The climate over both South and Southeast Asia (SA and SEA) is characterized by rainy summers and cold dry winters. During summer, warm and moist air originating from the Indian Ocean is transported northeast towards the Indian subcontinent and the Indochina peninsula where the air mass organizes into the South and Southeast Asian monsoon (SAM and SEAM, respectively) circulations. Apart from the external forcing of the land–sea heating contrast that is essential for initiating SAM and SEAM (Webster et al., 1998), internal feedback processes, which have been recognized as changing over the recent decades (Mishra et al., 2018; Endo et al., 2009), play a crucial role in amplifying their strength. Since monsoon systems are characterized by a strong temporal variability (Ding, 2007; Kutzbach, 1987; Prell and Kutzbach, 1987), the limited modern observational datasets are insufficient to constrain the expected future response of these systems to climate change. Projections of monsoon evolution from climate modeling of future conditions remain highly uncertain (Huang et al., 2020). Therefore, analysis of past changes in monsoon circulation systems under significantly different forcing conditions for both SAM and SEAM is potentially of great value in contributing to the understanding of these monsoons and their expected future changes.

The mid-Holocene (MH; 6 ka) is characteristic of a typical interglacial period during which the insolation distribution was significantly different from that of the present day (Fig. 1). During the MH, differences in the Earth's orbital configuration compared to the present day altered Northern Hemisphere insolation, which in the latitudinal range of these monsoons induces a top-of-the-atmosphere insolation increase of ∼20 Wm-2 in summer (Fig. 1; Laskar et al., 2004; Berger, 1978). Significant changes in the strength of the Asian monsoon during the MH have been revealed by various paleoclimatic reconstructions, such as those based on paleoceanographic evidence (Hutson and Prell, 1980; Prell, 1984a, b; Cullen and Prell, 1984; Prell and Van Campo, 1986), Tibetan ice cores (Thompson et al., 2000), Chinese Loess Plateau deposits (An, 2000; Porter, 2001), and stalagmites (Wang et al., 2001; Dykoski et al., 2005). During this period, the Indian monsoon penetrated deeper into the Indian subcontinent (Wang et al., 2010; Herzschuh, 2006; Kutzbach and Street-Perrott, 1985), although detailed knowledge is still limited by the absence of high-resolution proxy information. Furthermore, owing to a lack of paleoclimate proxies from SEA, little is known about the incursion and variability of the SEAM.

To better understand orbital-scale monsoon changes, various numerical experiments have been performed, primarily using global climate models (GCMs) to simulate the large-scale atmospheric circulation and the related monsoon circulation (Jalihal et al., 2019; Jiang et al., 2013; Dallmeyer et al., 2013). The earliest modeling studies of the MH monsoons focused only on the direct effect of insolation forcing (Kutzbach and Otto-Bliesner, 1982; Kutzbach and Guetter, 1986; Kutzbach and Gallimore, 1988). These studies generally produced strengthened monsoons and increased precipitation in Asia in response to the increased land–sea surface temperature contrast in summer caused by the enhanced seasonal cycle of MH insolation. Although the early GCM simulations were based on prescribed sea surface temperatures (SSTs) and land vegetation, recent studies have shown that land and ocean feedbacks can enhance the orbitally forced monsoon changes (Zhao and Harrison, 2012; Kutzbach et al., 2001). In addition, the strengths of SAM and SEAM circulations and their onset, maintenance, and withdrawal are greatly influenced by the contiguous Tibetan Plateau (TP; Fig. 1), which serves an elevated heat source for the atmosphere in summer that intensifies the thermal contrast between the continent and ocean in the region influenced by the Asian monsoons (Wu et al., 2007). However, due to the coarse horizontal resolutions of the global models, the TP, as well as the local mountains over SA and SEA, is poorly represented in GCMs. As a result, GCMs are incapable of realistically capturing local-scale atmospheric circulation and precipitation processes, which are strongly influenced by the orography. In order to respond to the need for high-quality climate information on regional scales while maintaining the computational tractability of the problem, this study employs the same dynamical downscaling pipeline described in Huo and Peltier (2021) to dynamically downscale MH global simulations.

During the MH, northern Africa (Fig. 1) was considerably wetter than today and was covered to a great extent by a mixture of shrubland, grassland, trees, and wetlands – a period referred to as the Green Sahara (GS; Claussen et al., 2017; Pausata et al., 2020; Holmes and Hoelzmann, 2017; Chandan and Peltier, 2020) or the African Humid Period. While it is well established that the Sahara existed in a state very different from that of today during the MH (Hély et al., 2014; Harrison and Bartlein, 2012) and had far-reaching local and global climatic influences (Chandan and Peltier, 2020; Pausata et al., 2017a, b), most studies investigating the Asian monsoon variability during the MH have employed the same global vegetation distribution as that characteristic of the preindustrial (PI) period and have been demonstrated to significantly underestimate the strength and extent of the SAM and SEAM (Zhao and Harrison, 2012; Braconnot et al., 2012). Since other key factors, including remote changes in vegetation cover, have been shown to modify the climate response to the insolation forcing (Piao et al., 2020; Sun et al., 2019), we have taken care in the current study to incorporate GS boundary conditions (Chandan and Peltier, 2020) in the analyses to be described herein in order to compare with the standard MH simulations and to capture interactions between the African Humid Period monsoon and that in India and SEA. Our intention is to construct a high-resolution representation of the teleconnection between these two regions. Pausata et al. (2017b) and Sun et al. (2019) pointed out that the presence of GS conditions in northern Africa shifts the Walker circulation westward through changes in equatorial Atlantic SSTs and warms the Indian Ocean, which enhances the SAM and SEAM. Griffiths et al. (2020) have also suggested that the transition out of the African Humid Period led to an amplification of middle to late Holocene megadroughts in mainland Southeast Asia via ocean–atmospheric teleconnections. Moreover, some recent studies have projected future increases in Sahelian precipitation in addition to a surface greening in the Sahel (Giannini and Kaplan, 2019; Evan et al., 2016). Hence, understanding the influence of African vegetation on the SAM and SEAM may also help us better appreciate the changes in the Asian monsoon system to be expected from ongoing climate change associated with human activity.

This paper presents an analysis of the MH monsoon using high-resolution coupled regional climate model (RCM) simulations. The key questions are the following. (1) How does SAM and SEAM precipitation respond to the MH orbital forcing on regional scales? (2) Do simulations produced by the coupled modeling system agree both qualitatively and quantitatively with proxy data during the monsoon season? (3) What are the impacts of the GS boundary conditions on SAM and SEAM? The coupled model system and the simulation configurations are described in Sect. 2. A synthesis of major monsoon changes is provided in Sect. 3, which also includes a discussion of the sensitivity of the SAM and SEAM to MH Sahara vegetation changes. In Sect. 4 we offer a summary of our conclusions.

The dynamical downscaling methodology employed for the analyses to follow (Fig. 2) is the same as that in Huo and Peltier (2021), which further developed the dynamical downscaling “pipeline” originally introduced in Gula and Peltier (2012) and then applied in d'Orgeville et al. (2014), Erler and Peltier (2016), and Peltier et al. (2018) for the purpose of investigating the expected precipitation changes over the Great Lakes Basin of North America and over western Canada, upstream and downstream of the Rocky Mountains topographic barrier. In the present version of this pipeline, a regional climate model – the Weather Research and Forecasting (WRF) Model – is coupled to a regional ocean model – the 3D Coastal and Regional Ocean Community model (CROCO) – and both are forced with the output from a GCM – the University of Toronto version of the NCAR Community Climate System Model version 4 (UofT-CCSM4) – at the lateral boundaries (Fig. 2). This pipeline has been applied in the projection of modern global warming impacts for SEAM, and results are verified to be in good agreement with the general monsoon patterns compared to instrumental era observations (Huo and Peltier, 2021). Similar simulations have also been carried out for the Indian monsoon region for present-day and future climate projections by Huo and Peltier (2019, 2020) with a somewhat simplified pipeline that did not include the CROCO regional ocean model and utilized an older version of WRF (V3.9.1).

The regional atmospheric model component of the pipeline is the WRF model version 4.1 (Skamarock and Klemp, 2008), with the same choice of major physics parameterizations as in Huo and Peltier (2021) except for the cloud microphysics parameterization. Unlike Huo and Peltier (2021), who employed two different microphysics schemes, this study uses a single microphysics parameterization scheme (the single-moment six-class scheme; Hong and Lim, 2006) for all experiments because the choice of microphysics scheme was found to have a much smaller impact on the simulated monsoon precipitation compared to the cumulus scheme in the modern-day simulations of both SAM (Huo and Peltier, 2020) and SEAM (Huo and Peltier, 2021). The same four cumulus parameterization schemes as in Huo and Peltier (2021), namely the Tiedtke scheme (Tiedtke, 1989), the Grell–Freitas ensemble scheme (GF; Grell and Freitas, 2014), the Betts–Miller–Janjic scheme (BMJ; Janjic, 1994), and the Kain–Fritsch scheme (KF; Kain, 2004), are employed in this study in order to construct a mini-physics ensemble for the SAM and the SEAM, which enables us to study the sensitivity of model performance to different cumulus parameterizations and thereby to estimate the uncertainty associated with these parameterizations in the simulated MH climate. In the radiation module of WRF, we have applied trace gases and orbital parameters based on the Paleoclimate Modeling Intercomparison Project Phase 4 (PMIP4; Otto-Bliesner et al., 2017) protocol for the MH simulations. The WRF domain covers most of the Asian landmass (shaded in Fig. 3) at 30 km resolution. Note here that most PMIP4 models have a resolution of approximately 1∘ (Otto-Bliesner et al., 2017), which is close to that of our GCM (the UofT-CCSM4 model) and is considerably coarser than the resolution of our regional model.

The 3D CROCO ocean model has also been coupled to the WRF using the OASIS 3 coupler (Valcke, 2013) through which hourly prognostic fields are exchanged (Fig. 2). This study employs the same ocean domain as in Huo and Peltier (2021), which covers most of the oceanic area in the WRF domain including the Bay of Bengal (BOB), the eastern part of the Arabian Sea, and the western part of the South China Sea (the white rectangular area in Fig. 3). It operates at a horizontal resolution of 0.2∘ (approximately 20 km) with 32 vertical levels.

Both WRF and CROCO are forced at their lateral boundaries by the atmospheric and oceanic fields obtained from simulations with a global model. The global simulations have been generated using a fully coupled atmosphere–ocean–land–sea–ice GCM: the University of Toronto version of the National Center for Atmospheric Research (NCAR) Community Climate System Model version 4 (UofT-CCSM4; Peltier and Vettoretti, 2014; Chandan and Peltier, 2017, 2018, 2020). It is based on the standard CCSM4 model (Gent et al., 2011) with modifications of the ocean component for paleoclimate simulations. The UofT-CCSM4 simulations were produced in its coupled configuration with a horizontal resolution of 1∘. For the analyses reported herein, WRF–CROCO was driven by three distinct GCM simulations, including one for the PI climate and two for the MH climate, the latter being forced by the PMIP4-mandated concentrations of greenhouse gases (GHGs) and orbital parameters (Otto-Bliesner et al., 2017). One of the MH experiments employs a preindustrial land surface and a vegetation prescribed to PI values, and it therefore differs from the PI only with regards to the orbital configuration and trace gas concentrations; the other MH experiment also incorporates GS boundary conditions by including vegetation, soil, and lake modifications over Africa and Arabia. Both of these simulations were recently described in detail in Chandan and Peltier (2020), where they were referred to as MHREF and MHVSL, respectively. Each UofT-CCSM4 experiment was integrated for several hundred years, and the last 15 years of output of each were downscaled using WRF–CROCO for the purpose of the analyses to be reported below. The two sets of MH experiments, both including results for each member of the four-member mini-ensemble, can be compared to investigate the role of the Saharan landscape in altering the SAM and SEAM. We will also compare the simulated changes in precipitation in the two sets of experiments to pollen-based climate reconstructions (Bartlein et al., 2011; Herzschuh et al., 2019) to help us evaluate the degree to which the incorporation of GS boundary conditions will lead to a diminution of the previously obtained misfits of global model-based climate reconstructions of SAM and SEAM rainfall. In the following analysis, the set of MH experiments (both global and regional) including no specific land surface changes over North Africa is referred to as the reference MH simulation and denoted as MHREF, while the other set of MH simulations, which incorporates GS boundary conditions, is referred to as MHGS.

We begin by discussing the sea surface temperature (SST) and surface air temperature anomalies simulated for MHREF before turning our attention to precipitation changes realized in the same simulation. Following this, we discuss the impact of including the full GS boundary conditions on the simulated MH SAM and SEAM. All spatially averaged anomalies reported here are calculated over the land surface of the Indian subcontinent or mainland SEA south of the TP (the two black rectangles in Fig. 3). These two analysis regions are identical to the inner WRF domains in Huo and Peltier (2020, 2021) wherein two levels of downscaling were employed.

In this study, we have reported on paleoclimate experiments with dynamical downscaling simulations that we have conducted using the coupled WRF–CROCO model driven by the output from the global coupled UofT-CCSM4 GCM to investigate the MH Asian regional climate. We have focused attention on the summer monsoon circulations of this region, reconstructing these at much higher resolution than is possible with a global model by employing a dynamical downscaling procedure. By employing a downscaling pipeline in which the reginal climate model WRF is coupled to the regional ocean model CROCO we have been able to not only capture the effects of the complex surface orography and land–sea contrast of the region but also achieve a much higher-fidelity data–model comparison than has been achieved previously. Furthermore, and in contrast to the common protocol for climate simulations of the MH Asian monsoon system (Otto-Bliesner et al., 2017), which only includes the impact of solar insolation changes and ignores the potential for teleconnection effects on the Asian monsoons, our study also incorporates a set of downscaled simulations that include Green Sahara (GS) boundary conditions through which we have been able to assess the remote response of the MH SAM and SEAM systems to the greening of the Sahara. These effects have been shown to be of first order in importance. The magnitude of the impacts is significantly enhanced by the high spatial resolution delivered by the downscaling pipeline.

Compared to the preindustrial, all physics ensemble members downscaled from the reference MH climate, which does not include a GS, consistently exhibit a strengthened summer monsoon intensity during the MH over both SA and SEA. The increase in summer rainfall is particularly strong over northern India and northern Indochina, while monsoon precipitation decreases over the southern part of SEA and the western BOB. MH monsoon intensity is weaker during early summer, particularly over SEA, which is likely linked to the lower SSTs in the eastern Arabian Sea and the BOB. A late monsoon onset and a delay in the monsoon withdrawal over both SA and SEA are also found in our analysis. All of these changes are stronger in the downscaled simulations than in the GCM.

Comparison with paleoclimatic reconstructions demonstrates that over northern Indochina and along the Himalayas, the simulated annual rainfall anomalies agree well with the reconstructions from a qualitative perspective. However, over the eastern TP, the model simulations are characterized by a drier climate, while wetter conditions are indicated by proxy data for the MH. Overall, the WRF results are in better agreement with proxy inferences, especially in the northern part of SEA and southwestern China, than the predictions of the UofT-CCSM4 global model, which is much coarser in resolution and does not resolve the regional orography accurately. Given that climate models using coarse spatial resolutions generally reproduce the direction and large-scale patterns of the MH monsoon system changes but tend to underestimate the regional response (Hargreaves et al., 2013; Zhao and Harrison, 2012; Braconnot et al., 2012), this study clearly demonstrates the added value of using dynamical downscaling in paleoclimate studies of the Asian monsoon system. However, since proxy data are scarce both geographically and temporally, particularly over SEA and the southern part of SA, more reconstruction work based on various proxies and methods will be required to draw more robust conclusions.

Comparison between simulations with and without GS boundary conditions demonstrates that greening of the Sahara amplifies the response of SAM and SEAM and leads to a further increase in MH monsoon precipitation, which is likely related to increased heating over the interior of the Eurasian continent. Over SA under GS conditions, the water vapor flux from the Arabian Sea to the northern Indian subcontinent is enhanced compared to the case when only the orbital forcing is considered. Over the southern part of SEA, although a Green Sahara weakens the low-level westerly moisture flux, it also reduces the moisture divergence. With a GS, the JJAS rainfall increases are more than twice as large as those caused by the orbital and GHG forcings in all WRF–CROCO physics ensemble members over SA. The precipitation increase in response to the GS forcing is associated with a drop in surface temperature over SA and SEAM (Fig. 7), which reduces the sensible heat flux. On the other hand, the substantially increased SAM and SEAM precipitation leads to a release of latent heat, which warms the middle and upper troposphere and adds to the temperature difference between land and ocean (Fig. 7), thus driving stronger winds and moisture advection, which in turn leads to enhanced precipitation. The duration of the monsoon season is also lengthened by several months over both regions by extending its withdrawal phase. Comparison of downscaled results including a GS with paleoclimate reconstructions points to significant improvements in simulating precipitation, especially over the TP and North China. These results highlight the climate sensitivity of Saharan vegetation changes via ocean–atmosphere teleconnections and emphasize the importance of incorporating MH vegetation feedbacks.

Finally, the precipitation anomalies reported in this study are subject to uncertainties associated with the parts of the GCM simulations used to drive WRF–CROCO. Since the Asian monsoon precipitation is strongly influenced by the El Niño–Southern Oscillation (ENSO), further simulations forced by different parts of the global simulations are necessary to characterize the internal variability and confirm the robustness of the current study. In addition, in this study we have only considered the impact of land surface changes over northern Africa on SAM and SEAM using one coupled regional modeling system. However, some recent studies also argue for an impact on the MH Asian monsoon arising from reduced dust over northern Africa (Pausata et al., 2020; Sun et al., 2019). Additionally, proxy data from the MH indicate widespread vegetation changes everywhere, particularly greater vegetation coverage over Eurasia (Tarasov et al., 1998) and South and East Asia (Zhang et al., 2014). It has already been shown that the higher Eurasian forest cover during the MH can shift the intertropical convergence zone northward and has a remote impact on global climate (Swann et al., 2014). These results call for further investigations using both diverse numerical models and more realistic vegetation distributions as well as dust concentrations to enable further improvements in reconstructions of the MH Asian monsoon systems.