Mid-Pliocene West African Monsoon rainfall as simulated in the PlioMIP2 ensemble

The mid-Pliocene warm period (mPWP; ∼ 3.2 million years ago) is seen as the most recent time period characterized by a warm climate state, with similar to modern geography and ∼ 400 ppmv atmospheric CO2 concentration, and is therefore often considered an interesting analogue for near-future climate projections. Paleoenvironmental reconstructions indicate higher surface temperatures, decreasing tropical deserts, and a more humid climate in West Africa characterized by a strengthened West African Monsoon (WAM). Using model results from the second phase of the Pliocene Modelling Intercomparison Project (PlioMIP2) ensemble, we analyse changes of the WAM rainfall during the mPWP by comparing them with the control simulations for the pre-industrial period. The ensemble shows a robust increase in the summer rainfall over West Africa and the Sahara region, with an average increase of 2.5 mm/d, contrasted by a rainfall decrease over the equatorial Atlantic. An anomalous warming of the Sahara and deepening of the Saharan Heat Low, seen in >90 % of the models, leads to a strengthening of the WAM and an increased monsoonal flow into the continent. A similar warming of the Sahara is seen in future projections using both phase 3 and 5 of the Coupled Model Intercomparison Project (CMIP3 and CMIP5). Though previous studies of future projections indicate a west–east drying–wetting contrast over the Sahel, PlioMIP2 simulations indicate a uniform rainfall increase in that region in warm climates characterized by increasing greenhouse gas forcing. We note that this effect will further depend on the long-term response of the vegetation to the CO2 forcing.


Introduction 45
The mid-Pliocene Warm Period (mPWP; 3.264-3.025 Ma; also known as the mid-Piacenzian Warm Period) is considered to be the most recent historical warm climate state, with average global temperatures several degrees above pre-industrial (PI) levels (1.4 -4.7 °C; Haywood et al., 2020) and atmospheric CO2 concentrations of ~400 ppmv (Badger et al., 2013;Bartoli et al., 2011;Dowsett et al., 2010;Haywood et al., 2020Haywood et al., , 2013Martínez-Botí et al., 2015;Pagani et al., 2010;Raymo et al., 1996;Seki et al., 2010;Tripati et al., 2009;50 Zhang et al., 2013). Paleoenvironmental reconstructions indicate a warm and humid climate during the mPWP, with elevated sea surface temperatures (SSTs) and surface air temperatures (SATs), especially at high latitudes (Dowsett et al., 2010;, forests and grassland expanding into areas previously covered by tundra, and savanna and woodland expanding at the expense of deserts (Salzmann et al., 2008). While much of the research on the mPWP climate focused on global large-scale patterns and the high latitudes (Haywood et al., 2013(Haywood et al., , 55 2020De Nooijer et al., 2020), several studies have emphasized the implications of the warm climate state for tropical climate, showing e.g. an enhancement of the East Asian Summer Monsoon (Wan et al., 2010) and a drying of the Southern Hemisphere tropics and subtropics (Pontes et al., 2020). Analysis of e.g. dust records of the coast of West Africa also indicates a strengthened West African Monsoon (WAM) during the mPWP as well as wetter conditions over West Africa and the Sahara region (Kuechler et al., 2018;Salzmann et al., 2008). 60 With a paleogeography and atmospheric CO2 concentrations similar to today (Dowsett et al., 2010), the mPWP has long been considered an interesting analogue for near-future climate projections (Chandler et al., 1994;Jiang et al., 2005) and been the focus of many modelling studies (e.g. Haywood and Valdes, 2004;Salzmann et al., 2008). To increase our understanding of the dynamical drivers of the warm climate state, several model simulations have been performed as part of the Pliocene Modelling Intercomparison Project (PlioMIP; Haywood et al., 2010Haywood et al., , 2011. While previous studies have shown that the high-latitude warming has reduced the equator-pole temperature gradient (Haywood et al., 2013) and weakened tropical circulation such as the Hadley Circulation (Corvec and Fletcher, 2017), the terrestrial warming during the mPWP has been shown to strengthen the WAM and increase 75 the summer rainfall over the Sahel region by more than 1 mm/day (Haywood et al., 2020;Zhang et al., 2016). A similar rainfall increase over Sahel is seen in future projections for both CMIP3 and CMIP5 ensembles, though with a drying located over western Sahel (Roehrig et al., 2013). However, models have been shown to inaccurately capture rainfall variability and change over West Africa and the Sahel region (Berntell et al., 2018;Roehrig et al., 2013), and there is still little confidence in future projections of the summer rainfall (Biasutti et al., 2008;Cook, 80 2008;Roehrig et al., 2013). West Africa is a region sensitive to hydrological variability which experienced extended droughts during the 1970s and 1980s (Berntell et al., 2018;Held et al., 2005;Nicholson et al., 2000), and there is a large need to increase the confidence in future projections in order to support adaption strategies in the region.
The similarity to modern conditions, as well as the high amount of paleogeological and environmental data from 85 the mPWP, has made it well suited to both evaluate the models' ability to capture a warm climate state and further our understanding of the effects of greenhouse gas forcing on the global climate system (Haywood et al., 2020;Haywood and Valdes, 2004). In this article we will evaluate the representation of the WAM within the PlioMIP2 ensemble, qualitatively compare it to palaeohydrological reconstructions and discuss the implications for the WAM in a near-future warm climate state with increasing greenhouse gas forcing. 90 2. Data and method 2.1. Participating PlioMIP2 models To examine the behavior of the WAM during the mPWP, data produced by 17 different general circulation models as part of the PlioMIP2 was used (Table 1). Simulations produced within PlioMIP2 are run for at least 500 years (Haywood et al., 2016) towards an equilibrium state, and the last 100 years of the simulations are then used for 95 analysis. In the experimental set-up the CO2 levels are set to 400 ppmv, and the remaining concentrations of trace gases and aerosols are set to pre-industrial levels (Haywood et al., 2016). The simulations are run using enhanced boundary conditions as described in Haywood et al., (2016), with changes to e.g. the topography, bathymetry and land ice cover. COSMOS uses dynamic vegetation , while the remaining 16 models use  Salzmann et al. (2008). As the models have different horizontal resolutions, the 100 data from the models was bilinearly interpolated onto a 1º x 1º grid using the software CDO (Climate Data Operators, Schulzweida, 2019) to facilitate multi-model analysis.  The progression of the WAM creates different seasonal cycles of rainfall depending on the region, where northern latitudes in West Africa have one clear peak while more southern regions have a wider or bimodal rainy season.
We have therefore divided West Africa into two sub-regions, Sahel (10-20° N, 20° W-30° E) and Coast of Guinea 125 (5-10° N, 20° W-30° E). The seasonal cycle of terrestrial rainfall is calculated for each ensemble member and presented together with the MMM for the PI and mPWP simulations separately, as well as for the Pliocene anomaly (mPWP-PI) (Fig. 1). The "modern" seasonal cycle is plotted together with the PI cycle for reference, based on 1901-1930 CRU TS v4 data (Harris et al., 2020). Pre-Industrial (mm/day) In agreement with PI observations, the PI MMM shows a seasonal cycle with an August peak in rainfall over Sahel 130 at 3.1 mm/day. The individual models mainly exhibit the same seasonal cycle; however, four models exhibit highest levels of rainfall shifted to July (HadCM3) or September (CCSM4-Utrecht, NorESM-L and NorESM-F) rather than August. The magnitude of summer rainfall seen in CESM1.2 and MIROC4m is at 5.1 and 5.2 mm/day respectively comparable to modern conditions ( Fig. 1), while the other 15 ensemble members remain within a span of 2-4 mm/day which is considerably below modern levels. The mPWP MMM shows an increase in monsoon rainfall, 135 with the maximum rainfall doubling and reaching 6.1 mm/day in August. All models show an increase in rainfall in the July-October period, with the largest increase occurring either in August, September or October, resulting in a lengthening of the WAM. Keeping with previous studies (e.g. Berntell et al., 2018;Giannini et al., 2003;Mohino et al., 2011;Roehrig et al., 2013), we will however still base our spatial analysis of the WAM on the July-September period. The largest increase is shown in EC-Earth3-LR at 7.3 and 7.5 mm/day in August and September, making it 140 reach a maximum of 9.2 mm/day in Pliocene Sahel. As with the PI, the highest level of Pliocene rainfall in the PlioMIP2 ensemble is seen in MIROC4m with 11.5 mm/day in August.
Over the Coast of Guinea, the PI simulations show higher levels of rainfall through most of the Northern Hemispheres spring, summer and fall, with the ensemble mean showing a maximum of 8.1 mm/day occurring in August ( Fig. 1). 16 of the 17 members have maximum levels of rainfall spanning between 5.9 mm/day and 9.8 145 mm/day, while MIROC4m again supersedes the remaining models with rainfall reaching 11.9 mm/day in July. The MMM of the mPWP simulations again shows an increase of monsoon rainfall compared to the PI, with positive anomalies throughout the seasonal cycle but showing highest values in October and a secondary peak in July.
However, while no individual models showed negative anomalies during the monsoon season in Sahel, CCSM4-NCAR, MIROC4m and NorESM1-F show decrease in rainfall over the Coast of Guinea in July-September. The 150 remaining models show both increasing and decreasing rainfall during April-June, but mainly positive anomalies from July-November. To see the changes in the WAS rainfall during the mPWP we look at the JAS rainfall anomalies (mPWP-PI, Fig. 2). The MMM shows a clear dipole pattern with a latitudinal transition at 7°N stretching from the Atlantic Ocean 160 to the eastern part of Northern Africa (Fig. 2a). The robust signal of rainfall increase is centered on Sahel and southern Sahara, covering most of northern Africa and reaching from the Coast of Guinea into northern Sahara. The negative anomalies cover an area stretching from 7°N and continuing south over the Equatorial Atlantic, with the largest decrease located along the Gulf of Guinea.
The large-scale features of the rainfall anomalies are consistent over the individual models, with the rainfall 165 increase centered at 10-15° N and reaching up into southern Sahara, and negative values located over the Gulf of Guinea ( Fig. 2a and 2b-r). The results are less consistent along the Coast of Guinea with models indicating slightly different locations of the transition from negative to positive rainfall anomalies. Some models exhibit a rainfall decrease reaching up to 9°N (MIROC4m, GISS-E2-1-G) while other models limit the negative values to only cover the Equatorial Atlantic and Central Africa (CCSM4-UoT, HadCM3). EC-Earth3-LR and CCSM-UofT show the 170 highest pattern correlation to the MMM at R=0.95 and R=0.92 respectively, while GISS-E2-1-G has the lowest correlation (R=0.50). The different models show the largest spread over Sahel and southern Sahara (standard deviation of 2-4 mm/day, not shown). This is a region where all models indicate an increase in rainfall, but the simulated magnitude differs largely, from over 8 mm/day in EC-Earth3-LR and MIROC4m to around 1 mm/day for GISS-E2-1-G and IPSLCM5A2. A spatial mean of the rainfall anomalies over Sahel (Fig. 3) shows a similar 175 spread, with the highest values for EC-Earth3-LR and MIROC4m (6.6 and 5.8 mm/day) and the lowest for GISS-E2-1-G and IPSLCM5A2 (0.5 and 0.8 mm/day). The remaining 13 models all show an increase of 1-4 mm/day over Sahel with a MMM of 2.7 mm/day.    Sea level pressure anomalies for the monsoon season (JAS, mPWP-PI) are shown in Fig. 5 for the individual PlioMIP2 models. All models except MRI-CGCM 2.3 (Fig. 5n) show a deepening of the low-pressure area across the Sahara region (negative anomalies) and a strengthening of the negative latitudinal pressure gradients between Sahara and the Equatorial Atlantic. EC-Earth3-LR and CCSM4-UofT ( Fig. 5f and 5q Associated with the deepening of the Saharan Heat Low and strengthening of the latitudinal pressure gradients is an anomalous cyclonic flow and strengthened westerly/southwesterly horizontal winds at the 850 hPa level, going from the Equatorial Atlantic into Sahel and Sahara (Fig. 5 a-q). This is seen in all models, although at different magnitudes, with the highest increase in wind speed seen in CCSM4-Utrecht, EC-Earth3-LR and MIROC4m (Fig.   5c, 5f and 5m), and the lowest increase for GISS-E2-1-G, IPSLCM5A and IPSLCM5A2 (Fig. 5g, 5j and 5k). 215

Fig. 6. July-September (JAS) mean near surface temperature anomalies (ΔSAT, mPWP-PI) for the PlioMIP2 ensemble members (a-q).
The JAS near surface temperature anomalies (ΔSAT, mPWP-PI, Fig. 6) shows a strengthened north-south temperature gradient between the Sahara Desert and the Equatorial Atlantic for all models except MRI-CGCM 2.3 (Fig. 6n). The temperature increase either stretches relatively uniformly across Sahara as in EC-Earth3-LR, 220 COSMOS and CCSM4-UoT, or exhibits two separate centers, one in Western Sahara and one in Eastern Sahara, as for MIROC4m, NorESM-L and NorESM1-F. MRI-CGCM 2.3 (Fig. 6n) has positive temperature anomalies located mainly outside Sahara, both centered along the western coast of Sahara and over eastern Sahara and the Arabian peninsula. An area of negative temperature anomalies is located over the Mediterranean region, and its surrounding areas in northern Sahara exhibit a weaker warming than the neighboring areas of the Sahara Desert. The mid-Pliocene Warm Period is often used as an analog for near-future climate change due to its similar-tomodern paleogeography and high concentrations of CO2 in the atmosphere (Corvec and Fletcher, 2017;Sun et al., 2013), and both marine and terrestrial proxy reconstructions indicate a climate with higher sea 235 surface and surface air temperatures than present Salzmann et al., 2008). Model/data comparison using PlioMIP1 indicated that the models underestimated the high latitude warming by up to 15 °C while overestimating the low latitude temperatures by 1-6 °C Haywood et al., 2013;. A comparison of atmosphere-only general circulation models (AGCM) and coupled ocean-atmosphere models (AOGCM) showed that AGCMs using prescribed SSTs based on paleo reconstructions produce a much 240 stronger WAM compared to models using a coupled ocean-atmosphere configuration, believed to be due to the overestimation of SST and SAT in the tropics in the AOGCM's (Zhang et al., 2016). Analysis of the PlioMIP2 ensemble by Haywood et al. (2020) indicates a widespread model/data agreement for SSTs and little systematic temperature bias in the tropics, suggesting a reduced underestimation of the WAM in the PlioMIP2, but the relatively low availability of palaeohydrological proxies covering West Africa makes it difficult to perform a 245 similar model/data comparison for the WAM and its related rainfall (Salzmann et al., 2008. However, several studies of proxy reconstructions across Northern Africa indicates a more humid climate during the mid-Pliocene. Palynological data records indicate a higher density of tree cover and an expansion of woodland and savanna in Northern Africa at the expense of deserts (Bonnefille, 2010;Salzmann et al., 2008) and multi-proxy studies analyzing e.g. plant-wax and dust records in marine sediment cores taken off-shore of West Africa indicate 250 wetter conditions during the mid-Pliocene (deMenocal, 2004;Feakins et al., 2005;Kuechler et al., 2018), which is qualitatively consistent with the results from the PlioMIP2 ensemble (Fig. 2). The expansion of forest into the Sahara region is also seen in the results from COSMOS , which is the only member of the https://doi.org/10.5194/cp-2021-16 Preprint. Discussion started: 26 February 2021 c Author(s) 2021. CC BY 4.0 License.
PlioMIP2 ensemble that is run with dynamic vegetation. It is also important to note that the PlioMIP2 ensemble is designed to simulate the MIS KM5c within the mPWP (Haywood et al., 2020(Haywood et al., , 2016, and while it represents a 255 useful comparison to modern conditions it might not represent the full climate variability within the mPWP, possibly affecting model-data comparisons .

260
High pattern correlations of JAS rainfall over West Africa (R>0.90; Table 2) between the PI simulations and climatologies based on observational data (CRU: 1901-1930(Harris et al., 2020) for all models indicate that the models are able to sufficiently reproduce the WAM rainfall pattern. However, looking at the absolute values ( Fig.   1) it is clear that while they capture the general seasonal cycle with rainfall peaking in July-September, most models still underestimate the magnitude of the modern summer rainfall over Sahel by 1-3 mm/day, the only exceptions 265 being CESM1.2 and MIROC4m with >5 mm/day of rainfall in August. This is consistent with our general understanding that models struggle to capture West African rainfall (e.g. Roehrig et al., 2013).
The MMM shows a clear increase in summer rainfall in the Sahel region, consistent with a strengthened WAM during the mPWP (Fig. 1). The anomalies are centered on mid to late summer (August-September), which indicates a later withdrawal of the WAM and a lengthened monsoon season. The monsoonal rainfall over the (terrestrial) 270 Coast of Guinea also exhibits larger positive anomalies over the later months of the summer rainfall, further suggesting an intensification of the WAM rainfall towards the end of the monsoon season as well as a later withdrawal during the mid-Pliocene.
There is a large consistency within the ensemble regarding the general features of the mPWP WAM (Fig. 2) changes are statistically robust and consistent with previous studies on both PlioMIP1 and 2 where the tropics, particularly the Northern Hemisphere monsoon regions, are identified as a region with a robust rainfall signal during the mid-Pliocene (Haywood et al., 2020;Li et al., 2018;Pontes et al., 2020;Zhang et al., 2016). The signal is markedly stronger in the PlioMIP2 compared to PlioMIP1, where the MMM shows a doubling of the rainfall 280 increase over Sahel from 1-2 mm/day in PlioMIP1 (Zhang et al., 2016) to 2-4 mm/day in PlioMIP2 (Fig. 2), although the use of June-August as the monsoon season in Zhang et al. (2016) might also have contributed to the discrepancy, especially given the rainfall increase seen over the later part of the monsoon season (Fig 1). The weakest rainfall increase in Sahel is seen in GISS-E2-1-G (Fig 2), which is consistent with the model's low global rainfall response to the CO2 changes (Haywood et al., 2020). Models which were identified as having a larger 285 land/sea rainfall anomaly contrast with a larger rainfall enhancement over land compared to the ocean (Haywood et al., 2020), are also the models which show a larger rainfall increase in Sahel (EC-Earth3-LR, HadCM3, MIROC4m, NorESM1-F, NorESM-L and CCSM4-UoT). However, COSMOS, which did not show a clear land enhancement globally, exhibits similar levels of rainfall increase in Sahel, and even slightly more than NorESM1-F (2.32 and 2.30 mm/day respectively). 290 Haywood et al. (2020) also suggests that, in general, models exhibiting large SAT sensitivity (i.e., high global mean ΔSAT) also exhibit a larger rainfall change (globally), but there is still uncertainty in changes in more regional patterns. While it is consistent with the results from EC-Earth3-LR, which has both one of the highest increase in Sahel rainfall and global SAT (De Nooijer et al., 2020), there is less consistency within the remaining ensemble.
MIROC4m and IPSLCM6A both exhibit similar global ΔSAT (De Nooijer et al., 2020), but their rainfall change 295 differs by close to a factor of 3 (Fig. 3). The PlioMIP2 models however show a consistent JAS warming of the Sahara Desert (Fig. 6), and if the region is limited to the Sahara (10°W-10°E, 20-30°N) a clear link between the ΔSAT and the rainfall increase can be observed (R=0.50, 95% significance). The warming of the Sahara Desert and strengthened latitudinal temperature gradient between the Sahara region and the equatorial Atlantic leads to a deepening of the thermally induced Saharan Heat Low (Fig. 5) (Lavaysse et al., 2009). This deepened Saharan Heat 300 Low induces low-level convergence and strengthens the southwesterly flow, bringing moisture from the equatorial Atlantic into the continent, leading to increased moisture availability and rainfall over Sahel and parts of Sahara and indicating a strengthened WAM (Fig. 5). warm climates, such as the Mid-Holocene and Last Interglacial period (Gaetani et al., 2017;Otto-Bliesner et al., 305 2020), but given the boundary conditions in the mid-Pliocene simulations this warming over Sahara is most likely due to the changes in the atmospheric CO2-concentration. Studies of model simulations as well as observational data has shown that greenhouse gas forcing leads to a land-ocean warming contrast, with a larger temperature increase over land (Byrne and O'Gorman, 2013;Haywood et al., 2020;Lambert et al., 2011). The contrast is a result of the lower moisture availability over land influencing the lapse rate and leading to a higher warming 310 compared to the ocean (Byrne and O'Gorman, 2013), which is consistent with the strong response over the arid Sahara region (Fig. 6). Studies show that this land/ocean warming contrast is present in both equilibrium and transient simulations (Lambert et al., 2011), and future scenarios of climate change show a continued land/ocean contrast and warming of the Sahara region (Boer, 2011;Sutton et al., 2007), leading to strengthened latitudinal temperature gradients. 315 The enhanced vegetation over West Africa in the PlioMIP2 ensemble (Haywood et al., 2020;Salzmann et al., 2008) might also have contributed to the strengthening of the WAM through a vegetation feedback, which has been shown to strengthen the response of the WAM to external forcing in other past climates (e.g. Braconnot et al., 1999;Claussen and Gayler, 1997;Messori et al., 2019).
As the latitudinal land-ocean temperature gradient is central to the development and strength of the WAM through 320 the development of the Saharan Heat Low (Lavaysse et al., 2009), the results have strong implications for future scenarios. Unlike the results in PlioMIP2, and previously in PlioMIP1 (Zhang et al., 2016), which exhibit a uniform rainfall increase over West Africa, both CMIP3 (SRES A2) and CMIP5 (RCP8.5) model ensembles show a drying over western Sahel and a rainfall increase over central and eastern Sahel (Roehrig et al., 2013). As analysis of both CMIP3 and CMIP5 ensembles show a large spread in projected rainfall change in the Sahel region which weakens 325 its confidence in future projections (Roehrig et al., 2013), our results support a future strengthening of the WAM and rainfall increase over West Africa and Sahel in a high CO2 scenario.

Conclusion
The PlioMIP2 ensemble shows a clear rainfall increase over West Africa, with the largest increase located over Sahel, and a strengthening of the WAM leading to the rainfall reaching farther in over the continent. These results 330 https://doi.org/10.5194/cp-2021-16 Preprint. Discussion started: 26 February 2021 c Author(s) 2021. CC BY 4.0 License. are consistent with geological evidence which suggests a more humid climate during the mid-Pliocene (Kuechler et al., 2018;Salzmann et al., 2008). Some regional differences occur among the ensembles, mainly along the coast of Guinea where some models indicate drier conditions while other indicate a rainfall increase. The largest intermodel variability is centered along Sahel, where the magnitude of the rainfall increase varies largely between the models. The strengthened WAM is driven by the warming of the Sahara region and subsequent deepening of the 335 Saharan Heat Low, most likely due to the greenhouse gas forcing and land/ocean warming contrast. The deepened Saharan Heat Low leads to anomalous cyclonic flow and increased moisture flux into the Sahel region, resulting in a northward shift and intensification of the rainbelt. Given the potential for using the PlioMIP2 as an analogue for near-future scenarios, these results suggest a more uniform rainfall increase over West Africa and the Sahel region, unlike the east-west contrast seen in both CMIP3 and CMIP5 future projections (Roehrig et al., 2013). 340 Data availability. The model data can be downloaded from PlioMIP2 data server located at the School of Earth and Environment of the University of Leeds, an email can be sent to Alan Haywood (a.m.haywood@leeds.ac.uk) for access.
Author contributions. Ellen Berntell and Qiong Zhang designed the work, Ellen Berntell did the analysis and wrote the manuscript. All co-authors provided the PlioMIP2 model data and commented on the manuscript. 345 Competing interests. The authors declare that they have no conflict of interest.
Acknowledgements. This research has been supported by the Swedish Research Council (Vetenskapsrådet, grant no. 2013-06476 and2017-04232). The model simulations with EC-Earth3 and data analysis were performed by resources provided by ECMWF's computing and archive facilities and the Swedish National Infrastructure for Computing (SNIC) at the National Supercomputer Centre (NSC) partially funded by the Swedish Research Council through grant agreement no. 2016-07213.

350
GL and CS acknowledge computational resources from the Computing and Data Centre of the Alfred-Wegener-Institute -Helmholtz-Centre for 460 Polar and Marine Research towards generation of the COSMOS PlioMIP2 simulation ensemble. GL acknowledges funding by the Alfred Wegener Institute's research programme PACES2. CS acknowledges funding via the Helmholtz Climate Initiative REKLIM and the Alfred Wegener Institute's research programme PACES2.