Diols were analyzed by silylation of the polar fraction with 10 µL N,O-Bis(trimethylsilyl)trifluoroacetamide (BSTFA) and 10 µL pyridine, heated for 30 min at 60 ∘C and adding 30 µL of ethyl acetate. Diol analysis was performed using a gas chromatograph (Agilent 7990B GC) coupled to a mass spectrometer (Agilent 5977A MSD; gas chromatography–mass spectrometry, GC–MS) and equipped with a capillary silica column (25 m × 320 µm; 0.12 µm film thickness). The oven temperature regime was as follows: held at 70 ∘C for 1 min, increased to 130 ∘C at 20 ∘C min-1, increased to 320 ∘C at 4 ∘C min-1, held at 320 ∘C for 25 min. Flow was held constant at 2 mL min-1. The MS source temperature was held at 250 ∘C and the MS quadrupole at 150 ∘C. The electron impact ionization energy of the source was 70 eV. The diols were quantified using selected ion monitoring (SIM) of ions m/z 299.4 (C28 1,14-diol), 313.4 (C28 1,13-diol, C30 1,15-diol), 327.4 (C30 1,14-diol), and 341.4 (C32 1,15-diol; Versteegh et al., 1997; Rampen et al., 2012).

The fractional abundance of the C32 1,15-diol is expressed as a percentage of the total major diols as follows: FC321,15=[C321,15][C281,13]+[C301,13]+[C301,15]+[C321,15]×100.

GDGTs in the polar fractions of the extracts of the VA and LOCO core-top sediments were analyzed on an Agilent 1100 series LC/MSD SL following the method described by Hopmans et al. (2016). The BIT index was calculated according to Hopmans et al. (2004). We calculated the #ringstetra as described by Sinninghe Damsté et al. (2016) and the cyclization branched tetraether (CBT) index and soil pH as described by Peterse et al. (2012): #rings  tetra=GDGTIb+2×GDGTIcGDGTIa+GDGTIb+GDGTIc,CBT=log⁡GDGTIb+GDGTIIbGDGTIa+GDGTIIa,pH=7.9-1.97×CBT.

FC321,15 in surface sediments across the Mozambique Channel varies from 2.3 to 12.5 % (Fig. 1d, f) with one of the highest values in front of the Zambezi River mouth (10 %). The core tops located in front of other minor northern rivers (Licungo and Ligonha rivers) are also characterized by values of F1,15-C32 (> 7.5 %) higher than those further away from the coast (< 5 %). The major diol in all Mozambique surface sediments is the C30 1,15-diol (57.5 ± 9.9 %) with lower amounts of the C30 1,14-diol (21.1 ± 6.0 %) and the C28 1,14-diol (13.2 ± 4.9 %; Fig. 1f).

The values for the BIT index in surface sediments across the Mozambique Channel vary from 0.01 to 0.42 (Fig. 1c). BIT values are highest in the most northern region (0.4) and in front of river mouths (0.2–0.3) compared to values found close to the coast of Madagascar (< 0.04). Following Sinninghe Damsté (2016), we calculated the #ringstetra (the relative abundance of cyclopentane rings in tetramethylated branched GDGTs) to determine if the brGDGTs are produced in situ in the surface sediments or derived from the continent. The #ringstetra has an average of 0.39 ± 0.03 with higher values in front of the river mouths (with the highest values close to the Madagascar rivers) and shows a clear decrease towards the open ocean (Fig. 1d). The low #ringstetra indicate that there is likely limited in situ sedimentary production of brGDGTs in the sediments of the Mozambique coastal shelf area except for the samples closest to the Madagascar coast where high #ringstetra values and low BIT values indicate in situ production of brGDGTs. However, for the Mozambique shelf, the brGDGTs are mostly derived from the continent, confirming the use of the BIT index as a tracer for riverine input in this region.

In the sediments of the Mozambique Channel core 64PE304-80, FC321,15 shows a wide range; it varies from 2.4 to 47.6 % (Fig. 2). Between 44 and 39 ka the values are relatively stable (average of 27.6 ± 4.5 %); then they rapidly decline between 39 and 36 ka to 11 %. From this point on they gradually increase, reaching 37.4 % at 17 ka. FC321,15 then rapidly decreases until it reaches the lowest values of the record after 12 ka (average of 4.9 ± 1.4 %). Holocene sediments (0–11 ka) show relatively low and constant values of FC321,15 (5 ± 1.5 %), similar to the values found in the surface sediments of the area, i.e., 3.5 ± 1.6 % (Figs. 1 and 2d).

The BIT index record (data from Kasper et al., 2015) shows similar changes as that of FC321,15. Between 44 and 39 ka, the average BIT value is 0.43 ± 0.06; then the BIT value decreases to 0.36 at 36 ka, followed by an increase until 17 ka to reach a value of 0.6, while the Holocene values are constant and average at 0.1 ± 0.02. The #ringstetra of branched GDGTs is constantly low (average 0.15 ± 0.01; Fig. S1a) between 44 and 15.5 ka, then increases to 0.4 at 8 ka, and stays constant until the end of the Holocene (average 0.34 ± 0.03). Overall, these values are low and do not approach the values (0.8–1.0) associated with in situ production of branched GDGTs in coastal marine sediments (Sinninghe Damsté, 2016). The #ringstetra also shows a negative correlation with the BIT index throughout the record (R2 = 0.74, p < 0.05), indicating that when BIT values are high, #ringstetra is low. Therefore, high BIT values can definitely be associated with terrestrial brGDGT input. If we assume that the in situ production of brGDGTs in the river (e.g., De Jonge et al., 2015; Zell et al., 2015) is minimal, we can then infer sources of soils from the different catchment areas by reconstructing the soil pH via the CBT index (see Eqs. 3 and 4, Peterse et al., 2012). This showed a constant soil pH (average 6.2 ± 0.1) from 43 to 15 ka followed by a slight increase to 7 at 8 ka and then a constant value (average 6.8 ± 0.08) at the end of Holocene (Fig. S1b).

In eastern Mediterranean sediment core GeoB 7702-3, FC321,15 ranges from 3.9 to 47.0 %. Between 24 and 15 ka the values slowly decrease from 41 % at 24 ka to 7 % at 15 ka. Subsequently, FC321,15 rises sharply until 11.7 ka (44 %) followed by a sharp decrease down to 16 % at 10 ka. FC321,15 increases again up to 30 % until 7.5 ka, followed by a slow decrease in the late Holocene towards values as low as 6 % (Fig. 3a). The BIT index (data from Castañeda et al., 2016) varies in a similar way to FC321,15. It is constant between 24 and 17 ka (average 0.37 ± 0.05) and then decreases to 0.13 at 14.5 ka. It subsequently increases between 15.6 and 9 ka before it decreases after 9 ka and stays constant in the Holocene (average 0.17 ± 0.05). The #ringstetra of the brGDGTs (Fig. S1c) is constant from 24 to 15 ka (0.37 ± 0.05), then shows lower values from 15 to 7 ka (0.29 ± 0.04), and, finally, increases again during the late Holocene (0.40 ± 0.05). The BIT index and #ringstetra do not show a clear negative correlation as observed for the Mozambique core. However, the values of #ringstetra are well below 0.8–1.0, suggesting that in situ production of brGDGTs does not play an important role, in line with the depth from which the core was obtained, which is well below the zone of 100–300 m where in situ production is most pronounced (Sinninghe Damsté, 2016). During parts of the record, low #ringstetra are associated with high BIT values, indicating that between 24 and 7 ka, the brGDGT are mainly terrigenous. For the oldest part of the core, the soil pH shows a stable period from 24 to 14.8 ka (average 6.94 ± 0.07) and then increases to 7.3 at 15 ka, followed by a large decrease (pH reaching 6.5 at 8.5 ka). As the in situ production of brGDGT is likely to be minimal in the latest part of the Holocene, and assuming that riverine production of brGDGTs is minimal, the soil pH can be reconstructed via the CBT index and shows a stable pH (average of 6.8 ± 0.1).

In the surface sediments of the Mozambique Channel, FC321,15 is relatively low overall (< 10 %) in comparison with other coastal regions with substantial river input (Fig. 1f), where values can be as high as 65 % (De Bar et al., 2016; Lattaud et al., 2017). Moreover, the BIT values are also relatively low at 0.01–0.42. Further confirmation of the low amount of terrestrial input in the analyzed surface sediments comes from the low C / N values (between 4.2 and 8.9 for the VA surface sediments; Schulz et al., 2011), characteristic of low terrestrial OM input (Meyers, 1994). Nevertheless, the slightly higher values of both the BIT index and the FC321,15 near the river mouths indicate that both proxies do seem to trace present-day riverine input into the Mozambique Channel in line with earlier findings of other coastal margins influenced by river systems (De Bar et al., 2016; Lattaud et al., 2017).

We compared the record of FC321,15 with previously published proxy records, in particular the BIT index (Kasper et al., 2015) and log(Ca / Ti) (van der Lubbe et al., 2016). These two proxies show the same pattern as FC321,15 (Fig. 2). Indeed, the BIT index and FC321,15 are strongly correlated (r2= 0.83, p < 0.001). Since the #ringstetra of brGDGTs varies between 0.06 and 0.4 (Supplement Fig. S1a), and is significantly negatively correlated with the BIT values, the brGDGTs are predominantly derived from the continent (cf. Sinninghe Damsté, 2016), and thus the BIT likely reflects riverine input into the marine environment. Furthermore, FC321,15 also shows a significant negative correlation with log(Ca / Ti) (r2= 0.43, p < 0.0001, van der Lubbe et al., 2016). This is another proxy for riverine input since Ti is mainly derived from the erosion of continental rocks transported to the ocean through rivers, whereas Ca derives predominantly from the marine environment.

The records of FC321,15 and the BIT index show three major variations: a steep drop from 19 to 10 ka, a slow increase from 38 to 21 ka during the last glacial, and a steep decrease between 40 and 38 ka. The largest change in the BIT index and FC321,15 is between 19 and 10 ka, i.e., a major drop which coincides with an interval of rapid sea level rise (Fig. 2b). Following Menot et al. (2006), we explain the drop in the BIT index, and consequently also the drop in FC321,15, by the significant sea level rise occurring during this period. Rising sea level flooded the Mozambique plateau, moving the river mouth further away from the core site and establishing more open-marine conditions. This most likely resulted in lower FC321,15 and BIT values, conditions that remained throughout the Holocene. The decrease in the delivery of terrestrial matter is also seen in element ratios (Fe / Ca) and organic proxies (BIT) in nearby core GeoB9307-3 (Schefuß et al., 2011), which is located closer to the present-day river mouth in the Mozambique plateau (Fig. 1a). Likewise, the gradual increase in the BIT index and FC321,15 between 38 and 21 ka occurred at a time when sea-level was decreasing (Fig 2b., Grant et al., 2014; Rohling et al., 2014), and thus the river mouth came closer to our study site. The decrease in BIT values and FC321,15 during 40–38 ka coincides with Heinrich Event 4 (H4), a cold and dry event in this part of Africa (Partridge et al., 1997; Tierney et al., 2008; Thomas et al., 2009), with dry conditions likely leading to a reduced riverine input into the ocean and thus a reduced input of brGDGTs and the C32 1,15-diol.

Interestingly, there are two periods where BIT and FC321,15 records diverge (Fig. 2a): during the YD (12.7–11.6 ka) and H1 (17–14.6 ka) with the BIT index decreasing ca. 1 ky later than FC321,15. Comparison with the Ca / Ti ratio shows that both during H1 and the YD, the Ca / Ti ratio increased at the same time as the C32 1,15-diol but earlier than the BIT index, suggesting that the latter was influenced by other parameters. The BIT index is the ratio of brGDGTs (produced mostly in soil or in situ in rivers in this area based on the low values for #ringstetra; Sinninghe Damsté, 2016) to crenarchaeol (produced mainly in marine environments; Schouten et al., 2013, and references cited therein). As both the Ti / Ca ratio and FC321,15 indicate a decrease in riverine input, a constant BIT index can be explained in two ways: a simultaneous decrease in crenarchaeol (marine) production or a change in soil input with higher brGDGT concentrations eroding into the river. The concentration of crenarchaeol during H1 is relatively stable, but there is a slight decrease in crenarchaeol during the YD (Fig. S2b). Thus, the difference between BIT and FC321,15 during the YD can be partly explained by decreased crenarchaeol production together with a decrease in branched GDGTs due to a reduced river flow leading to relatively stable BIT values. In contrast, crenarchaeol and brGDGT concentrations are relatively stable during H1, and thus the lower river input, as indicated by the Ca / Ti and FC321,15, apparently did not lead to a decrease in brGDGT input. This could be due to a shift in the sources of soil which are eroded in the river; i.e., if in this period there is a shift towards soils with relatively higher brGDGT concentrations, the BIT index would remain high despite decreased river flow.

A shift in soil sources may be due to two major changes that happened during this period (and also during the YD), i.e., a shift in the catchment area of the Zambezi River (Schefuß et al., 2011; Just et al., 2014) and a shift in the relative influence of the Zambezi River versus northern Mozambique rivers (van der Lubbe et al., 2016). The shift in catchment area is evident from the higher influx of kaolinite-poor soil into the marine system during H1 and the YD (Just et al., 2014) coming from the cover sands of the coastal Mozambique area (Fig. 3d, blue circle), relative to the kaolinite-rich soils of the hinterlands (Fig. 3d, red circles). If the brGDGT concentrations from the latter region are higher, then this change in soil input could lead to a stable brGDGT flux into the marine environment, despite decreasing Zambezi River runoff. Support for a shift in soil sources comes from the soil pH record reconstructed from brGDGTs, which during the YD shows a shift towards more acidic soils. However, no change in soil pH is observed during H1.

The influence of other rivers (Lurio and Rovuma rivers) relative to the Zambezi River (Fig. 3d, green circle) was inferred from neodymium isotopes by Van der Lubbe et al. (2016); i.e., more radiogenic rocks are found in the northern river catchments in comparison to the rocks in the Zambezi catchment (Fig. 2b). These authors found that during H1 and the YD, the relative contribution of the northern rivers is lower than normal, likely due to drought conditions north of the Zambezi catchment area (Tierney et al., 2008, 2011; Just et al., 2014). These northern rivers run through a catchment containing mainly humid highstand soils, which are different soil types than those observed in the catchment area of the Zambezi River (van der Lubbe et al., 2016). We hypothesize that higher brGDGT concentrations in the soils of the catchment areas of the Zambezi River can potentially explain the discrepancy between BIT and FC321,15; i.e., during H1 and the YD, there is more input of brGDGT-rich soils from the Zambezi than brGDGT-poor soils from the northern rivers, leading to constant BIT values despite a dropping riverine input. Further research examining the brGDGT contents of soils in the different river catchment areas as well as surface sediments from offshore of these northern rivers is required to distinguish between the different hypotheses.

With the eastern Mediterranean Sea core, we compared FC321,15 in core GeoB7702 with other proxies including the BIT index, log(Ca / Ti) and strontium isotopes, the latter to infer the relative importance of the Blue Nile and the White Nile as source regions (Fig. 4c–e). The BIT values (data from Castañeda et al., 2010) show a significant positive correlation with FC321,15 (r2=0.38, p < 0.05), while log(Ca / Ti) shows a negative correlation to FC321,15, again supporting a continental origin of the C32 1,15-diol. FC321,15 and BIT records show much lower Holocene values (12 ± 6 %) compared to the pre-Holocene (27 ± 11 %), which again can be attributed to the sea level rise occurring during the last deglaciation; i.e., our study site was further away from the river mouth and the amount of continental-derived OM reaching the site decreased. Both records show low values during H1 comparable to the Holocene. These low values can be attributed to extreme aridity in the Nile River catchment (Castañeda et al., 2016), which we hypothesize led to a lack of vegetation and enhanced soil erosion but also to a severely reduced river flow, thereby decreasing the net amount of river-borne OM reaching our core site.

In this core, there are three major discrepancies observed between the BIT index and FC321,15: (1) during the LGM, between 22 and 19 ka, where the C32 1,15-diol shows a decrease while the BIT index remains constant; (2) during the onset of the deposition of S1 (6.1–10.5 ka; Grant et al., 2016) where the BIT index decreases later than the C32 1,15-diol; and (3) after 2 ka when the BIT index increases while the C32 1,15-diol decreases. For the LGM FC321,15 is decreasing and log(Ca / Ti) is as well, but the BIT index remains constant (and the brGDGT concentration is also low; see Fig. S3), indicating that there is no significant decrease in terrigenous OM reaching the core site at that time. During the LGM, there is no significant change in continental climate, based on the findings of Castañeda et al. (2016), suggesting no change in vegetation cover or river flux. This suggests that the change in FC321,15 is not due to a change in the input of C32 1,15-diol but in other, mainly marine-derived, diols, in particular the C30 1,15-diol. If this hypothesis is true, then an increase in this marine diol will lower the FC321,15, but if the amount of crenarchaeol is not changing at the same time, the BIT values will remain unaffected.

The deposition of S1 is described as a period of increased freshwater input leading to stratification and anoxia (Rossignol-Strick et al., 1982). However, an increased river input is neither reflected in FC321,15 nor in the BIT index; in fact both of them are asynchronously decreasing. Castañeda et al. (2010) showed that the decrease in the BIT index is due to a large increase in crenarchaeol (Fig. S3), much larger than the increase in brGDGTs, due to increased productivity and preservation. A similar scenario may apply for the diols; i.e., the marine diols (in particular the C30 1,15-diol; data not shown) are, at that time, also increasing more substantially than the C32 1,15-diol, thus lowering FC321,15. However, there is a difference in timing; i.e., the BIT index decreases slightly later than the C32 1,15-diol (9.1 and 10.5 ka, respectively). The decrease in FC321,15 coincides with a substantial increase in sea level (Fig. 4b), which would cause the distance between the core site and the river mouth to increase, thereby decreasing the amount of terrigenous material reaching the site. This terrigenous decrease is also visible to some extent in the log(Ca / Ti) but not in the BIT index. Possibly, like with the Mozambique Channel, the brGDGT concentrations in the river were much higher at that time. Indeed, the Sr isotopic record suggests a major shift from a Blue Nile to a White Nile source at 10.5 ka, with the latter possibly containing more eroded soils with high brGDGT concentrations. This shift in soil sources is also shown in the change towards more acidic soil pH during that period based on the CBT index (Fig. S1d).

For the most recent part of the record (0–5 ka), the BIT index increases, while FC321,15 slightly decreases. The δDleaf waxes (Fig. 4d) shows it was period of mild aridity which likely led to a decreased riverine runoff and thus decreased river input. The reason the BIT index is increasing rather than decreasing is due to an increase in brGDGT concentrations (Fig. 3b), despite evidence for a decrease in river runoff. This can possibly be linked to the amount of vegetation in the Nile catchment; i.e., at that time there was a decrease in vegetation cover (Blanchet et al., 2014; Castañeda et al., 2016), which led to more soil erosion and thus potentially a higher brGDGT concentration in rivers and a higher BIT index. This hypothesis is supported by the log(Ca / Ti) (Fig. 4c), which decreases at this time, suggesting that soil runoff was increasing.

Our results from both the Nile and Mozambique Channel cores illustrate that FC321,15 provides a suitable proxy for reconstructing past riverine input into coastal seas. Although some discrepancies are noted with other terrigenous proxies for both cores, FC321,15 generally agrees well with these proxies. However, our interpretation of the C32 1,15-diol record relies on the assumption that production of this diol in rivers does not change with different hydroclimate fluctuations on land, something that needs to be tested. However, De Bar et al. (2016) showed that FC321,15 in the Tagus River in Portugal did not significantly change over the course of a year, suggesting that this assumption might be valid. Since the C32 1,15-diol is mainly produced in rivers itself, it is not impacted by vegetation abundance and soil composition, in contrast to other proxies like the BIT index and lignin concentrations. This may make it a potentially more reliable proxy to trace past river input into marine environments.