Introduction
The hydrologic cycle is of vital importance for the global climate system,
owing to its function in regulating the heat and moisture balance (Lindzen,
1990, 1994). This is mainly achieved through the atmospheric and ocean
circulation, which tend to equalize the differences in solar heating between
low and high latitudes. The importance of this heat redistribution is
illustrated by the fact that around 50 % of the annual energy budget of
high latitudes originates from lower latitudes (Peixoto and Oort, 1992),
peaking in winter, when polar night conditions prevail at high latitudes
(Davis and Brewer, 2011). In this way, atmospheric water vapour acts as the
dominant greenhouse gas, exerting an important feedback on global warming
(e.g. Held and Soden, 2000). Thus a good understanding of the dynamics of the
hydrological cycle is crucial. However, focusing on the instrumental record
only would limit the state of our knowledge (Brohan et al., 2006). To assess
the dynamics of a changing climate, palaeoclimate studies provide means to
investigate long-term developments of the monsoon (Mohtadi et al., 2016).
Monsoon systems are the most prominent parts of the global hydrological
cycle. Monsoon variability on orbital timescales is mainly attributed to
variations in low-latitude summer insolation, which is dominated by 19 and
23 kyr periodicities of the precession cycle (e.g. Kutzbach, 1981;
Rossignol-Strick, 1983; Tiedemann et al., 1994; Larrasoaña et al., 2013;
Mohtadi et al., 2016). Consequently, maximum insolation intensities are
expected to increase monsoon circulation and thus rainfall, comparable to
modern seasonal variations. Recent sensitivity studies show that the effect
of precession on tropical climate is far stronger than that of obliquity
(Bosmans et al., 2015a). Nevertheless, obliquity (41 kyr) cycles have
also been detected in sequences of Mediterranean sapropels (Lourens et al.,
1996), which formation depends on monsoonal freshwater discharge from Africa
(Rossignol-Strick, 1983; Colleoni et al., 2012; Larrasoaña et al., 2013).
Equally, dust flux rates at Ocean Drilling Program (ODP) Site 659 (Tiedemann
et al., 1994) and pollen flux rates at ODP Site 658 (Dupont et al., 1989)
show obliquity cycles in records of tropical Africa, which are difficult to
explain, since obliquity has a negligible effect on low-latitude insolation
(Laskar, 1990).
A plausible mechanism that incorporates the obliquity signal and the climatic linkage between low and high latitudes is found in the
latitudinal insolation gradient (LIG; Davis and Brewer, 2009). This intra-hemispheric insolation gradient leads to differential heating
between the cold polar regions and the hot tropics and creates the latitudinal temperature/pressure gradient, which ultimately controls
the poleward heat and moisture transport, a mechanism that has been invoked in the so-called “gradient hypothesis” by Raymo and
Nisancioglu (2003) to explain the strong obliquity rhythm of glacial–interglacial cycles between 0.8 and 3.0 Ma (the
“41 kyr world”). Another gradient hypothesis is favoured by Bosmans et al. (2015b), who suggest that the inter-hemispheric
insolation gradient drives winds and associated cross-equatorial moisture transport deep into the African continent linked with an
intensification of the Hadley cell during winter (Reichart, 1997). The same mechanism has been invoked by Leuschner and Sirocko (2003)
defining the Indian summer monsoon index. The concept of insolation gradients is supported by recent climate model studies (Mantsis
et al., 2014) simulating an enhanced mid-latitude eddy circulation (which is important for the heat and moisture transport from low to
high latitudes) in response to low obliquity (resulting in a strong LIG during summer), and a concomitant shift of the poleward
boundaries of the tropical rain belt towards the Equator. In addition, weakening of the inter-hemispheric gradient leads to diminished
cross-equatorial heat transport and a weaker Hadley circulation during winter (Reichart, 1997; Bosmans et al., 2015a). Further model
simulations corroborate the importance of an obliquity-induced insolation gradient forcing for West African monsoon variability
(Rachmayani et al., 2016).
Both gradient hypotheses stay in contrast to the traditional view, in which the obliquity signal in the tropics is linked to changes in
the extent of glaciations. Northern Hemisphere ice sheets and sea ice would affect the monsoon through the impact on the atmospheric
circulation and moisture advection (Tiedemann et al., 1994; DeMenocal, 1995; Mohtadi et al., 2016). Using Pliocene records (between 5
and ∼3 Ma) that register palaeoclimate prior to the intensification of the Northern Hemisphere glaciation, we can test if
the monsoon system responds to obliquity independently of Northern Hemisphere ice growth and decay.
Another good reason to study the warmer-than-today climate of the Pliocene is that many boundary conditions are similar to today
(Dowsett et al., 2013; Haywood et al., 2013). This last warm period of the geological record is often referred to as future analogue. For
the assessment of Pliocene climate conditions, focus has been on sea surface (Dowsett et al., 2013) and continental temperatures
(Salzmann et al., 2013). Global mean annual temperatures were more than 3 ∘C warmer and sea level was about 22±10 m higher than at present (Dowsett et al., 2013; Haywood et al., 2013). However, data–model mismatches have been detected
for terrestrial temperature estimates in the tropics likely due to limited proxy data (Salzmann et al., 2013). Moreover, a model
inter-comparison revealed inconsistencies in tropical precipitation estimates (Haywood et al., 2013), indicating that more data on
tropical hydrology are needed.
Monthly precipitation (blue), δD of precipitation (black relative to Vienna Standard Mean Ocean Water) and temperature
(grey) at Bamako (Mali; 12∘41′24′′ N, 7∘59′24′′ W; 381 m)
exemplifying the amount effect (Dansgaard, 1964) in West African hydrology (IAEA/WMO, 2014). This relationship is shown for monthly
means, but also holds true on an annual basis (Rozanski et al., 1993).
Hydrogen isotopes (deuterium) of plant leaf waxes (δDwax) provide a proxy for palaeohydrologic variations (Fig. 1)
related to the isotopic composition of precipitation (δDrain; Sachse et al., 2012). In tropical regions,
δDrain is negatively correlated to precipitation amounts (Dansgaard, 1964; Rozanski et al., 1993; Risi et al.,
2008), whereas δDwax also incorporates secondary effects of evapotranspiration, plant physiology, and photosynthetic
pathway (Sachse et al., 2012). In areas with C4 grasses, the stable carbon isotopic compositions of the same compounds
(δ13Cwax) provide additional information about vegetation changes and can be used to differentiate between
contributions from grassy (C4) and woody (C3) vegetation (e.g. Vogts et al., 2009).
Location of Ocean Drilling Program (ODP) Sites 659 (star; 18∘05′ N,
21∘02′ W;
3070 m water depth) and 658 (dot; 20∘45′ N, 18∘35′ W; 2263 m water depth)
offshore of West Africa (Ruddiman et al., 1987) and the location of Bamako in Mali. Main vegetation zones (dashed lines; simplified
after White, 1983), dominant photosynthetic pathways (right), atmospheric trajectories (arrows). The thickness of the arrows marks
the different altitudes of the African Easterly Jet (AEJ; >3000 m) and the north-east trade winds (NETW; <1000 m).
Our study focusses on ODP Site 659, situated offshore of West Africa and centred beneath the main wind trajectories (Fig. 2). Its dust
record provides evidence for the persistence of orbital arid–humid cycles during the last 5 Myr (Tiedemann et al.,
1994). Coupled δDwax and δ13Cwax analyses from this site for the last glacial cycle
have demonstrated their ability to record shifts in West African hydrology and vegetation (Kuechler et al., 2013). However,
δDwax studies from West Africa covering the Pliocene have not yet been published. Here we provide orbitally
resolved δDwax and δ13Cwax records for two time intervals (5.0–4.6 Ma and
3.6–3.0 Ma) to evaluate hydrologic changes in Pliocene West Africa. In addition to the well-established precession forcing, we
discuss further insolation mechanisms to explain the long-term evolution of the West African hydrologic cycle.
Seasonal extremes in vegetation (a, b, NDVI, normalized difference vegetation index available at
www.earthobservatory.nasa.gov) and rainfall (c, d, precipitation in mm per day available at http://www.cdc.noaa.gov/)
of northern and West Africa in September 2004 (a, c) vs. March 2004 (b, d).
Modern climate and vegetation
The amount and distribution of precipitation in West Africa is a function of latitude (Nicholson, 2009; Fig. 3). Tropical rainfall is
tightly linked to atmospheric dynamics at higher altitudes, involving the African Easterly Jet (AEJ) and the Tropical Easterly Jet
(Nicholson, 2009). Highest precipitation amounts are found in the equatorial regions, which are marked by a double rainfall maximum,
occurring during the transition seasons, which is mostly attributed to the latitudinal migration of the Intertropical Convergence Zone. Further north a single
rainfall maximum occurs during summer (July/August), when a low-pressure system is formed over the Sahara due to stronger heating of
the northern African continent relative to the adjacent ocean. This low-pressure system is situated between the NE trade winds and the
moist “SW monsoon”. Precipitation during the wet season in Sahara and Sahel moderates the surface temperatures, thus enhancing the
temperature contrast to the hyperarid Sahara in the north, which strengthens the AEJ during its northward displacement after rainfall
(Nicholson and Grist, 2003). Precipitation in the temperate regions of North Africa is influenced by the mid-latitude westerlies
falling mainly during the winter season. The Sahara separates the summer and winter precipitation regimes and thus, exhibits
a complex pattern. Much of the rainfall in the desert occurs during the transition seasons and has an extra-tropical origin related to
the mid-latitude westerlies interacting with the tropical easterlies (Nicholson, 1981, 2000).
The modern vegetation distribution and composition in West Africa (White, 1983) reflects the latitudinal migration of the tropical rain
belt (Fig. 3). Close to the Equator, high precipitation rates lead to dense vegetation cover, consisting of tropical rain forests and
woodlands (mostly C3). Further to the north, towards the Sahara, increasingly open grasslands (mostly C4) occur
as rainfall decreases and wet season length shortens. North of the Sahara, the Mediterranean vegetation is composed of scrublands,
steppes, and forests, which consist exclusively of C3 plants.
Material and methods
ODP Site 659 is located on top of the submarine Cape Verde Plateau at 3070 m water depth (Fig. 2; Ruddiman et al., 1987). The
siliciclastic fraction is considered to be of purely aeolian origin due to low carbonate concentrations in the dust composition and its
distal location on a submarine rise excluding fluvial input (Tiedemann et al., 1994). Two Pliocene age models have been established,
one based on variations in dust accumulation (Tiedemann et al., 1994) and the other based on stable oxygen isotopes
(δ18O) of benthic foraminifers (Clemens, 1999).
Stable isotope analyses of hydrogen (δD) and carbon
(δ13C) were carried out on 230 samples of cores 10, 11, 14,
and 15 of Hole A. The analysed three Pliocene intervals range from 3.00 to
3.27, 3.31 to 3.62, and 4.63 to 5.00 Ma) with an average temporal
resolution of ∼4 kyr, comparable to the dust record of the same
site (Tiedemann, 1991; Tiedemann et al., 1994). A sampling gap between
3.31–3.27 Ma is related to a core break between cores 10 and 11 of
Hole A. A detailed description of methods, including lipid extraction,
quantification and stable isotope analyses, is given in Kuechler
et al. (2013). δD values are reported relative to Vienna Standard Mean
Ocean Water (VSMOW). An external n alkane standard yielded a precision
(1σ) and accuracy of 2 ‰, and squalane as internal standard
yielded values of 2 and 1 ‰, respectively. The mean precision
(1σ) of replicates for the n-C29–33 alkanes is
2 ‰. δ13C values are reported relative to the Vienna
Pee Dee Belemnite (VPDB) standard. The external standard yielded a precision
(1σ) of 0.3 ‰ and an accuracy of <0.1 ‰. The
values for internal standard are 0.2 and 0.3 ‰, respectively.
Replicates yielded a mean precision (1σ) for the
n-C29–33 alkanes of 0.2 ‰. Data sets are stored
online at PANGAEA (10.1594/PANGAEA.875694; Küchler et al., 2017).
Statistical analysis has been carried out using the software package PAST 3.0 (Hammer et al., 2001). To calculate correlation
coefficients data sets have been re-sampled at steps of 4 kyr (close to the original average temporal resolution). In order to
illustrate the temporal evolution of the dust and δDwax records, we applied continuous wavelet transform
(Morlet) after Torrence and Compo (1998). For this spectral analysis, the Pliocene data sets were linearly interpolated to an even
spacing of 5 kyr, and the Pleistocene data set to 3 kyr. Significance (p=0.05) is tested after a chi-squared test with
the null hypothesis of a white noise model.
Correlation coefficients (r) and (p) for plant-wax-specific stable isotopes (δD, δ13C), long-chain
n alkane (n-Cx) and dust percentages at ODP Site 659.
n-Cx
LGC
p
mPWP
p
pre-mPWP
p
EP
p
δD
29/31
0.80
<0.01
0.85
<0.01
0.70
<0.01
0.70
<0.01
31/33
0.74
<0.01
0.83
<0.01
0.58
<0.01
0.63
<0.01
29/33
0.73
<0.01
0.79
<0.01
0.52
<0.01
0.56
<0.01
δ13C
29/31
0.86
<0.01
0.87
<0.01
0.82
<0.01
0.60
<0.01
31/33
0.78
<0.01
0.70
<0.01
0.64
<0.01
0.51
<0.01
29/33
0.62
<0.01
0.66
<0.01
0.73
<0.01
0.41
<0.01
δD vs. δ13C
29
-0.08
0.67
0.42
<0.01
0.20
0.08
0.12
0.24
31
-0.55
<0.01
0.36
<0.01
-0.04
0.76
-0.23
0.03
33
-0.70
<0.01
0.01
0.91
-0.10
0.38
0.05
0.61
δD31 vs. dust %
0.34
0.02
0.41
<0.01
0.21
0.06
0.23
0.02
n-C27–35 vs. dust %
0.36
0.02
0.48
<0.01
0.16
0.16
0.70
<0.01
For statistical analyses (PAST 3.0; Hammer et al., 2001), the late Pliocene
or Piacenzian interval was split into two periods corresponding to the
mid-Piacenzian Warm Period (mPWP; 3.273–3.007 Ma) and a preceding one
(pre-mPWP; 3.616–3.309 Ma) separated by a stratigraphic gap of
∼36 kyr in the sedimentary record. The early Pliocene or
Zanclean interval (EP) ranges from 4.997 to 4.625 Ma. Data sets were re-sampled
at evenly spaced 4 kyr, which is close to the original average temporal
resolution. Values for the last glacial cycle (LGC; last 130 kyr after
Kuechler et al., 2013) are given for comparison. Most r values exhibit p<0.01 (in bold); less significant values are denoted in italics.
Results
Plant-wax-derived long-chain n alkane concentrations from ODP Site 659 (Fig. 4) mostly vary between 0.01 and
0.80 µgg-1 dry sediment and are correlated (Table 1) with the dust record (Tiedemann, 1991). Carbon preference index
values mostly exceed 3, indicating terrestrial plant contributions (Eglinton and Hamilton, 1967).
Results.
(a–c) Stable carbon (δ13C) and hydrogen (δD)
isotope compositions of
C29-C33 n alkanes in per mil vs. Vienna Pee Dee Belemnite (VPDB) and Vienna Standard Mean Ocean Water (VSMOW),
respectively; n alkane C27–C35 concentrations in microgram per gram dry sediment, and dust percentage after
Tiedemann (1991) against time in millions of years (Ma). Age model after Clemens (1999). For visibility, δ13C29
values (green) are shifted minus 0.5 ‰ and δD33 values (red) minus 10 ‰. Results of the C31
n alkane for the last glacial cycle (LGC) from Kuechler et al. (2013) are plotted to the left for comparison. mPWP: mid-Piacenzian
Warm Period.
Spectral analysis of stable carbon (δ13C31) and hydrogen (δD31) isotope compositions of
the C31 n alkane using a Morlet wavelet for three periods of the Pliocene (4.995–4.625,
3.615–3.310, and
3.270–3.005 Ma). Data are interpolated to 5 kyr steps. Signal power is expressed in colour shadings with the
significance level (p=0.05) in black contours. The cone of influence (thick black line) identifies the area of boundary
effects. Periodicities per thousand years on the vertical axes (log2 scale). Horizontal lines denote orbital periodicities of 19
and 23 kyr (precession), 41 kyr (obliquity), and 100 kyr. Bottom: eccentricity and precession for the same
periods based on Lascar et al. (2004) http://vo.imcce.fr/insola/earth/online/earth/online/. Boxes indicate periods with
reduced eccentricity and low precession amplitudes.
Stable carbon isotope compositions of n alkanes fluctuated between -27.3
and -24.7 ‰, -26.5, and -24.4 ‰, and between -26.1
and -23.3 ‰ for the n-C29, n-C31, and
n-C33 alkanes, respectively. A trend to more enriched values
towards the present started at 3.2 Ma. Isotopic signatures of the
major homologues significantly correlate (Table 1). Therefore, we focus on
the n-C31 alkane as the most abundant compound, attributed to the
prevailing C4 grass input (Vogts et al., 2009). Spectral analysis
(wavelet analysis) indicates that the variability in the isotope record is
not stable (Fig. 5). During the periods between 4.7 and 4.65 Ma and
between 3.1 and 3.0 Ma both significant precession and obliquity
variability are found in the δ13C31 record. Additional
precession variability is found between 3.48 and 3.38 Ma. These
periods are characterized by large eccentricity and thus large precession
cycle amplitudes. On the other hand, this type of orbital variability breaks
down during periods with low precessional variability (4.917–4.775,
3.520–3.475, 3.405–3.360, and 3.236–3.190 Ma; see boxes in
Fig. 5).
Hydrogen isotope compositions of n alkanes fluctuate between -160 and
-124 ‰, -171 and -133 ‰, and between -177 and
-127 ‰ for the n-C29, n-C31, and
n-C33 alkanes, respectively. Again, isotopic signatures of the
homologues significantly correlate (Table 1) and we thus focus on
δDwax values of the n-C31 alkanes, too.
Pliocene δD31 values range from -171 to -133 ‰
(Fig. 4) and show alternating arid and humid conditions in which
δD31 maxima (less negative) indicate aridity corresponding
with the dust maxima (Tiedemann, 1991). Spectral analysis indicates some
precessional variability around 4.7 Ma, between 3.55 and 3.52, 3.35
and 3.31, 3.15 and 3.19, and between 3.04 and 3.0 Ma (Fig. 5). During
low-eccentricity periods little precession variability is found.
A positive correlation is found between δ13C31 and
δD31 for the period between 3.3 and 3.0 Ma,
corresponding to the mid-Piacenzian Warm Period (mPWP;
3.264–3.025 Ma after Haywood et al., 2013). This is in contrast to
the negative correlation found for the last glacial cycle. The earlier
Pliocene results do not reveal a significant correlation between stable
carbon and hydrogen isotope compositions of the n-C31 alkanes
(Table 1).
Discussion
Plant-wax provenance and vegetation sources
The good correspondence between the dust record of ODP Site 659 (Tiedemann, 1991) and the n alkane concentrations indicate
predominance of aeolian transport of plant waxes probably in the form of coatings on dust particles. Pliocene
δ13C31 values display a narrow range around the average of -25.4 ‰, clearly below the enriched values of
up to -23 ‰ as observed for the last glacial cycle (Fig. 4; Kuechler et al., 2013). The low δ13C31
variations are attributed to a relatively stable wind system (Tiedemann et al., 1994) and the integration of a large source area. The
Pliocene record shows a different pattern than the last glacial cycle, when C3 plant-wax material relatively increased during
arid phases, which is attributed to enhanced contributions by the trade winds in combination with no or sparse vegetation cover in the
Sahara (Kuechler et al., 2013). This suggests that only small amounts of plant waxes derived from Mediterranean sources because trade
winds were weak, in line with the pollen records of ODP Sites 658 (Leroy and Dupont, 1994) and 659 (Vallé et al., 2014). Thus,
depleted δ13C31 values of the Pliocene indicate higher C3 plant coverage at the latitude of the Sahel
compared to today suggesting generally more humid conditions. Compared to the last glacial cycle the dominance of C4 plants,
estimated up to 90 % (Kuechler et al., 2013), is much less prominent during the Pliocene.
Differences in leaf anatomy, rooting depth and photosynthetic pathway may
contribute to the final plant-wax δD signal (Sachse et al., 2012).
Overall, C4 grasses are deuterium-depleted by ∼20 ‰
relative to C3 trees (McInerney et al., 2011) and deuterium-enriched
by ∼20 ‰ relative to C3 grasses (Smith and Freeman,
2006). However, the absolute variability of δ13C31 is
small (∼2 ‰) and would correspond to a vegetation shift
between C3 vs. C4 plants of less than 20 %, when using
the δ13C end-members of -35.2 ‰ for C3
plants and -21.7 ‰ for C4 plants (Castañeda et al.,
2009). The corresponding difference in hydrogen isotopic fractionation would
be less than ±4 ‰. Compared to the large variability in the
plant-wax δD record (between -171 and -133 ‰ for the
n-C31 alkane) such plant-dependent variations are minor.
Considering the large uncertainties in the estimation of
C4/C3 plant coverage using stable carbon isotopes, we
refrain from applying such a correction.
Pollen records from ODP Sites 658 (Leroy and Dupont, 1994) and 659 (Vallé
et al., 2014) indicate that Pliocene sub-Saharan savannahs have no modern
analogue, which harbours the potential for further uncertainties in the
plant-wax δD record. Nevertheless, δDwax
studies from offshore of NW Africa covering the African Humid Period (∼15–5 ka) yield a robust humid signal among different records
(Niedermeyer et al., 2010; Collins et al., 2013; Kuechler et al., 2013;
Tierney et al., 2017), although the vegetation of this African Humid Period
had no modern analogue (Watrin et al., 2009). Watrin et al. emphasize that
instead of a homogenous latitudinal shift of vegetation zones as a whole,
individual plant species likely have an advantage over others. The robustness
of the Holocene δDwax records implies that this proxy
for palaeohydrology is apparently not strongly affected by a
“non-analogous” vegetation composition.
Multiple isolation forcings. (a–f) Stable oxygen isotopes of benthic foraminifers
(δ18Obenthic) from ODP Site 659 (Tiedemann et al., 1994) compared to the global stack LR04 (Lisiecki and
Raymo, 2005) in per mil vs. Pee Dee Belemnite (PDB); dust percentages after (Tiedemann, 1991); summer latitudinal insolation
gradient (LIG) calculated from the 21 June insolation at 30∘ N minus that of 60∘ N; stable hydrogen isotope
composition δD31 in per mil vs. Vienna Standard Mean Ocean Water (VSMOW) for the last glacial cycle after Kuechler
et al. (2013) and for the Pliocene (this study) – stars indicate δD31 minima associated with Mediterranean sapropels;
summer insolation (June, July, August) at 20∘ N; cross-equatorial summer inter-tropical insolation gradient (SITIG)
calculated from the 21 June insolation at 23∘ N minus that of 23∘ S and occurrence of Mediterranean sapropels
(Emeis et al., 2000). Insolation values from http://vo.imcce.fr/insola/earth/online/earth/online/ (Laskar et al., 2004). Yellow
shading denotes periods with low precession variability.
Spectral analysis of hydrogen (δD31) isotope
compositions of the C31 n alkane (from Kuechler
et al., 2013) using a Morlet wavelet for the last glacial cycle (LGC). Data are interpolated to 3 kyr steps. Signal power is
expressed in colour shadings with the significance level (p=0.05) in black contours. The cone of influence (thick black line)
identifies the area of boundary effects. Periodicities per thousand years on the vertical axes (log2 scale). Horizontal lines
denote orbital periodicities of 19 and 23 kyr (precession), and 41 kyr (obliquity). Bottom: eccentricity and
precession for the same period. The box indicates the period with reduced eccentricity and low precession amplitudes.
Choice of age model
We choose the age model based on stable oxygen isotopes advocated by Clemens (1999) over the original one of Tiedemann et al. (1994),
since the former is independent from the dust record and produces a better fit with the global benthic δ18O stack
(Lisiecki and Raymo, 2005), especially for the Zanclean (Supplement Fig. S1). In general, both age models (dust and
δ18O) contain the same orbital frequencies (Supplement). However, the dust age model assumes precession
as the main forcing and accordingly, Pliocene dust peaks are tuned to insolation minima. Using the dust age model would introduce
age-model-dependent precession variability. The alternative δ18O-based age model also has a strong precession signal,
but only for periods with large precession amplitudes, while obliquity has a stronger impact during times of low precession
variability. Moreover, tuning of the dust record would introduce some circular reasoning in our argument as the dust record partly
depends on precipitation.
Other effects on the sedimentary δDwax signature
Aridity may exert a considerable influence on the apparent fractionation
(ε) between plant waxes and meteoric water via evapotranspiration
and associated deuterium-enrichment (Polissar and Freeman, 2010; Douglas
et al., 2012; Kahmen et al., 2013a, b). It was found that such an effect is
less pronounced in lake sediments compared to soils, likely due to the higher
potential of lakes to integrate large catchment areas and the small-scale
variability of soils related to differences in microclimate and vegetation
(Douglas et al., 2012). Niedermeyer et al. (2016) could not detect the effect
of evapotranspiration comparing high-resolution sediment data off West Africa
with instrumental records. Our records of the terrigenous fraction in marine
sediments also integrate huge catchment areas since large parts of the Sahara
and the Sahel can be considered sources of the material of primarily aeolian
origin (Tiedemann et al., 1994; Vallé et al., 2014). Moreover, Gao
et al. (2014) investigated aerosols from arid and humid subtropical
environments and found only minor deviations in ε (<10 ‰), attributed to a possible compensation of isotopic enrichment
due to decreasing relative humidity by a shift from trees to grasses in the
vegetation. Similar results were found in marine surface sediments forming
a transect running west along the central African and southern African coast (Vogts et al., 2016).
The δD record of the n-C31 alkane was adjusted for changes
in global ice volume (Tierney and DeMenocal, 2013). The Last Glacial Maximum
was scaled to 1 ‰ (Schrag et al., 2002) in the LR04 global benthic
δ18O stack (Lisiecki and Raymo, 2005) relative to the present
(0 ‰), and subsequently converted into δD values, using the
global meteoric water line. The procedure determined the use of the revised
age model (Clemens, 1999), which is based on δ18O variability
at ODP Site 659 and thus better matches the LR04 stack (Lisiecki and Raymo,
2005). However, the δD adjustment for the Pliocene period is almost
negligible (<2 ‰) and results are not shown.
Finally, plant waxes extracted from modern dust samples offshore of West Africa yielded a radiocarbon age of 650±150 years
(Eglinton et al., 2002) hinting at residence and transport times of several centuries. Since our research questions concern orbital
timescales (and our records have a ∼4 kyr resolution), intermediate storage on the continent and mixing of plant waxes
are considered to have a negligible effect on our interpretations.
Aridity and humidity in Pliocene West Africa
In contrast to the stable carbon record, which shows much less variability for the Pliocene than in the last glacial cycle, the
fluctuations in the deuterium record are of the same amplitude, indicating large variations between arid and humid periods in West
Africa occurred long before the intensification of Northern Hemisphere glaciations. Correspondence between the δD31
and the dust record of ODP Site 659 support previous conclusions concerning the generation of dust during arid periods (Tiedemann
et al., 1994) and the use of δD31 as humidity proxy. With it we present the first proxy record of West African
humidity and precipitation during the Pliocene. This record is also important because river discharge and run-off from northern Africa
(including the eastern parts of West Africa) is crucial for the development of Mediterranean sapropel layers. Almost all recognized
sapropels can be correlated with maxima in West African humidity (Fig. 6; δD31 minima). In general, our findings
highlight the influence of the West African monsoon for Mediterranean sapropel formation, and corroborate a recurrent greening of the
Sahara, which might have intermittently allowed hominin migration through otherwise hostile territory (Larrasoaña et al., 2013).
Precipitation in West and northern Africa depends on the West African monsoon mainly varying with low-latitude summer insolation, which
is dominated by 19 and 23 kyr periodicities of the precession cycle (e.g. Kutzbach, 1981; see also Introduction). However,
variations in low-latitude insolation cannot explain the entire variability in the δD31 and dust records. For
instance, four dust percentage maxima between 4.91 and 4.77 Ma correspond to four obliquity minima. Examples of humid periods
(strongly depleted δD31) during insolation minima were found shortly before 4.90 Ma, centred at 4.79 and
3.50 Ma, or just after 3.20 Ma. Conversely, examples of arid periods (less depleted δD31) during
insolation maxima are centred at 4.88, 4.78, 3.48, and 337 Ma and just before 3.20 Ma (Fig. 6). Moreover, spectral analysis
of both Pliocene (Fig. 5) and Pleistocene (last glacial cycle; Fig. 7) series indicate a more complicated pattern with additional
periodicities unrelated to precession. These periods, in which humidity changes do not follow low-latitude insolation, might be linked
to obliquity. Shifts in variability already occurred prior to the intensification of the Northern Hemisphere glaciation at the
beginning of the Pleistocene (2.58 Ma), but are most prominent during the last glacial cycle.
The mechanistic basis for the obliquity signal in low-latitude climate records is a matter of debate. Modelling studies show
contrasting results, both including (Tuenter et al., 2003; Weber and Tuenter, 2011) and excluding (Bosmans et al., 2015a; Rachmayani
et al., 2016) an influence of the high latitudes. On the one hand, obliquity in the tropics is attributed to higher moisture transport from
mid/high latitudes (Tuenter et al., 2003) or varying ice sheets and greenhouse gases (Weber and Tuenter, 2011) that influence the
monsoon intensity and timing. On the other hand, recent modelling studies revealed an increase in monsoonal rainfall due to a stronger
atmospheric pressure contrast between the western African continent (low pressure) and the South Atlantic (high pressure) under conditions
of low precession (i.e. maximum local insolation) and high obliquity. Based on this, the interhemispheric (or cross-equatorial)
insolation gradient (summer inter-tropical insolation gradient) at lower latitudes was suggested to drive winds and associated moisture
transport (Bosmans et al., 2015a, b). In general, insolation gradients shape the climate system by differential heating and thus
determine spatial temperature patterns, which ultimately steer atmospheric pressure gradients (Nicholson, 2009).
On orbital timescales, the periodicity of glacial–interglacial cycles with the high-to-low-latitude contrast in insolation
(i.e. latitudinal insolation gradient) forces moisture fluxes poleward via strong gradients, thus triggering ice sheet growth (Raymo
and Nisancioglu, 2003). In this context, most studies focus on high latitudes, and studies considering a latitudinal gradient forcing of
tropical monsoon systems are limited to the last glacial cycle (e.g. Davis and Brewer, 2009). However, summer weakening of the
latitudinal gradient when obliquity is strong would lead to a northward shift of the tropical monsoon front (Davis and Brewer,
2009). This was recently supported by another climate model, indicating a tight link between the gradient-controlled mid-latitude eddy
circulation (driving the heat and moisture transport from low to high latitudes) and the poleward boundary of the tropical rain belt
(Mantsis et al., 2014). Thus, a northward shift of the tropical monsoon front is related to increased mid-latitude eddy circulation
during summer when the latitudinal temperature gradient linked to the summer latitudinal insolation gradient (dominated by obliquity)
is large. For West Africa the strongest monsoonal influence would thus be expected during summer when the latitudinal gradient is weak,
shifting the tropical rain belt northward, instead of being purely triggered by local insolation. For the Pliocene, both proxy data and
model results confirm reduced meridional and zonal temperature gradients when the tropical monsoon was enhanced (Dowsett et al., 2013;
Haywood et al., 2013). Solely low-latitude insolation forcing cannot explain the variations in the δD31 and dust
records.
Towards a more comprehensive understanding of the drivers of the West African monsoon
Based on our results from the Pliocene and comparison with the last glacial cycle we propose a modified gradient forcing
mechanism. Wavelet analyses show repeated shifts in the periodicity of the δD31 records from the obliquity to the
precession band and vice versa, depending on the amplitudes in precession and obliquity (Figs. 5 and 7). We infer a consistent pattern
with generally two modes: during periods with strong precession cycles (i.e. high eccentricity), the influence of summer LIG on monsoon
variability is superimposed by changes in local insolation. In contrast, during weak precession cycles, especially those of the last glacial cycle, summer LIG forms the primary forcing of the West African monsoon and its influence on the northward distribution of
atmospheric moisture plays the decisive role in the West African climate. This is also supported by climate models, which show a strong
dependency of the obliquity-induced precipitation response to precession, but not vice versa (Tuenter et al., 2003). The indication for
LIG forcing during weak precession cycles of the Pliocene is less pronounced than during the Late Pleistocene suggesting that the
obliquity forcing of the summer LIG is reinforced by the intensification of the Northern Hemisphere glaciation. LIG-induced climate
shifts during the last glacial cycle are more severe than during the Pliocene due to increased ice–albedo feedbacks (Raymo and
Nisancioglu, 2003). In addition, the duration of humid–arid periods seems to increase in the course of the last glacial cycle,
reflecting the shift from one dominant forcing to another, i.e. from precession to obliquity. These findings caution against a too
simplified view on orbital forcing mechanisms of the (tropical) hydrologic cycle and hence related records.