Millennial-Scale Environmental Variability in Late Quaternary Deep-Sea Sediments from the Demerara Rise, NE Coast of South America

: We carried out a rock magnetic study of two deep-sea gravity cores from the Demerara Rise, NE South America. Our previous studies provided radiocarbon and paleomagnetic chronologies for these cores. This study presents detailed rock magnetic measurements on these cores in order to characterize the rock magnetic mineralogy and grain size as indicators of the overall clastic fraction. We measured the magnetic susceptibility, anhysteretic remanence, and isothermal remanence and demagnetized the remanences at several alternating ﬁeld demagnetization levels. The magnetic intensities estimate the magnetic material concentration (and indirectly the overall clastic fraction) in the cores. Ratios of rock magnetic parameters indicate the relative grain size of the magnetic material (and indirectly the overall clastic grain size). Rock magnetic intensity parameters and rock magnetic ratios both vary systematically and synchronously over the last 30,000 years in both cores. There is a multi-millennial-scale cyclicity, with intervals of high magnetic intensity (high magnetic and clastic content) with low magnetic ratios (coarser magnetic and clastic grain size), alternating in sequence with intervals of low magnetic intensity with high magnetic ratios (ﬁner grain size). There is also a higher-frequency millennial-scale variability in intensity superposed on the multi-millennial-scale variability. There are nine (A–I) multi-millennial-scale intervals in the cores. Intervals A, C, E, G, and I have high magnetic and clastic content with coarser overall magnetic and clastic grain size and are likely intervals of enhanced rainfall and runoff from the NE South American margin to the coastal ocean. In contrast, intervals B, D, F, and H have lower clastic ﬂux with ﬁner overall grain size, probably indicating lower continental rainfall and runoff. During the Holocene, high rainfall and runoff intervals can be related to cooler times and low rainfall and runoff to warmer times. The opposite pattern existed during the Pleistocene, with higher rainfall and runoff during interstadial conditions and lower rainfall and runoff during stadial conditions. We noted a similar pattern of Pleistocene multi-millennial-scale variability in a transect of deep-sea sediment cores along the NE Brazilian margin, from the Cariaco Basin (~10 N) to the NE Brazilian margin (~1 ◦ N–4 ◦ S). However the NW part of this transect (Cariaco Basin, Demerara Rise, Amazon Fan) has an out-of-phase relationship with the SE part of the transect (NE Brazilian margin) between warm–cold and wet–dry conditions. One possible cause of the high–low rainfall and runoff patterns might be oscillation of the Intertropical Convergence Zone (ITCZ), with higher rainfall and runoff associated with a more southerly average position of the ITCZ and lower rainfall and runoff associated with a more northerly average position of the ITCZ.


Introduction
The circulation pattern within the western Equatorial Atlantic Ocean and the large sediment supply from the Amazon River create a complex sedimentation system off the coast of NE South America. The Amazon continental shelf accumulates about five hundred million tons of sediment annually (Kuehl et al., 1986) [1], carried by about 20% of the world's R REVIEW 3

Regional Ocean Circulation
The ocean circulation pattern in the western Equatorial Atlantic Ocean is dominated by the Atlantic Meridional Overturning Circulation (AMOC) (Berger and Wefer, 1996;Schmitz and McCartney, 1993) [13,14]. This system brings warm water into the North Atlantic Ocean, where it is transformed into North Atlantic Deep Water (NADW), which then migrates to the South Atlantic Ocean (Metcalf and Stalcup, 1967;Kirchner et al., 2009)

Regional Ocean Circulation
The ocean circulation pattern in the western Equatorial Atlantic Ocean is dominated by the Atlantic Meridional Overturning Circulation (AMOC) (Berger and Wefer, 1996;Schmitz and McCartney, 1993) [13,14]. This system brings warm water into the North Atlantic Ocean, where it is transformed into North Atlantic Deep Water (NADW), which then migrates to the South Atlantic Ocean (Metcalf and Stalcup, 1967;Kirchner et al., 2009) [15,16]. The principal surface ocean current along the northeastern margin of South America is the North Brazil Current (NBC) (Figure 1). The NBC forms off the easternmost margin of Brazil, where the westward-flowing South Equatorial Current (SEC) bifurcates to form the northwardflowing NBC and the southward-flowing Brazil Current (Stramma et al., 1995) [17]. The NBC crosses the equator as a warm-water current, where it connects to the North Equatorial Current (NEC), Caribbean Current, and Gulf Stream (Schmitz and McCartney, 1993) [14]. The NBC plays an important role in the cross-equatorial transfer of heat to the North Atlantic Ocean (Metcalf and Stalcup, 1967) [15]. The NBC also distributes the sediment plume associated with Amazon River discharge northward along the continental shelf and Demerara Rise-Abyssal Plain (Ruhlemann et al., 2001) [18].
Three deep-ocean currents flow along the NE margin of South America below the NBC (e.g., Huang et al., 2014) [12]. Antarctic Intermediate Water (AAIW) flows northward at depths of~500-1500 m, directly below the NBC, and is an important source of nutrients and carbon to the North Atlantic Ocean. NADW flows southward at depths of~2000-4000 m below AAIW. The deepest parts of the Demerara Abyssal Plain may see northward flow of Antarctic Bottom Water (AABW) at depths below 4000 m.

Regional Deep-Sea Sedimentation
All of the currents carry varying amounts of suspended sediment, depending partly on flow velocities and partly on advection of sediment at the surface by aeolian processes or coastal runoff. The most significant source of advected sediment is the Amazon River discharge. This large flux of sediment creates the Amazon Fan, the third largest modern deep-sea fan, and an expanded continental shelf. The Amazon delta and adjacent shelf currently trap perhaps two-thirds of the discharged sediment (Figueiredo and Nittrouer, 1995;Nittrouer et al., 1995) [19,20], however the finest sediment fraction is probably dispersed beyond the shelf by the NBC. This flux of Amazon River water and sediment beyond the shelf environment brings vast amounts of nutrients to the surface waters of the western Equatorial Atlantic Ocean, causing large blooms of phytoplankton (Smith and DeMaster, 1996) [21].
The western Equatorial Atlantic has two major modes of sedimentation caused by sealevel changes. Higher sea levels associated with interglacial periods foster the deposition of most sediment from the Amazon River onto the north Brazil continental shelf (Nittrouer and DeMaster, 1986;Nittrouer et al., 1995) [20,22]. A lesser portion of the sediment is transported as a suspended load either by longshore currents toward the northwest, where it is deposited along the continental shelf as far as French Guinea, or by the NBC, where it is deposited onto the Amazon Fan and adjacent Demerara Rise-Abyssal Plain (Lund et al., 2018) [23]. Lower sea levels associated with glacial periods cause sediment to bypass the exposed continental shelf and be deposited directly into the NBC and head of the Amazon Submarine Canyon (Damuth and Fairbridge, 1970) [5] and onto the fan.
The "on-off" supply of sediment to the Amazon Fan results in distinctive sediments for glacial versus interglacial intervals. Glacial sediments on the abyssal plain of the most recent glacial period consist of mottled and dark grey hemipelagic clays, a low concentration of carbonate and foraminifera, and occasional redeposited terrigenous sand and silt layers. Above the glacial sediments, Holocene pelagic sediments consist of tan foraminiferal marls or oozes with diverse foraminifera assemblages (Damuth, 1977;Damuth and Flood, 1985) [7,24]. Holocene and interglacial sediments in the western Equatorial Atlantic are characterized by lower concentrations of magnetic material (mostly magnetite) and by finer-grained magnetic material (Meinert and Bloemendal, 1989) [25]. In contrast, the glacial sediments have higher concentrations of magnetic material (mostly magnetite) that is coarser-grained.
Climate also plays a major role in the amount and distribution of sediment deposited on the Amazon Fan, adjacent continental shelf, and more distal Demerara Rise-Abyssal Plain at shorter timescales. The seasonal migration of the Intertropical Convergence Zone (ITCZ) creates distinctive wet and dry seasons and has a large influence on environmental patterns of the coastal region (e.g., Wainer and Soares, 1997) [26]. Boreal summer placement of the ITCZ near 10 • N causes dryness in the Amazon Basin and strong precipitation further to the north, including the Cariaco Basin (Haug et al., 2001) [27]. During the boreal winter, the ITCZ moves over the Amazon Basin (0 • -10 • S), creating a pronounced wet season (austral summer monsoon from December to February).
There is evidence at millennial timescales that the average annual placement of the ITCZ may be biased in its position. During the Early-Middle Holocene (approximately 8000-4000 years BP), there is widespread evidence that the climatic conditions in the tropical Andes were significantly drier than present (e.g., Mayle et al., 2004;Mayle and Power, 2008) [28,29]. More southerly areas experienced similar increased aridity in the Early Holocene. For exam-ple, the driest climatic conditions at Lake Junin (11 • S), Lake Titicaca (14-17 • S), and Sajama Mountain (18 • S) were centered at ca. 10,000, 5500, and 4000 years BP, respectively [30][31][32][33] ( Thompson et al. 1998;Seltzer et al., 2000;Abbott et al., 2003;Bush et al., 2005). In contrast, more northerly sites in Venezuela and Columbia (including Cariaco Basin) showed significant increases in rainfall in the Early Holocene, as the ITCZ migrated even more northerly during summer months (Haug et al., 2001) [27]. Overall, the sediments of the northern South American margin and their stratigraphic variability provide a correlative record of Amazon Basin climate and environmental variability (e.g., , Lund et al., 2018 [9,23].

Core Descriptions
Cruise KNR197-3 collected a sequence of cores at~500 m to~3000 m water depth in a transect outward from the South American margin (Figure 1) as part of a paleoceanographic study of Atlantic Meridional Overturning Circulation (AMOC) in the Atlantic Ocean. Typically, multicores (MC), gravity cores (GGC), and long piston cores (CDH) were collected to provide a complete, composite sedimentologic record along the transect. The focus of our rock magnetic study is 9GGC (405 cm) and 25GGC (352 cm) gravity cores with companion 10CMC (31 cm) and 25DMC (32 cm) multicores ( Figure 1, Table 1). We also make use of a 46CDH piston core because it has a large number of radiocarbon dates and XRF measurements that can be transferred into the gravity cores based on magnetic susceptibility (CHI) profile correlations. The radiocarbon dates (Huang et al., 2014) [12] indicate that our paleomagnetic or rock magnetic study covers the last~19 ka in core 25GGC and~31 ka in core 9GGC.

Rock Magnetic Measurements
U-channel samples were collected from both gravity cores for our rock magnetic studies. The multicores were sampled discretely with contiguous 2 × 2 × 2 cm cubes because the cores were too short (<50 cm) for good-quality u-channel measurements. Paleomagnetic measurements of these cores have been previously described by Lund et al. (2017) [11]. All cores were subjected to whole-core magnetic susceptibility measurements while onboard the ship. The u-channeled sediments were given a series of rock magnetic measurements after completion of the paleomagnetic studies. All samples were given an artificial anhysteretic remanent magnetization (ARM) sequence with a bias field of 0.05 milliTesla (mT) and an alternating field of 100 mT. After sample measurement, all samples were subjected to a sequence of ARM af demagnetization and measurements at 10 mT, 20 mT, 40 mT, and 60 mT. Finally, all samples were given another artificial isothermal remanent magnetization (IRM) sequence with a pulsed 1 T field. (This 1-T artificial remanence is usually referred to as a saturation IRM (SIRM), because all magnetite and titanomagnetite grains, which hold almost all of the natural remanence (NRM), are saturated.) After sample measurement, all samples were subjected to a sequence of SIRM af demagnetization and measurements at 10 mT, 20 mT, 40 mT, and 60 mT. One u-channel from each of the gravity cores also had their ARMs and SIRMs subjected to 80 mT and 100 mT af demagnetization. Lund et al. (2017Lund et al. ( , 2018 [11,23] summarized the AF coercivity patterns of the NRM, ARM, and SIRM. Those papers also considered variations in the ARM/Chi ratio, which showed no evidence of fine-grained (bacterial) magnetite. These studies both suggest that detrital magnetite is the dominant magnetic mineral in these sediments. The resulting CHI, ARM, and SIRM intensities for both gravity cores are shown in Figures 2 and 3. The magnetic intensity patterns in the two multicores and neighboring multicores (Lund et al., 2018) [2] are consistent with the pattern in the uppermost 50 cm of the gravity cores. This provides corroborating evidence that the gravity cores did not lose significant surface sediment in the coring process.
The magnetic intensity variations reflect changes in the percentage of detrital magnetic material (and presumably total clastic material) in the sediment. Magnetite is the most common magnetic mineral noted in several other studies of deep-sea sediments from the Equatorial Atlantic-Amazon Fan region (Meinert and Bloemendal, 1989;Franke et al., 2007;Bendle et al., 2010) [9,25,34,35]. Our previous rock magnetic studies of deep-sea sediments from the neighboring Demerara Abyssal Plain also support this view (Lund et al., 2018) [23]. These studies also indicate that ferric oxides (mostly hematite and goethite) are present in varying small amounts in the sediments. The magnetic intensity variations may be due to varying clastic flux (presumably mostly from the Amazon drainage) or varying dilution by biogenic components due to variable upwelling and biological productivity.
Our ARM and Chi measurements and ratios of ARM or SIRM under different levels of demagnetization (SIRM20/SIRM0; SIRM40/SIRM20, ARM20/ARM0, ARM40/ARM20) estimate the grain size distribution of the magnetic mineral fraction (e.g., Stacy and Banerjee, 1974; Bloemendal et al., 1988) [36,37]. For all rock magnetic ratios, higher (lower) values indicate a finer-grained (coarser-grained) set of magnetic grains. Variations in magnetic grain size may be due simply to a finer-grained or coarser-grained source of overall clastic material; they might also reflect changes in the distribution of magnetic minerals if more than one is significant in concentration.    [23]. These studies also indicate that ferric oxides (mostly hematite and goethite) are present in varying small amounts in the sediments. The magnetic intensity variations may be due to varying clastic flux (presumably mostly from the Amazon drainage) or varying dilution by biogenic components due to variable upwelling and biological productivity.
Our ARM and Chi measurements and ratios of ARM or SIRM under different levels of demagnetization (SIRM20/SIRM0; SIRM40/SIRM20, ARM20/ARM0, ARM40/ARM20) estimate the grain size distribution of the magnetic mineral fraction (e.g., Stacy and Banerjee, 1974; Bloemendal et al., 1988) [36,37]. For all rock magnetic ratios, higher (lower) values indicate a finer-grained (coarser-grained) set of magnetic grains. Variations in magnetic grain size may be due simply to a finer-grained or coarser-grained source of overall clastic material; they might also reflect changes in the distribution of magnetic minerals if more than one is significant in concentration.
Selected samples from gravity core 9GGC were also given an extra series of magnetic hysteresis measurements. The samples were cycled through a MicroMag up to 1 Tesla (T). Values of remanent magnetization (Mr), saturation magnetization (Ms), coercivity field (Hc), coercivity of remanence field (Hcr), and field intensity (H) were calculated to Selected samples from gravity core 9GGC were also given an extra series of magnetic hysteresis measurements. The samples were cycled through a MicroMag up to 1 Tesla (T). Values of remanent magnetization (Mr), saturation magnetization (Ms), coercivity field (Hc), coercivity of remanence field (Hcr), and field intensity (H) were calculated to estimate the grain size distribution and percentage of ferric magnetic minerals (goethite, hematite) versus titanomagnetite minerals (primarily magnetite). H represents the amount of magnetic material that acquires a remanence between 0.3 and 1 T (ferric magnetic minerals, primarily hematite). The primary purpose of these measurements was to assess the percentage of ferric magnetic minerals that primarily develop in Amazon Basin soils and reflect the relative importance of erosion from the Andes (primarily magnetite and titanomagnetite) versus the Amazon Basin itself (a more significant portion of goethite and hematite) on total Amazon drainage to the coastal ocean.

Development of a Revised Magnetic Chronostratigraphy Technique
Lund et al. (2017) [11] developed chronostratigraphies for cores 9GGC and 25GGC by first correlating the two cores based on environmental magnetic and paleomagnetic variability and then using that correlation to assess the radiocarbon dates in cores 9GGC and 25GGC. (Selected radiocarbon dates from 46CDH were also tied into the two gravity cores using magnetic susceptibility correlations.) The final chronostratigraphies for 9GGC and 25GGC were then derived from a selection of radiocarbon dates tied to each core individually and connected with linear interpolation. We then identified 52 paleomagnetic and rock magnetic features that both cores have in common. We estimated the ages of each feature based on the chronologies outlined by Lund et al. (2017) [11]. We found age differences between the same correlatable features in both cores of as much as 1000 years because of differences in linear interpolation between radiocarbon dates in the two cores.
We developed a revised magnetic chronostratigraphy technique for these two cores that provides more consistent ages for environmental and paleomagnetic features. The starting point was to date the 52 environmental magnetic or paleomagnetic correlation features in the two cores (summarized in Supplementary Table S1) based on the original chronos-tratigraphies developed by Lund et al. (2017) [11]. The 52 feature ages in 9GGC and 25GGC were then averaged. Those feature ages were then used to rebuild chronologies for both cores. The results are shown in Figure 4. Figure 4 shows the original Lund et al. (2017) [11] chronologies based on calibrated radiocarbon dates (large open circles in Figure 4), with linear interpolation and the revised chronology based on the 52 feature ages summarized in Supplementary Table S1. This procedure makes PSV and environmental variability more synchronous between the two cores. It does not, however, improve the overall chronology of the sediments. It simply makes the two records chronologically consistent. There are two underlying assumptions in this procedure that we think are warranted. First, we presume that the sediment remanence acquisition process is comparable in both cores. Second, we presume that the sedimentation process near our study cores is not time transgressive. Both assumptions seem reasonable based on the strong (±2 cm) correlation between both PSV and rock magnetic variability among the two cores over their entire lengths.
The revised chronostratigraphies were used to estimate sedimentation rate variability in both cores. The results are shown in Figure 5. There is a notable overall pulse of high sedimentation around the Pleistocene-Holocene transition (~10-16 ka). Superposed on this are multi-millennial variations in sedimentation rates that we can tie to our rock magnetic variability described below. In particular, note that colder intervals of the Late Pleistocene (Younger Dryas (YD) and Heinrich events H1 and H2) have lower sedimentation rates than the neighboring warmer intervals (Bolling-Allerod (B/A) and Dansgard-Oeschger (D/O) interstadial events 2-4).
Oceans 2021, 2, FOR PEER REVIEW 9  [11] with linear interpolation to dates the cores. We built a revised chronology, as summarized in Supplementary Table S1, to create the more detailed chronologies shown here (dark lines).
The revised chronostratigraphies were used to estimate sedimentation rate variability in both cores. The results are shown in Figure 5. There is a notable overall pulse of high sedimentation around the Pleistocene-Holocene transition (~10-16 ka). Superposed on this are multi-millennial variations in sedimentation rates that we can tie to our rock magnetic variability described below. In particular, note that colder intervals of the Late Pleistocene (Younger Dryas (YD) and Heinrich events H1 and H2) have lower sedimentation rates than the neighboring warmer intervals   [11] with linear interpolation to dates the cores. We built a revised chronology, as summarized in Supplementary Table S1, to create the more detailed chronologies shown here (dark lines).

Environmental Variability
Figures 6 (9GGC) and 7 (25GGC) display the pattern of magnetic intensity variation as a function of time over the last 30 ka. The two gravity cores have the same pattern of intensity variation, with six multi-millennial-scale intervals in common (A-F in Figures  6 and 7). Core 25GGC also displays three more intervals (G-I) that are older than the data in 9GGC. Each of these intervals lasts ~1500-4000 years. The intervals oscillate between high and low intensity. The intensity variations could be associated with variations in the rate of clastic flux to the coastal ocean or with variations in biological productivity causing biogenic carbonate or silica dilution.
Figures 8 (9GGC) and 9 (25GGC) display the pattern of magnetic coercivity variations over the last 30 ka. The coercivity varies between high (finer-grained clastics) and low (coarser-grained clastics) values and is the same pattern in both cores. The timing of coercivity variations is also consistent with the pattern of intensity variations. High intensities generally go with lower coercivities. This relationship between intensity and coercivity strongly suggests that clastic flux variations in amount and grain size are the primary cause of the observed multi-millennial-scale variability. Higher rates of clastic flux probably indicate stronger outflow from the Amazon delta; this should also yield overall coarser clastic flux. This pattern is consistent with the sedimentation rate variations noted in Figure 5. High intensity-coarser clastic flux intervals that suggest stronger Amazon outflow are also intervals of higher than average sedimentation rate. Additionally, the lower intensity-finer clastic intervals that suggest lower Amazon outflow are also intervals of lower than average sedimentation rate.
Several other deep-sea sediment studies discussed below use Ti/Ca or Fe/Ca from XRF studies to assess clastic flux variability. Delia Oppo (unpublished) [38] supplied us with an XRF and magnetic susceptibility record for core 46CDH (Figure 1, Table 1). The results are plotted in Figure 10. We correlated our multi-millennial-scale variability (intervals A-I) to the Ti/Ca results using the magnetic susceptibility records. The high   Figures 6 and 7). Core 25GGC also displays three more intervals (G-I) that are older than the data in 9GGC. Each of these intervals lasts~1500-4000 years. The intervals oscillate between high and low intensity. The intensity variations could be associated with variations in the rate of clastic flux to the coastal ocean or with variations in biological productivity causing biogenic carbonate or silica dilution. Figure 8 (9GGC) and Figure 9 (25GGC) display the pattern of magnetic coercivity variations over the last 30 ka. The coercivity varies between high (finer-grained clastics) and low (coarser-grained clastics) values and is the same pattern in both cores. The timing of coercivity variations is also consistent with the pattern of intensity variations. High intensities generally go with lower coercivities. This relationship between intensity and coercivity strongly suggests that clastic flux variations in amount and grain size are the primary cause of the observed multi-millennial-scale variability. Higher rates of clastic flux probably indicate stronger outflow from the Amazon delta; this should also yield overall coarser clastic flux. This pattern is consistent with the sedimentation rate variations noted in Figure 5. High intensity-coarser clastic flux intervals that suggest stronger Amazon outflow are also intervals of higher than average sedimentation rate. Additionally, the lower intensity-finer clastic intervals that suggest lower Amazon outflow are also intervals of lower than average sedimentation rate.

Environmental Variability
Several other deep-sea sediment studies discussed below use Ti/Ca or Fe/Ca from XRF studies to assess clastic flux variability. Delia Oppo (unpublished) [38] supplied us with an XRF and magnetic susceptibility record for core 46CDH (Figure 1, Table 1). The results are plotted in Figure 10. We correlated our multi-millennial-scale variability (intervals A-I) to the Ti/Ca results using the magnetic susceptibility records. The high (low) Ti/Ca intervals (relatively high clastic flux) are consistent with our high-intensity-coarser (low intensity-finer) rock magnetic intervals.
Intervals A-I can be easily associated with known regional to global climate variations over the last 30 ka. Intervals A-C can be associated with Late Holocene ( There is also a higher-frequency scale of variability in these records that is more subdued. This pattern is indicated by the solid lines in Figures 6-9. The lines indicate local intensity peaks. The relationship with coercivity is more complicated and subdued at best or perhaps there is no relationship at all. These cycles occur in all of the longer-duration intervals noted above. They suggest a higher-frequency oscillation in Amazon outflow, with the black lines indicating times of higher flow (higher clastic flux). These more subtle oscillations in intensity are consistent between cores 9GGC and 25GGC. However, the lack of any notable coercivity variability makes their interpretation more questionable. Table 2 presents the intensity peak ages and the durations between successive peaks. The average duration is~1750 ± 450 years.
We finally assess the evidence for variations in ferric (primarily hematite)/ferrous (magnetite and titanomagnetite) ratios in the clastic sediments. These ratios indicate the relative importance of ferrous versus ferric components to the overall sediment magnetization. Hysteresis measurements of selected samples from 9GGC are shown in Figure 11. The rock magnetic data suggest that the magnetization is primarily magnetite-titanomagnetite in the size range of~5-15 microns. This distribution could have varying percentages of coarser (MD) or finer (SD) grains and could still be within the pseudo-single domain (PSD) range for magnetite ( Figure 11A). The hysteresis data let us estimate the amount of magnetization that is acquired with increasing magnetic fields; magnetic contributions between 0.3 and 1.0 T are usually referred to as hard components (H), which is associated with ferric minerals (primarily hematite). The ratio of total magnetic remanence to H (H/SIRM; Figure 11B) estimates the relative portion of ferric magnetization in any sediment sample. The time interval 6-9 ka has the lowest ratio of the last 25 ka. Bendle et al. (2010) [35] and Lund et al. (2018) [23] have both documented low H/SIRM ratios in the Early-Middle Holocene (6-9 ka) of deep-sea sediments from the Demerara Abyssal Plain; our results corroborate this. They all indicate that the Amazon Basin itself was relatively dry during the Early-Middle Holocene, with lower than average erosion of Amazon Basin soils (relative to contributions from the Andes).
(low) Ti/Ca intervals (relatively high clastic flux) are consistent with our high-intensitycoarser (low intensity-finer) rock magnetic intervals.  (Figure 1, Table 1), situated directly between gravity cores 9GGC and 25GGC. Top-magnetic susceptibility record; the variability can be directly correlated to data from 9GGC ( Figure 2) and 25GGC (Figure 3). Below are Ti and Ca counts from unpublished XRF data on 46CDH (Delia Oppo, personal communication). The rock magnetic intensity zonation is consistent with the XRF data. High chi goes with high Ti (a clastic indicator; low chi goes with high Ca (a biologic indicator).  (Figure 1, Table 1), situated directly between gravity cores 9GGC and 25GGC. Top-magnetic susceptibility record; the variability can be directly correlated to data from 9GGC ( Figure 2) and 25GGC (Figure 3). Below are Ti and Ca counts from unpublished XRF data on 46CDH (Delia Oppo, personal communication). The rock magnetic intensity zonation is consistent with the XRF data. High chi goes with high Ti (a clastic indicator; low chi goes with high Ca (a biologic indicator).  1900  1500  1700  3200  1300  2700  1200  2950  1250  4600  1400  4200  1500  4400  1450  6100  1500  6300  2100  6200  1800  8000  1900  8100  1800  8050  1850  10,200  2200  10,000  1900  10,100  We finally assess the evidence for variations in ferric (primarily hematite)/ferrous (magnetite and titanomagnetite) ratios in the clastic sediments. These ratios indicate the relative importance of ferrous versus ferric components to the overall sediment magnetization. Hysteresis measurements of selected samples from 9GGC are shown in Figure 11. The rock magnetic data suggest that the magnetization is primarily magnetitetitanomagnetite in the size range of ~5-15 microns. This distribution could have varying percentages of coarser (MD) or finer (SD) grains and could still be within the pseudosingle domain (PSD) range for magnetite ( Figure 11A). The hysteresis data let us estimate the amount of magnetization that is acquired with increasing magnetic fields; magnetic contributions between 0.3 and 1.0 T are usually referred to as hard components (H), which is associated with ferric minerals (primarily hematite). The ratio of total magnetic remanence to H (H/SIRM; Figure 11B

Comparison with Other Records
The first evidence for millennial-scale variability in the tropical Atlantic region was noted by Grimm et al. (1993) [40] in Lake Tulane (27.6° N) in Florida and Curry and Oppo (1997) [41] in the Ceara Rise (5° N), also on the NE South American continental margin. The Lake Tulane study documented enhanced rainfall during Pleistocene North Atlantic cold (Heinrich) events. The Ceara Rise study linked coherent multi-millennial-scale tropical sea-surface temperature and high-latitude North Atlantic temperature variations. Locally, we can compare our deep-sea sediment records with others from the NE margin of South America. Lund

Comparison with Other Records
The first evidence for millennial-scale variability in the tropical Atlantic region was noted by Grimm et al. (1993) [40] in Lake Tulane (27.6 • N) in Florida and Curry and Oppo (1997) [41] in the Ceara Rise (5 • N), also on the NE South American continental margin. The Lake Tulane study documented enhanced rainfall during Pleistocene North Atlantic cold (Heinrich) events. The Ceara Rise study linked coherent multi-millennial-scale tropical sea-surface temperature and high-latitude North Atlantic temperature variations. Locally, we can compare our deep-sea sediment records with others from the NE margin of South America. Lund et al. (2018) [23] studied rock magnetic variability of Holocene deep-sea sediments from the Amazon Fan-Demerara Abyssal Plain. Our results from this study are the same. Our Pleistocene results (~7 • N) can be compared with three other studies along the NE South American margin: the Cariaco Basin (~10 • N), Demerara Rise-Amazon Fan (~4 • N-7 • N), and outboard of the NE Brazilian margin (~1 • N-3 • S). All of these studies show strong evidence of multi-millennial-scale variability from 12-30 ka, dominated by colder phases associated with the Younger Dryas and Heinrich events 1 and 2 versus warmer phases associated with the Bolling-Allerod and DO cycles 1, 2, and 3.
Our results are consistent with results from the Cariaco Basin (Haug et al. 2001;Deplazes et al., 2013) [27,42] showing lower sedimentation rates and clastic flux during the Younger Dryas, H1, and H2 associated with dryer conditions along NE South America, while the intervening intervals have higher clastic flux associated with heavier rainfall and runoff. These variations are associated with shifts in the long-term position of the Intertropical Convergence Zone (ITCZ). Haggi et al. (2017) [43] studied the isotopic composition of plant waxes in a deep-sea sediment core from the same region (7 • N) as our cores. They suggested that the Amazon rainforest was affected by intrusions of savannah or open framework vegetation types during H1 and H2 associated with drier conditions. The intervening intervals indicated higher Amazon rainfall. This pattern is completely consistent with our Pleistocene multimillennial-scale variability. They attributed these long-term variations to changes in atmospheric circulation. Crivellari et al. (2018) [44] studied the H1 interval in more detail in the same deep-sea core as Haggi et al. (2017) [43]. They identified an early part of the H1 interval (HS1a: 17-18 ka) that had higher influx of sediment to the site, however the core of HS1 (HS1b: 15-17 ka) showed a decrease in terrestrial flux. Our results do not show evidence of this two-part structure, however our study and the study by Haggi et al. (2017) [43] do show a broadly drier response to H1.
A group of four deep-sea sediment studies further to the south (1 • N-4 • S) along the NE Brazilian margin also show the same pattern of multi-millennial-scale environmental variability but with mostly an opposite phasing: colder intervals had higher clastic flux and warmer intervals had lower clastic flux. However, there is some disagreement among these four studies that could be due to chronology issues. Arz et al. (1998) [45] used Ti/Ca XRF ratios to document multi-millennial-scale pulses of significant clastic sedimentation into the Atlantic slope along the NE corner of Brazil (3.5 • S). One pulse correlates to H1, but two pulses correlate to warm Bolling-Allerod and DO interstadial 2 phases; the intervening Younger Dryas and H2 cold intervals had low clastic flux. Nace et al. (2014) [46] studied a marine core along the NE Brazilian margin at 0.3 • N. They also identified multi-millennialscale pulses of high clastic flux associated with Ti/Ca XRF data, which they associated with increased precipitation or weathering on the adjacent continent. They correlated these events to multiple (cold) Heinrich events. However, their data actually showed high clastic pulses associated with H1 (~16 ka) and (warm) D/O cycles 2-3 (~30 ka) and low clastic flux at~26 ka (cold H2). This pattern is consistent with the data from the study by Arz et al. (1998) [45]. Zhang et al. (2015) [47] studied Fe/Ca XRF data from a deep-sea core at 1.6 • S and noted higher clastic flux during all three late Pleistocene cold intervals, namely the Younger Dryas, H1, and H2 events. Mulitza et al. (2017) [48] studied a deepsea sediment core (2 • S) in the same region and concluded that both the Younger Dryas and H1 cold intervals were intervals of increased rainfall and higher clastic flux based on Fe/Ca XRF data. These results are opposite to the Younger Dryas and H2 results of Arz et al. (1998) [45] from the same region. Overall, these records do generally support an opposite phase of high clastic flux in cold intervals than we noted for the Demerara Rise.
Speleothem records from the Amazon Basin display similar complexity in Late Pleistocene rainfall variability. Cheng et al. (2013) [49] suggested that broadly the eastern Amazon region had lower rainfall during the late Pleistocene versus the western Amazon region. This might support our observation that NW marine records (Cariaco Basin, Demerara Rise) had an out-of-phase relationship for rainfall and runoff with SE marine records (NE Brazilian margin). Cheng et al (2013) [49] also noted that both regions had multi-millennial-scale variability on top of longer-duration orbital-scale variability with wetter intervals in NE Brazil (Rio Grande do Norte record) during the H1 and H2 stadials. This is consistent with the NE Brazilian margin marine records noted above. Wang et al. (2017) [50] also noted a similar pattern of multi-millenial-scale rainfall variability in their Paraiso speleothem record (~3 • S) of NE Brazil.

Discussion
Cores 9GGC and 25GGC are dominated by millennial-to multi-millennial-scale variability throughout their entire extent. It is worthwhile noting that the timescales of variability do not significantly change between the Holocene and the Pleistocene. The dominant variability seems to follow a simple oscillatory multi-millennial pattern of wet versus dry conditions in the Amazon Basin (stronger and faster Amazon outflow with coarser clastic grains during wetter conditions versus lower and weaker Amazon outflow with finer clastic grains during drier conditions) in NE South America. Intervals A, C, E, G, and I are associated with relatively wet conditions (and strong Amazon flow), while intervals B, D, F, and H are associated with relatively dry conditions (and weaker Amazon flow). The sediments reach the Demerara Rise from the Amazon outflow under control of the North Brazil Current after passing the local shelf. Coarser (finer) clastic grains are more (or less) likely to reach the North Brazil current under conditions of higher (or lower) Amazon flow. The deep-sea sediment records south of the Amazon Fan (NE Brazilian margin, 1 • N-4 • S) suggest an out-of-phase relationship with our Demerara Rise results and suggest that NE Brazil to the east of the Amazon River delta receives most rainfall and runoff when the Demerara Rise and Cariaco Basin are drier. This also suggests that when we consider Amazon River outflow versus NE Brazil outflow, we are talking about western versus eastern Amazon Basin patterns as a whole.
Multi-millennial-scale intervals A-C are of the Holocene-Glacial transition in age. This interval seems to associate low rainfall and Amazon flow to warmer conditions (Middle Holocene warmth-B) and higher rainfall and Amazon flow to cooler conditions (Late Holocene-A; glacial transition-C). On the other hand, the Pleistocene intervals (e.g., Stuiver and Grootes, 2000 [51]; Cronin, 2009 [52]) D-I have an opposite relationship between Amazon flow and rainfall and temperature. Intervals D (Younger Dryas), F (H1), and H (H2) are all times of low rainfall and Amazon flow with colder conditions, while intervals E (Bolling-Allerod), G (D/O interstadial 2), and I (D/O interstadials 3-4) are times of strong rainfall and Amazon flow and warmer conditions. On this basis, we argue that multi-millennial-scale cyclicity in the Holocene is out of phase with multi-millennial-scale variability in the Pleistocene.
The higher-frequency magnetic intensity variability indicated by vertical black lines in Figures 6-9 suggests another level of wet-dry oscillation associated with Amazon Basin hydrology and Amazon outflow. However, the data do not show a consistent associated variation in coercivity, and so the interpretation of this more subtle higherfrequency variability is more questionable. Even so, there is a great deal of evidence for millennial-scale environmental variability (e.g., Bond et al., 1997 [53]), most dominantly with~1500 year cyclicity. The average duration of our successive millennial-scale high intensity peaks is 1794 ± 482 years ( Table 2). All of our cycles that are common to both 9GGC and 25GGC have an average of~1600 years. The pre 20 ka interval in 9GGC alone shows three intervals with~2600 year cyclicity (underlined in Table 2). As noted previously, our millennial-scale intensity cycles are not as notable as our longer-term cycles. If we are simply missing three cycles in the pre 20ka interval of 9GGC then it seems that our millennial-scale intensity variability is consistent with previous estimates of 1500 year cyclicity.
Several of the published studies discussed above have commented on possible mechanisms for tropical environmental variability. One process that fits the overall data is multi-millennial-sale variation in the placement of the Intertropical Convergence Zone (ITCZ). More southerly average placement will enhance NE South America and Amazon Basin rainfall. More northerly placement will lead to drier conditions. Haug et al. (2001) [27] suggested this mechanism caused a wet early Holocene wet interval in the Cariaco Basin (10.5 • N) further to the northwest along the NE South American margin. This occurred at the same time as our interval B (early Holocene dry conditions), which was also noted in other nearby cores of the Demerara Rise by Lund et al. (2018) [23]. This pattern is also consistent with Lake Tulane wet intervals during Pleistocene Heinrich (cold) events versus our dry events in cold intervals D, F, and H.
The dominance of millennial-to multi-millennial-scale environmental variability in the tropical North Atlantic Ocean is also interesting to note given the recent study by Obrochta et al. (2012) [54] on variability of a similar scale in the high-latitude North Atlantic Ocean (ODP Site 609). They focused on the Late Quaternary ice-rafting evidence for a "1500-year cycle" in climate variability over the last glacial cycle. They suggested that much of that cyclicity might be better interpreted as an admixture of~1000 and 2000 year cycles (Obrochta et al., 2012) [54]. It seems clear that the entire Atlantic region has been subject to continuing millennial-to multi-millennial-scale variability over the Late Quaternary, however the cause(s) of that regional climate pattern and the relationship between tropical and high-latitude North Atlantic variability are still not certain.

Conclusions
We have carried out a rock magnetic study of two deep-sea gravity cores (9GGC and 25GGC) from the Demerara Rise, NE South America. Previous studies (Huang et al., 2014 [12]; Lund et al., 2017 [11]) provided radiocarbon and paleomagnetic chronologies for these cores. In this study, we performed detailed rock magnetic measurements on these cores to characterize the rock magnetic mineralogy, magnetic concentration, and magnetic grain size as an indicator of the overall clastic fraction of the cores. We measured the magnetic susceptibility (chi), anhysteretic remanence (ARM), and isothermal remanence (SIRM) and demagnetized the remanences at several levels of af demagnetization. The magnetic intensities estimate the proportion of magnetic material (and indirectly overall clastic fraction) in the cores. The ratios of rock magnetic parameters (ARM/chi, ARM 10 /ARM0, SIRM 10 /SIRM0) indicate the relative grain size of the magnetic material (and indirectly the overall clastic grain size). Selected hysteresis measurements also gave us a sense of the variation in ferric (goethite, hematite)/ferrous (magnetite, titanomagnetite) ratios for the magnetic minerals.
The rock magnetic intensity parameters (chi, ARM, SIRM) and the rock magnetic ratios (ARM/chi, ARM 10 /ARM0, SIRM 10 /SIRM0) both vary systematically and synchronously over the last 30,000 years in both cores. There is clear evidence of a multi-millennialscale pattern of cyclicity, with intervals of high magnetic intensity (high magnetic and clastic content) and low magnetic ratios (coarser magnetic and clastic grain size) alternating in sequence. There is also evidence of finer millennial-scale variability in intensity superposed on the multi-millennial-scale variability, with an average cycle duration of 1800 ± 500 years. There are nine (A-I) multi-millennial-scale intervals in the cores. Intervals A, C, E, G, and I have high magnetic and clastic content with coarser overall magnetic and clastic grain size and are likely intervals of enhanced rainfall and runoff from the NE South American margin to the coastal ocean. Alternatively, intervals B, D, F, and H represent periods of lower clastic flux with finer overall grain size, probably indicating lower rainfall and runoff from the continental margin. During the Holocene, high rainfall and runoff intervals can be related to cooler times and low rainfall and runoff to warmer times. The opposite pattern is true during the Pleistocene, with higher rainfall and runoff occurring during interstadial conditions and lower rainfall and runoff occurring during stadial conditions. The highest ferric/ferrous ratios occur during Holocene interval B (low rainfall and runoff) during the Early Holocene warm interval~5-9 ka. We associated that interval with drought conditions in the Amazon region, consistent with previous estimates of Amazon flow in this interval (Maslin and Burns, 2000 [10]; Bendle et al., 2010 [35]; Lund et al., 2018 [23]).
We noted a similar pattern of Pleistocene multi-millennial-scale variability in published marine deep-sea core studies from Cariaco Basin (~10 • N) to the NE Brazilian margin (~1 • N-4 • S). These studies documented clastic flux variability using XRF measurements of Ti and Fe. However, it seems likely that the NW part of this transect (Cariaco Basin, Demerara Rise, Amazon fan) has out-of-phase warm-cold wet-dry relationships that support the notion of the eastern Amazon Basin and western Amazon being alternately wetter or drier. One possible cause of the high-low rainfall-runoff patterns might be oscillation of the Intertropical Convergence Zone (ITCZ), with higher rainfall and runoff associated with a more southerly average position of the ITCZ and lower rainfall-runoff associated with a more northerly average position of the ITCZ. Author Contributions: Writing-S.L., Data acquisition-S.L. and E.P., data analysis-S.L. and E.P.