Oxygen and carbon isotopes were measured on tests of Cassidulina teretis, a cold-water species of endobenthic foraminifera that is generally abundant in the samples and common in finely grained sediment and at relatively low salinities (Mackensen and Hald, 1988; Rosoff and Corliss, 1992). Because of their endobenthic habitat, they record isotope compositions of pore waters, which leads to somewhat reduced (δ13Cb) values compared to the overlying bottom waters. Since the amount of material from the sidewall cores is limited, the isotope data are only produced for the cored intervals with the principal aim to confirm the phase relationship described by Kuhlmann and Wong (2008) between facies and climate. Preservation was based on a visual inspection and assignment of a relative preservation scale of 1–5, after which the poorest two classes were discarded because primary calcite was nearly absent. The best-preserved specimens (cat. 1) had shiny tests (original wall calcite) and showed no signs of overgrowth. Category 2 specimens showed signs of overgrowth but were not recrystallized and cat. 3 specimens were dull and overgrown by a thin layer of secondary calcite. Between ∼ 20 and 50 µg of specimens per sample was weighed, after which the isotopes of the carbonate were measured using a Kiel III device coupled to a 253 Thermo Finnigan MAT instrument. Isotope measurements were normalized to an external standard “NBS-19” (δ18O =-2.20‰, δ13C =1.95‰).

In modern oceans, dinoflagellates are an important component of the (phyto)plankton. About 15–20 % of the marine dinoflagellates form an organic walled cyst (dinocyst) during the life cycle that can be preserved in sediments (Head, 1996). Dinocyst distribution in marine surface sediments has shown to reflect changes in the sea surface water properties, mostly responding to temperature (e.g., Rochon et al., 1999; Zonneveld et al., 2013). Down-core changes in dinocyst assemblages are widely used in reconstructing past environmental changes in the Quaternary (e.g., de Vernal et al., 2009), but also in the Neogene and Paleogene (e.g., Versteegh and Zonneveld, 1994; Head et al., 2004; Pross and Brinkhuis, 2005; Sluijs et al., 2005; Schreck et al., 2013; De Schepper et al., 2011, 2013; Hennissen et al., 2017).

Here we use the preference of certain taxa for cold-temperate to arctic surface waters to derive SST trends. The cumulative percentage of the dinocysts Filisphaera microornata, Filisphaera filifera, Filisphaera sp., Habibacysta tectata, and Bitectatodinium tepikiense in the total dinocysts represents our cold surface water indicator (Versteegh and Zonneveld, 1994; Donders et al., 2009; De Schepper et al., 2011). Interestingly, Bitectatodinium tepikiense, the only extant dinocyst among our cold-water species, has been recorded from the mixing zone of polar front oceanic waters with cold brackish meltwaters from glacier ice (e.g., Bakken and Dale, 1986) and at the transition between the subpolar and temperate zones (Dale, 1996). The combined abundance of Lingulodinium machaerophorum, Tuberculodinium vancampoae, Polysphaeridium zoharyi, and Operculodinium israelianum is used here to indicate coastal waters, although they generally also relate to warmer conditions. In particular, high percentages of L. machaerophorum are typically recorded in eutrophic coastal areas where reduced salinity and (seasonal) stratification due to runoff occur (Dale, 1996; Sangiorgi and Donders, 2004; Zonneveld et al., 2009). At present, T. vancampoae, P. zoharyi, and O. israelianum are also found in lagoonal euryhaline environments (Zonneveld et al., 2013) and hence could be used to indicate a more proximal condition relative to L. machaerophorum (Pross and Brinkhuis, 2005).

At present, Protoperidinioid (P) cysts are mostly formed by heterotrophic dinoflagellates and the percentage of P cysts may be used as an indicator of high eukaryotic productivity (see Reichart and Brinkhuis, 2003; Sangiorgi and Donders, 2004; Sluijs et al., 2005). Here we use the percentage of P cysts (Brigantedinium spp., Lejeunecysta spp., Trinovantedinium glorianum, Selenopemphix spp., Islandinium spp., Barssidinium graminosum, and B. wrennii) to indicate eukaryotic productivity.

Terrestrial palynomorphs (sporomorphs) reflect variations in the vegetation on the surrounding land masses and provide information on climate variables such as continental temperatures and precipitation (e.g., Heusser and Shackleton, 1979; Donders et al., 2009; Kotthoff et al., 2014). A ratio of terrestrial to marine palynomorphs (T / M ratio) is widely used as a relative measure of distance to the coast and thereby reflects sea level variations and depth trends in the basin (e.g., McCarthy and Mudie, 1998; Donders et al., 2009; Quaijtaal et al., 2014; Kotthoff et al., 2014). Morphological characteristics of late Neogene pollen types can, in most cases, be related to extant genera and families (Donders et al., 2009; Larsson et al., 2011; Kotthoff et al., 2014). In A15-3 and A15-4, the relatively long distance between the land and the site of deposition means that the pollen assemblage is not only a reflection of vegetation cover and climate but also includes information on the mode of transport. Assemblages with a relatively high number of taxa, including insect-pollinated forms, are indicative of substantial pollen input through water transport (Whitehead, 1983), whereas wind-transported pollen typically show a low diversity. Sediments of a location proximal to a river delta likely receive a majority of pollen that is water transported, while distal locations are dominated by wind-transported pollen and particularly bisaccate taxa (Hooghiemstra, 1988; Mudie and McCarthy, 1994). To exclude these effects, the percentage of arboreal pollen (AP), representing relative terrestrial temperatures, was calculated excluding bisaccate forms. The non-arboreal pollen (NAP; mainly Poaceae and also Artemisia, Chenopodiaceae, and Asteraceae) consist only of nonaquatic herbs. High AP percentages indicate warm, moist conditions, whereas open vegetation (NAP and Ericaceae) is indicative of cooler, drier conditions consistent with a glacial climate (Faegri et al., 1989).

The samples were processed using standard palynological procedures (e.g., Faegri et al., 1989) involving HCl (30 %) and cold HF (40 %) digestion of carbonates and silicates. Residues were sieved with 15 µm mesh and treated using heavy liquid separation (ZnCl, specific gravity 2.1 g cm-3). The slides were counted for dinocysts (with a minimum of 100 cysts) and pollen (with a preferable minimum of 200 grains). The dinocyst taxonomy follows Williams et al. (2017). Resulting counts were expressed as percent abundance of the respective terrestrial or marine groups of palynomorphs.

We applied three measures for the relative marine versus terrestrial hydrocarbon sources. The carbon preference index (CPI), based on C25–C34n-alkanes, originally devised to infer thermal maturity (Bray and Evans, 1961), has high values for predominantly terrestrial plant sources (Eglinton and Hamilton, 1967; Rieley et al., 1991). Values closer to 1 indicate greater input from marine microorganisms and/or recycled organic matter (e.g., Kennicutt et al., 1987). Furthermore, peat mosses like Sphagnum are characterized by a dominance of the shorter C23 and C25 n-alkanes (e.g., Baas et al., 2000; Vonk and Gustafsson, 2009), whereas longer-chain n-alkanes (C27–C33) are synthesized by higher plants (e.g., Pancost et al., 2002; Nichols et al., 2006). Here we express the abundance of Sphagnum relative to higher plants as the proportion of C23 and C25 relative to the C27–C33 odd-carbon-numbered n-alkanes. Finally, the input of soil organic matter into the marine environment was estimated using the relative abundance of branched glycerol dialkyl glycerol tetraethers (brGDGTs), produced by bacteria that are abundant in soils, versus that of the marine Thaumarchaeota-derived isoprenoid GDGT crenarchaeol (Sinninghe Damsté et al., 2002), which is quantified in the branched and isoprenoid tetraether (BIT) index (Hopmans et al., 2004). The distribution of brGDGTs in soils is temperature dependent (Weijers et al., 2007; Peterse et al., 2012). Annual mean air temperatures (MATs) were reconstructed based on down-core distributional changes of brGDGT and a global soil calibration that uses both the 5- and 6-methyl isomers of the brGDGTs (MATmr; De Jonge et al., 2014a). Cyclization of branched tetraethers (CBT) ratios was shown earlier to correlate with the ambient MAT and soil pH (Weijers et al., 2007; Peterse et al., 2012). The much improved CBT' ratio (De Jonge et al., 2014a), which includes the pH-dependent 6-methyl brGDGTs, is used here to reconstruct soil pH. The total organic carbon (TOC) and total nitrogen measurements are used to determine the atomic C / N ratio that in coastal marine sediments can indicate the dominant source of organic matter, with marine C / N values at ∼10 and terrestrial between 15 and 30 (Hedges et al., 1997).

Organic geochemical analyses were limited to the core and sidewall core samples. For TOC determination ∼0.3 g of freeze-dried and powdered sediment was weighed and treated with 7.5 mL 1 M HCL to remove carbonates, followed by 4 h of shaking, centrifugation, and decanting. This procedure was repeated with 12 h of shaking. Residues were washed twice with demineralized water and dried at 40–50 ∘C for 96 h after which weight loss was determined. About 15 to 20 mg of ground sample was measured in a Fisons NA 1500 NCS elemental analyzer with a normal Dumas combustion setup. Results were normalized to three external standards (Bureau Communautaire de Référence BCR-71, atropine, and acetanilide) and analyzed before and after the series, and after each 10 measurements. % TOC was determined by % C x decalcified weight divided by the original weight.

For biomarker extraction ca. 10 g of sediment was freeze-dried and mechanically powdered. The sediments were extracted with a dichloromethane (DCM) : methanol (MeOH) solvent mixture (9:1, v/v, three times for 5 min each) using an accelerated solvent extractor (ASE, Dionex 200) at 100 ∘C and ca. 1000 psi. The resulting total lipid extract (TLE) was evaporated to near dryness using a rotary evaporator under near vacuum conditions. The TLE was then transferred to a 4 mL vial and dried under a continuous N2 flow. A 50 % split of the TLE was archived. For the other working half, elemental sulfur was removed by adding activated (in 2M HCl) copper turnings to the TLE in DCM and stirring overnight. The TLE was subsequently filtered over Na2SO4 to remove the CuS, after which 500 ng of a C46 GDGT internal standard was added (Huguet et al., 2009). The resulting TLE was separated over a small column (Pasteur pipette) packed with activated Al2O3 (2 h at 150 ∘C). The TLE was separated into an apolar, a ketone, and a polar fraction by eluting with n-hexane : DCM 9:1 (v/v), n-hexane : DCM 1:1 (v/v), and DCM : MeOH 1:1 (v/v) solvent mixtures, respectively. The apolar fraction was analyzed using gas chromatography (GC) coupled to a flame ionization detector (FID) and gas chromatography mass spectroscopy (GC-MS) for quantification and identification of specific biomarkers, respectively. For GC, samples were dissolved in 55 µL of hexane and analyzed using a Hewlett-Packard G1513A autosampler interfaced to a Hewlett-Packard 6890 series GC system equipped with a FID, using a CP-Sil 5 fused silica capillary column (25 m × 0.32 mm, film thickness 0.12 µm), with a 0.53 mm pre-column. The temperature program goes from 70 to 130 ∘C (0 min) at 20 ∘C min-1 and then to 320 ∘C at 4 ∘C min-1 (hold time 20 min). The injection volume of the samples was 1 µL.

Analyses of the apolar fractions were performed on a Thermo Finnigan TRACE GC ultra interfaced to a Thermo Finnigan TRACE DSQ MS instrument using the same temperature program, column, and injection volume as for GC analysis. Alkane ratios are calculated using peak surface areas of the respective alkanes from the GC-FID chromatograms.

Prior to analyses, the polar fractions, containing the GDGTs, were dissolved in n-hexane : propanol (99:1, v/v) and filtered over a 0.45 µm mesh PTFE filter (ø 4 mm). Subsequently, analyses of the GDGTs were performed using ultra-high-performance liquid chromatography mass spectrometry (UHPLC-MS) on an Agilent 1290 infinity series instrument coupled to a 6130 quadrupole mass selective detector with settings as described in Hopmans et al. (2016). In short, separation of GDGTs was performed on two silica Waters ACQUITY UHPLC HEB HILIC (1.7 µm, 2.1 mm × 150 mm) columns, preceded by a guard column of the same material. GDGTs were eluted isocratically using 82 % A and 18 % B for 25 min and then with a linear gradient to 70 % A and 30 % B for 25 min, where A is n-hexane and B =n-hexane : isopropanol. The flow rate was constant at 0.2 mL min-1. The [M + H]+ ions of the GDGTs were detected in selected ion monitoring mode and quantified relative to the peak area of the C46 GDGT internal standard.

The glacial–interglacial (G–IG) range in Cassidulina teretis δ18O (δ18Ob) is ∼1 ‰ between MIS 98 and 97 and ∼ 1.3 ‰ between MIS 95 and 94, but with considerably more variation in especially MIS 95 (Fig. 3). The δ13Cb data covary consistently with δ18Ob and have a G–IG range of ∼ 1.1 ‰, in addition to one strongly depleted value in MIS 94 (-3.5 ‰). The MIS 95 δ13Cb values are less variable than the δ18Ob, pointing to an externally forced signal in the latter. The δ18Ob confirms the relation between glacial stages and finely grained sediment as proposed by Kuhlmann et al. (2006a, b). Although the data are somewhat scattered, the A15-3 and A15-4 phase relation to the sediment facies is in agreement with the high-resolution stable isotope benthic foraminifera record of the onshore Noordwijk borehole (Noorbergen et al., 2015). The glacial-to-interglacial ranges are very similar in magnitude with those reported by Sosdian and Rosenthal (2009) for the North Atlantic but are on average lighter by ∼ 0.5 ‰ (δ18Ob) and ∼ 1.8 ‰ (δ13Cb).

Palynomorphs, including dinocysts, freshwater palynomorphs, and pollen, are abundant, diverse, and well preserved in these sediments. Striking is the dominance of conifer pollen. Angiosperm (tree) pollen are present and diverse, but low in abundance relative to conifers. During interglacials (MIS 103, 99, 97, 95, and 93) the pollen record generally shows increased and more diverse tree pollen (particularly Picea and Tsuga) and warm temperate Osmunda spores, whereas during glacials (MIS 102, (100), 98, 96, and 94) herb and heath pollen indicative of open landscapes are dominant (Fig. S2 in the Supplement). The percentage of arboreal pollen (AP; excluding bisaccate pollen) summarizes these changes, showing maximum values of > 40 % restricted to just a part of the more coarsely grained interglacial intervals (Fig. 3). The percentage record of cold-water dinocysts is quite scattered in some intervals but indicates generally colder conditions within glacial stages and minima during % AP maxima (Fig. 3). After peak cold conditions and a TOC maximum (see below), but still well within the glacials, the percentage of Protoperidinoid consistently increases. Some intervals (e.g., top of MIS 94) are marked by influxes of freshwater algae (Pediastrum and Botryococcus), indicating a strong riverine input; these data, however, do not indicate a clear trend. This robust in-phase pattern of G–IG variations is also reflected by high T / M ratios during glacials, indicating coastal proximity, and low T / M during (final phases of) interglacials. The G–IG variability in the T / M ratio is superimposed on a long-term increase. The coastal (warm-tolerant) dinocyst maxima are confined to the interglacial intervals and their abundance increases throughout the record. Successive increases in coastal inner neritic Lingulodinium machaerophorum, followed by increases in coastal lagoonal species in the youngest part, mirror the shoaling trend in the T / M ratio, which in time corresponds with the gradual progradation of the Eridanos delta front (Fig. S1).

The lowest TOC contents are reached in the clay intervals and typically range between 0.5 % in glacials and 1 % in interglacials (Fig. 3). Nitrogen concentrations are relatively stable, resulting in C / N ratios primarily determined by organic carbon content, ranging between ∼ 8 and 9 (glacials) and ∼ 14 and 17 (interglacials). The carbon preference index (CPI) is generally high, reflecting a continuous input of immature terrestrial organic matter. Minimum CPI values of ∼ 2.8–2.9 are reached at the transitions from the coarser sediments to the clay intervals, after which they increase to maxima of 4.5–5.0 in the late interglacials. The n-C23+25 Sphagnum biomarker correlates consistently with the T / M ratio, % AP, and cold-water dinocysts (Fig. 3), while the variation in the CPI index is partially out of phase; it is more gradual and lags the % TOC and other signals. Generally lower BIT index values during interglacials (Fig. 3) indicate more marine conditions, i.e., larger distance to the coast and relatively reduced terrestrial input from the Eridanos catchment (see Sinninghe Damsté, 2016). As both brGDGT input (runoff, soil exposure, and erosion) and sea level (distance to the coast) vary across G–IG timescales, for example during deglaciation and subsequent reactivation of fluvial transport (Bogaart and van Balen, 2000), the variability in the BIT index is somewhat different compared to the T / M palynomorph ratio (Fig. 3), but is generally in phase with gradual transitions along G–IG cycles. The MATmr-based temperature reconstructions vary between 5 and 17 ∘C, reaching maximum values in MIS 97. However, in the MIS 99 / 98 and MIS 96 / 95 transitions the MATmr shows variability opposite to the identified G–IG cycles and the signal contains a lot of high-order variability. Low values during interglacials generally coincide with low CBT'-reconstructed soil pH of < 6.0 (Fig. 3).

The source area study by Kuhlmann et al. (2004) indicated the Eridanos paleoriver as the principal source of the terrestrial deposits. The detailed seismic interpretations indeed show the advancing Eridanos delta front from the east toward the sites, especially between 2.44 and 2.34 Ma (Fig. S1). This trend is captured by the long-term increases in the T / M ratio and the proportion of coastal dinocysts (Fig. 3). Bisaccate pollen is the component most sensitive to differential transport processes, yet regardless of whether it is included in the T / M index (Fig. S5 in the Supplement) the same patterns are recorded, indicating no direct influence of differential transport on the T / M ratio in this dataset. During MIS 103, 99, 97, 95, and 93 the percentage of AP increases indicate generally warmer and more humid conditions than during MIS 102, 98, 96, and 94 (Fig. 3). The cold-water temperature signal based on dinocysts is more variable than the terrestrial cooling signals from the percentage of AP. Pollen assemblages represent mean standing vegetation in the catchment and also depend on dominant circulation patterns and short-term climate variations (Donders et al., 2009). Due to exclusion of bisaccate pollen, the percentage of AP is generally low but eliminates any climate signal bias due to the direct effect of sea level changes (Donders et al., 2009; Kotthoff et al., 2014). In the record there are small but significant time lags between proxies, which have important implications for explaining the forcing of G–IG cycles. In the best-constrained MIS transition (98 to 97), the G–IG transition is seen first in decreases of the cold-water dinocysts and n-C23+25 n-alkanes predominantly derived from Sphagnum. Subsequently, the BIT decreases, and MATmr and the percentage of AP increase, and finally the δ18Ob and T / M ratio decrease with a lag of a few thousand years (Fig. 3). Changes in the CPI record are more gradual but are generally in line with T / M. The percentage of AP and T / M proxies have the most extensive record, and detailed analysis of several G–IG transitions shows that the declines in percentage of AP consistently lead the T / M increases by 3–8 kyr based on the present age model (Fig. S2). The T / M ratio variability corresponds well to the LR04 benthic stack (Fig. 3), which is primarily an obliquity signal. Within the constraints of the sample availability, our record captures the approximate symmetry between glaciation and deglaciation typical of the early Pleistocene (Lisiecki and Raymo, 2005).

The high variability and strongly depleted values in δ18Ob during MIS 95 occur during peak coastal dinocyst abundances, suggesting high runoff during maximum warming phases. During cold-water dinocyst maxima, the high abundance of Protoperidinioids indicates high nutrient input and productive spring–summer blooms, which point to strong seasonal temperature variations. This productivity signal markedly weakens in MIS 94 and 92 and the gradual T / M increase is consistent with the basin infill and gradually approaching shelf-edge delta (Fig. S1). As Protoperidinioid minima generally occur during TOC maxima there is no indication for a preservation overprint since selective degradation typically lowers relative abundances of these P cysts (Gray et al., 2017). Combined, the high TOC and CPI values, coastal and stratified water conditions, and intervals of depleted δ18Ob document increased Eridanos runoff during interglacials. These suggest a primarily terrestrial organic matter source that, based on mineral provenance studies (Kuhlmann et al., 2004) and high conifer pollen abundance documented here, likely originated from the Fennoscandian Shield. The finely grained material during cold phases is probably transported by meltwater during summer from local glaciers that have developed since the late Pliocene at the surrounding Scandinavian mainland (Mangerud et al., 1996; Kuhlmann et al., 2004).

Whereas the BIT index reflects the G–IG cycles consistently, the MATmr record, which is based on GDGTs, has a variable phase relation with the G–IG cycles and high variability. The use of MATmr in coastal marine sediments is based on the assumption that river-deposited brGDGTs reflect an integrated signal of the catchment area. As the Eridanos system is reactivated following glacials, glacial soils containing brGDGT are likely eroded, causing a mixed signal of glacial and interglacial material. The lowest MATmr and highest variability is indeed observed during periods of deposition of sediments with a higher TOC content and minima of CBT'-derived pH below 6 (Fig. 3), consistent with increased erosion of acidic glacial (peat) soil. Additional analysis of the apolar fractions in part of the samples reveals a relatively high abundance of the C31 17α, 21β-homohopanes during these periods, which in immature soils indicates a significant input of acidic peat (Pancost et al., 2003). This suggests that the variability in the MATmr record is not fully reliable due to (variable) erosion of glacial soils or peats. Alternatively, the terrestrial brGDGT signal may be altered by a contribution of brGDGTs produced in the marine realm. BrGDGTs were initially believed to be solely produced in soils, but emerging evidence suggests that brGDGTs are also produced in the river itself (e.g., Zell et al., 2013; De Jonge et al., 2014b) and in the coastal marine sediments (e.g., Peterse et al., 2009; Sinninghe Damsté, 2016). Based on the modern system, the degree of cyclization of tetramethylated brGDGTs (#ringstetra) has been proposed to identify a possible in situ overprint (Sinninghe Damsté, 2016). The #ringstetra in this sediment core is <0.37, which is well below the suggested threshold of 0.7, and thus suggests that the brGDGTs are primarily soil derived. However, a ternary diagram of the brGDGT distribution shows some offset to the global soil calibration that decreases with increasing BIT values (Fig. S6 in the Supplement), pointing to some influence of in situ GDGT production when terrestrial input is relatively low. Finally, selective preservation in the catchment and during fluvial transport may have affected the brGDGT signal, although experimental evidence on fluvial transport processes indicates that these do not significantly affect initial soil-brGDGT compositions (Peterse et al., 2015).

The classic Milankovitch model predicts that global ice volume is forced by high northern summer insolation (e.g., Hays et al., 1976). Raymo et al. (2006) suggested an opposite response of ice sheets in both hemispheres due to precession forcing, canceling out the signal and amplifying obliquity in the early Pleistocene. That hypothesis predicts that regional climate records for both hemispheres should contain a precession component that is not visible in the sea level and deep-sea δ18Ob record and is supported by evidence from Laurentide Ice Sheet melt and iceberg-rafted debris of the East Antarctic ice sheet (Patterson et al., 2014; Shakun et al., 2016). Alternatively, a dominantly obliquity-forced G–IG cycle is supported by a significant temperature component in the deep-sea δ18Ob temperature record (Sosdian and Rosenthal, 2009) and dominant 41 kyr variability in North American biomarker dust fluxes. Our results show that the regional NH climate on both land and sea surface vary on the same timescale as the local relative sea level, which, with the best possible age information so far (Fig. S3), mirrors the global LR04 δ18Ob record. The temperature changes lead the local sea level by 3–8 kyr, which is consistent with a NH obliquity forcing scenario as cooling would precede ice buildup and sea level change. Contrary to the model proposed by Raymo et al. (2006), this suggests that the NH obliquity forcing is the primary driver for the G–IG in the early Pleistocene, although we cannot exclude precession forcing as a contributing factor. Various studies indicate the importance of gradual CO2 decline in the intensification of NHG (Kürschner et al., 1996; Seki et al., 2010; Bartoli et al., 2011) combined with the threshold effects of ice albedo (Lawrence et al., 2010; Etourneau et al., 2010) and land cover changes (Koenig et al., 2011). Simulations of four coupled 3-D ice models indicate that Antarctic ice volume increases respond primarily to sea level lowering, while Eurasian and North American ice sheet growth is initiated by temperature decrease (de Boer et al., 2012). The latter dominate the eustatic sea level variations during glacials. Our observations agree with the modeled temperature sensitivity of NH ice sheet growth. The dominant obliquity signal further suggests a seasonal aspect of the climate forcing. The combination of high summer productivity, based on increased Protoperidinioid dinocysts, and increased proportions of cold dinocysts during the glacials in the SNS record indicate a strong seasonal cycle. This confirms similar results from the North Atlantic (Hennissen et al., 2015) and is consistent with an obliquity-driven G–IG signal in a midlatitudinal setting, likely promoting meridional humidity transport and ice buildup.

The southward migration of Arctic surface water masses indicated by increases in cold-water dinocysts (Fig. 3) is furthermore relevant for understanding the relation between the Atlantic meridional overturning circulation (AMOC) intensity and ice sheet growth (e.g., Bartoli et al., 2005; Naafs et al., 2010). Mid-Pliocene increased heat transport and subsequent decrease during NHG due to AMOC intensity changes has been invoked from many proxy records but is difficult to sustain in models (Zhang et al., 2013). Our results indicate that the NW European early Pleistocene climate experienced significant cooling in all temperature-sensitive proxies during sea level lowstands, which is consistent with southward displacement of the Arctic front and decreased AMOC (Naafs et al., 2010). The MATmr indicates a 4–6 ∘C G–IG amplitude, although the timing is offset relative to the other proxies. The data–model mismatch in AMOC changes might be due to dynamic feedbacks in vegetation or (sea) ice (Koenig et al., 2011; de Boer et al., 2012) that are prescribed variables in the model comparison by Zhang et al. (2013).

In addition, our SNS record provides a well-dated early Pleistocene G–IG succession integrating marine and terrestrial signals improving on the classic terrestrial Praetiglian stage. While conceptually valid, the earliest Pleistocene glacial stages defined in the continental succession of the SE Netherlands (Van der Vlerk and Florschütz, 1953; Zagwijn, 1960) and currently considered textbook knowledge are highly incomplete and locally varied (Donders et al., 2007). This shallow marine SNS record provides a much more suitable reflection of large-scale transitions and trends in NW Europe and merits further development by complete recovery of the sequence in a scientific drilling project (Westerhoff et al., 2016).