A circumpolar dust conveyor in the glacial Southern Ocean

The increased flux of soluble iron (Fe) to the Fe-deficient Southern Ocean by atmospheric dust is considered to have stimulated the net primary production and carbon export, thus promoting atmospheric CO2 drawdown during glacial periods. Yet, little is known about the sources and transport pathways of Southern Hemisphere dust during the Last Glacial Maximum (LGM). Here we show that Central South America (~24‒32°S) contributed up to ~80% of the dust deposition in the South Pacific Subantarctic Zone via efficient circum-Antarctic dust transport during the LGM, whereas the Antarctic Zone was dominated by dust from Australia. This pattern is in contrast to the modern/Holocene pattern, when South Pacific dust fluxes are thought to be primarily supported by Australian sources. Our findings reveal that in the glacial Southern Ocean, Fe fertilization critically relies on the dynamic interaction of changes in dust-Fe sources in Central South America with the circumpolar westerly wind system.

A ir trapped in Antarctic ice indicates that glacial atmospheric CO 2 concentrations were lower by about 80-100 p.p.m. compared to interglacial levels over the past 800,000 years 1 . Past variability in the atmospheric CO 2 concentrations was synchronous with variations in air temperature and dust deposition over Antarctica on glacial-interglacial to millennial time scales [1][2][3] .
It has been hypothesized that physical and biogeochemical feedbacks in the Southern Hemisphere enhanced the net primary productivity and carbon export during glacials playing a crucial role for the observed coupling between dust, CO 2 , and temperature changes [4][5][6][7] . The deficiency of the micronutrient iron (Fe) is considered a limiting factor for phytoplankton growth in the Southern Ocean [8][9][10] . Martin 8 proposed that the increased input of Fe-bearing mineral dust to the Southern Ocean would lead to increased primary productivity and enhanced oceanic sequestration of CO 2 during past glacial periods. This relationship has been studied extensively in the modern Southern Ocean 9,10 where the export of organic carbon from the surface to the deep ocean can be particularly efficient 11 . However, the effects on net carbon export remain equivocal, partly due to the temporal and spatial limitations of artificial Fe fertilization experiments 10,12 . More continuous natural Fe fertilization in the Southern Ocean suggests higher carbon export, but spatial limitations remain 10 .
Evidence for large-scale natural Fe fertilization experiments is preserved in the marine geological record and confirms the mechanistic link between Southern Ocean dust deposition, primary productivity, CO 2 , and temperature during the late Pleistocene glacial-interglacial cycles [5][6][7] . Specifically, the dust-Feinduced increases in primary productivity enhanced nutrient consumption and export productivity (i.e., export of organic carbon) in the Southern Ocean Subantarctic Zone (SAZ) during past glacials 5,6,13 . This effect was suggested to account for a net drawdown of atmospheric CO 2 of up to~40 p.p.m. representing almost half of the total glacial-interglacial change 6,7,14 .
The data-based estimates of dust-Fe-induced drawdown of atmospheric CO 2 critically rely on reconstructions of particle fluxes and nutrient consumption from the South Atlantic and extrapolation of the results to the entire Southern Ocean 5,6 . These estimates do not take into account contributions from different dust sources in the Southern Hemisphere 15,16 . However, the solubility and bioavailability of dust-borne Fe in the surface ocean is controlled by the complex interaction of multiple factors, including source area particle mineralogy [17][18][19][20] , atmospheric transport (organic complexation, (photo)chemical reactions and pH, and particle sorting) 18,21,22 , sea-ice processing 23,24 , surface ocean photochemistry, and seawater biogeochemistry (Fe chemistry, biotic processing, abundance and type of organic ligands, and particle interactions) 21,22,25 . Changes in dust provenance are primarily linked with variations in source area mineralogy and the characteristics of atmospheric particle transport influencing particle-liquid interaction in the atmosphere and in the surface ocean 18,22 .
In the present day, the largest contributions to the total dust flux in the individual sectors of the Southern Ocean are immediately upwind in Australia, South America, and South Africa following the circumpolar flow of the Southern Hemisphere Westerly Winds (SWW) 15,26,27 (Supplementary Fig. 1). Air parcel trajectories suggest that dust emissions from all continental dust sources in the mid-latitude Southern Hemisphere, including New Zealand, can contribute to the total dust flux to the Southern Ocean 16 . However, Australia and South America are typically considered the most prominent dust sources in the Southern Hemisphere, hosting multiple dust source regions in their (semi) arid continental interior and on the eastern side of the Andes, respectively [26][27][28][29] . The modern dust emissions from terrestrial source regions in the Southern Hemisphere have been traced downwind using their distinct geochemical fingerprints 28,[30][31][32][33][34][35] .
Reconstructions of dust provenance in past environments benefit from the geochemical differences between individual potential dust source areas (PSAs) and revealed that southern South American sources, and in particular Patagonia (south of 38°S), dominated the dust supply to East Antarctica during glacials 28,31,[36][37][38][39] . Yet, currently available Antarctic ice-core data do not unambiguously resolve specific terrestrial sources due to geochemical similarities between important dust sources in Australia, New Zealand, and South America 31,33,37,40 , and/or analytical limitations resulting from the low abundance of lithogenic dust particles in Antarctic ice 35,38 . Previous work from the South Pacific SAZ showed that the general pattern of glacial-interglacial variability in dust deposition resembles the characteristics of Antarctic ice-core records, with higher input during glacials compared to interglacial periods 41 . The increased glacial dust supply to the South Pacific SAZ was ascribed to Australian and/or New Zealand sources synchronized with Patagonian dust emissions by large-scale common climate forcing 20,41,42 . However, relatively little is known about sources and transport of dust to the glacial South Pacific SAZ 42 comprising the largest area for oceanic CO 2 sequestration through Fe fertilization in the Southern Hemisphere. Hence, characterizing the sources of dust input to the South Pacific SAZ during the Last Glacial Maximum (LGM) is paramount to further understanding the role of the Southern Hemisphere dust cycle in the glacial drawdown of atmospheric CO 2 . Here we use a set of complementary, but independent geochemical tracers including rare earth elements (REEs), strontium (Sr), neodymium (Nd), and lead (Pb) isotopes to constrain sources and transport paths of dust delivered to the mid-latitude South Pacific during the LGM. Our data reveal remarkable changes in the geochemistry and spatial distribution of dust between the LGM and the Holocene, refining the existing picture of the glacial Southern Hemisphere dust cycle.

Results and discussion
Isotope signatures of South Pacific fine fraction sediments. For this study, we selected 18 locations between~2 and 5 km water depth covering the mid-latitude South Pacific from~174°E tõ 75°W and from~45°S to~63°S across the Antarctic Circumpolar Current (ACC) (Fig. 1). For each location, we typically processed five samples of the < 5 µm sediment fraction across the LGM interval between 18,000-24,000 years before present (i.e., 18-24 ka BP) (n = 85), which was identified using available age constraints (see Supplementary Data File 1).
We have chosen the < 5 µm size fraction for our dust provenance study, because the long atmospheric lifetime of this size fraction 43 enables inter-continental (long-range) airborne transport 44 and thus the dominant role of < 5 µm particles in the lithogenic sediment fraction of the South Pacific SAZ 45 . Furthermore, moderate changes in grain-size composition during atmospheric transport 43 reduce possible bias from grain-size effects in the < 5 µm fraction, thus allowing direct comparison to many geochemical data sets from terrestrial PSAs. An additional asset is that the high cohesiveness of this size fraction minimizes resuspension and post-depositional transport by bottom currents 46 . The details of sample preparation and analyses are outlined in the "Methods" section. The results are reported as averages per sampling location with 2 SD across the LGM time slice (n = 2 -7).
We obtained Nd isotope compositions (expressed as ε Nd ; see "Methods" for more details) between ε Nd = −3.7 ± 0.5 and −5.3 ± 0.9, and Sr isotope compositions ranging from 87   NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-18858-y ARTICLE large range of Nd-Sr isotope compositions of Southern Hemisphere PSAs in Australia, New Zealand, South America, South Africa, and Antarctica (Fig. 2). Of these PSAs, Antarctica is the only source region which releases significant amounts of lithogenic material that can reach the mid-latitude South Pacific by multiple modes of transport including rafting ice, (re)suspension by bottom currents, turbidity flows, and dust 24,34,42,47,48 . Identifying lithogenic input from Antarctic sources is therefore pivotal to determine the mode of transport of fine fraction material to our sampling locations, in particular because our Nd-Sr isotope data show overlap with the compositions of shelf sediments in the Pacific sector of Antarctica (Fig. 2). Lithogenic material from Antarctica can reach the mid-latitude South Pacific via the main routes of ice rafting and bottom water export from the Ross Sea 24,42,47,49 . Sediment provenance tracers, which are less sensitive to grain-size effects, such as Nd and Pb isotopes 28,39,42 (see Supplementary Note 4), support Sr isotope evidence (see Supplementary Note 5) showing that the Ross Sea area was not a major source of < 5 µm fraction sediment supply to our sample locations during the LGM (Figs. 3 and 4). Likewise, northwards penetrating turbidity flows are unable to supply significant amounts of lithogenic fine fraction material to the mid-latitude South Pacific 34 . Accordingly, the zonal diversity of Nd-Sr isotope compositions of West Antarctic shelf sediments 50 is not reflected in our LGM fine fraction data (Fig. 2). The confined icefree regions along East Antarctica's Ross Sea coast were shown to be a significant source of dust to the modern Southwest Ross Sea 24 and at the nearby Taylor Glacier during the Holocene 48 . However, the dust signal from older ice of the Taylor Glacier was ascribed primarily to input from sources outside of the Antarctic continent 48 . Taking into account that LGM dust fluxes were two orders of magnitude higher in the mid-latitude South Pacific (~120 −200 mg cm −2 ka −1 ) 41 than contemporaneous fluxes at the Taylor Glacier (<~1.2 mg cm −2 ka −1 ) 48 , we consider contributions from Antarctic dust sources to our study area insignificant.
Consequently, the lithogenic < 5 µm sediment fraction was delivered to our South Pacific core locations primarily by airborne transport from terrestrial sources outside of the Antarctic continent, thus supporting previous suggestions that the lithogenic fluxes in the mid-latitude South Pacific reflect predominantly dust input 41,42 . Therefore, our LGM < 5 µm sediment fraction samples are also referred to as dust fraction.
Two main dust signatures in the glacial South Pacific. The key characteristic of our data set is a distinct latitudinal distribution of Pb isotope compositions of the dust fraction (Fig. 3). This pattern requires mixing of two dominating components including one component with radiogenic Pb isotope compositions and one with less radiogenic Pb isotope compositions dominating in the SAZ and AZ, respectively (Fig. 3b). Importantly, the relatively invariable Nd isotope signatures (ɛ Nd of −4.9 ± 0.6, 2 SD, n = 70) associated with the two different Pb isotope signals in our samples limit the possible PSA endmember compositions to a narrow range between ɛ Nd of~−4 and~−6 ( Fig. 4). This narrow Nd isotope range excludes a possible role for emissions from active volcanoes showing typically highly radiogenic ɛ Nd of~6.5 51 . Moreover, the combined evidence from Pb, Nd, and Sr isotopes excludes South Africa as a significant dust source to our study area, consistent with existing model results 15

(Figs. 2 and 3).
Australia hosts geochemically diverse source areas 29,32,40 dominating the dust deposition in the mid-latitude South Pacific in the present day 15,32,34 . The Nd isotope compositions of PSAs in West (ɛ Nd of −17 ± 8.3, 2 SD, n = 21) and South Australia (ɛ Nd of −12.2 ± 6.1, 2 SD, n = 38) (Supplementary Table 2) 29,40 are inconsistent with the observed endmember ɛ Nd range between− 4 and~−6. The large Murray and Darling river basins in East Australia have been considered as important dust source regions during the last glacial period 29,52,53 . The Pb isotope compositions are variable and relatively unradiogenic in the two subbasins 29,38,54 (Supplementary Table 2 and Fig. 3b), but systematic Nd isotope differences exist between the Murray (ɛ Nd of −9.1 ± 3.4, 2 SD, n = 24) and the Darling Basin (ɛ Nd of −2.6 ± 4.2, 2 SD, n = 19) 29,55 (Supplementary Table 2  Murray and Darling Basins could be explained by a~1:1 mixture of sources in the two basins to comply with ɛ Nd values of −4.9 ± 0.6 in our LGM dust samples (Fig. 4). However, a 1:1 mixture would imply a large Sr isotope offset between the hypothetical endmember signal ( 87 Sr/ 86 Sr of 0.7212) (Supplementary Table 2) and our South Pacific dust sample compositions ( 87 Sr/ 86 Sr of 0.7111 ± 0.0015). Notwithstanding that the Murray-Darling data were obtained from clay fraction sediments (< 2 µm) 55 (Supplementary Table 3) showing typically elevated 87 Sr/ 86 Sr relative to larger grain sizes 56 (see also Supplementary Note 4), the magnitude of this offset is difficult to reconcile with documented bias from grain-size effects in dust deposits 28,30,33,56 . Furthermore, it would be surprising if dust flux changes of up to approximately one order of magnitude 52,53 yielded a relatively stable endmember mixing from geochemically distinct source regions in East Australia during the LGM. Instead, we suggest that the Lake Eyre Basin in central Australia was the primary source of long-traveled dust carrying unradiogenic Pb to our core sites in the South Pacific during the LGM. A prominent role for this source region has been suggested previously based on Nd and Sr isotope compositions of bulk sediment samples from the central South Pacific 42 . Supporting evidence is provided by the remarkable resemblance of Nd and Sr isotope compositions in the Lake Eyre Basin (ɛ Nd of −4.1 ± 2.0 and 87 Sr/ 86 Sr of 0.7112 ± 0.0038, 2 SD, n = 17) 29,40 with < 10 µm fraction samples from the eastern Tasman Sea (ɛ Nd of −5.1 ± 1.4 and 87 Sr/ 86 Sr of 0.7115 ± 0.0012, 2 SD, n = 2) 40 (Supplementary  Table 2) and our LGM dust samples (Fig. 2). We emphasize that parts of the Darling Basin share geochemical characteristics with the Lake Eyre Basin, which could have contributed to the integrated signal of Australian dust emissions recorded at our core sites in the South Pacific (Figs. 2 and 4).
Similarity exists also between the Nd and Sr isotope compositions of our LGM dust samples (ɛ Nd of −4.9 ± 0.6 and 87 Sr/ 86 Sr = 0.7111 ± 0.0015) and < 5 µm fraction sediments from New Zealand's South Island (ɛ Nd of~−5.4 ± 3.7, 2 SD, n = 10 and 87 Sr/ 86 Sr of 0.7142 ± 0.0103, 2 SD, n = 10) 31 (Supplementary  Table 2) (Fig. 2b), which has been considered an important dust source during glacial periods 16,20,41,42 . Based on Nd, Sr, and Pb isotope compositions, it is not possible to unambiguously differentiate between the integrated Australian dust signal and New Zealand PSAs (Supplementary Table 2). Complementary REE data conflict with an important role of dust supply from New Zealand to our study area (Supplementary Note 2 and Supplementary Fig. 3). We note that the published dust fraction REE data from New Zealand are very limited and may not reflect the full range of New Zealand PSA REE compositions ( Supplementary Fig. 3). Based on the currently available evidence, we use central Australian Pb 38 and Tasman Sea Nd isotope data 40 to constrain the composition of the dust endmember signal exported from Australia during the LGM (Supplementary  Table 2). This central Australian source can explain up to 100% of the total dust deposition in the South Pacific AZ (Fig. 4).
The second component dominating in the South Pacific SAZ requires a dust source with radiogenic Pb isotope compositions of 206 Pb/ 204 Pb > 19, which are not characteristic for Australian PSAs 38,54 ( Fig. 3b and Supplementary Table 2). The dust fraction of LGM sediments from the New Zealand margin yielded 206 Pb/ 204 Pb of 19.31 ± 0.09 (2 SD, n = 5), which could represent a possible source of radiogenic Pb to the South Pacific SAZ (Fig. 3). However, these sediments are distinct from the PSA signatures on land and reflect a local signal limited to the New Zealand continental shelf and margin environment (see Supplementary Note 3), probably related to mineral sorting during riverine, coastal, and/or submarine sediment transport 57 . Terrestrial PSAs with sufficiently radiogenic Pb isotope compositions that could account for the second endmember in our Pb isotope array are rare in the Southern Hemisphere. South America is often considered the major source of long-traveled dust in the Southern Hemisphere during glacial periods 28 Table 2), which seems inconsistent with the composition of our South Pacific dust samples (Fig. 2b). However, this can be reconciled with changes in the PSA weathering regime and/or different proportions of clay minerals in dust deposits relative to their source soils (see Supplementary Note 4). We suggest that the entrainment, long-range transport and/or deposition of dust discriminated against clay minerals with typically radiogenic Sr isotope compositions (high 87 Sr/ 86 Sr) 33,56 , thus inducing a bias towards lower 87 Sr/ 86 Sr at the site of deposition. For example, particle size distributions of Central South American dust trap and source soil samples indicate a depletion of clay-sized particles in airborne dust relative to the source soils 60 . Such depletion may be related to the emission of clay minerals in aggregates and their preferential removal during transport relative to smaller, fully dispersed clay minerals 43,61 . Supporting evidence for this hypothesis is provided by the low abundance or absence of the clay minerals smectite and kaolinite in East Antarctic ice cores 44,62 , which are important components of soils in the PSAs of South America and Australia 63,64 .
Using the most radiogenic Pb isotope signatures from Central South America (Supplementary Table 2), mixing calculations reveal that Central South America contributed up to~80% to the glacial dust deposition in the South Pacific SAZ, whereas the South Pacific AZ received up to 100% of dust from central Australia (Figs. 3b and 4). This finding is surprising, because Patagonia (south of 38°S) is typically considered the major source of far-traveled dust coming from South America 6,31,37,38 , although Central South America has been invoked as an important dust source to the South Atlantic and East Antarctica 28,33,39 . The pattern is also fundamentally different to the modern situation that is characterized by Australian dust dominating in the entire study area 15,34 and provides specific information about entrainment, transport, and deposition of airborne mineral dust during the LGM.
Glacial dust entrainment and circumpolar transport. We show that dust in the LGM South Pacific AZ was primarily derived from Australia. High emissions from central Australian dust sources are consistent with previous work on modern and past environments 29,35,40,52,53,65,66 . The combination of river systems supplying sediments to extensive alluvial plains and ephemeral lake plays could support efficient deflation of fine material in the expanded (semi)arid zone of central Australia during the LGM 29,53 . Australian dust activity typically follows a seasonal cycle peaking during the spring and summer seasons, as a result of the complex interaction of migrating atmospheric pressure systems with sediment availability and erosivity in the dust source areas 26,27,29,53 . In analogy to the modern situation, the distribution of mineral dust from Australia across the Tasman Sea 32,40 and the South Pacific was orchestrated by the eastward moving frontal systems of the SWW belt during the LGM.
The situation in the South Pacific SAZ is different. The prominent role of Central South America as a major dust source to the South Pacific SAZ during the LGM has not previously been shown. Modern observations and model simulations show pronounced dust activity in Central South American PSAs near 30°S 27,60,67 . Importantly, Central South American PSAs feature a number of environmental preconditions for the efficient production of fine particles including high relief, (semi)arid conditions with intermittent fluvial activity 26,27,33,67,68 , and increased glacier activity during glacial intervals 69,70 .
One mechanism to transport dust from South America to the South Pacific is westward by the trade winds. However, our study area is south of the low-latitude trade wind system today ( Supplementary Fig. 1) and the Pb isotope composition of dust transported by the (sub)tropical Pacific trade winds during the LGM ( 206 Pb/ 204 Pb of 18.69 ± 0.05, 2 SD, n = 7 including replicates) 71 is inconsistent with the dust signal recorded in the South Pacific SAZ ( 206 Pb/ 204 Pb of 19.09 ± 0.17, 2 SD, n = 28) (Fig. 3). Therefore, the only viable route of Central South American dust to the South Pacific SAZ was eastward on a circum-Antarctic path within the mid-latitude SWW belt (Supplementary Figs. 1 and 9).
Here we propose that increased SWW activity in Central South America north of~30°S facilitated efficient dust entrainment during the LGM, in particular during the winter season when the SWW regime reaches its northernmost extension 68,72 . In the present day, the mid-latitude SWW regime intersecting South America spans the lower troposphere westerly winds and the high-altitude jet streams 33,[72][73][74] . The westerly winds are typically associated with strong dry foehn-like winds promoting dust entrainment on the eastern side of the Andes 28,30,33,39,60,67,75 . The high-altitude westerly jet circulation features a mid-latitude jet stream in the Southern Hemisphere, which is characterized by a weakened subpolar (~60°S) and a strengthened subtropical branch (~30°S) over the South Pacific during austral winter 73 (Fig. 5a). Modern observations, trajectory modeling, and geochemical evidence show that the subtropical jet stream plays an important role in the export and long-range transport of dust from high-altitude PSAs in Central South America 33,60 (Supplementary Fig. 9). Recent work suggested that the subtropical jet stream was instrumental to enhance fluvial input at the Chilean continental slope near 27.5°S off the southern Atacama Desert during past glacials 68 .
We therefore conclude that SWW-induced orographic winds and intensified subtropical jet stream circulation 28,33,39,60,68,72 , possibly augmented by enhanced availability of fine material from glacier and (ephemeral) fluvial activity 33,68-70 , enabled efficient dust entrainment in and export from Central South American PSAs during the LGM. This scenario is consistent with the provenance of Argentinean loess and dust samples collected downwind of our suggested source region 28,30,39,60,75 , and with independent provenance constraints derived from heliumthorium isotope compositions of sediments in the South Atlantic 76 . The long-distance circumpolar transport of dust ( Supplementary Fig. 9) was likely promoted by reduced wet deposition and increased lifetime of atmospheric dust during the LGM 15 . The increased abundance and prolonged residence time of < 5 µm particles from Central South America would have enhanced scattering of incoming solar radiation, thus contributing directly to glacial cooling 43 . Different from the modern situation 34 and LGM model simulations 15 , these boundary conditions allowed Central South American dust to outpace the deposition from Australian sources in the South Pacific SAZ during the LGM (Fig. 5a and Supplementary Table 4). The interplay of these two important terrestrial PSAs may explain the increased Pb isotope variability in particular in the South Pacific SAZ during the LGM interval (Fig. 3) and could have been involved in the geochemical variability of dust samples from Antarctic ice cores 28,31,33,38,39,44,48 .
Dust deposition in the glacial South Pacific. The two distinct dust depositional environments in the LGM South Pacific are characterized by a narrow transition zone (Fig. 5a), a pattern that cannot be controlled by variable atmospheric dust transport alone. Therefore, we invoke local oceanic processes modulating the settling of atmospheric dust fallout to the seafloor. The delineation between the two dust depositional zones shows remarkable correspondence with the position of the modern polar front and the LGM winter sea-ice cover 77 (Fig. 5a). A primary control by hydrodynamic processes seems unlikely given the lacking correspondence between dust fraction geochemistry and the ACC fronts in the present day 34 . A characteristic feature of Southern Ocean sea-ice cover is a pronounced seasonal cycle 77 such that the dust deposition at locations under the LGM winter sea-ice cover was reduced when dust transport from Central South American PSAs was presumably most efficient due to the enhanced northward expansion of the SWW/jet stream regime during the LGM winter season (see above). The offshore motion of wind and sea ice 24,49 would move the wintertime South American dust fallout out of the ice-covered ocean areas of the South Pacific AZ (Fig. 5a, b). A sea-ice control on the deposition of lithogenic particles is also evident from modern sediment trap data showing reduced lithogenic fluxes in the South Pacific AZ and increased fluxes in the SAZ 78  The LGM wintertime scenario. The northward expanded Southern Hemisphere westerly wind (SWW) system (STJ present in the South Pacific as in a) delivers dust efficiently on a circumpolar trajectory from Central South America to the study area. The extensive sea-ice cover reduces the dust deposition in the South Pacific AZ. IRD: ice-rafted detritus. Numbers in yellow box as in a. c The LGM summertime scenario. The STJ transporting dust from Central South America is absent, the SWW move closer to Antarctica and deliver predominantly Australian dust to the South Pacific AZ and SAZ north of the summer sea-ice limit. Numbers in green box as in a. A comprehensive overview of tracer evidence constraining our proposed LGM scenario of circumpolar dust transport and deposition in the study area is provided in Supplementary example, settling of particles is typically enhanced by aggregation 80,81 explaining also the close correspondence of biogenic and lithogenic particle fluxes in the Southern Ocean 79 . Dust particles released from sea ice to the AZ surface ocean during the melting season 23,24 would be subject to further north (east)ward dispersal by the surface currents 49 . The earlier resumption of the primary productivity in the ice-free SAZ in spring 78,79 could enhance the transfer of wintertime dust to the seafloor. During the rest of the year, the SWW/jet stream regime would not provide an efficient dust conveyor from Central South America to the South Pacific. Then, Australian dust fallout could dominate over the entire study area, including the region that is seasonally ice-free, but without outpacing the (winter) input from Central South America in the South Pacific SAZ (Fig. 5c). As a result, the South Pacific SAZ was dominated by highly seasonal dust input from Central South American sources accounting for~50-80% of the total deposition and hence, for large parts of the threefold increase in glacial South Pacific SAZ dust fluxes 41 . We suggest that the seasonality of atmospheric dust supply from Central South America and sea-ice cover played important roles in modulating particle settling in the glacial South Pacific, thus inducing the distinct north-south trend in dust provenance evident from our data set (Fig. 5).
Dust provenance changes and Southern Ocean Fe supply. Previous estimates of a dust-induced~40 p.p.m. lowering of glacial atmospheric CO 2 concentrations critically rely on reconstructions from the South Atlantic SAZ, extrapolation of the results to the entire Southern Ocean, and a uniform value of 2% Fe solubility from mineral dust fallout 5,6 . However, the distinct pattern of LGM South Pacific dust provenance implies a change in source area particle mineralogy and in dust transport conditions, both of which can directly affect dust-Fe solubility 17,18,20,21 with possible cascading effects on biogeochemical reactions in the atmosphere and in seawater 18,22,25 . Recent laboratory experiments with natural dust samples showed a close relationship between diatom growth and the content of more soluble Fe(II)rich silicate minerals typical for glaciogenic dust sources 19 . Reconstructions estimate a relatively uniform~15-fold increase of Fe(II) supply to the glacial Southern Ocean in comparison to interglacial periods 20 . The similar phasing and magnitude of the glacial Fe(II) increase in the South Pacific and South Atlantic was related to the synchronized activity of chemically pristine (low degree of chemical weathering) glaciogenic mineral dust emissions from New Zealand and Patagonia, respectively 20 . Nonglaciogenic Australian PSA sediments show variable degrees of chemical maturity 64 , but dust from these sources might host less soluble Fe than chemically more pristine dust 17,21 . The contribution of Australian dust to the South Pacific could reduce the efficiency of a dust-induced glacial drawdown of atmospheric CO 2 to less than the proposed~40 p.p.m. Yet, our new data set suggests that the increased supply of primary Fe(II) silicates to the South Pacific SAZ 20 was largely driven by mineral dust from chemically more pristine sources in (semi)arid Central South America 28,39,59,82 (Figs. 4 and 5a). The solubility of dust-borne Fe from Central South America was probably further increased during the long circum-Antarctic transport 22 and by (winter) seaice processing in the study area 23 . A possible reduction of Fe(II) supply by the contribution of more mature Australian mineral dust is then rendered negligible by the large-scale deposition and sea-ice processing of long-traveled Central South American dust. As such, our data can explain the zonally symmetric increase in Fe(II) supply to the Southern Ocean SAZ 20 , which stimulated primary productivity 19 , nutrient consumption 6,13 , and export production 5 , thus supporting the increased sequestration of carbon in the glacial deep ocean 4,14 .
As a corollary, dust-borne Fe input as a critical component of the glacial-interglacial Fe feedback in the Southern Ocean was characterized by the interaction of the SWW system with changes in specific Central South American PSAs. This implies that Southern Ocean dust-Fe fertilization is highly sensitive to the environmental conditions in this important dust source region. Our findings are consistent with recent reconstructions of Fe supply to the subpolar Southern Ocean, but revise previous hypotheses of the source of the dust-borne Fe supply 5,6,20,41,42 . It may be speculated whether the enhanced dust emissions from these Central South American PSAs were systematic and linear during the past glacial-interglacial cycles or if there was threshold behavior. The precessional forcing of the South Pacific subtropical jet stream dynamics 68 would suggest that the LGM may not reflect whole glacial averages. Rather, our proposed link between dust activity in the source regions, the variable SWW system and the input of dust-borne Fe to the South Pacific SAZ indicates that the Southern Hemisphere dust-Fe supply can operate as a dynamic feedback on multiple time scales.

Methods
Grain-size separation and sediment leaching. For this study, we sampled about three cubic centimeters of wet bulk sediment for wet-sieving and subsequent Stokes-based separation of the < 5 µm grain-size fraction in glass settling tubes at the Alfred Wegener Institute Bremerhaven. The < 5 µm fraction was then subjected to a sequential leaching procedure at the Institute for Chemistry and Biology of the Marine Environment (ICBM) in Oldenburg to extract the lithogenic silicate fraction 34 (see also Supplementary Note 1 for more details). The < 5 µm fraction samples were processed in a total of eight batches. One batch contained typically twelve samples, two procedural blanks, and the two United States Geological Survey (USGS) rock reference materials AGV-1 and BCR-2. All chemical and analytical sample processing was carried out at the ICBM using high-purity reagents 34 if not indicated otherwise.
An aqueous solution with~5% H 2 O 2 was used to eliminate organics, followed by a two-step 1 M HCl leach (3 h and overnight) to remove carbonates and ferromanganese (oxy-)hydroxides. During a final step, the samples were exposed to 0.03 M EDTA (99.995% trace metals basis, Sigma-Aldrich ® ) to sequester any remaining free and loosely adsorbed metal ions. After all steps the supernatant reagent was removed and the samples were triple-rinsed with Milli-Q ® H 2 O. Then, the samples were freeze-dried and~50-100 mg (depending on opal content) of freeze-dried sample material was weighed into ultra-clean polytetrafluoroethylene (PTFE) vessels fitting the PicoTrace DAS 30 pressure digestion system. To break down any remaining organic compounds, the samples were exposed to aqua regia at 130°C overnight, followed by pressure digestion at a nominal temperature of 230°C (180°C measured in the digestion unit) using a mixture of concentrated HF-HNO 3 -HClO 4 . After complete digestion, the samples were converted to chloride form, redissolved in 6 M HCl, and split for REE analyses (see Supplementary Methods) and sequential wet-chemical extraction of target elements from the sample matrix following previously published procedures 34 . In brief, the sample aliquot was converted to bromide form for HBr-HNO 3 -based Pb extraction using Biorad ® AG1-X8 resin 83 . The matrix wash fraction was collected to separate the alkaline earth metals from the REE using Biorad ® AG50W X-8 resin 84 . Strontium was extracted from the remaining sample matrix using Eichrom Sr resin 85 and Nd was isolated from the LREE with TrisKem Ln resin-based chemistry 86 .
Radiogenic isotope analyses. The Pb, Nd, and Sr isotope compositions were determined using a Thermo Scientific TM Neptune Plus TM multi-collector ICP-MS at the ICBM in Oldenburg. All reported AGV-1 and BCR-2 reference material results were obtained on leached residues (see above). For Nd isotope analyses, mass bias was corrected for using 146 Nd/ 144 Nd = 0.7219 and an exponential law. Isobaric interferences of 142 Ce and 144 Sm on 142 Nd and 144 Nd were monitored and corrected for using 140 Ce and 147 Sm, respectively. The JNdi-1 reference material was measured every four samples to correct for the instrumental offset of the mass bias corrected 143 Nd/ 144 Nd ratios of the samples to JNdi-1 reference ratio of 0.512115 ± 0.000007 87 . The external reproducibility of normalized 143 Nd/ 144 Nd ratios of acid-leached BCR-2 and AGV-1 rock powders was 0.512641 ± 0.000013 (2 SD, n = 29) and 0.512794 ± 0.000020 (2 SD, n = 32), respectively. These results are in excellent agreement with literature 143 Nd/ 144 Nd of 0.512637 ± 0.000013 for acid-leached BCR-2 residues 88 and 0.512791 ± 0.000013 for unleached AGV-1 rock powder 89 . Repeat analyses of sample C7 yielded 143 Nd/ 144 Nd = 0.512387 ± 0.000011 (2 SD, n = 16). All sample Nd isotope results are expressed in epsilon notation: ε Nd = ( 143 Nd/ 144 Nd sample )/( 143 Nd/ 144 Nd CHUR ) − 1] × 10 4 where CHUR is the Chondritic Uniform Reservoir 90 .

Data availability
All data presented in this paper are included in this published article and its Supplementary Data files.