheric forcing layerindex), (e.g. plays waters, a high role or which as low well.ice is Table NAO- exports, through Framsalinity Strait, changes in that the lead Labrador to Sea. positive ice export events stabilises theocean. local Deep stratification of convection the suppressed is 1–2 years decreased after highdenced or ice by even exports. a This totally considerable isearly reduction spring evi- (February, in March, April) the mixedup late layer to 700 depth winter/ m in theconvection Labrador Sea. depth Compared to of the mean 1,500duced m, by almost mixed half (Fig. layerit depth is decreased is by re- 200 m.the This position mainly reflects of a deep changethere in convection. is a After deepening low in the iceincreased mixed exports, convection layer is depth and obvious thus in an thean Labrador interannual Sea. On timescale, notween significant correlation the be- salinity in the Labrador Sea and the <0

the subpolar gyre (estimated from Rasmusson and Mo >0 S S D D 1996). It will be seen that it is mainly the ice export ) but is <0 part entering the subpolar gyre that influences the ther- 1994 S <0.5 D S D mohaline circulation. Figures 4b,c show these exports ). However, for with the H1993 ice exports. The GSA runs capture well 0.5< are given in standard À 1999 S

the variability in ice exports as simulated by the limited- D 0.5 <1 0< À S

area model with more realistic wind forcing. D < kkinen ( S ¨ D ) but positive salinity changes using ) 1< 1 left À b. Horizontal circulation changes right 1 >1 0.5< S À D < ) within 3 years after positive Fram Strait ice The excess ice mass from the Greenland Sea enters S S ) ice exports D D the Labrador Sea during the period of deep mixing in ) and low ( right left Same as Table Composite analysis for wintertime, thus disrupting the convective processes that 2020 CESM Workshop, CVCWG, W. M. Kim ([email protected]) Distribution of the maximum year-to-year decrease in 3 are directly responsible for the formation of the LSW. Composite analysis for the ice export, through it, is probably underestimated. of surface waters with deeper layer waters, which is Number years% 13 6 56.5 26.1 1 4.3 3 13.1 23 years with negative icedeviations export anomalies exceeding 1.5 standard T. Koenigk et al.: Variability of Fram StraitTable sea 2 ice export: causes, impacts and feedbacks in a coupled climate model 25 meridional overturning could beThis agrees found well with in results our ofcontrary Power model. to et al. findings ( oflonger Ha timescales the surfaceSea and salinity the meridional of overturning the arecorrelated. highly Labrador It positively takes severalvoir years of to deep water consume in the theduring reser- Labrador such Sea, 1 which year is with formed strongsingle deep year convection. with large Thus, or a withoutsignificant deep convection influence has on no thesalinity meridional anomalies overturning. could If persist for more than a few Fig. 9 late winter/early spring (February, March, April) mixed layer depth anomalies in metres 1 and 2 years afterand high low ( ( through Fram Strait 24 T. Koenigk et al.: Variability of Fram Strait sea ice export: causes, impacts and feedbacks in a coupled climate model export events.exceeding Thirty-nine 1.5 standard deviations are years used.over The the salinity with is central averaged Labrador icedeviations Sea and export the anomalies Number years 19% 8 48.7 20.5 8 20.5 4 10.3 LSW is a part of the Deep Western Boundary Current FIG.4.(a)IcemassanomalyintheGSA1andGSA2experimentsFig. 8 annual mean 10 m salinity anomalies in psu 1 andafter 2 high years ( ice exports through Fram Strait Nevertheless, about 20% of large negative salinity driven by atmospheric forcingthe ice export, throughNevertheless, it, is about probablyanomalies 20% underestimated. cannot(e.g. of be explainedlies large by through prior Fram negative export Strait. anoma- They salinityinflow occurhigh through due to the fresh other water rador three Sea boundaries of or theeastern dueor boundary, Lab- which to is not freshlarge associated ice with waterlow exports previous inflow through Fram through Strait. The the local mixing NAO-Table 1 10 m salinity ( (DWBC), so an interruption in LSW formation will in- together with Ha¨kkinen (1993) anomaly (shifted in y axis). (b) anomalies cannot be explained by prior export anoma- index), plays a role as well. Table 2 summarises that low fluence the DWBC by reducing its transport. The ad- GSA1/2 ice export anomalies at Fram Strait and at (c) Denmark Strait. lies through Fram Strait. They occur due to fresh water ice exports, through Fram Strait, that lead to positive AthinsolidlinereferstoGSA1,athicksolidlinereferstoGSA2, inflow through the other three boundaries of the Lab- salinity changes in the Labrador Sea. justment to the new buoyancy forcing and changes in and a dashed line refers to H1993. rador Sea or due to fresh water inflow through the The fresh water input in the Labrador Sea after large the DWBC are propagated farther downstream through eastern boundary, which is not associated with previous ice export events stabilises the local stratification of the Rossby topographic waves (fast adjustment time, prop- large ice exports through Fram Strait. The local mixing ocean. Deep convection is decreased or even totally agation to the equatorial region occurs within a couple suppressed 1–2 years after high ice exports. This is evi- of months) (Hallberg and Rhines 1996) and by advec- denced by a considerable reduction in the late winter/ Table 1 Distribution of the maximum year-to-year decrease in early spring (February, March, April) mixed layer depth 10 m salinity (DS) within 3 years after positive Fram Strait ice up to 700 m in the Labrador Sea. Compared to the mean export events. Thirty-nine years with ice export anomalies convection depth of 1,500 m, mixed layer depth is re- exceeding 1.5 standard deviations are used. The salinity is averaged duced by almost half (Fig. 9, left). In the Greenland Sea, over the central Labrador Sea and the DS are given in standard deviations it is decreased by 200 m. This mainly reflects a change in the position of deep convection. After low ice exports, FIG.6.(a)Meridionaloverturningcellatyear15inexperiment2,contourintervalis2Sv.(b)–(d)MOCanomalies(experiment2–GSA2) DS< 1 10 À À À À there is a deepening in the mixed layer depth and thus an at years 5–7 as annual averages with contour interval of 0.5 Sv. Time evolution of the anomalous MOC with flow referenced to density at increased convection is obvious in the Labrador Sea. On sigma-1 5 32.3 for GSA1 (dashed) and GSA2 (solid) (in Sv) (e) at 258Nand(f)at458N(abinomialfilterusedoncetosmooththedata). Number years 19 8 8 4 % 48.7 20.5 20.5 10.3 an interannual timescale, no significant correlation be- tween the salinity in the Labrador Sea and the adecreaseof1.0–1.5Svatmaximum(Fig.6e),which to the decreased production of LSW. North of the lat- lasts only a couple of months during year 5. The largest itude band, the simultaneous Gulf Stream–North Atlan- deviations in GSA2 are about 3–4 Sv, which dominate tic Current weakening compensates the changes in the summers of simulation years 5–7. This magnitude DWBC zonally. South of 158NtheMOCchangesin change is about 20% of the standard simulation over- GSA2 are smaller but do extend to the southern bound- turning cell. At 458N, closer to the source of the dis- ary, which will change the whole basin quantities such turbance, the deviations are much larger (Fig. 6f), reach- as the salt content (section 3f). In GSA1 the response ing over 6 Sv in both runs. The return of convection is is limited to the subpolar gyre and midlatitudes. The marked by negative anomalies at year 4, which linger influence of the GSA event on the NADW overflow was much longer in GSA1. The reasons for the return of insignificant in both of the simulations: in the control convective conditions are discussed in the section 3d. runs the overflow is 6.5–6.6 Sv, while at the end of the The anomalies in the z level and density level MOC are GSA1 and GSA2 runs there is a decrease less than 0.2 similar in both of the cases for the latitude band 158N– and 0.5 Sv, respectively. These amounts constitute less 308N; however, farther north the z-level MOC anomaly than 7% change compared to the control runs. does not register the considerable changes in the LSW It is apparent from Fig. 6c that MOC of GSA2 did volume transport. The midlatitude response can be in- not return to the strength of the parallel run but is about terpreted to reflect decreased upwelling of deep waters 0.5–1.0 Sv weaker. Lenderink and Haarsma (1994) and (mainly LSW) to the thermocline, which diminishes due Rahmstorf (1995a,b) have discussed the sensitivity of Motivations

• However, the winter NAO was overall negative during 1960s and was the record low in 1969 (when the shutdown of deep convection occurred) • NAO-related surface buoyancy forcing predominantly controls the strength of deep convection, thermohaline circulation (buoyancy-driven AMOC and subpolar gyre circulation), and thereby northward transport of heat and salt from subtropics.

1969

2020 CESM Workshop, CVCWG, W. M. Kim ([email protected]) 4 Motivations

• However, the winter NAO was overall negative during 1960s and was the record low in 1969 (when the shutdown of deep convection occurred) • It is well known that NAO surface buoyancy forcing predominantly controls the strength of deep convection, thermohaline (AMOC) circulation, and thereby northward transport of heat and salt from subtropics. • No modeling studies have so far systematically compared the relative contribution of FSSIE and NAO buoyancy forcing to GSA Ø Gelderloos et al. (2012) found roughly equal contributions using observations and 1-D mixed layer model • Also, the models used in previous studies (mostly early 2000s) are almost two decades old

2020 CESM Workshop, CVCWG, W. M. Kim ([email protected]) 5 Model Simulations (CESM)

• CESM1 (LENS) preindustrial control simulation (2200 year long) − Later 1800 years are used for composite analyses • 1° and 0.1° forced ocean – sea-ice simulations (FOSI-L and FOSI-H) − Forcing: JRA55-do (1958-2018; Tsujino et al. 2018) − FOSI-L: Long spin-up cycles (5 times) and 5th cycle is used for analysis − FOSI-H: Only the first cycle is available (possible drift) − anomalies are relative to the 1962-1976 reference period • CESM1 NAO surface heat flux forcing experiments (Kim et al. 2020) − Used to investigate the role of a decade long NAO surface buoyancy forcing • CESM1 physics- and initial condition-perturbed experiments (Danabasoglu et al. 2019) − Used to examine the dependency of the results on temperature bias in the Labrador Sea

2020 CESM Workshop, CVCWG, W. M. Kim ([email protected]) 6 Confidential manuscript submitted to Journal of Climate Simulated GSA in FOSIs

4000

(a) FOSI-L 3500 FOSI-H FSSIE ]

-1 3000 yr 3 2500 FOSI-L FOSI-H Obs

2000

FSSIE [km 3 Average 3 3 2220 km /yr 1500 2470 km /yr 1950 km /yr (1991-1998) (Kwok et al. 2004) 1000 3 (b) 1967-1968 3 3 2300 km 34.7 Ishii + 0.15 2070 km 1960 km Anomalies (Vinje 2001)

34.6

34.5 Salinity [psu]

34.4

1500 (c) Lazier (1980) 1200 Avg 900

600 Mar. MLD [m]

300 Lab Sea Sea Lab

500 4 (d) 400 2 ] -2

300 0

200 -2 DJFM NAO JFM SHF [W m 100 -4

0 -6 846 1966 1968 1970 1972 1974 1976

847 Figure 1. Time series of (a) annual FSSIE, and (b) annual upper 100 m salinity, (c) March MLD,2020 CESM Workshop, CVCWG,and W. M. Kim ([email protected]) 7 848 (d) JFM surface heat loss (SHF) averaged over the interior LS from FOSI-L (blue bars) and FOSI-H 849 (red bars) around the GSA (1965-1976) period. Overlaid in (b-d) with the black line are the upper 850 100 m salinity in the interior LS from the hydrography data by Ishii (2009), MLD estimated from 851 hydrography data at OWS-B (Lazier, 1980), and station-based DJFM NAO index by Hurrell (1995), 852 respectively. In (b), the observed salinity is plotted with a +0.15 offset. The interior LS is defined by 853 the region where the ocean depth is greater than 2000 m between 51°-60°N and 42°-60°W.

38 Composite Analysis Salinity (<100m; shading)/Mar. MLD (contours) Anomaly Composites

Jan-Jun (t=0) > 1.8 s.d. & Jan-Mar (t=-1) > 0

Jan-Mar (t=0) > 1.8 s.d. & Jan-Mar (t=+1) > 0

2020 CESM Workshop, CVCWG, W. M. Kim ([email protected]) 8 Freshening due to Suppressed Convection

Q ➝ Tu < Td Monthly Lab Sea Salinity Anomaly Profile Tu, Su T F FOSI-L 1968-1969 FOSI-H 1968-1969 SHF Composite (t=-1 to 0) Td, Sd Fresher Fresher Fresher

Saltier Saltier Saltier

Mean

Qa

u u Ta , Sa Ta Fa

d d Ta , Sa

u u Sa < S d d Sa > S Weaker Mixing

2020 CESM Workshop, CVCWG, W. M. Kim ([email protected]) 9 Downstream Advection Salinity (<100m) 2-yr Running Averages

2020 CESM Workshop, CVCWG, W. M. Kim ([email protected]) 10 Downstream Advection Salinity (<100m) 2-yr Running Averages

Lab Sea -NAO Heat Flux Forcing Experiments (a) NAO SHF (DJFM) (b) Total SHF (DJFM) (c) MLD (FM)

70°N

60°N

50°N

72°W 54°W 36°W 18°W 0° 72°W 54°W 36°W 18°W 0° 72°W 54°W 36°W 18°W 0° • Fully coupled CESM simulations • 10 ensemble-200 members -150 -100 -50 0 50 100 150 200 -800-600-400-200 0 200 400 600 800 [W/m 2 ] [m] • 20-year simulations with the forcing (d)applied DJFM for SHF the firstDifference 10 (e) FM MLD Difference years (winter only) • Originally performed to study 300 0AMV mechanisms and climate 200

] impacts (Kim et al. 2020, -2 https://doi.org/10.1175/JCLI-D-19- -50 0530.1) 100 [m] [W m

-100 0 NAO SHF Total SHF 2020 CESM Workshop, CVCWG, W. M. Kim ([email protected]) 11 -100 5 10 15 20 5 10 15 20 Model Year Model Year Confidential manuscript submitted to Journal of Climate

(a) FSSIE Composite 2.4

1.6

0.8

0 [Sv]

-0.8

-1.6

-2.4 -4 -2 0 2 4 6 8 t

(b) -NAO Exp. 2.4

1.6

0.8

0 [Sv]

-0.8

-1.6

-2.4 1 3 5AMOC 7 9 11 13 15 Simulation year

(c) FOSIs 2.4

FOSI-L 1.6 FOSI-H

0.8

0 [Sv]

-0.8 Confidential manuscript submitted to Journal of Climate

-1.6 Confidential manuscript submitted to Journal of Climate

-2.4 (a) FSSIE Composite 2.41964 1966 1968 1970 1972 1974 1976 907 (a) FSSIECalendar Composite Year 1.62.4 908 Figure 11. Anomalies of annual maximum AMOC strength at 45°N from (a) the FSSIE event 0.81.6 909 composite for t = –4 to t = 7 (3 years before to 7 years after the main event year), (b) the –NAO 910 experiments during the simulation years 1 to 14, and (c)0.80 FOSIs for the 1964-1975 period. The [Sv]

911 anomalies of FOSIs are relative to the respective 19650 -1979 average.

[Sv] -0.8

-1.6-0.8

-2.4-1.6 -4 -2 0 2 4 6 8 -2.4 t -4 -2 0 2 4 6 8 (b) -NAOt Exp. 2.4 (b) -NAO Exp. 1.62.4

0.81.6 48

0.80 [Sv]

0

[Sv] -0.8

-0.8 -1.6

-1.6 -2.4 1 3 5 7 9 11 13 15 -2.4 Simulation year 1 3 5 7 9 11 13 15 12 2020 CESM Workshop, CVCWG,Simulation(c) FOSIs W. year M. Kim ([email protected]) 2.4 (c) FOSIs 2.4 FOSI-L 1.6 FOSI-H FOSI-L 1.6 0.8 FOSI-H

0.8 0 [Sv]

0 [Sv] -0.8

-0.8 -1.6

-1.6 -2.4 1964 1966 1968 1970 1972 1974 1976 -2.4 Calendar Year 907 1964 1966 1968 1970 1972 1974 1976 907 Calendar Year 908 Figure 11. Anomalies of annual maximum AMOC strength at 45°N from (a) the FSSIE event 908909 Figure 1composite for 1. Anomaliest = –4 to of annual maximum AMOC strength at 45t = 7 (3 years before to 7 years after the main event year)°N from (a) the FSSIE event , (b) the –NAO 909910 composite for experiments during the simulation years 1 to 14, and (c)t = –4 to t = 7 (3 years before to 7 years after the FOSIs main event year)for the 1964-197, (b) the 5 period.– NAO The 910911 experiments during the simulation years 1 to 14, and (c)anomalies of FOSIs are relative to the respective 1965-1979 average. FOSIs for the 19 64-1975 period. The 911 anomalies of FOSIs are relative to the respective 1965-1979 average.

48 48 Revised GSA Mechanisms

4 I.M. Belkin et al. /Progress in Oceanography 41 (1998) 1–68 (1) 1) Enhanced Fram Strait sea-ice export (FSSIE) in the late 1960s 2) Freshwater anomaly advected to the Labrador Sea, shutting down deep convection during 1969-1971 2) The shutdown of deep convection and freshening in the interior Labrador Sea due to strong anomalous heat gain 3) FSSIE-induced fresh anomaly propagated along the boundary currents to the gyre boundary, but (2) (4) dissipated there (3) 3) Continued to move following the subpolar gyre and returned back to sub-arctic seas a decade later 4) The propagation of the fresh anomaly along the subpolar gyre coming from the subtropics due to Belkin et al. (1998) weaker thermohaline circulation

2020 CESM Workshop, CVCWG, W. M. Kim ([email protected]) 13

Fig. 1. A scheme by Ellett and Blindheim (1992, Fig. 6) showing transit dates for the minimal values of the “Great Salinity Anomaly” of the 1970s (by Dickson et al., 1988). The Krauss (1986) scheme of circulation has been modified in the Northeast Atlantic by Ellett and Blindheim (1992) after Meincke (1986) and others. Climatologies

March Sea-Ice Thickness March MLD

Figure S1. Geographical information on the LS domain, and the Fram and Denmark Straits used in this study. They are denoted as the red box and red strait lines, respectively, and labeled as LS, FS, and DS. The FS and DS sections shown here are along the model grid lines from FOSI-H and the sections in FOSI-L have slightly different orientations. Also shown in colors is the bathymetry (m) from FOSI-H with 2000 m contours in black. The interior LS defined in this study is the region where the ocean depth is greater than 2000 m within the LS box. Figure S5. Climatological March MLD from (a) FOSI-L, (b) CESM1-PI, (c) FOSI-H. The climatological periods are 1965-1979 in (a,c) and entire simulation years (400-2200) in (b). For reference, observational MLD climatology (IFREMER) of de Boyer Montegut et al. (2004) is also shown in (d). Note that the definition of MLD in IFREMER, based on a surface-to-depth density difference, is slightly different from the one used in the simulations, which is based on a buoyancy gradient E1 2020 CESM Workshop, CVCWG, W. M. Kim ([email protected] (Large et al. 1997). ) FSSIE 4000 FOSI-L FOSI-H

3000 /yr] 3 [km

2000

5

1000 1960 1970 1980 1990 2000 2010 Figure S2. Time series of annual FSSIE from FOSI-L (blue) and FOSI-H (red) for the full simulation period (1958-2018).

2 JFM SHF (shading)/SLP (con) Anomaly Composite

Confidential manuscript submitted to Journal of Climate

Individual MLD anomalies of the FSSIE events

(a) r = -0.26 (b) r = -0.81 600

400

200

0

-200 MLD Anom [m] -400

-600

-800 0 1000 2000 3000 4000 -150 -100 -50 0 50 100 150 3 -2 883 FSSIE Anom [km ] SHF Anom [W m ] 884 Figure 7. Scatter plots of interior LS MLD anomalies as a function of (a) the FSSIE and (b) LS 885 surface heat flux anomalies for the extreme FSSIE events, with the composite means indicated by 886 diamonds. The MLD and surface heat flux anomalies are the average over the next two years after 887 the main FSSIE event year (t = 1 to 2) and the FSSIE anomalies are accumulated over the main event 888 and the preceding years (t = –1 to 0). The linear regression line (black line) and regression 889 coefficient are also shown in each panel.

2020 CESM Workshop, CVCWG, W. M. Kim ([email protected]) E2

44 JUNE 1999 HA¨ KKINEN 1785

3). Altogether, the excess ice export through the Den- mark Strait was about 900 km3 during the 6-month pe- riod, October 1968–March 1969. Timing of this fresh- water pulse has a strong effect on convection down- stream because in late fall–early winter, the new deep waters are formed through densification driven by ther- mal fluxes with freshwater fluxes opposing the process. However, besides timing, the intensity of the pulse, 900 km3 of freshwater in a few months, is equally important. Comparison to Hakkinen (1999) Also noteworthy is that the Fram Strait ice export was much lower than average for several years before the 1968 event (H1993), which would make the 1968 export event nearly three times the preceding annual exports. Sea-Ice Export Total Fw Export With this guidance from the limited-area model, the

] 500 (a)

simulated ice export volume can be modulated in an -1 Fram Arctic–North Atlantic model by changing the ice shear Hakkinen (1999) 400 Denmark viscosity. Thus we can arrive at an idealized simulation mon 3 of the GSA and its effects downstream without requiring 300 afull-blownlong-termsimulationwithinterannually 200 varying forcing. The two simulations, GSA1 and GSA2, have an anomalous ice mass accumulation in the Green- 100 land Sea (Fig. 4a) shown against the limited-area model 0 ice mass anomaly (shifted in y axis). In H1993 the total ice export from the Arctic was 1700 km3 above the -100 equilibrium. In GSA1 and GSA2, the total ice export FSSIE Composite [km -200 from the Arctic is 3355 and 3207 km3 above the equi- t=-1 t=0 t=+1 librium (of 2664 and 1907 km3,respectively)forthe first year and is close to zero for the following years. 500 (b) These exports are much larger than in the limited-area 400 model, but most of this ice export does not exit to the ] -1 subpolar gyre, as the excess ice export to the Labrador 300 Sea is 1170 and 1240 km3 in GSA1 and GSA2, re- mon spectively during the 6-month period covering October– 3 200 December of the first year and January–March of the 100 second year. These ice melt anomalies constitute about 50% of the climatological freshwater (P 2 E)fluxto 0 FOSI-L [km the subpolar gyre (estimated from Rasmusson and Mo 1996). It will be seen that it is mainly the ice export -100 part entering the subpolar gyre that influences the ther- -200 mohaline circulation. Figures 4b,c show these exports with the H1993 ice exports. The GSA runs capture well 500 (c) the variability in ice exports as simulated by the limited- 400 area model with more realistic wind forcing. ] -1 300 mon b. Horizontal circulation changes 3 200

The excess ice mass from the Greenland Sea enters 100 the Labrador Sea during the period of deep mixing in 0 wintertime, thus disrupting the convective processes that FOSI-H [km are directly responsible for the formation of the LSW. -100 LSW is a part of the Deep Western Boundary Current FIG.4.(a)IcemassanomalyintheGSA1andGSA2experiments (DWBC), so an interruption in LSW formation will in- together with Ha¨kkinen (1993) anomaly (shifted in y axis). (b) -200 fluence the DWBC by reducing its transport. The ad- GSA1/2 ice export anomalies at Fram Strait and at (c) Denmark Strait. 1967 1968 1969 AthinsolidlinereferstoGSA1,athicksolidlinereferstoGSA2, justment to the new buoyancy forcing and changes in and a dashed line refers to H1993. the DWBC are propagated farther downstream through Figure S9. Monthly anomalies of total (solid plus liquid) freshwater export through the Denmark Rossby topographic waves (fast adjustment time, prop- (red line) and Fram (gray bars) Straits from (a) the FSSIE event composite for t = –1 to t = 1 (1 year agation to the equatorial region occurs within a couple before and after the main event year), (b) FOSI-L, and (c) FOSI-H for the corresponding period of months) (Hallberg and Rhines 1996) and by advec- 2020 CESM Workshop, CVCWG, W. M. Kim ([email protected](1967-1969). The monthly anomalies of FOSI) -L and FOSI-H are relative to monthly means for the E3 1965-1979 reference period. The liquid freshwater export across the straits (Fig. S1) is estimated for the upper 200 m only as − ∬ #(% − %&'( )/%&'(+,+-, where # is the ocean velocity perpendicular to the straits, % is salinity, and %&'( is the section mean salinity.

8 Dependency on Lab Sea Temperature Bias

• Because of the nonlinearity of the equation of state, a density change is greater at colder temperature and Lab Sea salinity bias is positive in CESM1 • Similar composite analysis performed from physics- and initial condition-perturbed experiments (8 experiments) using the same CESM1 (Danabasoglu et al. 2019)

Confidential manuscript submitted to Journal of Climate

Composite Analysis from Perturbed Experiments

(a) FSSIE MLD Composite vs. T Bias (b) SHF MLD Composite vs. T Bias 100 -200

y = 65.22x - 139.6, r = 0.41 y = 84.86x - 446.1, r = 0.69

50 -250

0

-300

-50 MLD Composite [m] -350 -100

-150 -400 1 1.2 1.4 1.6 1.8 2 1 1.2 1.4 1.6 1.8 2 912 T Mean Bias [°C] T Mean Bias [°C] 913 Figure 12. Scatter plots of interior LS MLD composite anomalies against the upper 400 m salinity 914 biases in the interior LS for (a) the FSSIE events and (b) the LS surface heat flux events from the 915 CESM1 sensitivity experiments. The biases are relative to the climatology from WOA13. The values 2020 CESM Workshop, CVCWG, W. M. Kim ([email protected]) E4 916 from the three 600-year segments of CESM-PI are indicated by bold markers. The regression line 917 (black line), regression coefficient, intercept, and correlation coefficient are shown in each panel.

49 Journal of Geophysical Research: Oceans 10.1002/2015JC011290

5. Discussion 5.1. Pathways of Greenland Freshwater Dukhovskoy et al. (2016) 5.1.1. Horizontal Propagation The pathways of Greenland freshwater − Realistic Greenland meltwater is released along with a propagation follow the general circula- tion in the sub-Arctic seas (Figure 1). passive tracer in three different models (1/12, 1/4 and The tracer experiments presented 1/2º) demonstrate relatively rapid spreading of the passive tracer in the sub-Arctic − Only 1-2% of the passive tracer ends up in the interior seas within 5–7 years. Estimated travel Lab Sea time of the tracer in the model experi- ments agrees well with the estimates of propagation rates of the GSAs in the sub-Arctic inferred from observations [Dickson et al., 1988; Belkin et al., 1998; Belkin, 2004; Yashayaev and Seidov, 2015]. Based on the observed cooling and freshening signals in the region, Dickson et al. [1988] traced the spread- ing of the 1970s GSA and suggested that it took the GSA 7–8 years to prop- agate from Fylla Bank (off the south- western Greenland coast) to the Norwegian Sea and 9–10 years to Spitsbergen. Upstream, propagation timescales for the GSA were 1 year lon- ger from the eastern Greenland coast (north of Iceland). The GSAs of the 1980s and the 1990s spread faster propagating from Fylla Bank to the Norwegian Sea in about 5–6 years [Bel- kin et al., 1998; Belkin, 2004]. The major inflow of the tracer into the Nordic Seas in AO-HYCOM and ICMMG occurs with the North Atlantic Current. This result is in agreement with other stud- ies where the dominant role of nega- tive salinity anomalies carried by the Figure 14. Mass fraction (%) of the volume-integrated tracer in three regions (NS, LS, and BB) relative to the total tracer mass integrated over the model domain Atlantic inflow on the freshening sig- (limited by 388N for NEMO-LIM2 and ICMMG). (a) AO-HYCOM; (b) NEMO-LIM2; nal in the Nordic Seas has been pro-  2020 CESM Workshop, CVCWG, W. M. Kim(c) ICMMG. ([email protected] Shown are mass fraction) values after 3, 6, 9, 12, and 14 years of the posed [e.g., Glessmer et al., 2014; E5 simulation. The colored bars inside the ‘‘Nordic Seas’’ and ‘‘Labrador Sea’’ illustrate the mass fraction of the tracer integrated over the interior boxes (blue—GS, red— Reverdin, 2014]. IS, and green—IL) relative to the total tracer mass (the values are listed in colored numbers). Also listed is the total mass fraction of the tracer within the three The discrepancies in the numerical sol- regions. utions are apparent in tracer propaga- tion into the interior regions of the sub-Arctic seas. There are markedly stronger horizontal gradients in tracer concentration fields simulated in ICMMG, compared to the other two models. In the coarse-resolution ICMMG simulation, the interior slowly fills with the tracer that tends to remain with the current following the boundary regions (Figure 7). For example, in ICMMG the tracer is advected into the Nordic Seas after 5 years of the simulation in agreement with the AO-HYCOM experiment, yet it takes another 7–8 years for the tracer to spread into the interior GS and IS in the ICMMG simulation compared to 2–3 years in the AO-HYCOM experiment (Figure 5). One possi- ble reason for the observed discrepancies in the tracer distribution between ICMMG and the other two

DUKHOVSKOY ET AL. GREENLAND FRESHWATER IN THE SUB-ARCTIC 894