Since the start of the Holocene, temperatures in the Arctic have steadily declined. This has been accredited to the orbitally forced decrease in summer insolation reconstructed over the same period. However, here we present climate modelling results from an Earth model of intermediate complexity (EMIC) that indicate that 17–40 % of the cooling in the Arctic, over the period 9–0 ka, was a direct result of the desertification that occurred in the Sahara after the termination of the African Humid Period. We have performed a suite of sensitivity experiments to analyse the impact of different combinations of forcings, including various vegetation covers in the Sahara. Our simulations suggest that over the course of the Holocene, a strong increase in surface albedo in the Sahara as a result of desertification led to a regional increase in surface pressure, a weakening of the trade winds, the westerlies and the polar easterlies, which in turn reduced the meridional heat transported by the atmosphere to the Arctic. We conclude that during interglacials, the climate of the Northern Hemisphere is sensitive to changes in Sahara vegetation type.
The Holocene is characterized by an early thermal maximum
(
The termination of the AHP and the associated vegetation–climate
interactions have been studied in numerous climate model studies (Claussen
et al., 1999; Renssen et al., 2003; Liu et al., 2006; Notaro et al., 2008;
Timm et al., 2010). However, these simulations were mostly focused on the
regional climate effects in northern Africa, implying that the impact of the
Saharan desertification on the global climate has not yet been addressed in
model studies. It is important to realise that the Sahara covers a large area
(approximately
We applied LOVECLIM, an Earth model of intermediate complexity (Goosse et
al., 2010), which is comprised of a coupled atmosphere, ocean, sea-ice and
vegetation model. The atmospheric component (ECBilt) is a quasi-geostrophic
model of the atmosphere with a horizontal T21 truncation and three vertical
layers, 800, 500 and 200
The oceanic component (CLIO) is a primitive-equation, free-surface general
ocean circulation model (Deleersnijder and Campin, 1995; Deleersnijder et
al., 1997) coupled to a thermodynamic-dynamic sea-ice model (Fichefet and
Morales Maqueda, 1997, 1999), and has a resolution of
LOVECLIM simulates a modern climate that is in reasonable agreement with
observations, including large-scale distributions of temperature and
precipitation, and sea-ice cover in both hemispheres (Goosse et al., 2010).
It has a climate sensitivity to a doubling of the atmospheric CO
In order to calculate the contribution of Sahara desertification to cooling
in the Arctic during the Holocene, a series of experiments were designed. The
experiments can be broken into two categories. Firstly, we performed a series
of transient Holocene simulations (9–0
An initial transient simulation from 9 to 0
The transient experiments OG and OGGIS provided simulated estimates of the
evolution of the Sahara over the Holocene. Using these transient simulations,
we could thus define the vegetation composition of the Sahara at 9, 6 and
0
All equilibrium simulations were run for 2500 years, allowing the model, in
particular the deep oceans, to reach a quasi-equilibrium state (Renssen et
al., 2006). The initial conditions for these simulations were derived from a
control experiment with pre-industrial forcings. For the OGGIS equilibrium
simulations, LIS and GIS topography were kept constant. However, LIS and GIS
melt fluxes were not included, as this would have resulted in continuous
freshening of the oceans during the 2500 years of model integration, implying
that the simulated climate would not reach a quasi-equilibrium, in particular
in the deep oceans. Since our analysis is focused on a comparison of the
equilibrium responses to forcings (see Appendix B), such analysis of LIS and
GIS melt fluxes would not be particularly meaningful. Although we recognise
this as a potential weakness of our 9 ka equilibrium simulations, it must be
noted that with an earlier version of LOVECLIM, experiments performed including LIS melt fluxes showed that the impact of these on the Nordic
seas was a modest 0.5
From the OG and OGGIS equilibrium experiments we were able to deduce the
contribution of Sahara desertification to Arctic cooling during the Holocene
(see Appendix B). Data presented are averages over the last 500 years of each
simulation. To investigate an “extreme” example of Sahara desertification
between the mid- and late Holocene, we also performed six equilibrium
simulations that had 100 % grass and desert in the Sahara at 9, 6 and
0
We separate our results into several distinct sections. Firstly we describe the results from our sensitivity experiments (Sect. 3.1). Following this we explain the mechanism that is responsible for the reduction in poleward heat transport during the Holocene (Sect. 3.2), and the role that sea-ice feedbacks play in the Arctic cooling we simulate (Sect. 3.3). We then introduce the results of additional sensitivity experiments (Sect. 3.4), which we performed to test the robustness of our initial findings. We follow this with a discussion of the uncertainties of our results (Sect. 3.5). To finish, we discuss these findings in the context of the limitations of our model and how these could be overcome in future work (Sect. 3.6)
Our results for both the OG and OGGIS equilibrium experiments can be
separated into two distinct phases, 9–6 and 6–0
Simulated temperature change in the Arctic for
Mean annual surface temperature (
In our OGGIS transient experiment, the 9 ka climate is relatively cold due
to the cooling effect of the LIS topography and albedo, and the GIS albedo,
which leads to a delayed thermal maximum over most of the Arctic domain. This
delayed onset of the thermal maximum for the early to mid-Holocene has been
observed in numerous proxy records (Kaufman et al., 2004) and is consistent
with previous experiments performed with LOVECLIM. In particular Renssen et
al. (2009), who simulated a
The second phase, 6–0
For 9 to 0
The model experiments indicate that part of the cooling in the Arctic over
the Holocene is a direct consequence of Sahara desertification, which invokes
a land–atmosphere teleconnection. Due to the prescribed desertification in
the Sahara in our equilibrium simulations, over the course of the Holocene
net albedo and radiative heat loss increases, leading to a decrease in
surface temperatures and an increase in surface pressure in the Sahara. The
decreasing temperatures in the Sahara cause an increase in the surface
pressure (Fig. 3) with an extension over the Tropical Atlantic, leading to an
easterly zonal shift, plus an expansion, of the Azores high. Additionally, we
observe a weakening of the low-latitude trade winds and a net overall
decrease in the atmospheric meridional heat flux (Fig. 4). In particular we
observe a weakening of the mid-latitude westerlies. This overall weakening of
the winds and atmospheric heat transport over the Atlantic Ocean is
consistent with a decrease in the meridional temperature gradient due the
relatively strong cooling over the Sahara. As a result, the Icelandic Low
stabilises, which in turns results in a weakening of the polar easterlies.
The Bjerknes compensation theory suggests that weakened atmospheric heat
transport would be compensated by an increase in oceanic heat transport
(Bjerknes, 1964). This is indeed what we observe (Fig. 4), however the
increase in oceanic heat transport (0.058 PW at
In the OGGIS simulation, we see the same long-range land–atmosphere–ocean
teleconnection present. However, the localised effects of the increase in
albedo between 9 and 6
Since the peak in the Holocene thermal maximum (
9k0kEQ–9k9kEQ_OG 800 hPa geopotential height, allowing the effects of changing vegetation in the Sahara, on pressure, to be visualised.
The (i) mean annual oceanic heat flux anomaly (PW) (green line),
(ii) mean annual atmospheric heat flux anomaly (PW) (black line) and
(iii) mean annual wind magnitude anomaly (
Mean annual sea-ice concentration (%) anomaly plots from the OG
equilibrium simulation from 9 to 0 ka showing the relative contributions of
Our results for the OG equilibrium simulations show that from 9 to
0
To evaluate the full potential of the impact of Saharan vegetation changes on
the Arctic climate, we performed seven additional sensitivity experiments in
which we prescribed either 100 % grassland or 100 % desert cover in
the Sahara (see Appendix D). Biome reconstructions of pollen and plant
macrofossils show that during the mid-Holocene the Sahara was fully covered
by grass and shrubs (Hoelzmann et al., 1998; Jolly et al., 1998). Hoelzmann
et al. (1998) produced estimated percentages of various tree types across
northern Africa and the Arabian Peninsula for 6
However, in our OG simulation from 9 to 0
Therefore, this suggests that the impact of Sahara vegetation on Arctic
cooling could be greater than the 17 % we estimated from our initial
sensitivity experiments. To account for this, and to constrain our results,
we simulated extreme, early (9
The results show that from a 9
One important requirement for our results is that the Sahara region was
warmer at 9 and 6
In LOVECLIM, the total upward and downward shortwave fluxes are calculated as
a function of the ISCCP D2 cloud cover data set consisting of modern day
observations. Whilst this is a reasonable first-order approximation for the
modern day Sahara, it is not so during periods when the Sahara was
vegetated, i.e. 9 and 6
For instance, if we assume that desert albedo is 0.4, and we have a 100 %
cloudless desert environment, then the downward solar radiation reaching the
surface of the Sahara is 342
To test this we devised a series of sensitivity experiments that would allow
for the addition of cloud cover over the Sahara at 9 and 6
As can be seen (Fig. 6), the Sahara was warmer during the mid-Holocene in both
the simulation with the original modern day cloud data set at 6
Temperature change between
As a final test of the sensitivity experiments, when we compare the
temperature changes in the Arctic at 9
The results presented here provide a first-order approach to understanding
the impact of long-term radiative forcing of the Sahara over the course of
the Holocene (9–0
The land–atmosphere teleconnection we have outlined is initiated by a change in vegetation cover in the Sahara region and then carried through the atmosphere from the equatorial region to the Arctic. However, this draws close scrutiny of the atmospheric component of LOVECLIM. Given the relative simplicity of the ECBilt model compared to GCMs, it would be easy to dismiss our findings. However, as the model uses relatively simple physics, any findings in such a model should also be detectable in a more complex model. To highlight this, we shall compare the response of LOVECLIM in other studies to more complex models. Specifically we shall highlight studies that have focused on interglacial climate only as this will allow us to show the response of LOVECLIM to different perturbations, over a variety of temporal and spatial extents.
The primary focus of our study is the cooling of the Arctic over the
Holocene. Therefore, it would be apt for us to investigate how the
temperature profile of LOVECLIM compares with more complex models. In this
study and previous studies (Renssen et al., 2005), it is shown that LOVECLIM
shows a gradual cooling over the Holocene as a response to reduced summer
insolation forcing and decreasing greenhouse gases. In a recent study (Liu et
al., 2014), the primary focus of which was trying to understand the apparent
discrepancy between a reconstructed global temperature study (Marcott et
al., 2013) and models, it was shown that the three models employed in the
study, LOVECLIM, CCSM3 and FAMOUS (the latter two being GCMs) all simulated a
robust annual mean global warming (0.5
In a study of the last interglacial, Nikolova et al. (2013) provide a detailed
analysis of a variety of climatic parameters, as simulated in both LOVECLIM
and CCSM3. In these experiments they perturbed the climate for forcings
equivalent to MIS-5e (127
In a multimodel-data study (Bakker and Renssen, 2014), which included six GCMs and three EMICs, the data showed that when comparing the simulated period of maximum warmth during the last interglacial period against proxy-based temperature reconstructions, LOVECLIM, along with the two other EMICs in the study, all overestimated the maximum temperature the least, across all latitudes. This is probably on account of models with lower resolution displaying less internal variability, when compared to GCMs (Gregory et al., 2005), leading to a more spatially coherent evolution of the climate.
In another series of sensitivity experiments performed for the period MIS-13
(
As we have outlined above when compared to GCMs, LOVECLIM has shown itself to be able to replicate similar results across a range of interglacial climates. Although just a selection are included here, it serves to demonstrate that despite its apparent shortcomings, LOVECLIM is an effective tool for climate investigations over a variety of spatial and temporal scales.
Another limitation in our study is due to the simplistic cloud parameterisation scheme that is employed within our model. As discussed in Sect. 3.4, we designed a series of sensitivity experiments to test this; although this is not a perfect solution, we were able to show that our original land–atmosphere teleconnection holds. However, without the evolution of clouds in response to changes in the climate system, particularly in response to surface temperature and sea-ice coverage changes, we are still missing potentially important feedbacks. However, this issue is not easily solved as it is a problem that persists through the entire spectrum of the modelling community, and it is widely accepted that this is a major source of uncertainty in climate model sensitivity experiments (Randall et al., 2007).
Another potential source of error in our results lies in the vegetation
fractions simulated in our transient simulations, which were then used to
define the Sahara vegetation fractions in our equilibrium experiments. As
previously noted (Sect. 3.3), the vegetation fraction in our model decreases
from 65 to 20 % from 9 to 0
As a result of this, our 9 ka simulations (with a 65 % tree fraction at
9
As a result, the temperature gradient between the early and late Holocene in our experiments is not as steep as it possibly would be. Therefore, it would be interesting to see in future experiments, with a model that captures a more realistic temperature gradient between the early and late Holocene, how the polewards transported heat flux over the course of the Holocene is impacted.
To summarise, compared to comprehensive GCMs, LOVECLIM has several simplifications, of which the fixed cloud cover and the relatively simple vegetation scheme are likely to be the most important for this study. However, model inter-comparison studies have shown that LOVECLIM produces similar interglacial climate responses to changes in forcings as GCMs, making it likely that our results will be confirmed in future studies with more comprehensive models.
In conclusion, our simulation experiments
indicate that over the course of the Holocene, the observed cooling in the
Arctic region is not only driven by local changes in insolation, but that the
effects of desertification in the Sahara initiate a long-range
land–atmosphere teleconnection. In our model experiments, this teleconnection
accounts for between 17 and 40 % of the observed Arctic cooling between 9
and 0
Equations to calculate the relative contributions of ORBG
Vegetation fractions (%) of trees and grass (desert
F. J. Davies, H. Renssen, and M. Blaschek are funded by the “European Communities 7th Framework Programme FP7/2013, Marie Curie Actions, under grant agreement no. 238111: CASEITN”. F. Muschitiello is funded by the Bolin Centre for Climate Research. All support is greatly appreciated. Edited by: E. Zorita