Desfile de la Victoria

The calculated African monsoonal runoff presented here (Fig. 5.2) represents the first detailed, continuous quantification of excess freshwater influx into the eastern Mediterranean for S5. All features of the African monsoonal runoff reconstruction over the S5 interval are responses solely to the input records, with the exception of an isotopically depleted Atlantic inflow that was imposed in the model from 132.6 to 130 ka to account for Atlantic freshwater entering the Mediterranean during HS11 (see section

5.2.2). For comparison, a second model run was conducted in which no isotopically depleted Atlantic inflow is added over the HS11 interval.

The generously calculated monsoonal runoff uncertainty allows for full propagation of errors from several different inputs to the model (i.e., sea-level, SST, δ18Oruber), some of

which have large uncertainties themselves. However, despite these uncertainties, a clear underlying signal remains in the monsoonal runoff reconstruction.

HS11

A negative monsoonal runoff volume as calculated for ~132.3 – 129 ka (Fig. 5.2) is not possible. However this does not immediately invalidate the results. Part of this interval (132.6 -130 ka) is forced with an Atlantic seawater δ18O calculated for the Iberian Margin (Skinner and Shackleton, 2006) to represent the Atlantic freshwater influence in the Mediterranean during HS11 (see section 5.2.2). The additional model run, which was not forced with an isotopically depleted Atlantic inflow, displays a much shorter period of negative monsoonal runoff from 130.5 to 129.3 ka. This may indicate that the imposed Atlantic influence during HS11 has been overestimated, and that part of the negative monsoonal input is needed for the model to compensate and hit the target δ18_O

ruber.

However, that the median monsoonal runoff is negative for a period of time in both model runs, with and without an imposed HS11 Atlantic freshening, suggesting that at least part of the negative runoff volume captures a real climatic signal.

The runoff calculated is essentially the difference in the Mediterranean evaporation- precipitation-runoff balance relative to the present day. The model estimates the precipitation and evaporation components and the remainder is accounted for in the model by a change in runoff. However, an unquantified change in the Mediterranean hydrological balance in the past may therefore not be fully accounted for in the evaporation-precipitation calculations in the model, and potentially falsely assigned as a change in monsoonal runoff. Hence, the negative monsoonal runoff calculated for ~132- 129 ka may be representative of increased aridity and thus evaporation throughout the Mediterranean basin at this time. To account for the median negative monsoon runoff calculated during the model run with imposed HS11 Atlantic freshening, evaporation over the Mediterranean would have to increase by ~12.5% between 132.3 and 129 ka. Without

the HS11 imposed freshening, evaporation would have to increase by ~7.7% between 130.5 and 129.3 ka. Especially high aridity has been observed in the Mediterranean region during Heinrich stadials (Combourieu Nebout et al., 2002; Niedermeyer et al., 2009; Collins et al., 2013; Torfstein et al., 2018), suggesting that such an increase in evaporation throughout the Mediterranean at this time may be possible.

Ensuring that the evaporation-precipitation balance in the Mediterranean box model can account for past changes in aridity/humidity in the Mediterranean region is a challenge, but an important step to improve the model. While the negative monsoonal runoff has highlighted the issue of a more arid past climate, likewise, the model may be unable to account for shifts to more humid climates. In this case, unaccounted for increases in precipitation (relative to evaporation) may cause biases to higher monsoonal runoff estimates.

Sapropel onset

The clearly defined increase in monsoonal runoff at the start of S5 deposition at ~128.3 ka (Fig. 5.2) indicates a very rapid development of basin conditions enabling sapropel preservation following increased freshwater influx to Mediterranean surface waters. Rapid onset of sapropel S5 following shifts to lower δ18Oruber, has been reported for

several eastern Mediterranean sites (e.g., Schmiedl et al., 2003; Morigi, 2009). However, later onset of S5 in other sites suggests that the start of sapropel deposition was staggered throughout the eastern Mediterranean (e.g., Cane et al., 2002; Schmiedl et al., 2003; Rohling et al., 2006).

It has been suggested that persistent meltwater discharge into the North Atlantic may have delayed the timing of increased African monsoonal runoff for sapropels, and caused longer lags in the start of sapropel deposition relative to insolation (Grant et al., 2016). Indeed, African monsoonal runoff during S5 only increases after HS11 at ~129.5 ka (Fig. 5.2). However, the ΔT for S5 is greater than 0 at the start of HS11 (~135 ka) (Fig. 5.1d), suggesting that G. ruber (w) started to occupy a shallower water habitat within the SML at least 7.7 ka before sapropel deposition started (~128.3 ka). This may suggest that increased surface water stratification in the eastern Mediterranean started to become established from ~135 ka, potentially pre-conditioning the basin for sapropel deposition.

Rapid sea-level rise likely contributed to the development of Holocene sapropel S1 by encouraging stratification (Grimm et al., 2015; Grant et al., 2016). Sea-level rise associated with Termination II occurred from ~133 ka (Grant et al., 2012), and my results suggest that sea-level rise potentially also played a role in preconditioning the basin for the apparent rapid onset of S5, which occurred once monsoonal runoff increased. Further investigation is required to validate this suggestion. Notably, my working hypothesis may have simplified the relationship between ΔT and the G. ruber (w) depth habitat, and there are also uncertainties around the exact seasonal and depth origin of the UK’37 SST proxy signal. Nonetheless, abundances of the lower photic zone coccolithophore Florisphaera profunda clearly start to increase from ~130 ka (Fig. 5.2; Grelaud et al., 2012), indicating the development of a deep chlorophyll maximum (DCM), which is associated with pycnocline shoaling and increasing stratification. This is 1,700 years before the start of the calculated African monsoonal runoff (at ~128.3 ka), and potentially supports a pre- conditioning of the eastern Mediterranean for sapropel deposition (Fig. 5.2).

Euxinic conditions in the water column during S5 are evidenced by high concentrations of isorenieratene, which indicates the presence of anaerobic, phototrophic green sulphur bacteria (Chlorobiaceae), which require both sulphide and light (Passier et al., 1999). Isorenieratene measurements over S5 indicate that euxinic conditions extended up toward the photic zone, reaching ~200m in the Aegean Sea (Site LC21; Figs. 5.2, 5.3a; Marino

et al., 2007) and the central eastern Mediterranean (Site ODP 971; Fig. 5.3a; Rohling et al., 2006). Euxinic conditions reached the photic zone in the Aegean at ~127.9 ka, just 400 years after the start of the increase in my calculated African monsoonal runoff median (Fig. 5.2). Based on modern oxygen utilisation rates, consumption of all oxygen below 500 m depth is estimated to take ~600 years following the collapse of deep water formation (Rohling, 1994). However, the evidence from F. profunda indicating the development of a DCM from ~130 ka (Grelaud et al., 2012), and increases in organic carbon (Corg) preserved in bottom sediments from ~129.5 ka (Marino et al., 2007), indicate increased levels of productivity in the eastern Mediterranean more than 1,000 years before the increase in freshwater runoff (Fig. 5.2). High productivity may therefore partly explain the rapid oxygen utilisation and spread of euxinic conditions up to ~200 m in the eastern Mediterranean.

African monsoonal runoff

Rohling et al. (2004) previously provided a quantitative indication of monsoonal runoff intensity during sapropel S5, in a box-model interpretation of planktic foraminiferal δ18_O records through S5 in core KS205. They estimated a monsoon intensification of between 1.6 and 3 times that of present-day runoff for the lower lobe (definition of present-day runoff same as in this study; section 5.3), and between 1.2 and 2 times the present day for the upper lobe. These S5 monsoonal intensification estimates are lower than those calculated in this study (Fig. 5.2), with the most probable monsoon intensification estimates reaching around 4 times the present day in the lower lobe, and 2 to 3 times the present day in the upper lobe. Indeed, data-model comparisons during Mediterranean box model development in Chapter 4 suggested that the monsoonal runoff estimates of Rohling et al. (2004) were not enough to account for the δ18Oruber depletions observed

(section 4.4.1.3). The estimates of Rohling et al. (2004) were likely lower than those obtained here because they were based on the δ18Oruber record from KS205, while this

study accounts for KS205 together with three additional eastern Mediterranean cores (section 5.2.2). The multi-site δ18Oruber comparison (Fig. 5.2) shows that KS205 δ18Oruber

is not depleted to the same extent as in cores ODP 967 and ODP 971. This highlights the advantage of using multiple sites to obtain a more comprehensive understanding of the eastern Mediterranean dynamics during S5.

While that was the only other quantitative estimation specifically made for African monsoonal runoff, several other proxies give broader insight into the hydrological conditions of the eastern Mediterranean region during S5 (Fig. 5.2).

The δ18Oruber for S5 in individual cores gives insight into the extent of the freshwater

influence between different regions (Fig. 5.2; Fig. 5.3a). All records were compiled to produce a δ18_O

ruber stack for the eastern Mediterranean (section 5.2.2), so the calculated

African monsoonal runoff is dependent on all of them. However, the contrasts between δ18_O

ruber signals between sites remain useful when considering freshwater sources and

regional effects, as discussed in Chapter 3. Now, with the African monsoonal runoff estimate, the freshwater influences in the surface waters of the different eastern Mediterranean sites during S5 can be better analysed.

The median African monsoonal runoff during the sapropel interruption (126.5 – 125.7 ka) is the same as the total present-day (pre-Aswan) runoff into the Mediterranean (Fig. 5.2). In terms of African contribution, the present-day runoff is almost entirely derived from the Nile (Struglia et al., 2004). If there were no major changes in European runoff sources to the Mediterranean during the S5 interval, the reduction to a present-day runoff volume during the sapropel interruption could be explained by a shut-down of the palaeo-drainage systems over the wider African margin, as all the runoff during this interval can be accounted for by Nile outflow. When African monsoonal runoff increases again, ending the sapropel interruption at ~125.7 ka, the δ18Oruber in ODP 971 has started to decrease,

suggesting a re-activation of the paleo-drainage systems along the North African margin. However, the ODP 967 δ18Oruber does not decrease again until ~124 ka, which suggests

that Nile discharge stayed at the same levels as during the interruption. Furthermore, eastwards surface circulation would be expected to transport freshwater from the African margin to ODP 967 (Fig. 5.3b), but there is no evidence of this in the ODP 967 δ18Oruber.

However, as seen in the SST diagnostic experiments (section 4.4.1), temperature also has an important influence on δ18Oruber, in particular during S5 when a temperature

concentration effect is observed for the thin, freshwater-diluted upper SML. The Δ47- based SST which is measured directly on G. ruber (w) indicates a lower temperature for the latter half of S5 (Fig. 5.1c), and in particular looking at the individual records of Δ47– based SST for the separate cores, SST for ODP 967 is 1-3°C lower than for LC21 from 126 to 121.5 ka (Rodríguez-Sanz et al., 2017). This may explain the observations of δ18_O

ruber for ODP 967, and also emphasises the importance of considering temperature

when studying δ18_O

ruber over sapropel events. The next step here is to apply the

Mediterranean box model to detangle the temperature and freshwater components of the S5 sapropel imprint on δ18_O

ruberfor separate cores. This would provide much greater

spatial insight into the freshwater influences throughout the eastern Mediterranean. Neodymium isotopic compositions (εNd) of planktic foraminifera have also been used as a freshwater tracer in the Mediterranean (Scrivner et al., 2004; Osborne et al., 2008, 2010), and are available for S5 in three of the cores used in this study (Fig. 5.2; Scrivner et al., 2004; Osborne et al., 2008, 2010). εNd has the added advantage of providing information on freshwater source region based on catchment geology (Goldstein and Hemming, 2003; Scrivner et al., 2004). ODP 971 has a distinct increase in εNd at the start of sapropel S5,

which is less clear in ODP 967 and not apparent at all in LC21 (Fig. 5.2). Both the εNd and δ18Oruber signals together strongly suggest that runoff from paleo-drainage systems

along the African margin was the greatest source of freshwater, and that northern-sourced freshwater was much less significant during S5 (section 3.5; Scrivner et al., 2004; Osborne et al., 2008, 2010). While the εNd records have been instrumental in identifying sources of freshwater to the eastern Mediterranean during S5 (Scrivner et al., 2004; Osborne et al., 2008, 2010), their use is limited by few data points and large uncertainties.

A ‘wet-dry index’ for North Africa (Fig. 5.2; Grant et al., 2017) combines a sapropel/monsoon run-off signal, derived from a principal component analysis of x-ray fluorescence (XRF) data from ODP 967 (Grant et al., 2017), with an ODP 967 Saharan dust record (Larrasoaña et al., 2003b) to provide an integrated index of North African climate variability. The close association of my modelled African monsoonal runoff with a prominent ‘wet’ interval over S5 implies that the runoff timing is strongly correlated to that of North African climate. However, the wet-dry index suggests that HS11 is also a ‘wet’ interval, in contrast to my monsoonal runoff estimate which suggests an arid climate during HS11 (Fig. 5.2).

The titanium/aluminium ratio (Ti/Al) is another indicator of aridity/humidity in North Africa (Fig. 5.2; Konijnendijk et al., 2014). Ti is sourced both in the form of Nile suspended matter and aeolian dust. However, Nile suspended particles that contain titanium are heavier than other suspended matter, and hence are preferentially deposited close to the river mouth. Therefore, sediments originating from the Nile at ODP 967 will be Ti-depleted, and the main Ti contribution to sediment will be from aeolian dust fluxes from the Sahara (Wehausen and Brumsack, 1999, 2000; Calvert and Fontugne, 2001; Lourens et al., 2001; Larrasoaña et al., 2006; Konijnendijk et al., 2014). S5 is associated with a distinct low in Ti/Al, indicating a humid interval with low aeolian fluxes. However, there is no sharp transition into more humid conditions at the start of S5 in Ti/Al, and the pre-S5 interval is not especially arid (high Ti/Al). This contrasts with my African monsoonal runoff estimate which indicates a distinct interval of aridity from ~132-129 ka, with a sharp increase in runoff volume coinciding with the start of S5 (Fig. 5.2). However, as Ti/Al is a ratio of two elements with different environmental controls, the signal may be due to either changes in aeolian or riverine components of the system (or

both) and is not straightforward to interpret. If the Nile discharge was reduced (or even stopped) as suggested by the African monsoonal runoff estimate for ~132-129 ka (Fig. 5.2), an important source of both Ti and Al to ODP 967 sediments would be lost, and would potentially bias the Ti/Al ratio is an unpredictable manner.

Another dust proxy available for ODP 967 over S5 is environmental magnetic parameter IRM0.9T@AF120mT (IRM imparted with an induction of 0.9 T and AF demagnetised at 120 mT) (Fig. 5.2; Penny, 2018). IRM0.9T@AF120mT is a measure of haematite content variations, which is a component of Saharan dust in eastern Mediterranean sediments (see section 2.2; Larrasoaña et al., 2003). As in the Ti/Al record, IRM0.9T@AF120mT displays a distinct low plateau during S5, indicating almost no Saharan dust reaching the Levantine basin (Fig. 5.2). Furthermore, the IRM0.9T@AF120mT is also close to 0 during HS11, indicating a virtually absent Saharan dust influx, almost the same as during S5.

The evidence from two independent Saharan dust proxies of Ti/Al and IRM0.9T@AF120mT (Konijnendijk et al., 2014; Penny, 2018), and thewet-dry index for North Africa (Grant

et al., 2017), imply a reduced influx of Saharan dust to the far eastern Mediterranean during HS11 (Fig. 5.2). This appears to suggest that HS11 was not as arid as is implied by my African monsoonal runoff estimate (Fig. 5.2). However, arid climates over North Africa and the Mediterranean during Heinrich stadials has been evidenced by many different studies (Combourieu Nebout et al., 2002; Niedermeyer et al., 2009; Collins et al., 2013; Torfstein et al., 2018). While it is possible that HS11 is an anomaly among Heinrich stadials, an alternative explanation may be that atmospheric circulation was different during this North Atlantic cold event, and Saharan dust was not transported towards the Levantine basin during HS11. However, to fully investigate this requires a more in-depth investigation of atmospheric circulation during HS11, which is outside the scope of this study.

5.5. Conclusion

Here, Mediterranean planktic foraminiferal δ18O has been deconvolved to produce an estimate of African monsoonal runoff for sapropel S5. The estimated African monsoonal runoff volume suggests that large increases in freshwater runoff to the eastern Mediterranean (up to 4 times that of the present-day (pre-Aswan) total runoff into the

Mediterranean) are required to account for the δ18Oruber signals during sapropel S5

deposition. However, the method also indicates that SST increase via a 'temperature concentration' effect accounts for a large part of the ‘sapropel imprint’ on δ18_O

ruber, and

cannot be ignored (Chapter 4).

While the African monsoonal runoff estimates are novel and useful for understanding past monsoon variability, there remain limitations. Uncertainties in all proxies and parameters used in the model have been accounted for as far as possible, but issues remain regarding our ability to represent more complex features in the model, such as Atlantic freshwater influences during Heinrich Stadials. In particular, the negative monsoonal runoff estimate (~132-129 ka; Fig. 5.2) that is likely due to an increase in aridity that is unaccounted for in the model, highlights the need to account for past changes in the evaporation-precipitation balance in the model. This is especially important as the model may also not be fully accounting for past intervals of increased humidity, which may be expressed as biases to higher monsoonal runoff estimates.

Following on from this study, the Mediterranean box model could be applied to deconvolve the δ18Oruber signals over S5 for individual sites in the eastern Mediterranean.

This would enable the isolation of the freshwater components of the δ18_O

ruber signal for

different sites, which would enable spatial estimates of freshwater influence on surface waters throughout the eastern Mediterranean. In particular, this would provide insight into the circulation and mixing of surface waters throughout the eastern Mediterranean during sapropel events.

The method developed here provides new insights into the hydrological conditions in the eastern Mediterranean and North Africa, and has the potential to be applied to other sapropel events. Furthermore, the ability to deconvolve δ18Oruber and isolate the

freshwater component during sapropel events is a step towards reconstructing Mediterranean sea-level over sapropel events (Rohling et al., 2014). This would be especially valuable because sapropels often occur at interglacial highstands, which are intervals of great interest for understanding sea-level behaviour during warm periods (including projections into the future).

6. The effects of Mediterranean SST

evolution on the Mediterranean sea-level

reconstruction

6.1. Introduction

The Mediterranean sea-level reconstruction for the past 5.3 Myr (Rohling et al., 2014) is the first continuous, millennially resolved reconstruction based on a planktic foraminiferal δ18_{O that extends beyond 0.5 Ma. The Mediterranean sea-level method} utilises a Mediterranean box model, coupled to a hydraulic control model for water exchange at the Strait of Gibraltar, to estimate the expected δ18O of calcifiers residing in