1 Introduction

Climate reconstructions of early- to mid-Holocene West African climate have shown a significantly different climate compared to present-day conditions, with paleo-lake-level reconstructions indicating more humid conditions and vegetation reconstructions and eolian and leaf vax reconstructions indicating vegetation covering parts of the Sahara region currently characterized by desert (Street-Perrott et al. 1989; Hoelzmann et al. 1998, 2001; Jolly et al. 1998; deMenocal et al. 2000; Gasse 2000; Prentice et al. 2000; Lézine et al. 2011; Hély et al. 2014). This period is often referred to as the African Humid Period (AHP; (Claussen et al. 2017)), and the later stage of the period, the mid-Holocene (6 ka), has received extensive interest from the modelling and proxy reconstruction communities alike (Bartlein et al. 2011; Brierley et al. 2020; Larrasoaña et al. 2013; Otto-Bliesner et al. 2017). However, in spite of the significant focus being put on the Mid-Holocene by the modelling community, General Circulation Models regularly struggle with recreating the strengthened West African Monsoon (WAM), the feature most associated with the more humid climate of the period. The results from several phases of the Paleo Modelling Intercomparison Project (PMIP) also show a clear underestimation of the Mid-Holocene rainfall enhancement across the Sahel and Sahara region compared to modern conditions (Brierley et al. 2020; Braconnot et al. 2012).

While the AHP has been shown to be an orbitally forced wet period in West Africa, driven by an increase in boreal summer insolation over the NH (Kutzbach and Liu 1997), modelling studies have shown that only orbital forcing is not sufficient to produce the northward shift of the rainfall indicated by proxy reconstructions (Braconnot et al. 2000; Joussaume et al. 1999). Instead, understanding what feedback processes and model features enhance and/or drive the strength and variability of the WAM and its representation in models is an important step in closing this model-proxy mismatch. Several modelling studies have been dedicated to investigating this over the last few decades, and have increased our understanding of the role ocean (Kutzbach and Liu 1997), land surface (Chandan and Peltier 2020; Kutzbach et al. 1996; Lu et al. 2018) and dust (Pausata et al. 2016; Thompson et al. 2019) feedbacks play in enhancing the rainfall over West Africa. Vegetation feedbacks received special attention after the devastating droughts that plagued the Sahel region in the 1970s (Charney et al. 1975; Charney 1975). It has been shown that a decrease in vegetation cover over West Africa suppresses rainfall in the region through albedo-vegetation feedback, while a greening strengthens the West African Monsoon (Messori et al. 2019). In addition to vegetation-albedo feedbacks, studies have demonstrated that feedback mechanisms related to the soil moisture, evapotranspiration and moisture recycling can play a crucial role in amplifying the monsoon response to external forcing, although there is no consensus on which feedback mechanisms is dominant (Patricola and Cook 2008; Rachmayani et al. 2015; Yu et al. 2017). In particular, the increase in surface latent heat fluxes and enhanced moisture availability can weaken the African Easterly Jet (AEJ). This, in turn, reduces the moisture transport away from the continent aloft and results in a strengthened monsoon rainfall over West Africa (Rachmayani et al. 2015; Patricola and Cook 2008). Similarly, past greening of the Sahara has been shown to reduce the dust fluxes by 70–80% (deMenocal et al. 2000; Egerer et al. 2016; McGee et al. 2013), which when included in modelling studies has been shown to further strengthen the WAM and shift the rainbelt northward (Pausata et al. 2016), though later studies have shown this dust-albedo feedback to be highly model dependent and possibly overestimated (Hopcroft and Valdes 2019). While most research on the impacts of a “green Sahara” has been focused on the Mid-Holocene, it can be equally important for deepening our understanding of future climatic change in West Africa (Pausata et al. 2020), where changes to the land cover could lead to a greener and less dusty West Africa (Mahowald and Luo 2003; Evan et al. 2016).

While existing studies applying approximate mid-Holocene land surface and dust boundary conditions have provided valuable insights into the strengthening of the WAM and enhance the rainfall closer to the levels seen in proxy reconstructions (Pausata et al. 2016; Chandan and Peltier 2020), they remain limited in scope. Specifically, these simulations utilize coarse, idealized parameters for land surface, vegetation cover and atmospheric dust, often based on a restricted set of proxy records predominantly from the west coast of Africa (Pausata et al. 2016; Tierney et al. 2017; Chandan and Peltier 2020). Furthermore, such studies tend to generalize the monsoon’s response to land-surface changes, without accounting for regional variations highlighted in literature (e.g., (Li et al. 2018). In contrast, paleoclimate reconstructions reveal a far more nuanced picture, characterized by both spatial and temporal complexities in the environmental conditions at the end of the AHP (Dallmeyer et al. 2020; deMenocal et al. 2000; Kröpelin et al. 2008; Shanahan et al. 2015). This complex nature highlights the necessity of accurately modelling environmental heterogeneity and dynamic vegetation responses, especially when simulating transient climate changes. The observed intricacies in the climate system, even if they represent a seemingly minor proportion of total variability, can have cascading effects that are crucial for understanding both past and future climate scenarios. Therefore, Earth System Models (ESMs) coupled with dynamic vegetation are particularly valuable for these studies.

To assess the suitability of using ESMs for such studies, it is imperative first to evaluate their ability in simulating the vegetation responses against well-documented, greener past climate states. Extensive research has been conducted on the AHP using both iteratively and fully coupled vegetation models, including studies that have reported significant enhancement of the monsoon rainfall and a notable northward expansion of the vegetation cover into Sahara (Braconnot et al. 1999; Claussen and Gayler 1997; Dallmeyer et al. 2020; Renssen et al. 2006) Nevertheless, it is worth noting that the most ESM simulations still underestimate both the strength of the monsoon and the extent of northward shifts in rainfall and vegetation, especially when compared to proxy reconstructions (Braconnot et al. 1999; Dallmeyer et al. 2020). The inadequacies in these models have been attributed to a range of factors: absent processes, poor feedback representation, and limited sub-grid scale processes. However, the low number of available simulations hampers a structured multi-model evaluation, accentuating the necessity for diverse model approaches to gain a comprehensive understanding of the intricate processes involved.

In this paper we contribute to the limited number of ESM-based simulations by employing EC-Earth3-veg, one of the models with highest resolutions amongst the PMIP4 models. Our objective is to investigate its ability to enhance the West African rainfall to mid-Holocene levels through including more realistic, dynamic vegetation feedbacks. In doing so, we aim to offer a more nuanced perspective on the potential and limitations of using dynamic vegetation in climate simulations, which has been highlighted as a crucial component in the existing literature.

2 Model description, experiment design, modelling process and reconstructions

2.1 Model description

The present study uses EC-Earth 3.3, a fully coupled Earth System Model developed by the European consortium of 27 research institutions across Europe. The atmospheric component is the Integrated Forecasting System (IFS) cycle 36r4 developed by the European Centre for Medium-range Weather Forecasts (ECMWF), which includes the revised Tiled ECMWF Scheme for Surface Exchanges over Land incorporating land surface hydrology (HTESSEL) land model. The convection parametrization scheme is based on (Tiedtke 1989) with enhancements of the convective closure implemented to improve the diurnal cycle (Bechtold et al. 2014). The detailed description can be found in the IFS documentation at https://www.ecmwf.int/en/publications/ifs-documentation. The ocean and sea-ice components consist of the Nucleus for the European Modelling of the Ocean (NEMO) (Madec NEMO team 2008; Madec et al. 2015) and the Louvain-la-Neuve Sea-ice model version 3 (LIM3) (Vancoppenolle et al. 2009). All the components are coupled through the OASIS coupler (Ocean, Atmosphere, Sea Ice, Soil) (Craig et al. 2017). The simulations either use prescribed or dynamic vegetation, with the dynamic vegetation simulated using the Lund-Potsdam-Jena General Ecosystem Simulator (LPJ-GUESS) dynamic vegetation and biogeochemistry model (Smith et al. 2014). The coupling to EC-Earth 3.3 allows LPJ-GUESS to receive daily meteorological input, such as surface air temperature, soil temperature, precipitation and net shortwave and longwave radiation from IFS/HTESSEL. The vegetation is simulated and classified into different plant functional types (PFT) (Smith et al. 2014), separated in high and low vegetation types (see Table 1). The dominant high and low vegetation types are calculated for each grid box at the end of each model year and passed on to IFS/HTESSEL where it is used in land-surface processes such as albedo, runoff and sensible and latent heat fluxes.

Table 1 Prescribed and simulated plant functional types (PFTs)
Full size table

EC-Earth 3.3, a key contributor to the sixth phase of the Climate Model Intercomparison Project (CMIP6), has been extensively used in climate research, providing insights into current and past climatic conditions (Döscher et al. 2022; Zhang et al. 2021a; Hazeleger et al. 2010, 2012). The EC-Earth 3.3 simulations for present-day and pre-industrial conditions have demonstrated its capability in capturing global precipitation patterns, the seasonal cycle and the spatial distributions of monsoon rainfall with a high degree of representation (Berntell et al. 2021; Döscher et al. 2022; Faye and Akinsanola 2022; Quenum et al. 2021). Nonetheless, the model exhibits a consistent dry bias in West Africa when compared to observational data (Döscher et al. 2022; Zhang et al. 2021b).

2.2 Experiment design

The analysis herein is based on three simulations; one pre-Industrial control simulation (PI), and two simulations of the Mid-Holocene with different vegetation set-ups (MHVEG and MHREF) (Table 2). MHVEG uses the fully coupled LPJ-GUESS for a dynamic vegetation, while PI and the mid-Holocene reference simulation MHREF use prescribed pre-industrial vegetation. All simulations have an atmospheric resolution of T255 (~ 80 km) and 91 vertical levels. The Pre-Industrial and Mid-Holocene simulation set-ups follow the PMIP4 experimental design described by (Kageyama et al. 2018) (Table 3), with the differences between the simulations being the orbital parameters and concentrations of trace gases in the atmosphere. The difference in orbital forcing between the pre-Industrial and Mid-Holocene simulations leads to seasonal and latitudinal re-distribution of insolation, creating a strengthened seasonal cycle during the Mid-Holocene with the Northern Hemisphere (NH) receiving approximately 5% increase in insolation during the boreal summer and decrease during the boreal winter (Berger 1978). Aerosols are set as PI climatologies for all simulations, and land-sea mask and topography are set as modern. The prescribed vegetation is the simulated 1850 CE vegetation created by an offline LPJ-GUESS simulation.

Table 2 Experimental set-up for EC-Earth 3.3 simulations. Spatial resolution of the atmosphere model indicated by grid cell extent (in degrees longitude x latitude) and number of vertical layers (L), treatment of vegetation in simulation, trend of global annual mean near-surface air temperature for the years used in the analysis
Full size table
Table 3 Orbital parameters and atmospheric trace gas concentrations used in EC-Earth 3.3 simulations
Full size table

2.3 Initial conditions, spin-up and production

The initial conditions for all simulations are taken from the end of an approx. 1000 years long pre-Industrial spin-up simulation using EC-Earth3-LR (the resolution for atmosphere is T159 and the resolution for ocean is the same). The PI simulation is run for 300 years, using the 100 last years for analysis, as the simulation reaches quasi-equilibrium after 200 years of spin-up. Both MHVEG and MHREF simulations run for approximately 400 years, with a 200 years of spin-up period and using the last 200 years of each simulation for analysis to account for a higher centennial variability. The equilibrium state of the simulations is determined by monitoring the trend, with the criteria of  <  ± 0.05 °C per century in global mean surface temperature (Kageyama et al. 2018) being fulfilled by all simulations (see Table 2). Following (Brierley et al. 2020) a calendar adjustment is not done, as it has been shown to have a limited impact on monsoon rainfall in Mid-Holocene simulations.

2.4 Mid-Holocene reconstruction data

The Mid-Holocene simulations are evaluated against continental mean annual temperature (MAT) and precipitation (MAP) reconstructions from (Bartlein et al. 2011), where the quantitative reconstructions are inferred from pollen and plant macrofossil proxy records and represent a mean 5.5–6.5 ka anomaly relative to the present- day (see (Bartlein et al. 2011)) for details). The simulated vegetation in the MHVEG simulation is qualitatively evaluated against reconstructed biomes based on the BIOME6000 dataset (Harrison 2017a), where a MHVEG grid box is considered vegetated if the vegetation cover reached 20% at some point during the 200 yr. production period.

3 Results

In this section, we present the changes in monsoon rainfall and dynamics resulting from implementing a dynamic vegetation. We then examine the changes in surface energy properties linked to convection before comparing the vegetation pattern produced in MHVEG to vegetation reconstructions produced by proxy records. The significance of the changes are evaluated using a two-tailed student t-test at the 95% confidence level.

3.1 Changes in West African monsoon rainfall and dynamics

The WAM is driven by the seasonally induced Sahara Heat Low, an extensive area of low surface pressure caused by surface warming across the Sahara region during the boreal summer (Thorncroft et al. 2011; Lavaysse et al. 2010, 2009). The seasonal evolution of the Sahara Heat Low is well represented in the model simulations, with the thermal low reaching its summer position in western Sahara during the July–September months (not shown). This drives the low-level westerly monsoonal flow, brings in moisture to the West African region from the Tropical Atlantic region and shifts the rainbelt northward into the Sahel region with a peak in rainfall occurring during the July–September (JAS) period (see Fig. 2 ab for MHREF JAS climatology). We therefore start by examining changes to the seasonal cycle of temperature and rainfall in the Sahara and Sahel regions respectively (Fig. 1). The changes to the orbital forcing between the PI and MH time periods result in an expected warming of near-surface air temperature in the Sahara of up to 2.5 °C during the Boreal summer (July–October) and a cooling during the year’s remaining months, which can be seen when comparing MHREF and MHVEG to PI in Fig. 1a. The MHVEG simulation exhibits a slight relative warming throughout the year compared to the MHREF simulation and a weakening of the cooling during the February to March season compared to PI (Fig. 1a). The MH summer rainfall in the Sahel is enhanced compared to PI in both MH simulations, with the MHVEG showing higher amounts of rainfall than MHREF in August with 174.7 mm/month and 159.3 mm/month respectively.

Fig. 1
figure 1

Seasonal cycle of (a) near-surface temperature in Sahara [unit: °C] and (b) rainfall in Sahel [unit: mm/month] for MHREF (blue), MHVEG (green) and PI (black). Seasonal cycle of temperature is shown relative to PI, which is therefore not shown. The Sahara region is defined as 10°W-20°E, 20–30°N, and the Sahel is defined as 20°W-30°E, 10–20°N, see areas outlined in Fig. 2 b

Full size image
Fig. 2
figure 2

July–September mean near-surface temperature, mean sea level pressure, rainfall and low-level horizontal wind at 850 hPa for all MH simulations; MHREF (a,b) climatologies, MHREF anomalies (c,d; MHREF—PI) and MHVEG anomalies (e,f; MHVEG—MHREF). Significant (95%) MHVEG anomalies are indicated by x (temperature and rainfall) and + (sea level pressure) stippling with significant anomalies for both temperature and sea level pressure indicated by a star, and only significant wind anomalies are shown

Full size image

Both MH simulations exhibit an enhancement of the JAS large-scale monsoon pattern compared to PI (MHREF – PI; Fig. 2cd), with a low-level warming and deepening of the low-pressure area in the Sahara region together with westerly 850 hPa horizontal wind anomalies from the Gulf of Guinea into the Sahel region and positive rainfall anomalies reaching from 8°N to 16°N and a decrease of rainfall over the tropical Atlantic (MHVEG—PI anomalies not shown). This pattern is further enhanced when comparing MHVEG to MHREF, where the Sahara region is warmed by up to an additional 0.4 °C and the low-pressure area is deepened by 0.2 – 0.3 hPa (Fig. 2e). The JAS rainfall in the Sahel is enhanced by 30–40 mm/month and the westerlies are strengthened by approx. 1 m/s (Fig. 2f). Following the commonly used threshold definition of the West African Monsoon in Sahel as having a zonally averaged rainfall exceeding 2 mm/day or 60 mm/month, and the northernmost extent of the monsoon as the northernmost latitude where the zonally averaged rainfall exceeds this threshold (Pausata et al. 2016; Zhang and Cook 2014), we can also see that the WAM has shifted northward from 13.0°N in PI to 15.8°N in MHREF and 16.5°N in MHVEG (Table 4). Additionally, the JAS rainfall in central Africa exhibits a checkerboard-like signal. This is a known issue in EC-Earth as well as other Earth System Models such as E3SMv2 (Hannah et al. 2022) with solutions being discussed within the modelling community.

Table 4 Northernmost extent of the monsoon and latitudinal location of the African Easterly Jet (AEJ) for the PI, MHREF and MHVEG simulations respectively, calculated for the peak monsoon month of August. Monsoon extent is calculated over the area 15°W-20°E, and jet stream calculated over a box 20°E-20°E and of an altitude of 500 hPa
Full size table

The thermally induced African Easterly Jet (AEJ) stretches across West Africa at an altitude of approximately 600–700 hPa, and develops in response to the meridional contrast in temperature and moisture between the Sahara region and the Gulf of Guinea (Cook 1999; Nicholson and Grist 2003a). The location of the WAM rainbelt is believed to be associated with a latitudinal band of convection, confined southward of the AEJ (Nicholson 2009, Nicholson 2013). The WAM reaches its peak as well as northernmost position in August, and in line with previous studies the following analysis will therefore be limited to August mean climatologies (Nicholson 2009). While the PI simulation shows an AEJ of 11.4 m/s located at 11.4°N and 700 hPa altitude, consistent with observed 8–12 m/s and 10–12°N under modern conditions (e.g., Nicholson and Grist 2003b), the MH simulations show a strengthening and northward shift of the AEJ located at 15.1°N and 15.4°N for the MHREF and MHVEG respectively and with its center at 500 hPa altitude (Fig. 3 and Table 4). The latitude of the AEJ is also consistent with the northernmost extent of the WAM (Table 4). The low-level African Westerly Jet (AWJ), mainly present during wet Sahel conditions, is influenced by the cross-equatorial pressure gradient (20°N to 20°S) and linked to rainfall in the Sahel region (Nicholson 2013). In the PI simulation, the AWJ exhibits low-level zonal winds of around 2–4 m/s at 5–10°N and an altitude of ~ 900 hPa (Fig. 3), in line with conditions in modern dry years (Grist and Nicholson 2001). The MH-simulations show a notable intensification and vertical extension of the AWJ, with maximum wind speed of 6–8 m/s and extending up to ~ 600 hPa altitude (Fig. 3). This intensification is further enhanced in the MHVEG, showing significant positive anomalies up to 500 hPa altitude (Fig. 3d). Similarly, the West African Westerly Jet (WAWJ, not shown) is in the PI simulation in line with present-day conditions (Pu and Cook 2010), showing a core speed of 4.7 m/s at 925 hPa altitude and 8°N. In the MH-simulations, the WAWJ is strengthened, with core wind speeds of 7 m/s in MHREF and experiencing further enhancement by 0.2–0.8 m/s in MHVEG.

Fig. 3
figure 3

August zonal wind climatology averaged between 10°W and 10°E [unit: m s−1] for (a) PI, (b) MHREF and (c) MHVEG, and zonal wind anomalies for (d) MHVEG – MHREF, with significant MHVEG anomalies indicated with stippling. Eastward direction (negative values) indicated in blue

Full size image

The latitudinal band of negative vertical wind (updraft), located between southward of the AEJ and linked to the location of the rainbelt, is markedly stronger in the MH simulations compared to PI, and expands northward following the northward shift of the AEJ (Fig. 4). The updraft is further enhanced in MHVEG compared to MHREF, with the most significant changes centered around 13–16°N (Fig. 4cd).

Fig. 4
figure 4

August vertical wind climatology averaged between 10°W and 10°E [unit: Pa s−1] for (a) MHREF and (c) MHVEG, and vertical wind anomalies for (b) MHREF – PI and (d) MHVEG – MHREF, with significant MHVEG anomalies indicated with stippling. Upward direction (negative values) indicated in blue, location of African Easterly Jet (AEJ) and Tropical Easterly Jet (TEJ) indicated with vertical line

Full size image

The majority of the increase in rainfall over Sahel comes from the convective rather than large-scale precipitation, and to further examine the moist convection over West Africa, we compute the Moist Static Energy (MSE) content of the atmosphere. The MSE is defined as the sum of geopotential, enthalpy and latent energy, and is a direct indicator of monsoonal rainfall as the conversion of enthalpy and latent energy into geopotential energy in the upper troposphere is a main signal of convection (Fontaine and Philippon 2000).

$${\text{MSE}} = gz + {C}_{p}T + Lq$$

where g is Earth’s gravitational acceleration, z is the geopotential height, Cp is the specific heat of dry air at constant pressure, T is the temperature, L is the latent heat of evaporation, and q is the specific humidity. Compared to PI, there is a clear increase of MSE content in the atmosphere in the Mid-Holocene simulations, with positive anomalies from approx. 11°N to 30°N (Fig. 5a, MHVEG – PI not shown). These positive anomalies are further strengthened in the MHVEG simulation compared to MHREF, with significant positive anomalies 13–30°N, reaching up into the atmosphere to an altitude of 200 hPa (Fig. 5b). However, while the MSE content has increased in MHVEG, the peak MSE area in the lower troposphere (at 1000 hPa) remains at a similar latitude and magnitude as for MHREF (342.1 kJ kg−1 and 341.8 kJ kg−1, respectively at 12–14°N) with the positive anomalies located to the north of this center (Fig. 5c). The complex topography in West Africa might impact this result, with the model output being interpolated from higher levels rather than simulated directly for grid boxes with a land surface located higher than this 1000 hPa level, but the latitude of the peak MSE at 1000 hPa level remains consistent with the results of additional pressure levels (925 hPa and 850 hPa, not shown).

Fig. 5
figure 5

August Moist Static Energy (MSE) in the atmosphere calculated as zonal means between 20°W and 30°E [unit: kJ kg−1]. (a) MHREF anomalies against PI and (b) MHVEG anomalies against MHREF. Significant anomalies indicated with stippling. (c) zonal mean MSE at 1000 hPa, averaged between 20°W and 30°E for MHREF, MHVEG and PI

Full size image

3.2 Changes in vegetation

Figure 6 shows the prescribed vegetation used for the PI and MHREF simulations and the dynamically simulated vegetation produced by LPJ-GUESS in MHVEG. The main differences between the prescribed and dynamic vegetation are seen in the low vegetation with an approx. 5.5° northward shift from 13.5°N to 19.0°N of the tall grass in the western and eastern Sahel region and a general increase of short grass reaching up to 18.5°N in central Sahel. For the high vegetation, there is a shift from deciduous to evergreen broadleaf vegetation and a northward expansion of the tree line moving from 8°N to 11°N combined with an increase of vegetation reaching up to 15.5°N in central Sahel. However, the vegetation in MHVEG is sensitive to climate variability, and with a requirement of a mean JAS vegetation cover above 20% the northward shift remains approximately 1–3° more equatorward. The mean bare soil fraction shows a similar pattern, with the vegetation cover expanding in central Sahel, reaching a 20% vegetation cover at around 15°N during the summer season (May–September) (Fig. 7). The PFTs in MHVEG are also compared to reconstructed Mid-Holocene biomes based on the BIOME6000 dataset (Harrison 2017b). The reconstructions show a clear dominance of grassland in the Sahara region, some presence of savanna in the Sahel and a combination of tropical and warm-temperate forests south of the Sahel and in central Africa. While some short grass is present in the southern Sahara in MHVEG, the JAS max vegetation cover remains below the threshold of 20% indicating an underrepresentation of this northward expansion of grassland in the simulation. South of the Sahara, there is a better agreement between the reconstructed warm-temperate forest biome and the shift towards a higher representation of evergreen broadleaf vegetation in central West Africa and Central Africa in MHVEG.

Fig. 6
figure 6

Prescribed a) low and c) high vegetation for MHREF and PI, and simulated, mean b) low and d) high vegetation for MHVEG. Low vegetation types are short grass (sh grass), tall grass (ta grass) and bogs and marshes (bog/marsh), high vegetation types are evergreen and deciduous needleleaf (ever needle, deci needle), evergreen and deciduous broadleaf (ever broad, deci broad) and mixed forest. The vegetation types shown represent the dominant high and low vegetation type for the 200-year simulation for each grid box, and soil is specified for grid cells with < 20% July–September max vegetation cover (indicated in white). Reconstructed biomes, shown in filled circles, based on the BIOME6000 dataset (Harrison 2017b)

Full size image
Fig. 7
figure 7

(a) Prescribed and (b) simulated mean May–September bare soil fraction for MHREF and MHVEG respectively

Full size image

4 Discussion

4.1 West African monsoon enhancement

The PI simulation captures a seasonal cycle over the Sahel with the majority of the annual rainfall (> 80%) falling in July–September (Fig. 1b), in agreement with observations (Nicholson 2009) and in line with previous PI simulations with other climate models (Chandan and Peltier 2020). MHREF also exhibits the enhanced summer rainfall expected from the increased NH summer insolation, with rainfall in August increasing by over 120% from 71.5 mm/month to 159.2 mm/month (PI and MHREF, respectively), similar to what has been seen in previous studies. The further enhancement of the monsoon rainfall seen in MHVEG (Fig. 1b and Fig. 2f) is in line with recent mid-Holocene simulations using state-of-the-art ESMs, but significantly lower than earlier simulations using an iterative coupling to a Biome model (Dallmeyer et al. 2020; De Noblet-Ducoudré et al. 2000; Braconnot et al. 1999). However, despite JAS rainfall in Sahel being 15% higher in MHVEG compared to MHREF, it is still well below what has been seen in studies using prescribed Mid-Holocene vegetation (Chandan and Peltier 2020; Pausata et al. 2016; Texier et al. 2000). The analysis reveals that the rainfall anomalies in MHVEG are centered around the 12°-15° N latitudinal band, which coincides with the northern boundary of the MHREF rainbelt (Fig. 2b). However, the monsoon’s spatial extent does not significantly shift northward compared to MHREF, with the difference being approximately the width of a single grid cell as outlined in Table 4. This marginal difference does not qualify as statistically significant. Additionally, the AEJ maintains its intensity and latitudinal position, consistent with findings from previous simulations with prescribed vegetation (Gaetani et al. 2017), and the difference in strength and extent of the AWJ between the PI and MH simulations is similar to what can be seen between modern dry and wet years, but exhibits only a weak further enhancement in MHVEG compared to MHREF (Fig. 3). However, linked to the statistically increased rainfall over West Africa in MHVEG is a statistically significant warming and deepening of the low-pressure area in the Sahara region and an enhancement of the low-level monsoonal flow which brings moisture to the WAM (Fig. 2ef). It indicates that, while the signal is weaker compared to previous studies with prescribed vegetation, the West African Monsoon was intensified through the inclusion of dynamic vegetation, consistent with our understanding of the different vegetation feedbacks (Charney et al. 1975; Patricola and Cook 2008; Rachmayani et al. 2015) and similar to what can be seen in previous simulations with dynamic vegetation (Dallmeyer et al. 2020; Levis et al. 2004; Braconnot et al. 1999).

At the same latitudinal band as the positive rainfall anomalies is a strengthened uplift (12°-15°N), located in the northern part of the convective belt between the AEJ and TEJ (Fig. 4cd). Changes to the convective rain, which is responsible for the vast majority of West African Monsoon rainfall, is examined by analyzing the Moist Static Energy content in the atmosphere. The MSE describes the conversion of enthalpy and latent energy near the surface into geopotential energy through convection, and the results show apparent positive anomalies through the atmosphere reaching from 12°N to 30°N, with the most significant anomalies centered at 15°-20°N. These anomalies, which are linked to increased sensible and latent heat flux at the surface, are located south of the anomalies in previous studies with prescribed vegetation (Gaetani et al. 2017) and markedly weaker (Fig. 5). While the MSE anomalies are statistically significant increased, they do not shift the low-level (1000 hPa level) MSE peak northward, something that has been suggested to favor the northward migration of the rainbelt (Gaetani et al. 2017). Additional monsoon dynamics, such as mid-atmospheric circulation and wind patterns over West Africa, has been investigated but no significant changes can be seen between MHREF and MHVEG and the analysis is therefore not included in the manuscript.

4.2 Vegetation feedback

Previous studies have shown that including prescribed land surface feedbacks, such as prescribed Mid-Holocene vegetation and soil, results in a warming of the Sahara region and a strengthening of the WAM compared to orbital-only simulations (Chandan and Peltier 2020). This relative warming is present all year with the largest warming occurs during late fall to mid spring, though some studies find that with an expansion of the monsoon rainfall a surface cooling instead occurs in Sahara as a result e.g., increased cloudiness (Patricola and Cook 2007). However, while our MHREF simulation exhibits a very similar seasonal temperature cycle anomaly relative to PI as e.g. (Chandan and Peltier 2020), the pronounced warming seen in simulations with prescribed vegetation is not present in MHVEG with a warming of 0.1–0.2 °C in February – April (Fig. 1a) compared to 1.0–1.5 °C seen in (Chandan and Peltier 2020) and a warming over Sahara in the monsoon season of up to 0.3 °C (Fig. 2) compared to 1–7 °C seen in simulations with prescribed vegetation (Pausata et al. 2016). This is noteworthy as spring warming is a critical factor in driving the SHL, which subsequently influences the WAM (Lavaysse et al. 2009; Biasutti et al. 2009).

Previous studies have identified the vegetation-albedo feedback as a key process in past enhancement and northward shift of the WAM, producing a year-round warming of the Sahara (Broström et al. 1998; Doherty et al. 2000; Vamborg et al. 2011). The lack of strong warming in the Sahara in our MHVEG simulation is indicative of a limited vegetation response, which in turn results in only minor changes in albedo. Unlike findings in the Chandan and Peltier (2020), where prescribed land-surface changes led to enhanced spring warming and, consequently, a more robust SHL and monsoon, out simulation failed to reproduce similar enhancements. This lack of warming and albedo change could be a main explanation for the lack of northward shift of WAM in MHVEG simulation when compared to previous studies employing prescribed vegetation. This weak response is further seen by computing the change in low-level atmospheric thickness (LLAT) as a representation of the strength of the SHL, calculated as the geopotential height difference between two pressure levels (700 and 925 hPa) following the method by (Lavaysse et al. 2009), where the positive anomalies, i.e. dilation, stretch across West Africa but is centered around 15°N and eastern Sahara (where surface temperature and sea level pressure anomalies are larger; Fig. 2e) and non-significant over the SHL summer position between 20°-30°N in western Sahara (Fig. 8). The sensitivity of the monsoon response to albedo changes in dynamic vegetation models has been discussed previously, highlighting the role of e.g., vegetation parameterization in driving the climate sensitivity to vegetation feedbacks (Doherty et al. 2000). Similarly, including changes to the below-canopy soil albedo as a result of e.g., increased soil organic matter has been shown significantly strengthen the vegetation-albedo feedback and monsoon rainfall (Vamborg et al. 2011; Levis et al. 2004), a process which is currently not included in EC-Earth3-veg.

Fig. 8
figure 8

Low-level atmospheric thickness (llat) MHVEG anomalies (MHVEG – MHREF). LLAT calculated as the the geopotential height differences between two levels (700 and 925 hPa) following (Lavaysse et al. 2009) [unit: m]. Hatching indicates significant anomalies

Full size image

In contrast, other studies have found the vegetation-albedo feedback to play no substantial role in driving a positive vegetation-rainfall feedback, instead pointing to soil moisture and surface latent heat fluxes as being the primary process through its impact on the AEJ (Patricola and Cook 2008; Rachmayani et al. 2015). Increased latent heat fluxes weakens the AEJ, decreasing the transport of moisture from the continent out over the Atlantic Ocean and resulting in more moisture available for rainfall over West Africa (Patricola and Cook 2008; Rachmayani et al. 2015). However, unlike previous studies of simulations with both prescribed and dynamic vegetation (Patricola and Cook 2008, 2007; Rachmayani et al. 2015; Gaetani et al. 2017; Jungandreas et al. 2023), no weakening of the AEJ is present in MHVEG (Table 4) which points to a lack of strong response to the vegetation feedback in EC-Earth3-veg.

The lack of strong warming in Sahara combined with the lack of dynamic feedback on the atmospheric circulation in MHVEG therefore explains the relatively weak enhancement of the monsoon rainfall and the lack of the WAM's northward shift, which can be seen when using a prescribed vegetation. This indicates that while vegetation has been shown to sustain an enhancement and northward shift of the WAM, the vegetation feedback is not on its own strong enough to enhance neither the monsoon nor the vegetation to mid-Holocene levels in our simulation. Indeed, when we compare the MHVEG anomalies (relative to the PI simulation; Fig. 9) to proxy records (Bartlein et al. 2011), we can see that the enhancement is underrepresented in temperature and rainfall. While MHVEG does capture the overall pattern (warming in Sahara and wetting across West Africa) and the agreement is closer than for MHREF, the signal is markedly weaker than the reconstruction, especially with missing the strong Sahara warming. This underrepresentation of the mid-Holocene wetting is again similar to what can be seen in previous simulations with other ESMs. Following the method for model-data comparison seen in (Dallmeyer et al. 2020) we can see that while the mean and median annual rainfall anomalies over West Africa, calculated for the grid points corresponding to the proxy records north of 10°N in (Bartlein et al. 2011), is significantly lower than the proxy reconstructions (median: 384 vs. 23 mm/day and mean: 350 vs. 108 mm/day for proxy reconstructions and MHVEG respectively), they are in line with simulations run using the ESMs MIROC, MPI.PMIP3 and HadGEM2. However, the change in method from the prescribed vegetation in PI to the dynamic vegetation in MHVEG might impact the anomalies along with the simulated orbital forcing and feedbacks. In addition, it is important to note that while our anomalies are calculated as MH-PI, the proxy reconstructions represent the difference between MH and modern climate, using CRU observational data when modern reconstructions are unavailable, which might contribute to differences in magnitude.

Fig. 9
figure 9

Mean annual near-surface temperature anomalies [unit: °C] and annual rainfall anomalies [unit: mm/year] for MHREF and MHVEG. Mid-Holocene anomalies (MH—PI) are shown together with proxy-inferred temperature and rainfall anomalies (6-0k) from (Bartlein et al. 2011)

Full size image

In light of the inadequacy of our simulations to fully capture the mid-Holocene WAM and corresponding vegetation cover, it is plausible to consider the influence of additional feedback mechanisms in enhancing these climate feature. For instance, changes in soil properties – such as soil texture and color, influenced by an increase in organic matter—have been empirically demonstrated to enhance the monsoon (Levis et al. 2004; Vamborg et al. 2011). Such additional feedbacks not only could strengthen the vegetation response but also have the potential to extend the vegetation cover northwards, thereby further strengthen the vegetation feedback and monsoon dynamics. This suggests that running simulations with prescribed vegetation in Sahara could indirectly include the impact of these known and unknown feedbacks in our simulations through their effect on the vegetation. The direct effects of such processes have been investigated in previous studies, where especially albedo-related feedbacks caused by changes in atmospheric dust concentrations and soil properties in Sahara have been shown to increase the monsoon rainfall and further shift the rainbelt and vegetation northward (Chandan and Peltier 2020; Pausata et al. 2016; Kutzbach et al. 1996; Levis et al. 2004). As these feedbacks occur as a result of the changes in vegetation, which creates a reduction of atmospheric dust and changes to the soil, coupling an interactive dust-aerosol model with the dynamic vegetation could strengthen the response of the monsoon and vegetation changes. Similarly, improving dynamic soil composition and albedo schemes could further enhance the WAM during the Mid-Holocene. These model improvements have been recommended for future ESM studies of past Green Sahara periods (Lu et al. 2018). However, current state-of-the-art global climate models with dynamic dust have been criticized for failing to reproduce fundamental aspects of dust emissions and transport, and instead introduce more uncertainty in model simulations (Kok et al. 2017; Evan et al. 2014; Zhao et al. 2022), which highlights the importance of choosing a model configuration most relevant to the research question in balance with running more complex, resource-heavy coupled simulations. (Chandan and Peltier 2020) also point out that while recent studies have argued for the important role of reduced dust in the Sahara when simulating the mid-Holocene climate (Pausata et al. 2016; Thompson et al. 2019), their study shows that agreement with proxies can be reached solely by including feedbacks from land surface changes, indicating that the pathways for reaching sufficient enhancement of rainfall might be model dependent. Similar results have been seen in simulations run with ESMs, where different models find different dominating feedback processes in driving monsoon variability (Levis et al. 2004; Rachmayani et al. 2015). Additionally, while the MHVEG does not exhibit the pronounced strengthening of the WAM that has been seen in simulations with prescribed vegetation, resulting in a dry bias and lack of vegetation similar to what has been seen in previous studies (Braconnot et al. 2019), other ESMs with dynamic vegetation have successfully been able to recreate a strengthened WAM in line with models using prescribed paleo vegetation (Stepanek et al. 2020; Berntell et al. 2021; Brierley et al. 2020; Dallmeyer et al. 2020), which again indicates a level of model dependency in the vegetation-rainfall-albedo feedback and highlights the need for additional multi-model studies to draw more robust conclusions about its strength. Several studies, including one that specifically evaluates LPJ, a different version of the vegetation model used in this study, have indicated that it tends to underestimate vegetation cover in low-rainfall areas such as Sahara. Notably, the model requires an annual rainfall amount more than double that observed to reach a 20% grass coverage (Hopcroft et al. 2017). If a similar model bias is inherent in our MHVEG simulation, it is likely that the vegetation-rainfall feedback would be significantly muted. Previous studies of the present-day climate using EC-Earth3 have also identified a dry bias in West Africa (Döscher et al. 2022), potentially suppressing the simulated vegetation response. This limitation could account for the observed lack of expansion of grassland seen in MHVEG, further emphasizing the necessity for multi-model evaluations to ascertain the reliability of these feedback mechanisms. Additionally, recent studies have suggested that climate models, which are tuned for modern-day climate conditions, might be unable to recreate the abrupt climate change of the past, and that substantial tuning and paleoclimate-conditioning might be necessary to overcome the underrepresentation of mid-Holocene rainfall (Hopcroft and Valdes 2021).

5 Conclusion

The Mid-Holocene simulation with dynamic vegetation (MHVEG) exhibits a significant increase in summer rainfall across West Africa compared to the reference simulation with prescribed, pre-Industrial vegetation cover (MHREF), in line with previous simulations with dynamic vegetation. In addition, the results show a weak enhancement of the West African Monsoon dynamics with a warming of Sahara, deepening of the Sahara Heat Low and strengthening of the low-level winds which bring moisture from the Atlantic to the West African continent. However, despite an enhancement of the uplift and an increase in moist static energy in the northern edge of the rainbelt, the increased rainfall and warming of the Sahara region are markedly lower than those in simulations using prescribed vegetation, and there is no apparent northward shift of the WAM. This observation suggests that the vegetation feedbacks, primarily driven by orbital forcing alone and excluding other feedback processes, are insufficient to trigger a notable shift in our model simulation. Inherent model limitations such as a dry bias and suppressed vegetation representation, may have contributed to this weak response.