Northern Hemisphere Land Monsoon Precipitation Increased by the Green Sahara During Middle Holocene

Changes in land cover and dust emission may significantly influence the Northern Hemisphere land monsoon precipitation (NHLMP), but observations are too short to fully evaluate their impacts. The “Green Sahara” during the mid‐Holocene (6,000 years BP) provides an opportunity to unravel these mechanisms. Here we show that during the mid‐Holocene, most of the NHLMP changes revealed by proxy data are reproduced by the Earth System model results when the Saharan vegetation cover and dust reduction are taken into consideration. The simulated NHLMP significantly increases by 33.10% under the effect of the Green Sahara. The North African monsoon precipitation increases most significantly. Additionally, the Saharan vegetation (dust reduction under vegetated Sahara) alone remotely intensifies the Asian (North American) monsoon precipitation through large‐scale atmospheric circulation changes. These findings imply that future variations in land cover and dust emissions may appreciably influence the NHLMP.


Introduction
Understanding the dynamics of Northern Hemisphere land monsoon (NHLM) and reliably projecting its future changes are vitally important for infrastructure planning, disaster mitigation, food security, and water resource management Wang et al., 2018). The changes in NHLM precipitation (NHLMP, defined in section 2.3) are driven by natural Stevenson et al., 2017;Sun et al., 2017;Sun et al., 2019) and anthropogenic forcings (Devaraju et al., 2015;Dong et al., 2019;Giannini & Kaplan, 2018;Lau et al., 2008;Vecchi et al., 2006) through the global sea surface temperature (SST) changes (Giannini et al., 2003) and are also affected by the low-frequency internal modes within the earth's climate system Wang et al., 2017). Observations show that NHLMP intensified over the past three decades (Wang et al., 2012) due to both internal variability and the effects of climate change. Current climate models project an increasing trend in NHLMP over the 21st century (Lee & Wang, 2014). The increased Sahelian precipitation in the coming decades (Biasutti, 2013;Monerie et al., 2016) probably leads to an extension of Sahelian vegetation (i.e., grassland, shrubland, and wetlands) and reduced natural dust emission. However, human-induced overgrazing, deforestation and mismanagement of cropland can induce desertification, which might slow down the greening of Sahel (Engelstaedter et al., 2006;Evan et al., 2016). The vegetation and dust feedbacks are important not only for past and present monsoons but also for future monsoon changes . However, the impact of vegetation and dust on global climate has not drawn enough attention, and the processes by which these feedbacks change the climate have yet to be elucidated.
During the early to middle Holocene (11,000 to 5,000 years BP), increased summer insolation strengthened the African monsoon system. The Sahara Desert became once covered to a great extent by a mixture of shrubland, grassland, variable trees, and wetlands (Hély et al., 2014;Holmes, 2008), and the dust emissions were much lower than today (deMenocal et al., 2000;McGee et al., 2013), which led to the so-called "Green Sahara (GS)" or African Humid Period. Proxy data show that precipitation had increased substantially over the Saharan region during the mid-Holocene (MH, 6,000 years BP; Shanahan et al., 2015;Bartlein et al., 2011). Nevertheless, the simulations of the mid-Holocene performed in the Paleoclimate Modeling Intercomparison Project (PMIP), in which the land cover and dust concentrations are similar to that in the preindustrial period, fail to reproduce both the magnitude and the northward expansion of precipitation in North African (NAF) monsoon (Harrison et al., 2014). This is likely due to the fact that these models did not include the important feedbacks associated with changes in vegetation cover and dust concentrations (Pausata et al., 2016;Tierney et al., 2017). Moreover, in the context of the NHLM regions, the PMIP simulations not only underestimate the strength and extent of the NAF (i.e., Braconnot et al., 2012;Pausata et al., 2016) but also the Indian summer monsoon, East Asian monsoon, and North American monsoon (NAM; Braconnot et al., 2012;Zhao & Harrison, 2012;Bird et al., 2014). Some recent studies have indicated that the Green Sahara and dust reduction (GSRD) can have remote influences on the Arctic climate Muschitiello et al., 2015), the El Niño-Southern Oscillation  and tropical cyclones (Pausata, Emanuel, et al., 2017). In the present work, we use a series of sensitivity experiments with a fully coupled ocean-atmosphere model in which Saharan vegetation, dust concentration, and orbital forcing (ORB) are changed in turn in order to further investigate the role of vegetation and dust on NHLMP.

Model Description and Experimental Design
The model used in this study is the version 3.1 of the climate model EC-Earth (Hazeleger et al., 2010). The atmospheric model is based on the Integrated Forecast System (cycle 36r4), including the H-TESSEL land model. The oceanic model is version 2 of the Nucleus for European Modeling of the Ocean (Madec, 2008), with a horizontal resolution of~1°and 46 vertical levels. The model is also coupled with the Louvain-la-Neuve Sea Ice Model version 3 (Vancoppenolle et al., 2008). The coupling component is performed by the OASIS 3 coupler (Valcke, 2006). The preindustrial (PI) experiment is performed at T159 horizontal spectral resolution (1.125°× 1.125°, approximately 125 km) with 62 vertical levels, which is higher than the resolution in other PMIP models (Table S1 in the supporting information). The historical CMIP5 run from 1979 to 2008 based on EC-Earth (Present) is also conducted to represent the present-day climatology.
Based on the preindustrial condition, the first idealized sensitivity experiment (Green Sahara during preindustrial, PI GS ) is carried out (Table 1), which imposes the prescribed shrub vegetation type over the Saharan domain (11-33°N, 15°W-35°E). In our model, the surface albedo is decreased from 0.3 (for desert) to 0.15 (for evergreen shrub), and the leaf area index is increased from 0.2 (for desert) to 2.6 (for evergreen shrub). Pausata et al. (2016) tested the impact of replacing the evergreen shrub with grassland (albedo = 0.25) over eastern North Africa, showing no large impact on the strength of the western African monsoon. The test of NAF precipitation to these values in this study is also conducted in the supporting information ( Figure S3), and the results are similar with Bonfils et al. (2001). The standard MH orbital forcing simulation (MH ORB ) is performed following the PMIP3 protocol (Braconnot et al., 2011), where the orbital value is set at 6,000 years BP. For the greenhouse gases (GHGs), methane is set at 760 ppb in PI and 650 ppb in MH ORB , and there is no change in CO 2 and other greenhouse gases. The third experiment (MH GS ) imposes the prescribed shrub vegetation type over the Saharan domain, which is similar to the PI GS experiment but under the MH orbital condition. We also perform the MH vegetated Sahara and reduced dust (MH GSRD ) experiment, where the Sahara land cover is also set to shrub but the dust concentrations are reduced by almost 80% in the troposphere over a broad area around the Sahara desert ( Figure S1 in Pausata et al., 2016), according to the 60%-80% dust flux reduction from the proxy evidence (deMenocal et al., 2000;McGee et al., 2013). This imposed dust reduction results in a decrease in the local dust aerosol optical depth of approximately 60% and a decrease in the global total aerosol optical depth of 0.02. The initial conditions for each experiment are taken from a 700-year PI spin-up run, and the simulations are then run for 300-400 years. The quasiequilibrium is reached after 100-200 years, depending on the experiment. This research focuses on the last 100 years of each experiment.
For simplicity, we use ORB and GS to represent the net effect of orbital/GHGs forcing (MH ORB -PI) and vegetation change (MH GS -MH ORB ) during the MH, respectively (Table 1). The effect named GS PI denotes the net effect of vegetation change under the PI condition (PI GS -PI). GSRD is used to represent the combined effect of the vegetation change and dust reduction (MH GSRD -MH ORB ) under the MH condition.

Observation and Proxy Data
The data set of the Global Precipitation Climatology Project version 2.3 (Adler et al., 2003), which provides global (land and ocean) coverage for the period of 1979-2017, was used to verify the model performance of the climatological pattern. The results of EC-Earth 3.1 are consistent with the observations in terms of land monsoon precipitation climatology compared to all of the PMIP3 models ( Figure S1).
We also collected the precipitation proxy data to validate the simulated precipitation changes during the MH (about 6 ka BP; Table S2). The choice of proxy data needs to meet several criteria and these data are compiled from the published literature. First, the proxy data must reflect the precipitation or moisture conditions (precipitation minus evaporation). The records only reflecting temperature are not included. Second, the temporal resolution of proxies must be sufficient to reflect century-to-millennial scale climate changes. The MH proxy data are compared to the present day (0 ka BP).

Definitions of NHLM Area and Precipitation
Following the definition of global monsoon used by Wang and Ding (2008) and Liu et al. (2009), the NHLM area is defined by the land regions where the local summer mean minus winter mean precipitation exceeds 2 mm/day and the local summer precipitation exceeds 55% of the annual precipitation. Here summer is May-September, and winter is November-March. The NHLMP change is measured by the sum of summer precipitation anomalies in the NHLM area computed by each experiment (Hsu et al., 2012), which can better distinguish each forcing's effect on summer monsoon precipitation.

Geophysical Research Letters
Due to the lack of proxy records that cover the entire monsoon domain (Figure 1), we use the weighted-area average annual mean precipitation over the same NHLM area to compare the proxy data and model results in Figure 1b. The NHLM area in Figure 1 is derived from the MH GSRD experiment because it can better capture the expansion of NAF revealed by proxies during the MH (Pausata et al., 2016).

Changes in the NHLMP
During the MH, most of the PMIP models simulate enhanced (reduced) Northern (Southern) Hemisphere monsoon precipitation due to the increased (decreased) summer insolation over the Northern (Southern) Hemisphere (Jiang et al., 2015;Zhao & Harrison, 2012). The MH ORB experiment shows the very similar NHLM area and precipitation changes compared with the multimodel ensemble mean results in the PMIP3 ( Figure S2).
The model results from the MH ORB and MH GSRD experiments are first compared with proxy data (Table S2) to check whether the simulated annual mean precipitation over the NHLM is improved under the imposed Saharan vegetation and dust reduction during the MH (Figure 1). When only the orbital/GHGs forcing (MH ORB -Present) is considered during the MH, the NHLM annual mean precipitation change relative to the present-day is considerably underestimated (Figure 1b), which is similar to the ensemble mean results in PMIP3 (Braconnot et al., 2012).
When the Saharan vegetation and dust reduction (MH GSRD -Present) are also considered, the annual mean precipitation change over most of the NHLM regions shows a closer agreement with proxy records (Figure 1b). The model overestimates the changes in NAF annual mean precipitation, compared to the proxy data. However, the simulated precipitation change shows a good agreement with proxy data between 15 and 30°N over the NAF, compared with that in the MH ORB experiment ( Figure S3a). Precipitation is mainly overestimated (3 mm/day) between 10 and 15°N, but recent proxy data suggest this increase could be plausible (Hély et al., 2014). The Asian monsoon (ASIA) precipitation is also enhanced, yielding a better agreement with the reconstructions (Figure 1b). Most of the proxy data indicate wetter conditions over the North American monsoon (NAM) for the Green Sahara period relative to current conditions (Figure 1a and Table S2), but the increased amplitude is smaller than that in the NAF and ASIA (Figure 1b). This increased annual mean precipitation over the NAM revealed by proxy data is still not reproduced under MH GSRD -Present. In the areas outside the NHLM domain, the MH GSRD experiment shows the increased precipitation over Europe and central Australia and the decreased precipitation over the central-eastern North America  Table S2. (b) Box-and-whisker plot of annual mean precipitation anomaly over NHLM regions from the pollen reconstructions (black), EC-Earth MH ORB experiment (blue) and MH GSRD experiment (red), relative to the present-day. The box whisker plots show the 10th, 25th, 50th, 75th, and 90th intervals, and the crosses denote the weighted regional mean precipitation change. (c and d) Annual mean precipitation anomalies (mm/day) in the MH ORB and MH GSRD experiments, respectively, relative to the present-day. Black lines in a, and blue lines in c and d represent the land monsoon regions defined by the MH GSRD experiment. The dots denote areas in which the changes are significant at the 95% confidence level using a two-tailed Student's t test.
and South America (Figure 1d), which is more consistent with the proxy data than that in the MH ORB experiment.
To further quantify the NHLM changes, the NHLM area and NHLMP are analyzed. In GSRD, the NHLM area and precipitation are enhanced by 28.0% and 33.1%, respectively (Figures 2d and 2e), while under ORB, they are only enhanced by 15.5% and 19.4%, respectively; the increases are more than 1.7 times as large as that caused by the orbital/GHGs forcing. This is in better agreement with the proxy data (Braconnot et al., 2012; Figure 1b). Among the monsoon subregions, the northward expanded and enhanced North African monsoon contributes most to the NHLM area and precipitation changes under GSRD (Figures 2c-2e).
Interestingly, in addition to its impact on local precipitation, GSRD also significantly enhances the precipitation over the NHLM (no NAF) by 7.5% (Figure 2e), especially for the Asian monsoon precipitation (8.0%); the increase in precipitation is significant and almost equal to that under ORB. Additionally, the North American monsoon precipitation is increased by 5.2% under GSRD, but this change is not significant due to large uncertainties (Figure 2e).
GS alone increases the NHLM area and precipitation by 19.9% and 26.9%, respectively, which means that the Saharan vegetation plays a greater role in strengthening the NHLM, compared with the dust reduction. The North African monsoon precipitation contributes most to the NHLM precipitation, followed by the significantly enhanced Asian monsoon precipitation (8.3%; Figure 2e). Suppressed precipitation is found over North America, but its amplitude is weak. We also use the GS PI (Table 1) to isolate the individual effect of Saharan vegetation. The result shows that the NHLM area and precipitation are enhanced by 20.4% and 29.2%, respectively, which is almost equal to the effect of vegetated Saharan in the MH (Figures S4c and  S4d). The distribution of precipitation anomalies over the monsoon subregions under GS PI is also very similar to that under GS (Figures S4a and S4b). This means that the Saharan vegetation control the precipitation difference between MH GS and MH ORB , instead of the insolation or the nonlinear changes between vegetation and insolation. We also check the net effect of dust reduction under vegetated Sahara (GSRD-GS).
The results show that it can further substantially enhance the North African monsoon precipitation by 19.9% and significantly increase the North American monsoon precipitation by 9.1% (larger than one standard deviation).
Therefore, the Saharan vegetation and dust reduction not only strengthen the North African monsoon during the MH but also remotely enhance the NHLM (no NAF), and the influence of the vegetated Sahara plays a stronger role here, compared with the dust reduction.

Mechanism of the Green Sahara's Influence on NHLMP
Some previous studies consider that NHLMP changes are caused by the orbital-induced large-scale meridional temperature gradient and the land-ocean thermal contrast during the MH (Jiang et al., 2015;Zhao & Harrison, 2012). In this study, the summer anomalous zonal mean meridional temperature gradient, land-sea thermal contrast, and land-sea level pressure gradient are weaker over the NH (no NAF) under GSRD, compared with that under ORB ( Figure S5). However, a significant increase in the NHLM (no NAF) precipitation is found under GSRD (Figure 2d). Thus, it can be inferred that the vegetated Sahara and dust reduction affect the NHLM (no NAF) mainly through other mechanisms.
The surface albedo is reduced over the vegetated Sahara, leading to a warming in the months preceding the monsoon and favoring a strong convection after that (Pausata et al., 2016). A significant tropical North Atlantic SST warming enhances the north-south thermal gradients (Figure 3f), strengthening the southwesterly anomalies, further enhancing the Sahelian precipitation (i.e., Kamae et al., 2017;Monerie et al., 2019). The surface cooling occurs between 10 and 23°N ( Figure S6), which is caused by the latent heat release and the increased cloud cover reflecting solar radiation (Pausata et al., 2016;Ramanathan et al., 1989). An albedo-induced warming over the northern Sahara develops throughout the summer, enhancing the northward expansion of the North African monsoon (Figure 3f). The substantial increased monsoon precipitation leads to a release of latent heat, warming the middle and upper troposphere ( Figure S6). This increases the atmospheric thickness and the upper-level geopotential height, inducing an anomalous anticyclone in the upper troposphere (Figure 3d). Then a noticeable baroclinic structure is exhibited in the entrance of the westerly jet ( Figure S7). However, ORB causes the weaker middle and upper troposphere warming over the North African region ( Figure S6), which induces a much weaker anomalous anticyclone in the upper troposphere (Figure 3c), compared with that under GS. GS also induces an intensification and westward extension of the Walker Circulation over the Pacific Ocean ( Figure S8) through changes in equatorial Atlantic SSTs, which is explained by  The changes in the Walker Circulation enhance the low-level southeasterly anomalies over the northern Indo-Pacific Ocean (Figure 3f), which enhances the South Asian monsoon (i.e., Ning et al., 2017;Wang et al., 2015). This intensified Indian summer monsoon can excite the anomalous upper-level west-central Asian high (Ding & Wang, 2005). Subsequently, two baroclinic structures (with the stronger one in North Africa and the weaker one in west-central Asia) are formed ( Figure S7), generating a Rossby wave train. This wave energy propagates downstream to regions along the waveguide, which induces the barotropic structure over the regions of East Asia, the North Pacific and North America, which resembles the circumglobal teleconnection (CGT) pattern (Ding & Wang, 2005). Nevertheless, in the case of ORB, these two baroclinic structures are much weaker in North Africa and west central Asia ( Figure S7).
Under GS, a barotropic structure located over Japan induces an anomalous low-level divergence center (Figures 3d, 3f, and S9). Anomalous southerlies over the west of this divergence center enhance the northward transport of water vapor to northern China, causing the increased precipitation ( Figure S9b). Anomalous easterlies over the North Pacific carry more moisture into Southern Asia, increasing precipitation there. This intensifies the Asian monsoon precipitation. Additionally, GS induces the anomalous surface warming over the west-central Asia and Northwest Pacific ( Figure S6b). This is because the waveinduced anticyclonic anomalies suppress the cloud cover and increase the incoming solar radiation. These two warming centers are conducive to enhance the northward moisture transport to the South and East Asia. However, the suppressed precipitation is found along the East Asian subtropical front due to the local descending motion. Thus, the Saharan vegetation indirectly enhanced the Asian summer monsoon through the upper-level Rossby wave train and a westward extension of the Walker Circulation.
In North America, an anomalous upper-level anticyclone covers most of the midlatitude region and induces the low-level divergent winds under GS (Figures 3d and 3f). Moreover, the strong heating over North Africa excites the Gill-type Rossby wave pattern (Gill, 1980), which induces the descending motion over the equatorial Atlantic Ocean and tropical South America, suppressing precipitation there. It also causes the descending motion over central-east North America (Figure 3b), which is located to the west of the North African heat source. These two descending motions may contribute to the decreased precipitation over the Western Hemisphere, which is observed in the proxy data ( Figure 1a) and causes the easterly anomalies over the eastern equatorial Pacific (Figure 3f), slightly weakening the North American monsoon.
To validate that this mechanism is caused only by the effect of Saharan vegetation and is not a direct response to insolation or a combined effect of vegetation change and orbital forcing, we verify the results under GS PI . The result shows that GS PI also induces the upper-level wave train ( Figure S10). A barotropic structure located near Japan causes a very similar atmospheric circulation pattern over the Asian monsoon region, compared with the results under GS. Descending motions also occur over tropical South America and central-east North America, suppressing North American monsoon precipitation. This additional experiment highlights the important role of the Saharan vegetation in changing the NHLMP.
Previous studies have shown that the dust reduction strengthens the vegetation feedback on radiative forcing, which enhances the North African monsoon (Gaetani et al., 2017;Pausata et al., 2016), but the mechanism of its impact on the North American land monsoon is unclear. The reduced dust concentration happens over the area of roughly 100°W-60°E, 10°S-40°N ( Figure S1 in Pausata et al., 2016), which increases the downward solar radiation and warms the tropical North Atlantic SST ( Figure S11). This causes the east-west temperature gradient from the eastern tropical Pacific to the western Atlantic Ocean, inducing the anomalous westerlies over the northeastern tropical Pacific, enhancing the North American monsoon precipitation. At the same time, there is also a dust reduction of about 40%-50% over the North American region (Pausata et al., 2016), which strengthens the local moist convection.

Discussion and Conclusions
Previous model studies showed an obviously underestimated NHLMP changes compared with the reconstructions in the MH (Braconnot et al., 2012;Jiang et al., 2015;Zhao & Harrison, 2012). They focused on the insolation changes and ignored Saharan vegetation and dust concentrations. Here we show that the vegetated Sahara and dust reduction can modulate the atmospheric circulation and affect the NHLMP. It should be noted that only the direct effect of dust reduction is considered in this model version, while the indirect aerosol effect (nucleation that results in the formation of more rain droplets) is not included. This may affect the results shown in this study as suggested by a recent study focusing on the indirect effect on the West African monsoon (Thompson et al., 2019).
In summary, our results show that the simulated annual mean precipitation change is significantly improved over most of the NHLM regions during the MH compared with the reconstructions when the vegetated Sahara and dust reduction are also taken into consideration. These forcings increase the NHLMP by 33.1%, which is more than 1.7 times the impact of the orbital/GHGs forcing. Among the monsoon subregions, the strengthened North African monsoon precipitation contributes most significantly, which is mainly caused by the increased moisture convergence under the effects of vegetation and dust reduction. The Saharan vegetation alone also leads to the increased Asian monsoon precipitation by 8.0% through the upper-level wave train and a westward extension of the Walker Circulation, while dust reduction under vegetated Sahara enhances the North American monsoon by 9.1% through the anomalous westerlies induced by the tropical North Atlantic warming. These results indicate the strong impact of the Green Sahara on the NHLMP during the MH. They also suggest this factor may have a significant influence on the NHLMP in the future, which is critical for the demands of infrastructure planning, disaster mitigation, agriculture, and water resource management.