Show Summary Details

Page of

Printed from Oxford Research Encyclopedias, Climate Science. Under the terms of the licence agreement, an individual user may print out a single article for personal use (for details see Privacy Policy and Legal Notice).

date: 01 October 2022

Theory and Modeling of the African Humid Period and the Green Saharafree

Theory and Modeling of the African Humid Period and the Green Saharafree

  • Martin Claussen, Martin ClaussenUniversity of Hamburg
  • Anne DallmeyerAnne DallmeyerMax Planck Institute for Meteorology
  •  and Jürgen BaderJürgen BaderMax Planck Institute for Meteorology


There is ample evidence from palaeobotanic and palaeoclimatic reconstructions that during early and mid-Holocene between some 11,700 years (in some regions, a few thousand years earlier) and some 4200 years ago, subtropical North Africa was much more humid and greener than today. This African Humid Period (AHP) was triggered by changes in the orbital forcing, with the climatic precession as the dominant pacemaker. Climate system modeling in the 1990s revealed that orbital forcing alone cannot explain the large changes in the North African summer monsoon and subsequent ecosystem changes in the Sahara. Feedbacks between atmosphere, land surface, and ocean were shown to strongly amplify monsoon and vegetation changes. Forcing and feedbacks have caused changes far larger in amplitude and extent than experienced today in the Sahara and Sahel. Most, if not all, climate system models, however, tend to underestimate the amplitude of past African monsoon changes and the extent of the land-surface changes in the Sahara. Hence, it seems plausible that some feedback processes are not properly described, or are even missing, in the climate system models.

Perhaps even more challenging than explaining the existence of the AHP and the Green Sahara is the interpretation of data that reveal an abrupt termination of the last AHP. Based on climate system modeling and theoretical considerations in the late 1990s, it was proposed that the AHP could have ended, and the Sahara could have expanded, within just a few centuries—that is, much faster than orbital forcing. In 2000, paleo records of terrestrial dust deposition off Mauritania seemingly corroborated the prediction of an abrupt termination. However, with the uncovering of more paleo data, considerable controversy has arisen over the geological evidence of abrupt climate and ecosystem changes. Some records clearly show abrupt changes in some climate and terrestrial parameters, while others do not. Also, climate system modeling provides an ambiguous picture.

The prediction of abrupt climate and ecosystem changes at the end of the AHP is hampered by limitations implicit in the climate system. Because of the ubiquitous climate variability, it is extremely unlikely that individual paleo records and model simulations completely match. They could do so in a statistical sense, that is, if the statistics of a large ensemble of paleo data and of model simulations converge. Likewise, the interpretation regarding the strength of terrestrial feedback from individual records is elusive. Plant diversity, rarely captured in climate system models, can obliterate any abrupt shift between green and desert state. Hence, the strength of climate—vegetation feedback is probably not a universal property of a certain region but depends on the vegetation composition, which can change with time. Because of spatial heterogeneity of the African landscape and the African monsoon circulation, abrupt changes can occur in several, but not all, regions at different times during the transition from the humid mid-Holocene climate to the present-day more arid climate. Abrupt changes in one region can be induced by abrupt changes in other regions, a process sometimes referred to as “induced tipping.” The African monsoon system seems to be prone to fast and potentially abrupt changes, which to understand and to predict remains one of the grand challenges in African climate science.


  • Future Climate Change Scenarios
  • Climate of Africa


The Sahara, the largest desert on Earth, is one of the most fascinating regions in the world. Today an extreme environment for humans, flora, and fauna, it appeared to be much more habitable several millennia ago. In 1850, Heinrich Barth discovered petroglyphs in the Erg Murzuq (Figure 1), and he discussed his discovery in the context of archaeology and past climate change (Barth, 1857).

Figure 1. Petroglyph in the Teli-ssarhe (approximately 25.8°N, 12°E, in the Erg Murzuq) discovered by Heinrich Barth during his travel through North and Central Africa between 1849 and 1855 (Barth, 1857, Vol. I).

This figure is taken from a facsimile reprint published by the Heinrich Barth Institute and the Fines Mundi Publisher with permission from the Heinrich Barth Institute, Cologne, Germany.

The Hungarian adventurer and desert researcher László E. Álmásy was perhaps the first who coined the term Green Sahara in the 1930s when interpreting his findings of rock paintings in the Gilf Kebir and Gabal Uweinat located in the Eastern Sahara (Álmásy, 1934/1997). Back then, many scientists questioned the existence of a humid and vegetated Sahara because earlier reports (e.g., Herodotus, Historia (Melpomene, 168–199), 440 bce; Strabon, Geographica (book 1, chapter 3), 23 ce; see also Hornemann, 1802/1997) were of a more anecdotal nature.

In the second half of the 20th century, however, more and more evidence became available showing that during the early and mid-Holocene some 11.7 to 4.2 ky bp (1000 years before present), and in some regions even earlier with the onset of the Bølling/Allerød around 15 ky bp, subtropical North Africa was much more humid than it is today (e.g., Nicholson & Flohn, 1980; Ritchie et al., 1985; deMenocal et al., 2000; Shanahan et al., 2015). Perennial lakes were abundant, and lake levels were much higher during this so-called African Humid Period (AHP) (e.g., Kutzbach & Street-Perrott, 1985; Yu & Harrison, 1996; Hoelzmann et al., 1998; Coe & Harrison, 2002; Hoelzmann et al., 2010). Pollen-based reconstructions reveal higher water availability (Bartlein et al., 2011, Francus et al., 2013).

The Sahara was indeed greener than it is today (e.g., Jolly et al., 1998; Prentice et al., 2000; Kröpelin et al., 2008; Lézine et al., 2011), although the term Green Sahara might be somewhat misleading. Certainly, lush gallery forests were abundant in the vicinity of lakes and rivers. In large areas, however, a mixture of more yellowish or brownish Saharan, Sahelian, and Sudanian plant groups, or phytochoria, prevailed (Hély et al., 2014). A large number of archaeological excavations in the Eastern Sahara have revealed close links between past climatic change and prehistoric occupation during the past 12,000 years. The southward shift of the desert margin at the end of the African Humid Period is supposed to have triggered the emergence of the pharaonic civilization along the Nile (Kuper & Kröpelin, 2006).

This article reviews the causes of the AHP and the Green Sahara, that is, the forcing and the feedbacks that led to the large-scale climate and ecosystem change in this region. Particular attention is given to the discussion of terrestrial processes and to interpretation of paleo data that indicate the possibility of abrupt changes, that is, changes in climate and ecosystem that were much faster than the changes in external forcing. Before addressing the dynamics of the AHP, a brief overview of the present-day North African monsoon system and its dynamics is given.

A brief sketch of the North African monsoon system

The expansion of the Sahara depends on the spatial pattern and annual amount of rainfall. Parts of Northern Africa, foremost the Sahara, are located below the descending branches of large-scale tropical atmospheric overturning circulations. The strong subsidence is enhanced by the radiative loss above the bright, nearly cloud-free desert—in the climatological mean, insolation is weaker than the outgoing long-wave radiation. This radiative cooling, relative to surrounding regions, is compensated by adiabatic warming of descending air masses. Because of the large-scale subsidence, convection and the formation of rain are suppressed. The precipitation distribution in the Sahel, the transition zone between the Sahara and the tropical Africa, is basically determined by the regional supply of moisture and by dynamic systems inducing vertical uplift that is strong enough to overcompensate the large-scale circulation and to foster convection-triggering precipitation.

Figure 2. Precipitation (shaded, [mm/day]) and sketch of 850 hPa surface winds (arrows) over North Africa during boreal summer (June–September). Shown also is the location of the Saharan heat low (red) and the African and Tropical Easterly Jet axes. The wind and surface-pressure data are based on ERA (reanalysis) data of 1958–2001 (Uppala et al., 2005); the precipitation data are taken from the GPCP (Global Precipitation Climatology Project) dataset for the years 1979–2006 (Adler et al., 2003).

During summer, moisture from the Atlantic Ocean is provided by the West African summer monsoon—that is, southwesterly, low-level winds that are embedded in the trade winds passing the equator when the Intertropical Convergence Zone (ITCZ) moves to the Northern Hemisphere, following the sun’s zenith point. The northward penetration of the monsoon winds onto the continent depends on the Saharan heat low that is centered in the western part of the Sahara during summer (Figure 2).

Above this shallow heat low, there exists a high-pressure zone, known as Saharan high. Embedded in the anticyclonic flow around this high, and strongly amplified by the meridional temperature and moisture gradient, an easterly wind band is formed: the African Easterly Jet (AEJ) (Cook, 1999). During the summer months, the AEJ is found at an altitude of some 600 hPa, roughly 4 km (Figure 3). The jet maximizes at a latitude of 14°N, in the climatological mean, but it moves back and forth with the seasons and both the strength of the jet and the latitude of the jet maximum vary from year to year.

Figure 3. Zonal wind averaged over the region 10°W to 10°E during June, July, August, and September in (m/s) based on ERA reanalysis data of 1958–2001. The shaded domains display the region with strongest ascent (darker color indicates stronger ascent). Marked also are the positions of the dynamic systems Saharan heat low (L), Intertropical Front (ITF), African Easterly Jet (AEJ), near-surface westerly monsoon flow (M), Tropical Easterly Jet (TEJ). The zonally averaged precipitation data for June to September are taken from the GPCP dataset for the years 1979–2006.

The AEJ plays an important role in the development of the systems that produce rainfall. Along the AEJ, on a northern and southern track, synoptic scale disturbances with a wavelength of some 2500 km, the so-called African Easterly waves (AEWs), propagate westward. Downstream of the AEW troughs, regions with a vertical ascent are induced in which squall lines and mesoscale convective systems form, bearing strong rainfall events (Janiga & Thorncroft, 2016). For the last decades of the 20th century, Lebel et al. (2003) estimate that some 12 percent of the total number of mesoscale convective systems produce 90 percent of the rainfall during the peak rainy season. Nesbitt and Zipser (2003) indicate that very intense mesoscale systems, which comprise some 3 to 4 percent of all rain events, produce up to 80 percent of the rainfall that occurs in the Sahel. But convection is only possible if the waves are fed with moist air at the ground (Nicholson & Grist, 2003). Therefore, the AEW track north of the AEJ has no substantial effect on precipitation in the present-day climate. The maximum of AEW activity is located on the southern track, 4°–5° south of the AEJ, and coincides with the rainfall maximum.

The rainfall maximum in West Africa is found between the AEJ and a second jet, the Tropical Easterly Jet (TEJ) (Figures 2, 3). The TEJ is known as a key player in the Indian summer monsoon system, emerging from the large thermal contrast between the Tibetan Plateau and the equatorial atmosphere over the Indian Ocean. The TEJ is strongest over the Indian Ocean but spreads over the East African continent with a position of the jet axis at 14°N and at a height of some 150–200hPa. During summer, a second, well-separated easterly wind maximum appears over the Guinean coast, centered at 7°N and at an altitude of some 200 hPa (Figure 2). Whether this maximum is an extension of the TEJ over the Indian Ocean or is governed by its own dynamics is still under discussion (e.g., Nicholson & Grist, 2003; Nicholson, 2009).

The region between the AEJ and the TEJ is characterized by a strong uplift (Figure 3). This uplift is partly caused by the release of latent heat and partly by the dynamics of the TEJ. Owing to the breakdown of the geostrophic wind balance at the entrance and the exit of any jet streak (i.e., area of maximum wind speed), a secondary meridional circulation is induced, leading to convergence in the left entrance and right exit, and to divergence in the left exit and right entrance of the jet streak. The specific constellation with two maxima in the easterly wind pattern at the TEJ level results in a broad zone of jet-streak-related upper-level divergence above West Africa under which a vertical ascent is forced and convection and precipitation are favored. To what extent the dynamics of the TEJ and the rainbelt interact and to what extent wave disturbances along the TEJ affect the precipitation pattern are still being debated (Nicholson et al., 2007; Nicholson, 2009). The vertical uplift related to the Saharan heat low and the Intertropical Front (ITF; i.e., the zone at around 20°N where the southwesterly monsoon flow converges with the near-surface easterlies) is not strong enough to yield much precipitation in present-day climate.

The strong relation between the two jets and the Sahel rainfall also becomes apparent in the present-day rainfall variability. Sahel rainfall varies considerably at interannual time scales (Figure 4). In observations, positive anomalies in Sahel rainfall are often associated with a weaker, more northerly located AEJ and/or a relatively stronger TEJ, while negative anomalies in Sahel rainfall often co-occur with a stronger, more southerly located AEJ and/or a weaker TEJ (e.g., Cook, 1999; Nicholson, 2009).

The sketch of the West African monsoon outlined in this section holds for most of the Sahara-Sahel region. However, over the eastern part, roughly east of 30°E, the atmospheric circulation and the seasonal cycle of precipitation are affected by the East African Highlands and the East African monsoon. East of 30°E, the AEJ is much less pronounced than over West Africa, while the TEJ is clearly connected to the TEJ over the Indian Ocean. Besides the ITF, a second near-surface convergence zone develops during boreal summer: the Congo Air Boundary, stretching from Central Africa in a northeasterly direction toward the Red Sea, separates the southwesterly monsoon flow of Atlantic air masses and the south easterly flow of air masses mainly from the Indian Ocean.

A specific and remarkable feature of the Sahel rainfall is the decadal to multidecadal persistence of its variability over a large region. Most prominent is the Sahel drought in the 1970s and 1980s, with rainfall anomalies in July, August, and September up to 100 mm lower than the approximately 300 mm average of the years 1940–2014. Since then, Sahel precipitation has increased (Figure 4).

Figure 4. Anomalies of observed and simulated July, August, September mean precipitation in the Sahel (10°W–30°E, 10°N–20°N) derived from the GPCC dataset (blue line) and computed from an ensemble of climate simulations with prescribed observed sea-surface temperature (black dashed line) (Park et al., 2016). Green line: standardized anomalies of the NDVI (Normalized Differential Vegetation Index), a measure of vegetation coverage and greenness, taken from Kaptué et al. (2015).

The decadal precipitation anomalies in the Sahel are likely to be driven by changes in the sea-surface temperatures (SSTs) (Figure 4) and are presumably amplified by changes in land cover and vegetation (e.g., Zeng et al., 1999; Xue et al., 2016). Several studies (e.g., Giannini et al., 2003, 2005; Bader & Latif, 2003; Biasutti et al., 2008; Park et al., 2015) have demonstrated that a positive north–south gradient in the tropical Atlantic SST between regions (70°W–20°W, 5°N–30°N) and (40°W–5°E, 30°S–5°N) (Figure 5) tends to shift the Atlantic ITCZ northward.

Figure 5. Composite difference of July, August, September mean sea-surface temperatures between wet and dry years in the Sahel region during the period 1940 until 2013. (Redrawn from Park, 2015).

Thereby the Sahel receives more moisture from the Atlantic via a low-level southerly air flow. Cooler than normal SST in the Indian Ocean and the tropical Pacific Ocean reduce the stability of tropospheric air in the tropics, which tends to enhance convection over the Sahel. Xue et al. (2016) found that the Indian Ocean SSTs produce an anomalous displacement of the ITCZ before the onset of the West African monsoon, while the Pacific Ocean SSTs mainly contribute to the summer drought in West Africa. Since the late 1990s, the Sahel rainfall recovery has likely been dominated by the Mediterranean SST (Park et al., 2016).

Otterman (1974) and Charney (1975) suggested that changes in the land surface cause variations in the Sahel rainfall, specifically the prolonged Sahel drought that emerged in the 1970s. They argued that a vegetated surface with low albedo reduces the radiative loss and therewith the large-scale subsidence of air above the Sahara and the strong suppression of convective precipitation. Xue and Shukla (1993) and Eltahir and Gong (1996), however, found that the albedo changes in the Sahel were too weak to cause a strong feedback and that mechanisms other than the desert-albedo feedback might dominate. Subsequent studies indicated that land-cover changes might not cause, but rather amplify, Sahel rain variability like a low-pass filter by enhancing decadal climate variability and suppressing interannual climate variability in the Sahel (Zeng et al., 1999; Wang & Eltahir, 2000a; Wang et al., 2004; Los et al., 2006; Vamborg et al., 2014). During the last decades, satellite data have, contrary to the persistent narrative of overgrazing and human-induced desertification, shown a greening of the Sahel in some regions (Figure 4), which Kaptué et al. (2015) attributed to increased rain-use efficiency of the Sahel vegetation.

Forcing and feedbacks

In present-day climate, the Sahel has experienced substantial changes in climate and ecosystems. These changes were even larger and more widespread during the Late Quaternary, roughly the last one million years. Hence, the question that arises is to what extent these long-term changes in the African monsoon system can be attributed to the changes in the forcing or the feedback processes.

Orbital forcing: Pacemaker of the African Humid Period

Spitaler (1921) was presumably the first to propose the hypothesis that the strength of global monsoon winds varies depending on the periodic changes of the Earth’s orbit. These changes lead to seasonal variations in the meridional insolation gradient and, thus, to changes in the temperature contrast between ocean and continents. During periods with large temperature contrast, barometric gradients between land and continent increase (Figure 6), thereby amplifying the monsoon winds. Unfortunately, Spitaler’s computation of the seasonal variations of the Earth orbit parameters was flawed and, hence, his work has almost been forgotten (John Kutzbach, personal communication).

Figure 6. Differences in near-surface temperature (in °C, left figure) and air pressure at sea level (SLP in hPa, right figure) for boreal summer (June, July, August) between mid-Holocene climate 6000 years ago and pre-industrial climate around 1850ce (Ganopolski et al., 1998). This figure has been produced by Andrey Ganopolski using CLIMBER 2.4 with the same boundary conditions as in (Ganopolski et al., 1998).

Evidence in favor of the orbital monsoon hypothesis emerged much later. The occurrence of sapropels (marine sediment layers rich in organic carbon) in the East Mediterranean was found to highly correlate with the frequency of orbital parameters (Rossignol-Strick, 1985; Emeis et al., 2000; Skonieczny et al., 2015). Changes in the water budget computed with climate models in which orbitally driven insolation changes are prescribed are qualitatively in line with evidence from paleo lake data (Kutzbach, 1981; Kutzbach & Otto-Bliesner, 1982; Kutzbach & Street-Perrot, 1985). Tuenter et al. (2007) and Rachmayani et al. (2016) demonstrated that the varying precession is the pacemaker of the African monsoon, being modulated by the changes in the obliquity and the eccentricity of the Earth’s orbit. The precession moves the vernal point (or spring equinox) relative to the perihelion (the point of the Earth’s orbit next to the sun) and, hence determines the length of the seasons and the strength of insolation during the seasons. The obliquity, or tilt of the Earth’s axis, determines the location of the tropics and the polar circles (e.g., Paillard, 2001).

Figure 7. Simulated changes in Saharan vegetation coverage for the last 125,000 years (green curve, upper part). The thin black line indicates the changes in the Rossignol-Strick (1985) monsoon index M normalized to its maximum and minimum value during this period, where M is approximated by M = 2 I(23)–I(0), I(23) and I(0), being the insolation during June, July, August at the latitude 23°N and 0°, respectively. The lower part of the figure shows the variations of the continental humidity index (ratio of nonaeolian and aeolian sediment). (Redrawn from Tjallingii et al., 2008).

If the varying precession is mainly the pacemaker of long-term monsoon changes, then the AHP and a greening of the Sahara should have occurred approximately every 20.000 years, perhaps even for the last seven to eight million years since the formation of the Sahara (Schuster et al., 2006). Larrasoaña et al. (2013) identified 230 Green Sahara Periods over the last eight million years in various continental and marine records. Paleo hydrological data from the West African coast and climate model simulations confirm that the humidity in Northern Africa varies with the precession for the last glacial cycle (Tjallingii et al., 2008). In the simulation (Figure 7 upper part), the imprint of the varying precession can be seen only if the insolation exceeds a certain threshold. Whether this threshold depends on the global climate state—that is, whether it differs between glacial and interglacial climate, has not yet been discussed. The fast variations in the humidity index (Figure 7, lower part) during the cold climate phase have been attributed to the rapid climate changes associated with the Heinrich events and Dansgaard-Oeschger events, rapid swings between cold (stadial) and relatively mild (interstadial) phases during the last glacial, which are absent in the model simulation (Tjallingii et al., 2008).

Ocean and vegetation dynamics: Amplifier of the African Humid Period

Orbital forcing and the appearance of AHPs are highly correlated. But does orbital forcing account for the full range of changes seen in data and climate model simulations? Comparisons of climate simulations using atmospheric circulation models driven with mid-Holocene and present-day insolation, respectively, have shown that differences between simulated mid-Holocene and present-day precipitation are well below some 200 mm/y. That is, they are too low to explain any widespread mid-Holocene vegetation coverage in the Sahara (Joussaume et al., 1999; Braconnot et al., 2012) (Figure 8).

Figure 8. Differences between mid-Holocene and present-day mean annual precipitation (MAP) [mm/y] in North Africa. The line in each box depicts the median value, the box shows the 25 to 75 percent range, and the whiskers show the total range of values. The light blue box indicates reconstructed MAP differences; the purple, dark-blue, and green boxes depict simulated MAP differences from atmosphere-only models in which present-day SST were prescribed: atmosphere-ocean models, and atmosphere-ocean-vegetation models, respectively. (Redrawn from Braconnot et al., 2012).

Therefore, orbital forcing can only be a trigger, but not the full cause, of the onset and termination of AHPs. Feedbacks within the climate system, specifically the interaction between the atmosphere, the ocean, and the land, must have amplified orbitally triggered changes in the North African monsoon.

A number of climate system models that include interactive coupling between atmospheric and oceanic dynamics show that the interaction between atmosphere and ocean tends to amplify orbitally triggered changes in the African monsoon (Kutzbach & Liu, 1997; Hewitt & Mitchell, 1998; Braconnot et al., 1999; Liu et al., 2004; Zhao et al., 2005; Braconnot et al., 2007a, 2007b, 2012) (Figure 8). On average, all models with interactive atmosphere–ocean coupling reveal a delay in the response of the SST to mid-Holocene insolation changes. Winter cooling and summer warming over the ocean are found to occur up to two months later in comparison to temperatures over land. Zhao et al. (2005) analyzed seven coupled atmosphere–ocean circulation models. They found that differences between mid-Holocene and present-day insolation cause warmer than present-day Atlantic SST north of 5°N and colder SST south of 5°N. This dipole-like SST gradient which favors positive Sahel rainfall anomalies in present-day climate (Figure 5) is enhanced by a wind-evaporation feedback and a stronger Ekman drift over the tropical Atlantic Ocean. Generally, the response of the Atlantic SST to orbital forcing and the subsequent feedbacks reinforce the West African monsoon due to a stronger land–sea temperature contrast and a stronger moisture advection.

According to the analysis by Zhao et al. (2005), not only Atlantic SST changes affect the African monsoon, but also the late summer warming of the Mediterranean Sea also contributes to increased precipitation over northern Africa as seen in present-day climate (Rowell, 2003). This is an interesting aspect, for Park et al. (2015) found that in a globally warming climate caused by an increase in greenhouse gases, the link between Sahelian rainfall and tropical SST changes ceases. Instead, Northern Hemisphere warming induces a significant increase in Sahelian rainfall. Specifically, the Mediterranean Sea develops as the region that becomes important for Sahel rainfall (Park et al., 2016).

As in the case of ocean–atmosphere modeling, various studies focusing on land-surface processes indicate an amplification of the West African monsoon caused by interaction between land surface and atmosphere (e.g., Kutzbach et al., 1996). Models with interactive vegetation dynamics reveal a substantial spatial reduction of the mid-Holocene Sahara (e.g., Claussen & Gayler, 1997; Texier et al., 1997; Ganopolski et al., 1998; Braconnot et al., 1999; Doherty et al., 2000; Hales et al., 2006; Schurgers et al., 2006; Vamborg et al., 2011; Rachmayani et al., 2015). The patterns of simulated vegetation differences between mid-Holocene and the present-day Sahara-Sahel region, however, substantially differ (Figure 9).

Figure 9. Differences in vegetation cover in the region of today’s Sahara between present-day and mid-Holocene climate as simulated by various coupled vegetation (or biome)-climate models. Results are shown from simulations by (A)Claussen and Gayler (1997), (B)Doherty et al. (2000), (C)Schurgers et al. (2006), (D)Liu et al., (2007), (E) Vamborg et al. (2011), and (F) Rachmayani et al. (2015). In A, B, C the green areas indicate a change in biomes. In E, F green areas indicate an increase in vegetated coverage by more than 0.1. In E, the large crosses depict areas in which brief green spells appear during the mid-Holocene but not in present-day climate. Small crosses mark areas in which green spells occur more often in mid-Holocene climate than in pre-industrial climate. Green dots indicate locations in which vegetation occurs in the region of the present-day Sahara during the mid-Holocene according to the biome reconstruction by Prentice et al. (2000). The yellow dots indicate locations in the Libyan Sand Sea in which a desert biome is reconstructed also for mid-Holocene climate.

Comparisons with reconstructions of mid-Holocene biomes by Prentice et al. (2000) suggest that large parts of the present-day Sahara were covered by steppe, savanna, and xerophytic woods and scrubs. Only in the Libyan sand sea, desert conditions presumably prevailed (e.g., Larrasoaña et al., 2013). Hence, despite amplification of simulated monsoonal rainfall due to terrestrial feedback, many models underestimate the reduction in desert area.

The strength of the land-surface–atmosphere interaction has been a subject of some controversy. Early studies (Ganopolski et al., 1998; Braconnot et al., 1999) demonstrated that the interaction between land surface and atmosphere presumably amplified mid-Holocene monsoon precipitation in West Africa more strongly than the interaction between ocean and atmosphere did, and that there likely was an additional amplification, or synergy, by including interaction between atmosphere, ocean, and vegetation. A model intercomparison by Braconnot et al. (2007a, 2007b) revealed that in some models, the inclusion of dynamic vegetation in atmosphere–ocean models leads to a decrease in mid-Holocene precipitation over West Africa. A systematic, comparative analysis of feedbacks was, however, not possible because of the different setup of models. The factor analysis by Rachmayani et al. (2015), in turn, supported the early results of an amplifying effect of vegetation dynamics for mid-Holocene West African monsoon rainfall.

In conclusion, incorporation of oceanic and terrestrial feedback in climate models tends to reduce, but not to eliminate, the discrepancy between model simulations and paleo data. Even fully coupled climate system models tend to underestimate the amplitude of mid-Holocene monsoon changes in North Africa by some 20 to 50 percent (Braconnot et al., 2012) (Figure 8). Hence, some feedback processes might be missing or be incorrectly represented in current climate system models.

The specific role of terrestrial processes

The interaction between the land-surface, that is, the greening of the Sahara, and the African monsoon circulation is governed by mainly two land-surface climate parameters: surface albedo and soil moisture availability. As mentioned earlier (see A Brief Sketch of the North African Monsoon System), the desert–albedo feedback proposed by Otterman (1974) and Charney (1975) is probably too weak to cause a strong climate–vegetation feedback during the observational period. However, the differences in albedo between the mid-Holocene and the present-day Sahara were presumably much larger than the albedo variations in today’s Sahel. Today, albedo values up to 0.5 and higher are measured over the surfaces once covered by the Mega Lake Chad (Knorr & Schnitzler, 2006). During mid-Holocene, much lower albedo values presumably prevailed, which should be close to values representative for today’s Central Africa. Consistently, the desert–albedo feedback was found to be an important factor amplifying changes in the Sahara-Sahel precipitation during the Holocene (Claussen & Gayler, 1997; Texier et al., 1997; Knorr & Schnitzler, 2006).

The importance of surface albedo processes for changing the mid-Holocene African climate was investigated in more detail by Vamborg et al. (2011). They have demonstrated that changes in the soil albedo below a vegetation canopy caused by organic matter in the ground and by litter, as well as standing dead biomass covering the ground, strongly contribute to the large albedo difference between the mid-Holocene and the present-day Sahara. Moreover, the dynamic interaction between vegetation coverage and biogenic soil albedo enhances the likelihood of occurrence and persistence of green spells in the Holocene Sahara (Figure 9E).

With changes in the surface coverage, not only the net-radiation surface budget, but also the evaporation and transpiration from soils and vegetation, respectively, vary. Simulations by Levis et al. (2004) indicate that differences in soil albedo affect mid-Holocene precipitation more strongly than do differences associated with enhanced evapotranspiration over vegetated surfaces. Simulations by Liu et al. (2007), Wang et al. (2008), and Notaro et al. (2008), however, suggest the opposite. In their simulations, a negative biogeophysical feedback exists because evaporation from the bare ground appears to be stronger than transpiration from grassland in wet conditions. Thereby, the drying of soil is stronger in the absence of plants. Rachmayani et al. (2015) demonstrated, however, that this weak, or even negative, feedback can be attributed to the neglect of canopy transpiration. If in their model, plants are allowed not only to transpire, but also to evaporate intercepted rainwater, then the feedback between vegetation and precipitation appears to be positive.

The interplay between soil moisture, Sahel rainfall, and the AEJ in present-day climate was found to be possibly important for explaining a large part of the differences between the mid-Holocene and the present-day African monsoon. In the simulations by Patricola and Cook (2007), the mid-Holocene monsoon rainfall increases as the AEJ strongly weakens. In the simulations by Rachmayani et al. (2015), the northward shift of the AEJ led to a reduction in moisture export. The stronger evaporation and transpiration related to more widespread vegetation cover in the mid-Holocene Sahel and southern Sahara led to more rain in this region. According to Rachmayani et al. (2015), the interaction between precipitation, vegetation, and the AEJ dynamics was more important than the desert–albedo feedback for explaining the AHP. To isolate the relative strength of both feedbacks is, however, difficult, because both feedbacks operate at the same time and in the same direction.

Related to changes in the soil moisture availability and the land cover is the emission of dust, which can alter the African monsoon in different ways. Dust in the atmosphere affects the radiation budget, cloud formation, and hence, precipitation. Hui et al. (2008) suggested that increasing dust concentration in the atmosphere over the Sahel region leads to decreasing precipitation. They argued that dust reflects solar radiation, which would then cool the surface and diminish convection. In addition, more dust in the atmosphere would provide more cloud nuclei. Because of the large numbers of cloud nuclei, a large number of cloud droplets of similar small size can develop. On one hand, this homogeneity in droplet size prevents an efficient coagulation of cloud droplets such that larger rain droplets hardly develop. On the other hand, enhanced scattering of solar radiation by dust tends to also enhance absorption of solar radiation. Thus, the increasing dust burden can lead to an “elevated heat pump” (Solmon et al., 2008; Lau et al., 2009) that favors monsoon circulation and moisture advection from the ocean. It is conceivable that these dust–precipitation feedbacks have contributed to the differences between mid-Holocene and present-day African rainfall. A systematic analysis, however, is currently still to be done.

Besides soil and vegetation, lake surfaces also modify the near-surface evaporation. Today, a few lakes still exist in the Sahara, while lakes were abundant during the AHP. The strength of a lake surface–atmosphere feedback is disputed. Coe and Bonan (1997) and Broström et al. (1998) found only a marginal increase in mid-Holocene monsoon precipitation when adding lake surfaces and wetlands to a vegetated Sahara. Krinner et al. (2012) detected a much stronger effect in their simulations. They blamed the weak amplification, apparent in the earlier simulations on the low sensitivity of Sahel rainfall, for changes in vegetation cover in those models. Additionally, the size of a lake matters. According to simulations by Contoux et al. (2013), precipitation over the mid-Holocene Mega Lake Chad was probably reduced above the lake surface because deep convection was inhibited by the overlying colder air. Convective activity around the big lake could have been enhanced, however, because of the increased wind speed over the flat lake surface and the increased moister air to the leeward shore of the lake.

Abrupt termination of the African Humid Period

A brief overview of data and simulations

How did the last AHP begin, and how did it end? In his “Ansichten der Natur,” Alexander von Humboldt (1849) suggested that the AHP terminated abruptly. He argued: “Maybe all these causes of drought and heat . . . would not have been sufficient to turn such huge parts of the African plains into a terrible sand sea, had not some revolution in nature, for example the inflowing ocean deprived this level open country of its vegetation cover and nutritious topsoil. When exactly this phenomenon occurred and which force caused the (ocean’s) intrusion is hidden in the dark of the past.” Current research provides a somewhat different view of the dynamics of the AHP.

Shanahan et al. (2015) put together hydrologic reconstructions from across Africa. They showed that over much of tropical and subtropical Africa, the monsoon changed synchronously during the last deglaciation, that is, the period after the peak of the last glacial around 21 ky bp to the onset of the Holocene some 11.6 ky ago. Strong and large-scale internal climate system processes such as the melting of the large ice sheets and the reorganizations of the overturning circulation in the Atlantic Ocean forced a nearly synchronous onset of the AHP around 14.8 ky bp. Otto-Bliesner et al. (2014) suggested that both insolation-driven changes in the physical climate system and postglacial increased atmospheric greenhouse gas concentrations contributed to the humid climate in Africa north of the equator. Mc Gee et al. (2013) found a later start of the AHP around 11.8 ky bp, which is close to the end of the Younger Dryas cold climate (12.9 ky bp to 11.6 ky bpBP). A similarly fast onset of the AHP also occurred in the eastern part of the Sahara, presumably a bit later, around some 10.5 ky bp (e.g., Pachur & Hoelzmann, 2000; Kuper & Kröpelin, 2006).

Figure 10. Timing of the onset and the termination of the African Humid Period as found in proxies of the West African monsoon. The red curve (a) shows the insolation changes at 30°N during June–July–August. Sub-figures b depicts the deposition of mineral dust (given in terms of percentage of terrigenous material) in the Atlantic Ocean at the Mauretanian coast at 20.8°N, c-e show the isotopic changes in leaf wax from the Senegal River at 15.5°N, Lake Bosumtwi, 6.5°N, and from the Congo fan at 5.5°S, respectively. The red dashed arrow indicates a southward shift of the rainbelt. (Redrawn from Shanahan et al., 2015).

After the last deglaciation and throughout the Holocene, any strong and fast internal climate system changes, except for a brief cooling event around 8.2 ky bp, were absent. Hence, not surprisingly, the termination of the AHP appears to be spatially and temporally more complex than the onset. In marine sediments off the coast of the Western Sahara, the deposition of Saharan dust abruptly increased around 4.9 ky bp (McGee et al., 2013). Other indicators of humid conditions in Africa suggest that the timing of the AHP termination occurred progressively later at lower latitudes (Shanahan et al., 2015) (Figure 10).

Whether the abrupt increase in the deposition of Saharan dust into the Atlantic Ocean, in comparison with the subtle insolation change, is directly linked to an abrupt change in the African monsoon circulation or to a change in land surface is still under discussion. Climate system simulations, which include a direct coupling between atmospheric circulation and emission, transport and deposition of mineral dust, clearly show that the differences in amplitude between mid-Holocene and present-day dust deposition seen in the marine sediment records can be attributed to changes in the Saharan land surface, while differences in atmospheric and oceanic circulation apparently contributed only marginally to the difference in amplitude of dust deposition (Egerer et al., 2016).

Figure 11. Normalized time series of changes in insolation at 30°N (dashed line), pollen of tropical taxa in Lake Yoa (Kröpelin et al., 2008) (black line), dust flux into the North Atlantic Ocean (deMenocal et al., 2000) (dark red line, scale inverted), hydrogen isotopic composition in leaf waxes near the Horn of Africa (Tierney & deMenocal, 2013) (blue line), and in stable carbon isotope composition reflecting the proportion of grasses and trees/shrubs in the Nile watershed (Blanchet et al., 2014).

The pollen record of Lake Yoa in the Ounianga Kebir in the northeastern part of Chad—the only complete pollen record in the Sahara covering the last several millennia—indicates abrupt changes for some taxa and more gradual variations in other taxa (Kröpelin et al., 2008). For the 1000 years furthest in the past, that is, from 6 to 5 ky bp, large and fast variations were detected for influx rates of pollen and spores of tropical plant taxa (Figure 11, black line) and mountain shrub type like Erica arborea. These taxa eventually vanished around 4.5 yr bp. Other taxa, such as Poaceae, declined more gradually and never vanished. Analysis of hydrological proxies points at a gradual decrease in precipitation (Francus et al., 2013). Tierney and deMenocal (2013) provided proxy evidence for an abrupt transition out of the AHP in Northeast Africa around 4 ky bp (Figure 11, blue line). This change, however, might just have been a sudden break in the humid conditions, as their proxy data reveal a fast recovery to an earlier, more humid climate in the centuries following. Blanchet et al. (2014) found a gradual decline in the Nile River runoff, but a rapid shift in vegetation and in erosion between 8.7 and 6 ky bp (Figure 11, green line).

Figure 12. Simulated transient development of vegetation fraction and climate in the Sahara. (a) Simulated changes in vegetation fraction on average over the entire Sahara by Claussen and Gayler (1997). (b) Simulated changes in vegetation fraction in the Western Sahara/Sahel region (14°W to 3°E, 17°N to 28°N) by Renssen et al. (2003). (c) Simulated changes in grass cover in the Eastern Sahara (11°E to 34°E, 18°N to 23°N) by Liu et al. (2007). (Redrawn from Claussen, 2009).

Before the first evidence of an abrupt termination of the AHP became available (deMenocal et al., 2000), an abrupt decline of Saharan vegetation coverage was predicted to have occurred around 4.5 ky bp based on dynamical system theory (Brovkin et al., 1998) and climate modeling (Claussen et al., 1999) (Figure 12a). Subsequent climate simulations have revealed, like the proxy data did, more complex types of transitions. Renssen et al. (2003) found a more gradual decline but with an enhanced variability around 6 ky bp (Figure 12b). Similar results are seen in the simulations by Schurgers et al. (2006). In the simulations by Liu et al. (2006) (Figure 12c), East Saharan grass cover declined rapidly and nearly disappeared abruptly around 5 ky bp. In the simulation by Fischer and Jungclaus (2011), Saharan vegetation declined smoothly and gradually without enhanced variability at the end of the mid-Holocene.

A conceptual view on climate—vegetation interaction in the Sahara

To assess the differences in model simulations and to reconcile data and model results, a conceptual model of climate–vegetation interaction in semiarid regions is discussed in this section. A graphical representation of the conceptual model is outlined in Figure 13; the mathematical version has been formulated by Brovkin et al. (1998) and modified by Wang (2004), Liu et al. (2006), and Claussen et al. (2013). A discussion of the model in a broader context can be found in Scheffer et al. (2001).

The formulation of the conceptual model was motivated by the observation that multiple equilibrium solutions appear to exist in a comprehensive climate–vegetation model. For present-day climate, Claussen (1994, 1997) found two solutions: if the coupled atmosphere-biome model was initialized with a present-day vegetation pattern, then this pattern did not change significantly. If the model was initialized with forest or grass all over the world, the present-day vegetation pattern recovered, except for North Africa where a much greener Sahara arose, mainly in the western part. Similar results were obtained for glacial conditions (Kubatzki & Claussen, 1998). Differences in insolation between glacial and present-day climate were negligibly small—too small to cause any significant difference in the monsoon circulation. For mid-Holocene conditions, however, only one solution of the model, the green Western Sahara was obtained regardless of initial vegetation patterns (Claussen & Gayler, 1997).

The existence of multiple equilibrium states were confirmed in some climate system models. Zeng and Neelin (2000) and Wang and Eltahir (2000b) found multiple states in simulated present-day Sahelian rainfall, depending on the initial conditions which they argue are related to observed decadal variations in the Sahelian rainfall and aridity. Using a zonally symmetric model of vegetation dynamics and air flow over West Africa, Irizarry-Oritz et al. (2003) obtained a bi-stability of the atmosphere–vegetation system for the mid-Holocene period. Rachmayani et al. (2015) did not see any multiple equilibrium states in Sahel rainfall in their mid-Holocene climate simulations, but they did not rule out the possibility of multiple equilibrium states for the present-day climate.

Figure 13. Stability diagram showing three cases of possible response of a system to a change in forcing as a function of the feedback strength in the system. A detailed explanation is given in the text.

The appearance of multiple equilibrium states of a dynamical system as a function of varying forcing suggests the potential of abrupt transitions between states, if external forcing changes with time. This has been schematically depicted in the stability diagram, Figure 13. Generally, a nonlinear system can reveal different stability characteristics depending on the strength of feedbacks operating in the system. The system could rather gradually respond to a transient forcing (see Case 1 in Figure 13). If perturbed away from the equilibrium, the system would ultimately return to the equilibrium (indicated by green arrows). A system with stronger feedbacks may be rather inert over certain ranges of forcing, while responding more strongly to forcing, if the forcing approaches a critical level (Case 2; the critical forcing level is located in the region indicated by A). In a system with even stronger feedbacks, two or more equilibrium solutions may exist (Case 3). In such a case, abrupt shifts from one equilibrium state to the other equilibrium state may occur, if the forcing increases beyond a critical level B2 or decreases from strong forcing to weak forcing below a level B1. In between the critical levels B2 and B1, any perturbation strong enough to cross the so-called repellor (dotted curve) can trigger an abrupt shift between equilibrium states.

Regarding the African monsoon system, the system response can be identified with the monsoonal rainfall and the abundance of vegetation in the Sahara. The forcing is the theoretical change in the monsoon strength directly driven by variations in the Earth’s orbit without internal feedbacks. Case 1 might be representative of the climate—vegetation system with plant types which are rather insensitive, or resilient, to changes in precipitation and, hence, do not exert a strong feedback with climate. Case 3 might be valid for a system with plant types that are very sensitive to changes in precipitation and that, therefore, lead to a strong feedback with monsoon rainfall. Figure 14 shows two examples of solutions for a system with interacting vegetation and precipitation which correspond to Cases 1 and 3, respectively. The relative vegetation cover of the system with a resilient plant type (blue line in Figures 14, 15, 16) declines gradually once the threshold of precipitation for a complete vegetation coverage is crossed. The system with a sensitive plant type reveals an abrupt transition from a green to a desert state (red line). If the system is run backward in time (dashed red line), an abrupt transition from a desert to a green state is realized, but at a different time than the transition from green to desert. This hysteresis emerges because of the difference in equilibrium solutions between critical points B1 and B2 (Figure 13).

Figure 14. Simulated transient changes in vegetation coverage. Shown are changes in vegetation in a system in which a resilient plant type (blue curve) interacts with precipitation, sketched as Case 1 in the stability diagram of Figure 13, and in which a sensitive plant type (red curve) interacts with precipitation, depicted as Case 3 in Figure 13. The forcing varies linearly over the entire time period of 3000 years. The corresponding curves for precipitation changes follow the curves for vegetation coverage. Please note that t < 0 indicates past climate changes, that is, t = -6000 y stands for the year 6 ky bp. The dashed line depicts a hysteresis, which arises when the model is run backward in time from –3500y to –6500y. (Adapted from Claussen et al., 2013).

Limits of predictability of climate—vegetation dynamics in the Sahara

In nature, climate fluctuations occur at all time scales, and the climate system appears to be a stochastic dynamic system. Monotonous transitions as shown in Figure 14 are unlikely to exist. In the mathematical version of the conceptual model (Liu et al., 2006; Claussen et al., 2013), climate or weather noise can simply be represented by adding random fluctuations in the precipitation which mimics interannual monsoon variability. Two realizations of transitions in such a stochastic system, one with weak and one with strong feedbacks, are depicted in Figure 15. In the case of a resilient plant type, vegetation coverage and precipitation decline rather gradually with randomly varying fluctuations. In the case of a sensitive plant type, the transition occurs in the form of large swings between the green and the desert state until the system settles in the desert state where precipitation is too low to yield any vegetation. The occurrence of large fluctuations during the transition is sometimes called “flickering” or “Lorenz noise” (Wang, 2004).

Figure 15. Simulated transient change in vegetation coverage of the same climate—vegetation system with the same stability characteristics as the system depicted in Figure 14. Here, however, small random fluctuations of precipitation which mimic interannual monsoon rainfall variability are added to the system each year. The thin lines indicate annual values; the thick lines represent 100-year running means. (Adapted from Claussen et al., 2013).

As an obvious consequence of the stochastic nature of the climate system, there is neither a smooth transition in vegetation coverage nor a single monotonous jump from a green state to a desert state. Instead, any transition is accompanied by weaker or stronger fluctuations. Moreover, it is extremely unlikely that identical transitions occur if the numerical experiment is repeated over and over again. This has an important implication when comparing proxy data and model results: it is extremely unlikely that proxy data and model results perfectly match. An agreement between data and model results can be expected only in a statistical sense, if the data points lie within the large ensemble of model simulations.

Not only the prediction of the precise date of a termination is blurred by climate variability, but also the interpretation of the strength of biogeophysical feedbacks cannot uniquely be determined from one time series—be it proxy data or model results. Liu et al. (2006) showed that in the case of a weak feedback (i.e., in the case of a resilient plant type interacting with precipitation), an abrupt transition in vegetation cover can happen by chance. They have referred to this case as a “stable collapse.” Furthermore, Liu et al. (2006) argued that a “stable collapse” can be differentiated from an “unstable collapse”—that is, an abrupt transition due to a strong feedback—by comparing the transitions in precipitation. In an unstable collapse, abrupt transitions should be found in both vegetation cover and precipitation. In a stable collapse, an abrupt transition should be seen in vegetation cover only, while precipitation declines more or less gradually. This statement, however, can only be true in a probabilistic sense. Claussen et al. (2013) demonstrated that, just by chance, abrupt transitions in both vegetation and precipitation can also occur in the case of a resilient plant type interacting with precipitation. Hence, discrimination between the two mechanisms—unstable versus stable collapse or weak versus strong feedback—is not possible when analyzing only one proxy record.

The notion of a weak and a strong feedback that leads to a gradual decline, stable collapse, or unstable collapse, respectively, becomes even more ambiguous if different plant types interact simultaneously with climate. The feedback between each individual plant type (red curves in Figures 14, 15, and 16) and climate could be strong. But in combination, these types could give rise to a more gradual decline, with stability properties that are similar to a stable system. Figure 16 showsa realization of such a system with two plant types interacting with climate. Instead of an abrupt change of areal coverage of the sensitive plant type and a gradual change for the resilient plant type, a fast and synchronous decline with large fluctuations is seen in the areal coverage of both (interacting) plant types, while precipitation decreases rather gradually.

Figure 16. Same as Figure 15, but for the case in which the two plant types interact simultaneously with the climate—see text. (Adapted from Claussen et al., 2013).

The simulated vegetation decline depicted in Figure 16 resembles the decline of pollen taxa found in Lake Yoa (Kröpelin et al., 2008; see also Figure 11 for tropical taxa of Lake Yoa). This is a somewhat fortunate coincidence because the model was set up for conceptual visualization of the possible effects of plant diversity on the stability of a subtropical climate–vegetation system. The model was extended, however, by Groner et al. (2015) and adjusted to recapture the mosaic-like environment during the AHP as reconstructed by Hély et al. (2014). Also with the more realistic model, Groner et al. (2015) observed a stabilizing effect of high functional diversity on vegetation cover and precipitation.

Spatial heterogeneity and abrupt transitions

From the studies by Claussen et al. (2013) and Groner et al. (2015) it has to be concluded that proxy data from regions rich in plant diversity show a response that looks like a weak feedback strength, while data from regions poor in plant diversity might reveal a strong feedback with the possibility of an abrupt climate and ecosystem change. As diversity may change in time, the possibility of abrupt transitions may also change in time. Hence, the feedback strength of the climate–vegetation system is not a universal property of a certain region but depends on the vegetation composition. What could be the implications of the spatial heterogeneity of subtropical North Africa then? In the Lake Yoa record, some pollen taxa revealed a rather gradual decline (Kröpelin et al., 2008). Further analyses of hydrological proxies of the Lake Yoa record by Francus et al. 2013) indicate a gradual decline in precipitation, too. But even if the data of Lake Yoa were indicative of a gradual termination of the AHP, do these data contradict the dust records recovered from Atlantic marine sediments (deMenocal et al., 2000; McGee et al., 2013)?

Figure 17. Simulated difference between mid-Holocene and present-day vegetation cover (black hatched area; from Claussen & Gayler, 1997, and green hatched area, from Renssen et al., 2003). The blue shaded area indicates the area in the simulation of Liu et al. (2006) with simulated abrupt changes in vegetation. The stars indicate the position of records from Lake Yoa (eastern location) and from ODP site 658 C (western location). (Redrawn from Brovkin & Claussen, 2008).

The modeling studies by Claussen and Gayler (1997) and Renssen et al. (2003) simulate a strong biogeophysical feedback mainly for the western part of the Sahara. The results of the simulations by Liu et al. (2006), which suggest a weak climate–vegetation feedback, fit the data of Lake Yoa much better, also from a geographical perspective (Figure 17). Hence, Brovkin and Claussen (2008) concluded that the Lake Yoa data do not invalidate earlier modeling results on strong land–atmosphere coupling in the Western Sahara for which the Lake Yoa record is far less representative.

Bathiany et al. (2012) studied the importance of spatial and temporal variability of the system for the dynamic stability of the climate–vegetation interaction in North Africa in more detail. They found that in the same model, abrupt changes in vegetation and climate can happen at different locations at different times. In their simulations, the spatial structure of the multiple equilibria changes with time such that abrupt changes in vegetation occur earlier, around 8000 ybp, in the southern Sahara and later, around 4500 y bp, in the northern Sahel. This result is in line with the synopsis of data by Shanahan et al. (2015)—which might be a fortuitous coincidence because Bathiany et al. (2012) did not assess the robustness of their model results.

Perhaps more importantly, Bathiany et al. (2012) showed that because of spatial interactions, abrupt changes in one region can be induced by critical transitions in the neighboring region. Hence, the feedback between vegetation and atmosphere in a region under consideration might be too weak to induce an abrupt change emerging from a loss of stability of the system. Nonetheless an abrupt change can occur in this region triggered by strong vegetation–atmosphere interaction in a neighboring region. This “induced tipping of ecosystems” further complicates the interpretation of individual proxy records (Bathiany et al., 2013a, 2013b).

Dust feedback and potential atmospheric tipping points

So far, this discussion has focused on the possibility of abrupt transitions in Saharan climate and ecosystems triggered by atmosphere–vegetation interaction. However, abrupt transitions may also be caused by other feedbacks, including feedbacks between the atmosphere and the emission of mineral dust from the land surface. Alternatively, the African monsoon circulation itself may be prone to abrupt changes.

The section Orbital Forcing: Pacemaker of the African Humid Period mentioned that the difference in the amplitude of mid-Holocene and present-day dust deposition found in Atlantic marine records can be directly linked to the differences in surface cover between mid-Holocene and present-day Sahara. This finding does not necessarily imply, however, that the abruptness of the dust deposition seen in the marine sediments can be attributed to an abrupt expansion of the desert. Bhattachan et al. (2013) found a highly nonlinear response of dust emission to changes in vegetation coverage in field experiments in the Kalahari dune land. The emission of mineral dust increased strongly after the vegetation coverage shrank to a small value. This nonlinear behavior is caused by the disproportional response of dust emission to changes in soil moisture and wind speed. Soil moisture can change very fast in a slowly drying climate. Since the hydraulic conductivity of soils depends on the soil moisture itself, an initially steady decline in soil moisture results in an effective blocking of the vertical moisture transport in the soil. Once a critical threshold is crossed, the uppermost soil layers dry out abruptly (e.g., Hillel, 1982). The uptake of mineral dust from a sandy surface is proportional to the power of the wind—that is, proportional to the cube of the wind speed. Both processes lead to a highly nonlinear response of dust emissions to changes in atmospheric conditions.

Hence, an abrupt increase in dust emission and subsequent dust deposition might reflect an internal threshold behavior of soils in response to a steady reduction in monsoon rainfall and a steady expansion of the desert. However, it remains to be proven whether this concept based on local observations is valid on average over a large region. Because of the spatial heterogeneity of the landscape and the temporal variability of the African monsoon rainfall, an abrupt increase in dust emission could occur in some locations while in other locations no emission takes place. As a consequence, the areally averaged dust emission could respond to a change in surface conditions much more gradually.

Dust in the atmosphere affects the radiation budget, cloud formation, and precipitation. Therefore, changes in the Saharan land surface and the emission of mineral dust are likely to have amplified the difference between mid-Holocene and present-day African monsoon precipitation (see Limits of Predictability of Climate—Vegetation Dynamics in the Sahara). Hence, it is conceivable that the atmospheric dust concentration alters the transient dynamics of the land-surface–atmosphere interaction. Yu et al. (2015) used a conceptual mathematical model similar to the model by Brovkin et al. (1998) and by Liu et al. (2006) to demonstrate the existence of multiple equilibrium solutions of a system in which dust and land-surface conditions interact with precipitation. They considered the case of local (endogenous) dynamics within the Sahel, whereby enhanced dust emissions resulting from a decrease in vegetation partly suppressed precipitation, thereby further reducing vegetation cover. Yu et al. (2015) accounted for teleconnections between the Sahel precipitation and exogenous (i.e., Saharan) dust emissions due to an increase in the Saharan wind speed in years of above-average Sahel precipitation. In both cases, they detected two stable states of the system, one with low precipitation and high concentration of atmospheric dust and the other with high precipitation and lower levels of atmospheric dust. How strong this interaction between atmospheric dust, land surface, and precipitation is in relation to and in combination with other feedbacks remains to be reassessed using a comprehensive climate system model which includes real topography.

Other possible hypotheses for abrupt changes in North African rainfall are related to changes in the regional atmospheric dynamics. Analysis of present-day climate data (Nicholson, 2009, 2013) shows that the frequency distribution of Sahel precipitation, location of the AEJ, and strength of the TEJ reveals a bimodal structure. Wetter conditions co-occur with a northerly location of the AEJ and a stronger TEJ, and vice versa. Nicholson (2009) suggested that this bimodality points to the existence of dry and wet modes of the African monsoon circulation, that is, modes in which the system prefers to stay over several years before it jumps to the other mode. If the difference between the wet and the dry mode is large enough and if the modes persist long enough such that the ecosystems in the Sahara and Sahel adjust to them, then it is conceivable that this bimodality contributes to or even triggers abrupt climate and ecosystem changes in this region.

Skinner and Poulsen (2016) found that in their model simulations, the mid-Holocene rainfall season in the Western Sahara is prolonged to October, contributing up to 30 percent of the simulated annual precipitation during the mid-Holocene. This enhancement of rainfall is mainly related to two mechanisms: (1) the enhanced summer monsoon during June–September, providing sufficient wet conditions in North Africa that continue into the fall season, and (2) extratropical upper-level troughs that track northern Africa and transport moisture from the tropics into the Sahara in the form of concentrated water vapor plumes. Knippertz (2003) showed that a similar type of tropical–extratropical interaction also happens, albeit infrequently, in today’s climate, but the frequency of tropical plumes may have been increased during humid states of northern Africa. Skinner and Poulsen (2016) supposed that the influence of tropical plumes is particularly large in times of strongly enhanced solar insolation during the early boreal fall season (e.g., the orbital conditions of the mid-Holocene). The decreases in the summer monsoon strength and in the fall tropical plume frequency may amplify each other and feed back to the Saharan aridification at the end of the mid-Holocene. Hence, it is imaginable that a persistent latitudinal shift of either the extratropical westerlies or the tropical easterlies could rather suddenly change the strength of the tropical–extratropical interaction or even break up the coupling, leading to a rather rapid and persistent change in Western Saharan climate.


The African Humid Period and the Green Sahara are fascinating examples of nonlinear climate system dynamics. The AHP was triggered by changes in the orbital forcing with the climatic precession as the dominant pacemaker—an idea that was already brought up in the early 20th century. Climate system modeling in the 1990s showed that orbital forcing alone cannot explain the large changes in the North African summer monsoon and subsequent ecosystem changes in the Sahara. Feedbacks between atmosphere, land surface, and ocean were shown to strongly amplify monsoon and vegetation changes. Forcing and feedbacks have caused changes far larger in amplitude and extent than is experienced today in the Sahara and Sahel.

Most, if not all, climate system models, however, tend to underestimate the amplitude of past African monsoon changes and the extent of the land-surface changes in the Sahara. Hence, it seems plausible that some feedback processes are not properly described, or even missing, in the climate system models. Likewise, the jury is still out on whether changes in surface albedo or in evapotranspiration are the dominant processes and to what extent the emission of mineral dust from the Sahara surface is an important part in the chain of feedbacks between terrestrial and atmospheric processes. Also, the spatial scale matters. The tropical atmospheric circulation quickly responds to anomalies in radiative heating, wind, and moisture. Small-scale processes such as rain-intensive squall lines and mesoscale convective systems are tightly coupled to the large-scale flow like the AEJ and TEJ. Therefore, climate system modeling and model-based interpretation of proxy data are likely to improve only if these small-scale processes are directly simulated in the models.

Perhaps even more challenging than explaining the existence of the AHP and the Green Sahara is the interpretation of data that reveal an abrupt termination of the last AHP. Based on climate system modeling and theoretical considerations in the late 1990s, it was proposed that the AHP could have ended, and the Sahara could have expanded, within just a few centuries—that is, much faster than orbital forcing. In 2000, paleo records of terrestrial dust deposition off Mauritania seemingly corroborated the prediction of an abrupt termination. However, as more paleo data have been uncovered, considerable controversy has arisen over the geological evidence of abrupt climate and ecosystem changes. Some records clearly show abrupt changes in some climate and terrestrial parameters, whereas others do not. In addition, climate system modeling provides an ambiguous picture.

The understanding and simulation of abrupt climate and ecosystem changes at the end of the AHP is hampered by limitations implicit in the climate system. Because of the ubiquitous climate variability, it is extremely unlikely that individual paleo records and model simulations completely match. They could do so in a statistical sense—that is, if the statistics of a large ensemble of paleo data and of model simulations converge. Likewise, the interpretation regarding the strength of terrestrial feedback from individual records is elusive. Plant diversity, rarely captured in climate system models, can obliterate any abrupt shift between green and desert state. Thus, the strength of climate–vegetation feedback is probably not to be a universal property of a certain region, but depends on the vegetation composition which can change with time. Because of the spatial heterogeneity of the African landscape and the African monsoon circulation, abrupt changes can happen in several, but not all, regions at different times during the transition from the humid mid-Holocene climate to the present-day, more arid climate. Abrupt changes in one region can be induced by abrupt changes in other regions, sometimes referred to as “induced tipping.” In the end, it can be concluded that the African monsoon system is prone to fast and potentially abrupt changes, which to understand and to predict remains one of the grand challenges in African climate science.


The authors appreciate the constructive comments of the reviewer and editor, Sharon Nicholson, Florida State University. The authors also thank Sylvia Houston (MPI-M) for editing; Andrey Ganopolksi (PIK, Potsdam) who re-did the computations and provided Figure 6; Jong-yeon Park (Princeton University/GFDL) who provided the precipitation data for Figures 4 and 5; and Norbert Noreiks (MPI-M) who edited and re-drew Figures 2, 3, 9, and 10.