Silicon isotopes in an EMIC’s ocean: sensitivity to

19 The isotopic composition of Si in biogenic silica (BSi), such as opal buried in the oceans’ 20 sediments, has changed over time. Paleo records suggest that the isotopic composition, 21 described in terms of δSi, was generally much lower during glacial times than today. 22 There is consensus that this variability is attributable to differing environmental con23 ditions at the respective time of BSi production and sedimentation. The detailed links 24 between environmental conditions and the isotopic composition of BSi in the sediments 25 remain, however, poorly constrained. In this study, we explore the effects of a suite of 26 offset boundary conditions during the LGM on the isotopic composition of BSi archived 27 in sediments in an Earth System Model of intermediate complexity. Our model results 28 suggest that a change in the isotopic composition of Si supply to the glacial ocean is suf29 ficient to explain the observed overall low(er) glacial δSi in BSi. All other processes 30 explored triggered model responses of either wrong sign or magnitude, or are inconsis31 tent with a recent estimate of bottom water oxygenation in the Atlantic Sector of the 32 Southern Ocean. Caveats, mainly associated with generic uncertainties in today’s pelagic 33 biogeochemical modules, remain. 34

The wind forcing during LGM consists of the same prescribed climatology (NCEP) 195 used for the preindustrial simulations in our rather simple ESCM (see Weaver et al., 2001, 196 for details). Rather weak dynamic wind feedbacks are parameterized as a function of sur-197 face temperature gradients and added to this climatology. The choice of building on a 198 present day climatology is pragmatic, since the wind conditions during the LGM are dis-199 cussed controversially (e.g. Kohlfeld et al., 2012;Sime et al., 2013).
where T denotes the spatial divergence of diffusive and advective transports. sms refers 208 to the source-minus-sinks term. The sms terms of the silicon module are adopted from where r denotes the diatom opal dissolution rate and P Si denotes the biogenic opal 220 production. r is temperature dependent: with the parameters A setting the dissolution rate and T c determining the tem-222 perature dependance of opal dissolution. The value of T c is adopted from Gao et al.

223
(2016) (see Table 1). The value of A has been determined in a tuning exercise (see 224 Section 2.2.2), P Si is calculated as a function of the production rate of particu- where, R Si:P denotes the molecular DSi to phosphate uptake ratio, associated with 228 BSi production, K DSi P HY denotes the half-saturation constant of DSi uptake (see 229   Table 1). r P ro is the maximum rate of BSi production under non-limiting condi- where, w is the sinking speed of BSi and w ∂BSi ∂z denotes the divergence of verti- We calculate the silicon isotopic composition δ 30 Si in units as a function of the 243 total DSi (or BSi) concentration and D 30 Si (or B 30 Si) following: supply of Si to the surface ocean by 20% in experiment LGMflush (LGMtrickle).

314
Because reliable data regarding the variability of the spatial distribution of this 315 input over time is sparse we distribute all input evenly over space.   Table 2. Depending on the 336 sign of changes in δ 30 Si in BSi we fit either to: where f (t, x, y) is the accumulated change in δ 30 Si of BSi at time t, longitude x and lat-

yr
LGMfe identical to LGM except for Si:N stoichiometric ratio reduced to 1, mimicking the effect of iron replete conditions end of LGM 10000 yr LGMbreezy identical to LGM except for a doubling in all winds, driving the oceanic circulation end of LGM 10000 yr LGMslack identical to LGM except for a halving of all winds, driving the oceanic circulation end of LGM 10000 yr LGMflush identical to LGM except for 20% increase in land-ocean DSi supply end of LGM 10000 yr LGMtrickle identical to LGM except for 20% decrease in land-ocean DSi supply end of LGM 10000 yr LGMlight identical to LGM except for a 1 decrease in δ 30 Si of land-ocean DSi supply end of LGM 10000 yr major aim is to dissect mechanisms that lead to reasonable agreement with the obser-351 vations and paleoarchive data. LGMslack-PI

LGM-PI
LGMfe-PI LGMbreezy-PI 30 Si LGMfe-PI, LGMslack-PI and LGMbreezy-PI, respectively. Magenta circles denote locations of observations of δ 30 Si in BSi as preserved in sediment cores (see Section 2.1). LGM and our preindustrial simulation PI. We find that the colder glacial climate, over-355 all, increases δ 30 Si of BSi deposited to the sediments (relative to PI). This is inconsis- In order to set a reference point for the following discussions (in this Section) we 360 dissect the processes that imprint the wrong sensitivity into simulation LGM: key to un-361 derstanding is that LGM features an oceanic DSi inventory that is 15% lower relative 362 to that in PI. This is puzzling because the export of BSi across 120 m depth (which con-363 stitutes the origin of all BSi sinking to depth) is also reduced by a substantial 30% dur-364 ing the colder LGM climate. Given that the riverine supply of Si is identical in LGM and 365 PI, this is counterintuitive. Further investigations revealed that the process behind this

(c)
LGMlight-PI 30 Si LGMflush-PI and LGMlight, respectively. Magenta circles denote locations of observations of δ 30 Si in BSi as preserved in sediment cores (see Section 2.1).
conundrum is the antagonistic effect of temperature on BSi sedimentation rate in our 367 model framework.
LGM features, consistent with observational evidence (e.g., Margo Project 368 Members, 2009), an average of 2 • C colder oceanic temperature than PI. In combination 369 with an increase in sea-ice cover during the LGM, which shields the ocean from photo-370 synthetically active radiation essential for autotrophic growth, this slows down the global 371 primary production and associated export of organic material from the sun-lit surface 372 to depth. Hence, less BSi is produced and less BSi is set on its way sinking to the sed-

373
iments. This reduction in BSi production is, however, overcompensated by a reminer-374 alization rate that is also slowed down by the lower temperatures such that more organic 375 material reaches the seafloor before it is remineralized and dissolved. A rough scaling, 376 assuming steady state (and horizontal uniformity which reduces the problem to one spa-377 tial dimension), puts the potential of this effect into perspective: the vertical flux of sink-378 ing BSi described in Equation 5 is, following the notation of Kriest and Oschlies (2008), 379 given by where F (z ) is the sinking flux at depth z defined as that distance between actual depth 381 and the depth of the euphotic zone. F 0 is the flux out of the euphotic zone. The sink-382 ing speed w is 10 m day −1 (see Table 1 The simulation LGMfe anticipates that more bioavailable iron was available dur-  LGMbreezy follow the overall patterns already discussed for LGM.

453
In summary, our model suggests that reduced winds are inconsistent with observed 454 δ 30 Si of BSi in the SO. In contrast, the effect of increasing winds appear to be more con-    LGMlight LGMlight

507
The two experiments LGMbreezy and LGMlight, out of our total of 7 numerical sen-508 sitivity experiments, feature the best fit with respect to the sign to the overall, lower (rel-tivity which, although of correct sign, is much lower than suggested by observational ev-512 idence. In addition LGMbreezy features very long manifestation timescales (when com-513 pared to LGMlight; cf., Sect. 4.1) which implies that the process of increasing winds is, 514 in reality, even harder to detect than the rather weak signal we find at the end of our 515 10 000 year long numerical time slice experiment suggests. This leaves us with simula-516 tion LGMlight being the most consistent with observations of δ 30 Si of BSi out of all con-517 sidered processes (listed, e.g., in Table B.1).

518
Even so, other processes such as changes to air-sea iron fluxes and wind fields may 519 also have been at play. According to our model results, however, these should manifest 520 themselves more prominently in metrics other than in the isotopic composition of BSi.

533
The difference among the Sectors in the SO is facilitated by a reduced (down to 40% rel-534 ative to LGM) Antarctic Circumpolar Current which reduces the zonal mixing between 535 the Sectors as a result of reduced winds supplying less momentum to the ocean. Please 536 note that a comprehensive analysis of oceanic deoxygenation, which must cover the role 537 of (preferably explicitly resolved) iron dynamics (see, e.g. Stoll, 2020) and more data (e.g.

538
Jaccard and Galbraith, 2011), is beyond the scope of this manuscript which focuses on 539 the isotopic composition of BSi in response to changing environmental conditions. Our 540 main conclusion here is that LGMbreezy is apparently inconsistent with sedimentary redox-541 sensitive trace-metal records. LGM-PI LGMslack-PI LGMbreezy-PI

597
In summary, we refer to a model that is capable of reproducing the effects of (prein-598 dustrial) circulation and isotopic fractionation during BSi production with a fidelity com-599 parable to existing non data-assimilated 3-dimensional coupled ocean circulation biogeo- parable to the period of glacial-interglacial cycles.

632
As a side aspect we find a simulated oceanic DSi inventory which is 10-20% lower 633 during the Last Glacial Maximum than today. This is somewhat counterintuitive because sinking down to the oceanic sediment. More comprehensive analysis shows that this ef-636 fect is outweighed by BSi dissolution rates that are also slowed down as a consequence 637 of colder temperatures such that more BSi escapes dissolution prior to sedimentation in 638 our model.   low it seems unlikely that a deficient circulation is the cause for these biases (although 687 this can not be ruled out). Hence, the SO nutrient trapping apparently relates strongly 688 to the biogeochemical model parameters. One conclusion from this may be that the bio-689 geochemical model is better tuned with respect to phosphate than to DSi. This is to be  active physical property can be much larger than the misfit of the rather passive (in terms however, not improve probably because we do not account for isotopic fractionation dur-762 ing BSi dissolution (see discussion in Section 4.3).

763
In summary, PI * features a more realistic DSi distribution, more realistic levels of