Mean climate state
In this section we explore the mean climate state, both throughout the full
precession cycle and in more detail for the two precessional extremes, both
with 280 and 400 ppm CO2 concentrations. Note that the
occurrence of precession minimum relatively close to the obliquity maximum
(as seen in Fig. 1b) is likely to enhance seasonality in the Northern
Hemisphere .
Surface air temperature response through a full precession cycle
The evolution of the modelled global annual mean surface air temperature
(SAT) throughout the full precession cycle is not significantly influenced by
the varying orbital parameters in either of the two hemispheres (Fig. 2a),
with changes within ∼ 0.3 ∘C. It also does not appear to be
related to the evolution of the annual mean incoming shortwave radiation,
perpendicular to the Earth's surface, at the top of the atmosphere (from now
on referred to as insolation), which is the same in the Northern and Southern
hemispheres (Fig. 2a). Variations in global mean annual insolation are only a
result of eccentricity, while precession and obliquity have no impact on it.
In the model, the mean global SAT is the result of a combination of the two
hemispheres balancing each other out, with neither of the two clearly
dominating the trend (Fig. 2a). Finally, global mean SATs in the Southern
Hemisphere are generally higher (by ∼ 2 ∘C) than in the
Northern Hemisphere in our late Miocene simulations. The present-day
configuration is the opposite of this, with the Northern Hemisphere on
average 1.5 ∘C warmer than the Southern Hemisphere
(, , and references therein). This
difference is caused by the open Panama Seaway in our late Miocene
simulations , leading to a weaker Atlantic Meridional
Overturning Circulation in the late Miocene, compared with that of the
present day. This maintains warmer temperatures in the Southern Hemisphere
and colder in the Northern Hemisphere in our palaeosimulations, as opposed to
modern temperatures.
Evolution throughout the precession cycle (indicated in the top
panel) of surface air temperature (blue lines) and incoming solar radiation
at the top of the atmosphere (green lines), both in the Northern and Southern
hemispheres and including global means in panel (a). (a) Annual mean,
(b) winter – DJF in the NH and JJA in the SH, (c) summer – JJA in the NH and DJF
in the SH, (d) spring – MAM in the NH and SON in the SH, (e) autumn – SON in
the NH and MAM in the SH. Note that different scales are used in each panel
and that in panel (a) the variation in incoming solar variation is much
smaller than in the other panels (less than 1 W m-2).
In the model, seasonal temperature variations are driven by orbital forcing
and related to changes in insolation, which in turn exhibit opposite phasing
between the two hemispheres in every season (Fig. 2b–e). In addition, in
both hemispheres the seasonal cycles cancel each other out in pairs and
therefore produce only small variations in the annual mean (as seen in Fig. 2a).
In winter, SAT in the Northern Hemisphere is in phase with local
insolation (in anti-phase with precession) but with a lead of
∼ 2 kyr (Fig. 2b). The same lead can be seen in the Southern
Hemisphere but with the difference that SAT is in anti-phase with insolation
(in phase with precession; Fig. 2b). In summer, modelled SAT is in phase
with insolation with a lead of ∼ 3 kyr, both in the Northern and
Southern hemispheres (Fig. 2c). In the model, the difference between maximum
summer temperatures and minimum winter temperatures in the Southern
Hemisphere is much lower (∼ 7 ∘C) than in the Northern
Hemisphere (∼ 20 ∘C), due to the more extended presence
of land in the Northern Hemisphere. A warmer winter season results in a
higher annual mean temperature in the Southern Hemisphere than in the
Northern Hemisphere (as seen in Fig. 2a).
In spring, modelled SAT is in phase with insolation in both hemispheres, with
a lead of ∼ 2 kyr in the Northern Hemisphere and of ∼ 2.5 kyr
in the Southern Hemisphere (Fig. 2d). In autumn, the Northern Hemisphere SAT
is in phase with insolation and it leads it by 3 to 4 kyr (Fig. 2e). The
same phasing can be observed in the Southern Hemisphere, with the difference
that SAT leads insolation by about 3 kyr (Fig. 2e).
Note, however, that in our simulations SAT is influenced not only by
precession but also by the varying obliquity, which changes from a maximum of
∼ 23.9∘ to a minimum of ∼ 22.8∘. In
contrast to precession, obliquity will have the same effect on both
hemispheres, with maximum obliquity resulting in stronger summer insolation
and weaker winter insolation, thus leading to an enhanced seasonal contrast.
Using idealised orbital transient simulations with an earth system model of
intermediate complexity, have shown that a simple mechanism
can explain how the leads and lags in the climatic response (e.g. surface air
temperatures) to insolation within a year can result in leads or lags in time
with respect to orbital parameters. However, the same simulations performed
with interactive vegetation indicated how quickly this mechanism can be
extensively modified, because of vegetation feedbacks and changes in the
albedo, as induced by different sea ice distributions . The
simulations analysed in this study are carried out using a more complex
general circulation model and using interactive vegetation. As such,
understanding the causes of leads and lags in the climate system is sometimes
challenging and the simple mechanism described by is no
longer applicable. However, we investigate in more detail the vegetation
responses to changes in orbital forcing for the North African monsoon region
(Sect. 3.4.1), as well as the spatio-temporal phasing of surface air
temperatures globally (Sect. 3.2).
Global climate response to orbital forcing: precession extremes
In the model, the DJF SAT anomalies between precession minimum and maximum
(pMIN-pMAX) are generally negative (i.e. cooler; Fig. 3a), especially in
north-central Asia and India (except adjacent to the Tibetan Plateau),
central-east North America, and most of Australia. Maximum cooling is found
around the Sea of Japan, which is probably a model artifact caused by the
enclosed nature of the basin in the model, which intensifies the signal. The
Nordic Seas exhibit much higher DJF temperatures during the precession
minimum. interpreted the substantial winter warming of the
Arctic as a response to insolation forcing, resulting from the “summer
remnant effect”. This effect has also been discussed by and
, who attributed the warming to delays in sea ice
formation during the winter season as a result of excess solar radiation
during the summer months. In this study, however, the location of the warm
anomaly is shifted further south, localised in the Nordic Seas area rather
than in the Arctic. This is possibly a consequence of using a late Miocene
palaeogeography, rather than the modern palaeogeographies applied by
and , and of the different sea ice
distribution in our experiments. In these regions, the main differences are
the presence of the Kara/Barents Sea landmass in our late Miocene simulations
(Fig. S1) and the higher concentration of sea ice in the Nordic Seas and in
the subpolar North Atlantic in general, compared to the preindustrial
(Fig. S5).
Anomaly plots of SAT between the two precession extremes, where the
difference is pMIN – pMAX, in (a) DJF, (b) JJA, (c) cold month mean (CMM),
(d) warm month mean (WMM) and (e) annual mean. Differences with significance
outside of the 99 % confidence interval (T test) are represented in white.
280 ppm CO2 concentrations.
The modelled JJA mean SAT differences exhibit globally warmer temperatures
during precession minimum (Fig. 3b), especially on land in the Northern
Hemisphere, with maximum warming (positive anomalies) over central Eurasia.
The exception is the cooling (negative anomalies) that occurs over the
monsoon regions in North Africa and India, due to the intensified cloud cover
as a consequence of enhanced monsoonal precipitation .
Maximum warming is generally centred over the main land masses rather than
the ocean, because of the ocean's greater heat capacity and potential for
latent cooling. However, the North Atlantic also shows significantly higher
SATs (up to 5.5 ∘C) at times of precession minimum.
The cold and warm month means (which at each model grid square represent the
SAT for the coldest and warmest months, respectively) exhibit clear
differences in the sign of the anomaly in each hemisphere (Fig. 3c, d). This
represents the opposite effect of precession on both hemispheres. In the
Northern Hemisphere, the cold month mean (Fig. 3c) mirrors DJF values
(Fig. 3a), with reduced warming in the Nordic Seas. The warm month mean
(Fig. 3d) largely mirrors JJA values (Fig. 3b), except the intensified
warming in the North Atlantic and the differences in the Northern Hemisphere
monsoon regions, where cooling is no longer visible.
The Southern Hemisphere's cold month mean (Fig. 3c) anomalies are mainly
positive during precession minimum, especially in northern Australia and in
the Southern Ocean along the Antarctic coast. Negative anomalies dominate the
warm month mean (Fig. 3d), with the exception of northern-central South
America and part of central-southern Africa, as a result of vegetation changes
modifying the albedo feedback. The mean annual temperature difference
(Fig. 3e) is characterised by maximum warming in the Nordic Seas and part of
the Arctic regions during the precession minimum, whereas cooling is found in
the African and Indian monsoon belts (and off the coast of South America,
around 10∘ S), with the most negative values in the sensitive area
around the Sea of Japan (which could be the result of a model artifact, as
previously discussed). Changes are generally small elsewhere (mostly within
∼ 1.5 ∘C) and this lack of a clear signal is mainly due to the
positive and negative forcing during summer and winter seasons balancing each
other out. The colder (negative anomalies) mean annual temperatures over the
monsoon regions are caused by the dominant JJA cooling, and the mean annual
warmer (positive anomalies) values in the Arctic and Nordic Seas are a result
of the overall warming in both seasons.
Absolute plots of modelled JJA precipitation at precession maximum (Fig. 4a)
and minimum (Fig. 4b) portray the distribution of enhanced precipitation in
the equatorial regions and clearly highlight the northward shift of the
inter-tropical convergence zone (ITCZ) during precession minimum, which is
most clearly seen over the monsoon regions. Coupled climate models are
typically affected by a split of the ITCZ over the tropical West Pacific
Ocean (e.g. Bellucci et al., 2010; Oueslati and Bellon, 2015), which leads to
large disagreement between models and observations for present-day
simulations. This is also clearly visible in our simulations, especially in
the absolute plots of precipitation (Fig. 4a, b).
Absolute values of summer (JJA) precipitation during (a) precession
maximum and (b) precession minimum, with 280 ppm CO2
concentrations and anomaly plots of precipitation between the two precession
extremes, where the difference is pMIN – pMAX, in (c) DJF and (d) JJA.
Differences with significance outside of the 99 % confidence interval
(T test) are represented in white.
Anomaly plots of SAT (top panels) and precipitation (bottom panels)
between the two precession extremes at different CO2
concentrations, where the difference is [(pMIN - pMAX)400 ppm -(pMIN - pMAX)280 ppm].
(a) SAT anomalies in DJF, (b) SAT
anomalies in JJA, (c) precipitation anomalies in DJF, (d) precipitation
anomalies in JJA. Differences with significance outside of the 99 %
confidence interval (T test) are represented in white.
Modelled precipitation in DJF shows small differences between the two
precession extremes at high latitudes in both hemispheres (Fig. 4d).
Prominent features include changes in the North Atlantic storm tracks which
take a more southerly route during precession minimum, leading to the
widespread spatial precipitation anomaly which extends over the Mediterranean
Sea and south-west Europe (Fig. 4d). Most of the significant changes in
precipitation patterns between the two precession extremes, both in DJF and
JJA, are found around the location of the ITCZ, depicting its migration
between the two hemispheres in response to changes in orbital forcing
(Fig. 4c, d). In JJA the ITCZ shifts northward, towards the warmer Northern
Hemisphere as a result of the higher insolation forcing in summer. This can
be clearly identified in the monsoon regions, especially in Africa and Asia,
which experience much higher summer precipitation (more than
3.5 mm day-1 increase) during precession minimum (Fig. 4c). In JJA
wetter (positive anomalies; up to 1.5 mm day-1) conditions during
precession minimum are also found north of ∼ 50∘ N in the
Northern Hemisphere, as well as across the Southern Ocean and over most of
Australia in the Southern Hemisphere (Fig. 4c). In contrast, significant
negative anomalies (up to 3.5 mm day-1) dominate the North Pacific,
North America and the North Atlantic between ∼ 10 and 40∘ N.
Finally, precipitation anomalies are small in Antarctica and across the
Arctic regions because of the reduced amount of precipitation over these
areas.
Global climate sensitivity to atmospheric CO2 concentrations
In addition to the full set of 22 simulations with preindustrial CO2
concentrations (280 ppm), two sensitivity experiments were carried out at
400 ppm for the two precessional extremes (pMIN400 and pMAX400), in order to
explore the global and local climatic response to varying CO2
concentrations and to assess the impact of CO2 on the model–data
comparison. This was necessary because of the uncertainty in reconstructed
CO2 concentrations for the late Miocene, as discussed in Sect. 1.
In line with previous modelling studies (e.g. Bradshaw et al.,
2012a, 2015), late Miocene climate warms significantly as CO2
concentrations increase, especially at high latitudes, and the greatest
warming is found on land in the Northern Hemisphere (not shown). For the
precession minimum simulation the mean annual global SAT is 14.6 ∘C
at 280 ppm while at 400 ppm it is 17 ∘C. Modelled SATs in DJF
(Fig. 5a) are more sensitive to orbital changes (where the “orbital
sensitivity” described here is the difference between precession minimum and
maximum) at 400 ppm CO2 (over 5 ∘C) in the subpolar North
Atlantic south-east of Greenland, in the regions around the Sea of Japan and
in the north-west Pacific Ocean. In contrast, some areas reveal DJF SATs with
increased sensitivity at 280 ppm (Fig. 5a), including the regions north of
India, Canada and other parts of North America, central Africa and most of South
America.
The most extended significant SAT anomalies in JJA (Fig. 5b) are found in
the subpolar North Atlantic, Nordic Seas and the Arctic Ocean, where modelled
temperatures are more sensitive to orbital changes at 400 ppm. This is also
true for the North Pacific and the area around the Sea of Japan. Higher
sensitivity at 280 ppm for JJA SATs is found locally in the Southern Ocean,
especially around the Ross Sea, and in the Northern Hemisphere over Greenland
(up to 4 ∘C). In both seasons, differences in and around the polar
regions are most likely to be linked to changes in sea ice distribution. In
fact, relatively warm initial conditions for the late Miocene simulations
lead to enhanced sea ice loss during precession minima, triggering a strong
positive sea ice feedback mechanism as CO2 concentrations increase
.
Phasing of SAT throughout a full precession cycle. Each colour
indicates the temporal offset from the maximum/minimum SAT per model grid
square for (a) warm-month maximum SAT (maximum SAT) and (b) cold-month
minimum SAT (minimum SAT). Simulations are indicated on the left and in
relation to the precession cycle, as shown in Fig. 1b.
In the model, the precipitation response to increasing CO2 is a moderate
increase at mid to high latitudes in both hemispheres (not shown), as
illustrated by Bradshaw et al. (2012a). The most
significant differences in orbital sensitivity of precipitation patterns are
found across the equatorial regions, largely driven by shifts in the ITCZ,
both in DJF and JJA (Fig. 5c, d). This is generally most pronounced over
the ocean, but nonetheless precipitation over central northern Asia, eastern
North America, and western Africa is significantly more sensitive to orbital
changes at 280 ppm in JJA. In contrast, over central Greenland, western
Europe, and central North Africa JJA precipitation is more sensitive at
400 ppm (Fig. 5d). In DJF the most significant changes in precipitation
sensitivity over land are found in the Southern Hemisphere, especially in
South America and central and southern Africa, in both cases with some regions
exhibiting higher sensitivity at 280 ppm but dominantly at 400 ppm
(Fig. 5b).
Spatio-temporal phasing of surface air temperatures
While comparison of orbital extremes is probably adequate to investigate the
links between climate and orbital forcing, we argue that it may not capture
the full variability and leads and lags between the orbital forcing and the
climatic response. Our results through a full late Miocene precession cycle
show that maximum warming and cooling are not spatially synchronous and
strongly vary in time across different regions (Fig. 6). Consequently, the
warmest or coldest SATs do not necessarily correspond to precession minima
and maxima, respectively.
Our experiments only capture the full variability of a single precession
cycle. Obliquity (and eccentricity) values also realistically vary in the
ensemble together with precession, but our simulations are not designed to
fully capture the variability of an entire eccentricity or obliquity cycle. A
detailed separation of the effect of precession and obliquity forcing is
beyond the scope of this work, but this could be addressed in a future study.
The effect of obliquity on seasonal insolation is especially significant at
high latitudes. Nonetheless, an obliquity-driven signal has been found in
some low-latitude proxy record for the late Miocene in the Mediterranean
region e.g., despite the small influence of
obliquity on low-latitude insolation. The influence of obliquity may explain
some of the leads and lags between modelled SAT and precession discussed in
this section.
(a)–(d)
Illustrative definition of model–data mismatch and overlap.
(e) Definition of orbital, model, and data ranges.
(f) Model–data mismatch is defined as the minimum possible distance
to overlap, but here we show that the maximum possible differences could be
much greater if the true values for both the model and the data were to lie
at the extremes of the uncertainty ranges (Bradshaw et al.,
2012a). Note that the relative contributions of
model and data uncertainties will vary depending on the variable analysed and
for each experiment. The real values are not indicated here as this figure is
schematic.
For example, there are regions showing largely synchronous warming or
cooling, especially in the Northern Hemisphere, but in other areas (even
neighbouring ones and in the same hemisphere) maxima and minima can be out of
phase with the precessional maximum or minimum by as much as 6 kyr
(Fig. 6a). This might be expected in the monsoon regions because of the
intensified cloud cover reached at times of minimum precession, but it is
less understandable for the other locations. Modelled SATs are more out of
phase over the ocean than on land, which may relate to the more direct link
between solar forcing and temperature over land than over the ocean. Maximum
SATs are consistently not synchronous (4–6 kyr out of phase) with
precession minimum/maximum in the eastern North Pacific Ocean, in the region
of the Indonesian Throughflow, and in the Southern Ocean (Fig. 6a). Given the
location and latitudinal extension across the Southern Ocean, here the lag
could be associated with changes in ocean circulation and linked to the
pathway of the Antarctic Circumpolar Current. Moderate out-of-phase behaviour
(2–3 kyr) is also found in northern and southern Asia, over central North
America, part of Greenland, in the Arctic regions, the Indian Ocean, the
South Atlantic and over several parts of the Pacific Ocean. In the Southern
Hemisphere, the monsoon regions in South America, southern Africa and
northern Australia are out of phase by 5 kyr or more.
(ia–id) Difference between model–data comparison including orbital
variability and using modern orbit with 280 ppm CO2 concentrations.
(iia–iid) Discrepancy between Messinian proxy data and model output including
orbital variability at 280 ppm. (iiia–iiid) Discrepancy between Messinian
proxy data and model output with 400 ppm (precession extremes only).
(iva–ivd) Difference between model–data comparison with 400 ppm (precession
extremes only) and 280 ppm (full precession cycle variability). A version of
this figure with alternative colours is available in the Supplement
(Fig. S6).
The obliquity maximum reached in experiment 2 will tend to shift SAT maxima
away from the precession minimum towards the obliquity maximum, when the
system is sensitive to obliquity and responds directly to summer insolation.
In that case, a maximum lead of 5 kyr with respect to precession (minima) can
be explained. Note, however, that 65∘N summer insolation varies in
anti-phase with precession (see Fig. 1) and is not shifted in the direction
of the obliquity extremes. The different response in the two hemispheres,
with stronger out-of-phase behaviour in the Southern Hemisphere, might also
be partially explained by the use of a modern calendar in these simulations
.
Minimum modelled SATs (Fig. 6b) are mostly not synchronous (4–6 kyr out of
phase) with precession minimum/maximum in the North Atlantic Ocean and Nordic
Seas, as well as part of the South Atlantic. Strong out-of-phase behaviour is
also found over Greenland, northern and central Asia, South America, south of
Africa and at several locations in the equatorial regions and in the North
Pacific. Moderately out-of-phase (2–3 kyr) temperatures extend over North
America, north and central Asia, part of the Arctic and in several locations
over the ocean both in the Northern and Southern hemispheres. Because of
their location, we suggest that the patterns observed across the North
Atlantic and North Pacific, with areas out of phase by up to 4 kyr, are
associated with the winter storm tracks. Overall, minimum temperatures
exhibit an even more complicated mosaic of patterns than the maximum ones.
The response of the climate system at high latitudes is more complex due to
vegetation, snow, and sea-ice albedo feedbacks . This could
therefore exacerbate leads and lags with the orbital forcing in these
regions.
These results further demonstrate the importance of considering orbital
variability in order to capture the entire magnitude of the warming/cooling
(or wettest/driest periods), especially locally and when considering
model–data comparisons. also found significant
out-of-phase responses when investigating peak warming around two Pliocene
interglacials. These authors argued that proxy-based reconstruction of
temperature time series that rely on cold/warm peak alignment and averaging
e.g. could potentially result in significant
temporal miscorrelations. This is confirmed by our results from a single late
Miocene precession cycle. The bias is relevant for all pre-Quaternary
model–data comparison studies, which require a methodology incorporating the
effect of orbital variability on climate . The more
traditional time-average approach must be avoided in order to compare model
results with proxy reconstructions robustly
.
Global terrestrial model–data comparison
carried out a quantitative terrestrial model–data
comparison using a late Miocene data set, which incorporated a conservative
estimate of uncertainties associated with both the model output and the data
reconstructions. As well as calibration uncertainties in the proxies, model
bias and interannual variability, their methodology also considered the
potential impacts of poor temporal constraint on determination of the data
palaeolocation (see , , for full
details of the model–data comparison methodology). The available late Miocene
terrestrial proxy record is biased by a sparse and patchy distribution, and
low temporal resolution. Despite these large uncertainties,
found significant discrepancies between the climate model
output and the available late Miocene terrestrial proxy record.
These authors applied a modern-day orbital configuration to their
simulations. Here, as described in Sect. 2, we use the same numerical model
and initial setup, but we take into account the full range of variability
through the analysed late Miocene precession cycle when undertaking the
model–data comparison. This is achieved by selecting the maximum and minimum
value through the orbital cycle from the 22 simulations, for every analysed
variable in each grid cell. Our definitions and estimates of model–data
agreement or mismatch (Fig. 7) and the uncertainties in the model and data
are the same as those described by but with an extension
to the envelope of model uncertainty to include orbital changes.
The methodology developed by includes a bias correction
which corrects for the offset between the model's simulated preindustrial
climate and preindustrial observations. This assumes that even if simulated
temperatures and precipitation are not necessarily accurate in an absolute
sense, there is a robust relationship between the late Miocene climate and
that of the present day . For the model, the uncertainty
associated with the natural interannual variability within the simulation is
also included. For each value this is calculated as 1 standard deviation of
the interannual variability of the last 50 years of the model simulation. In
addition, given that the observational data sets are characterised by a higher
spatial resolution than the model, in the model–data comparison all the model
grid cells adjacent to the ones containing the proxy data are considered,
where the minimum and maximum values from all of the eight adjacent cells, rather
than only the value on the specific grid cell, are used (nine grid cells in
total). This is a way to account for the poorly constrained age control on
the data, plate rotation uncertainties, and the location of the climate
signal recorded by the proxy record . Finally, the
calibration error for each proxy type is also included and calculated based
on modern proxies. Any remaining error must be due to model
structural or parametric uncertainties which could only be addressed through
multi-model inter-comparison studies.
Overlap or mismatch (Fig. 7) depends on whether the range between the maximum
possible model value (Mmax) and minimum possible model value
(Mmin) overlaps with the range between the maximum and minimum
data values (Dmax, Dmin). In our case, for each
variable and in each grid cell, Mmax is the maximum value out of
198 (22 × 9, where 9 are the 8 grid cells surrounding the data
location plus the grid cell itself, and 22 is the number of orbital
simulations) grid cells, plus 1 standard deviation of the interannual
variability, and similarly for Mmin. Finally, the bias correction
is applied.
In this way we are able to capture the entire range of variability simulated
by the model throughout the full precession cycle for each variable, allowing
us to check whether the proxy reconstructions would overlap with model
results at any point during the precessional cycle. We can therefore test
whether part of the mismatch obtained by may be explained
by orbital variability. Our model results are compared to mean annual SATs
and precipitation, and warm and cold month SATs from the Messinian,
reconstructed from proxy data. The data set used is the same compilation of
terrestrial proxy reconstructions as but with the
addition of palaeo-precipitation data for North America by
(Supplement, Table S1).
The comparison of our results at 280 ppm CO2 including orbital
variability with those of demonstrates an overall
reduction of the model–data mismatch almost everywhere, both for the mean
annual temperature and precipitation records (Fig. 8ia, ib). The only
exception is a single data point in the South American continent, showing
slight deterioration (see Fig. 7d, depicting the mismatch because of the
added orbital variability). We find 766 overlaps (Table 2) from a total of
1193 data points between our orbital ensemble and the Messinian terrestrial
proxy data, as opposed to the 610 overlaps found in the
simulation for the Messinian part of the data set. The cold month temperature
and annual precipitation over the Asian, African and North American
continents are well matched between our simulations and the data. However,
over most of the European continent the proxy record is still warmer and
wetter than the climate reproduced by our simulations, both in the warm month
and the annual means (Fig. 8iia, iib, iic). The Mediterranean region, where
there is the greatest density of observations, gives both the highest match
and mismatch between the model and the data. Even when considering the wider
envelope of model variability, the simulations still largely fail to capture
the magnitude of warming found in the Messinian data, exhibiting mostly
colder temperatures (both annual and warm month mean) especially in the
Mediterranean region (Fig. 8iia, iic), but with a good match (187 overlaps
out of 238 data points) for the cold month mean (Fig. 8iid).
June–July–August–September average of SAT and precipitation over the
northern and southern regions of North Africa.
PI280
LM280
pMAX280
pMIN280
pMIN400
SAT northern region (∘C)
33.9
28.8
30.0
31.5
37.3
SAT southern region (∘C)
26.3
26.6
26.8
26.4
29.1
Precipitation northern region (mm day-1)
0.57
0.32
0.21
0.35
0.65
Precipitation southern region (mm day-1)
6.78
2.99
3.53
5.06
6.95
As previously discussed, it is important to remember that in our realistic
late Miocene simulations we are only considering one specific precession
cycle. We, therefore, cannot capture the full variability of obliquity. In
addition, there are other higher-amplitude precession cycles in the Messinian
(higher eccentricity values), which means that we are not able to fully
capture the maximum amplitude of precessional variations either. Running
idealised simulations including the full orbital variability throughout the
late Miocene period (using absolute maximum and minimum values for all
orbital parameters) may result in an even better match with the proxy record.
Bradshaw et al. (2012a) also investigated the
impact of using different CO2 concentrations when modelling the late
Miocene climate and obtained a better match with the proxy record using
400 ppm (719 overlaps for the Messinian part of the data set) rather than
preindustrial values of 280 ppm (610 overlaps). We have therefore carried
out an additional model–data comparison, taking into account the variability
between the two precession extreme experiments at 400 ppm for each analysed
variable (Fig. 8iii, iv). Our simulations with higher CO2 concentrations
and including orbital variability also exhibit a significantly better match
with the Messinian observational record (Fig. 8iv) than the orbital ensemble
carried out at 280 ppm, both for mean annual temperature (MAT) and warm
month mean temperature (WMT). This is indicated by the presence of 172
overlaps for the MAT (Table 2) and 183 overlaps for the WMT, compared to the
86 MAT and 121 WMT overlaps obtained in the 280 ppm ensemble (Fig. 8iiia,
iiic). WMTs in the model at 400 ppm show a good agreement with the proxy data
(Fig. 8ivc), which is more improved than the match achieved in the 280 ppm
simulations, except for the north-east Asian region. All Messinian WMT
reconstructions overlap with the model results in the Mediterranean region
(Fig. 8ivc), and there is an almost complete overlap in this region also for
the cold month temperature (CMT). Modelled MATs (Fig. 8iva) exhibit both some warmer and colder data
points compared to the Messinian observational record, despite generally good
agreement over the European continent. There are no major differences in the
comparison between CMT and mean annual precipitation
(MAP) at 280 and 400 ppm CO2 concentrations. However, a slight
deterioration is found in the CMT (Fig. 8ivb) and in the MAT (Fig. 8ivd).
This is indicated by the presence of 185 overlaps for the CMT (Table 2) and
370 overlaps for the MAP at 400 ppm, compared to the 187 CMT and 372 MAP
overlaps obtained in the 280 ppm ensemble (Fig. 8iiib, iiid).
also discussed the reasons for model–data comparison
deterioration with higher CO2 concentrations in certain areas and found
that the best fit for mean annual precipitation occurred at 180 ppm
CO2, despite the best match for SATs resulting at 400 ppm. The reasons
for these discrepancies are still not clear, and our results show that these
cannot be reconciled by including orbital variability.
As the warmest or coldest temperatures do not necessarily correspond to
precession minimum and maximum, the 400 ppm precessional extreme
sensitivity experiments do not necessarily capture the full variability of
the precession cycle (refer to Fig. 6). At 280 ppm CO2, the model–data
comparison outputs for the true minimum and maximum resulting from the full
ensemble of simulations covering the whole precession cycle are almost
identical to the model–data comparison results for just the precession
minimum and maximum. In fact, there is a difference of only five overlaps
(Table 2), because the differences in the simulations are smaller than the
uncertainties in the proxy reconstructions. However, this may not be the case
for regions where well-constrained data are available, such as the
Mediterranean Sea.
To summarise, our results imply that accounting for orbital variability, when
combined with higher CO2 concentrations, reduces model–data mismatch by
more than 25 % as compared to previous experiments for the late Miocene
using a modern orbital configuration . In regions where
good agreement is obtained between model and data and where in addition
high-resolution and more precisely dated proxy records are available for our
specific Messinian modelled time period, it would also be possible to
estimate during which part of the precessional cycle the proxy reconstruction
has been generated (assuming that the model realistically simulates orbital
and seasonal variability). For instance, this can be applied to Messinian
micropalaeontological data from the Mediterranean Sea that have been sampled
on sub-precessional timescales.
(a) SAT and (b) precipitation difference between
minimum (pMIN) and maximum (pMAX) precession during the monsoon season
(JJAS). The dashed boxes in (a) illustrate how North Africa is split
in two areas, northern “box” and southern “box”, for analysis (where only
the land component is considered these are defined as northern region and
southern region). Latitudes and longitudes for the southern “box” are
defined according to Thorncroft and Lamb (2005) for the West African monsoon.
Absolute values for the monsoon season (JJAS) precipitation at
(c) precession minimum and (d) precession maximum.
African summer monsoon variability between precession extremes
The majority of the late Miocene terrestrial proxy data are concentrated
around the margins of the Mediterranean Sea. River discharge into the
Mediterranean today is dominated by the Nile. In the late Miocene
another North African river which is now dry, the Eosahabi, may also have
drained from Lake Chad into the eastern Mediterranean
. Changes in the discharge of these rivers is
driven by the summer North African monsoon, which is in turn influenced by
precession e.g. and, to a
lesser extent, by obliquity . We therefore analyse
the dynamics of the North African monsoon and its seasonal precipitation and
SAT changes throughout our full simulated precession cycle. Here, we consider
the North African monsoon system as the combination of both the present-day
West African and Central African monsoon dynamics, predominantly controlled
by the overriding north–south large-scale Hadley circulation.
Our model results highlight the prominent effect that different orbital
configurations have on the African summer monsoon. For instance, the minimum
precession simulation exhibits significantly higher SATs over Europe (and
generally the Northern Hemisphere, as shown in Fig. 3b), but lower values
over part of North Africa, ∼ 10–20∘ N (Fig. 9a), as a result
of increased cloud cover caused by major changes in precipitation patterns
over this area (Fig. 9b). The northward shift of the ITCZ is clearly visible
in the absolute changes in precipitation between the two precession extremes
(Fig. 9c, d). During precession minimum, precipitation
> 10 mm day-1 reaches as far north as ∼ 18∘N
and intensifies over land (Fig. 9d). By contrast, during precession maximum
higher precipitation (positive anomalies) occurs over the Atlantic and only
reaches ∼ 10∘ N (Fig. 9c).
As well as this land–sea contrast, the northernmost part of the northern African
continent exhibits very different patterns from the more southerly area.
These two regions (as defined in Fig. 9a and b) are therefore analysed
separately.
In the northern region (land-only component of the northern “box” in
Fig. 9a), modelled SATs show a very similar seasonal distribution and a
single seasonal peak around the month of July in all the simulations
presented in Fig. 10a: both extreme precession experiments with 280 ppm
CO2, in the two control runs (late Miocene and preindustrial
palaeogeography with present day orbital forcing), and in the minimum
precession simulation with 400 ppm. Considerably higher temperatures are
reached during precession minimum (over 35 ∘C at 280 ppm and close
to 40 ∘C at 400 ppm) while the lowest summer temperatures
(< 30 ∘C) occur in the precession maximum simulation
(Fig. 10a). Precipitation exhibits a bi-modal distribution, which is most
pronounced in the precession minimum simulation, when even in the drier parts
of North Africa precipitation reaches 0.8–1 mm day-1 in August
(Fig. 10b). Precipitation generally peaks around the months of June and
September, but during precession minimum this second and most pronounced peak
occurs about 1 month earlier in the season (August) and later (October) in
the late Miocene control. The winter months are characterised by extremely
dry conditions in all simulations, with precipitation consistently below
0.10 mm day-1 (Fig. 10b).
SAT and precipitation seasonal distribution over North Africa
(averaged over land in the northern and southern “boxes”, as indicated in
Fig. 8) for the two precession extremes (pMIN and pMAX) at 280 ppm,
precession minimum at 400 ppm and the two control experiments (late Miocene
and preindustrial at 280 ppm). Differences due to orbits, palaeogeography
and CO2 concentrations are highlighted by the vertical bars relative to
the month of August when the seasonal distribution is not varying. Note that
the scales in panel (b) and (d) are not the same, due to the strong differences
in the amount of precipitation. Dashed lines in panel (d) represent
precipitation in the northern region (from panel b) on the same scale as
precipitation in the southern region; the simulation–colour correspondence is
the same as in the other panels.
Modelled SATs in the southern region (land-only component of the southern
“box”, as defined in Fig. 9a, where latitudes and longitudes are defined as
in for the present-day West African monsoon) of North
Africa show a weak bi-modal distribution with peaks in April–May and
September–October, and this second peak is most pronounced in both precession
minimum simulations (Fig. 10c). These summer temperatures of
∼ 28 ∘C are considerably lower than those in the northern
region and are caused by the increased cloud cover during the monsoon season,
with little variation in the seasonal distribution between the different
simulations (Fig. 10d). However, considerably higher precipitation values are
reached in the precession minimum experiments (over 8 mm day-1)
irrespective of which CO2 concentrations are used (Fig. 10d). This may reflect a
non-linear relationship between changes in
African monsoonal precipitation and an
increase in CO2 concentrations, similarly to what has been demonstrated for present-day simulations
.
The only difference between the two control experiments (LMctrl and PIctrl)
is the palaeogeography, and this results in significantly different
precipitation values. For instance, in the late Miocene control experiment,
precipitation rates are up to ∼ 2 mm day-1 lower in southern
North Africa than they are in the preindustrial control run (Fig. 10d). There
are also smaller differences (< 0.3 mm day-1) in the northern
region (Fig. 10b). Across the whole of North Africa, the preindustrial
control experiment is on average warmer than the late Miocene control. This
is most pronounced in the northern region, where SATs are up to
∼ 2 ∘C greater (Fig. 10a), as a result of stronger local
summer insolation. Analysis of the different simulations demonstrates that in
the northern region the biggest influence on temperature range is orbital
variability (∼ 7 ∘C; Fig. 10a), while CO2 results in
∼ 3 ∘C temperature difference and palaeogeography
∼ 1 ∘C. The striking differences in precipitation in the
southern region are again most strongly influenced by orbital variability,
which contributes up to ∼ 2.5 mm day-1 to the August peak
(Fig. 10d). The June–July–August–September average of SAT and
precipitation for each of the experiments summarised in Fig. 10 can also be
found in Table 1. This highlights the extended length of the monsoon season
during precession minimum at 400 ppm CO2, resulting in increased
modelled precipitation in the month of September (but no change with respect
to the 280 ppm simulation in the month of August) and therefore for the
entire period.
Summer (JJAS) u and v components of low level winds (850 hPa) over
North Africa at pMIN (a), pMAX (b) and for the late Miocene CTRL experiment (c).
Vegetation fraction differences between precession minimum and
precession maximum simulations for different functional types: (a) C4
grasses, (b) bare soil, and (c) broadleaf trees at 280 ppm CO2
concentration. Vegetation fraction differences between precession minimum and
precession maximum simulations at 400 ppm CO2 minus the same
difference at 280 ppm CO2 for different functional types: (d) C4
grasses, (e) bare soil, and (f) broadleaf trees. The approximate location of
the Sahel region is indicated in panel (f).
Note that, to a lesser extent, obliquity forcing also has an impact on the
North African summer monsoon e.g.. Given
our experimental design, it is possible to compare experiments 1 and 22,
where obliquity is a maximum and minimum, respectively, and precession has
very similar values in both simulations (see Fig. 1). The seasonal
distribution of precipitation in the North African monsoon region (southern
“box”) is, however, very similar between the two extreme obliquity
simulations and so are the mean annual values, only showing significant
differences in the summer months (below ∼ 1 mm day-1;
Supplement, Fig. S7). This results in much smaller hydrologic changes
compared to the extreme precession simulation and these are, therefore, not
discussed in more detail in this study.
In the model, the variability in the African summer monsoon between the two
precession extremes can largely be explained by changes to the regional
circulation – for instance, in the strength of the African westerly jet, which
transports moisture into North Africa during precession minima. Because of a
greater land–sea temperature differential, low level winds are stronger
(> 10 m s-1) in the precession minimum simulation
(Fig. 11a) and weaker (< 4 m s-1) during precession maxima
(Fig. 11b), relative to the modern orbit late Miocene control experiment
(Fig. 11c). The importance of perturbations to the large-scale atmospheric
circulation is also shown by the differences in the strength of the Hadley
circulation between these three simulations (Supplement, Fig. S11). During
precession minimum, the ascending branch is much stronger than in the late
Miocene control run and it shows a northward propagation of
∼ 4∘. During precession maximum, the ascending branch is
significantly weaker than in the control and located ∼ 3∘
further south. This clearly indicates the shifts in the position of the ITCZ
during these three simulations.
Impact on vegetation
In our experiments, the substantially increased precipitation at times of
precession minimum (Fig. 9b, d) results in a greening of the areas south of
the Sahel region (Fig. 12). During the precession minimum C4 grasses shift to
the north (Fig. 12a), colonising areas around 15–20∘ N, which are
instead covered by the desert fraction (bare soil) during the precession
maximum (Fig. 12b). Further south, between ∼ 5 and 15∘ N, bare
soil is also partially substituted by broadleaf trees in the precession
minimum simulation (Fig. 12b, c). A similar amplified precession signal in
the monsoon and an extended seasonality within a year when interactive
vegetation is included has also been found in both transient and
time-slice simulations e.g.. A greening
around the Sahel region during this time period is also consistent with
geochemical and mineralogical studies , and a northward
displacement of the tree line during precession minima has also been observed
in an idealised modelling study . The permanent presence of
an extensive desert area in North Africa throughout the entire precession
cycle also appears realistic, since both observational and
modelling studies suggest that the formation of the Sahara
may have been initiated as early as the late Miocene. Vegetation
reconstructions for the Late Miocene are also consistent with this
hypothesis, indicating the presence of arid conditions starting at around
7 Ma .
We have also investigated the sensitivity of changes in vegetation
distribution to varying CO2 concentrations. However, since the
precipitation simulated by precession minima experiments with both 280 and
400 ppm CO2 are nearly identical over southern North Africa, where the
significant vegetation changes are found at 280 ppm (Fig. 12a–c), the small
difference in vegetation across this area is unsurprising (Fig. 12d–f). No
major changes are found over North Africa between the two experiments in the
expansion of the tree fraction (Fig. 12f), and the differences further south
are unrelated to the North African summer monsoon. Patchy differences in C4
grasses distribution increase with CO2 in the central part of North
Africa, where they cover areas that are desert at 280 ppm. C4 grasses
decrease to the western side, where they are substituted by the desert
fraction (Fig. 12d, e). The less predictable distribution of these changes is
also perhaps not unexpected, since CO2 and vegetation feedbacks do not
necessary combine linearly . More significant differences
in sensitivity may be found in regions outside North Africa, where vegetation
productivity is higher (Supplement, Figs. S8, S9 and S10). Exploring this
further is, however, beyond the scope of this work.
The recurrence of the so-called African humid periods has been intensively
studied both in observational (e.g. Larrasoaña et al., 2003, and
references therein) and modelling
e.g. investigations, especially
for the Quaternary period. The proxy record indicates that these periods were
characterised by a northward shift in precipitation as a result of a stronger
African summer monsoon, paced by astronomically forced insolation changes. To
date, modelling studies have largely failed to simulate the northward penetration of
the African summer monsoon beyond 21∘ N and increase precipitation
sufficiently to simulate the mid-Holocene “Green Sahara”
conditions
e.g.. These
conditions would allow savanna-like vegetation to expand northward, beyond
the central Saharan watershed .
hypothesised that the lack of interactive vegetation could be the main reason
for the insufficient precipitation over the Sahara in mid-Holocene
simulations. However, in our simulations which are coupled with a vegetation
model, the summer precipitation increase during precession minimum is still
confined south of 21∘ N in North Africa. Assuming that the monsoon
system in the late Miocene was similar to that of the Quaternary, this
indicates that even our fully coupled model still fails to represent relevant
processes driving precipitation in the Sahel regions. This is perhaps
suggesting the lack of relevant teleconnections in the model, such as those
found with North Atlantic dynamics e.g.. In
addition, this could also be due to the low sensitivity of the
land–atmosphere coupling which characterises models of the HadAM3 family
.
Seasonality of the African summer monsoon on sub-precessional timescales
Our experimental design allows us to analyse the seasonal distribution of
SATs and precipitation patterns over North Africa not only for the two
precessional extremes but also throughout the different stages of the
orbital cycle (Fig. 13). The highest SATs (up to 35 ∘C) are reached
in the northern region during the summer months (Fig. 13a). In the southern
region, SATs remain below 30 ∘C throughout the entire cycle
(Fig. 13b). The highest quantity of precipitation (up to
2500 mm day-1) is found in the southern region during the summer
months and especially around the precession minimum (Fig. 13d). In the
northern region, which is outside the area influenced by the summer monsoon,
drier conditions persist throughout the entire cycle and in all seasons
(Fig. 13c), with values consistently below the driest periods experienced in
the southern region (maximum 250 mm day-1).
SAT (a, b) and precipitation (c, d) evolution throughout the
simulated precession cycle, in the northern (left) and southern (right)
regions. (a–d) Annual means relative to the corresponding panel above. On the
horizontal axis is the geological time, represented by the 22 orbital
experiments plotted with respect to climatic precession. In panels (a) and
(d) the black dashed line highlights during which month the maximum value of
temperature or precipitation, respectively, is reached. Note that panel (c) is
not on the same scale as panel (d), as all the values are below
0.86 mm day-1 (lowest contours and orange colours in panel d). Also
note that the four annual mean panels are not on the same scale, as their aim
is to show the phasing with orbital forcing rather than to compare the actual
values.
The mean annual values show the correlation with the precession forcing,
which is positive for modelled precipitation and SAT in the southern region,
and negative for SAT in the northern region. However, some lags between
orbital forcing and the climate response can also be seen. For instance,
maximum precipitation in the southern region occurs at the same time as the
precession minimum, but minimum precipitation lags the precession maximum by
about 2 kyr (Fig. 13d). In the northern region, where precipitation rates
are an order of magnitude smaller than in the southern region, the phasing
with precession is less clear; maximum annual precipitation corresponds to
the precession minimum, but the signal flattens out in the remaining part of
the cycle around the precession maximum, and minimum values are reached
around simulations 13 and 20 (Fig. 13c). Note that part of the changes in the
northern region, characterised by low precipitation value, may be due to
interdecadal/centennial variability in the model.
Minimum annual SAT in the southern region occurs with a 1 kyr lag after the
precession minimum, while maximum SAT lags the precession maximum by
2–3 kyr (Fig. 13b). In this area, the SAT response to orbital forcing is
linked to the increased cloud cover at times of precession minimum (maximum
monsoon strength), as discussed in Sect. 3.4. In the northern region, maximum
annual SAT leads the precession minimum by ∼ 5 kyr, while minimum SAT
leads the precession maximum by ∼ 4 kyr.
Although maximum values for modelled SATs in the northern region and for
precipitation in the southern region can be correlated with the precession
minimum, the seasonal response is not “symmetrical”, but rather it exhibits an
elongated and slightly tilted structure (Fig. 13a, d). This asymmetrical
response around the precession minimum has also been observed in transient
idealised orbital simulations with a model of intermediate complexity
and has been explained by the extended length of the
monsoon season around the precession minimum. At this stage in the orbital
cycle, the North African monsoon can start up to 1 month in advance and end
a month later than average monsoon timing (Fig. 13d), in agreement with the
results of . One possible explanation for this phenomenon is
the presence of a larger vegetated area during precession minimum, which
modifies the albedo feedback and results in a longer monsoon season
.
Finally, modelled SATs in the northern region consistently peak during the
month of July (Fig. 13a), with the exception of simulations 11 to 15 (August)
and 20 to 22 (June). In the southern region, the month of August consistently
exhibits the highest values of precipitation (Fig. 13d), apart from
simulations 2, 3, and 22, where July is the wettest month, and simulation 21,
in which June has the highest precipitation rates. This differs from the
results of , whose idealised simulations consistently showed
maximum precipitation rates during the month of July, in all the precession
minimum experiments throughout their entire simulated time period. This
difference is likely due to the fact that our simulations use
realistically varying orbital parameters throughout one precession cycle and
that interactive vegetation is also included.