Impact of an acceleration of ice sheet melting on monsoon systems

. The study of past climates demonstrated the occurrence of Heinrich events during which major ice discharges occurred at the polar ice sheet, leading to significant additional sea level rise. Heinrich events strongly influenced the oceanic circulation and global climate. However, standard climate change scenarios (Representative Concentration Pathways or RCPs) do not consider such potential rapid ice-sheet collapse; RCPs only consider the dynamics evolution of greenhouse gases emissions. We carried out water-hosing simulations using the Institute Pierre Simon Laplace global Climate Model (IPSL-CM5A) 5 to simulate a rapid melting of the Greenland and Antarctic ice-sheets, equivalent to +1 and +3 m additional sea level rise (SLR). Freshwater inputs were added to the standard RCP8.5 emission scenario over the 21 st century. The contribution to the SLR from Greenland or from Antarctic ice sheets have differentiated impacts. The freshwater input in the Antarctic is diluted by the circumpolar current and its global impact is moderate. Conversely, a rapid melting of the ice-sheet in the North Atlantic slows down the Atlantic meridional overturning circulation. This slowdown leads to changes in winds, inter-hemispheric temperature 10 and pressure gradients, resulting in a southward shift of the tropical rain belt over the Atlantic and Eastern Pacific region. The American and African monsoons are strongly affected and shift to the south. Changes in the North American monsoon occur later, while changes in the South American monsoon start earlier. The North African monsoon is drier during boreal summer while the South African monsoon intensifies during austral summer. Simulated changes were not significant for the Asian and Australian monsoons. data Merged and Bias-corrected data for ISIMIP (EWEMBI) (Lange, 2016). To study the impact of each simulation on these monsoon indices, the difference in inter-annual variability for each simulation between the period 2041-2070 and the historical simulation 1976-2005 is then calculated as the difference between a year and the historical mean divided by the historical mean 250 and converted to a percentage: 100 . ( year i − mean hist ) /mean hist These results are presented in the form of whisker boxes for each region and each simulation.

to 415 ppm and they might reach 500 ppm or more by the end of the 21 st century. In addition, the last deglaciation occurring during the first half of the Holocene was strongly non-linear with an accelerated phase of melting (Mimura, 2013). Therefore, it is important to estimate the impact of such accelerated phases associated with Greenland and Antarctica melting (Hemming, 2004). The SLR predicted at the end of the 21 st century by the IPCC fifth Assessment Report (AR5) is ranging between 0.52 and 0.98 m (Church et al., 2013). Nevertheless, non-linear climatic processes may occur, as for instance DeConto and Pollard 100 (2016) have shown: a rapid ice sheet destabilization could lead to an additional SLR well exceeding 1 m.
GCMs are often not fully coupled with ice sheet models. Even if some GCMs include melting ice (Church et al., 2013), studies have shown that the melting predicted by these models is underestimated (Rignot et al., 2011;Gillet-Chaulet et al., 2012). Estimates of additional SLR are not directly available from GCMs outputs. Ice sheet models or regional models forced 105 by temperature and rainfall simulated by GCMs are used to estimate additional SLR off-line. (Fettweis et al., 2012). However, the addition of freshwater due to the melting of the ice sheet at high latitudes is having a global impact on the climate (Mimura, 2013).
A major release of freshwater, linked to a tipping point of the ice sheet, would not be without consequences. In Greenland, the inputs of fresh-water slow down the Atlantic Meridional Overturning Circulation (AMOC) (Bakker et al., 2016). A release 110 of freshwater in the North Atlantic leads to a cooling of the Northern Hemisphere and a southward shift of the inter-tropical convergence zone (ITCZ) in the Atlantic (Schiller et al., 1997;Vellinga and Wood, 2002;Jackson et al., 2015). This finding is consistent at different time scales: for paleo modelling studies (Kageyama et al., 2013;Marzin et al., 2013a), for pre-industrial historical simulations (Vellinga and Wood, 2002;Stouffer et al., 2006) or for future climate simulations considering an increase in GHGs emissions (Vellinga and Wood, 2008;Liu et al., 2020). A slowdown of the AMOC will have also affect the 115 Pacific meridional overturning circulation (PMOC) with potential changes in associated temperature and precipitation patterns at global scale (Liu and Hu, 2015). This is clearly demonstrated in paleoclimatology during Heinrich events; it is possible to find proxy of induced changed in temperature and precipitation in many places on earth (Clement and Peterson, 2008). Simulations of such a rapid melting have impacts on the Asian monsoon (Marzin et al., 2013b), the African monsoon (Mulitza et al., 2008;Marzin et al., 2013a) and might induce changes in European and American temperature and precipitation (Jackson 120 et al., 2015). These feedbacks and the magnitude of temperature and precipitation changes outside the North Atlantic region depend on the mean simulated climate (Swingedouw et al., 2009b). Although GCMs have biases, the consequences of an influx of freshwater into the North Atlantic, a cooling of the North Atlantic and a southward shift of tropical precipitation, has been shown in simulations conducted with different GCMs (Stouffer et al., 2006;Swingedouw et al., 2013). The melting of Antarctica can moderate the simulated rise in temperature in the Southern Hemisphere (Swingedouw et al., 2008). Competition 125 between the deep-waters formed in the North Atlantic (NA) and the Southern Ocean (SO) leads to a process called the bipolar oceanic seesaw (Stocker, 1998). Swingedouw et al. (2009a) show that the release of freshwater in the Southern Hemisphere, linked to the melting of the Antarctic, can impact NA Deep-Water (NADW). This effect occurs because of three processes, the deep-water adjustment which strengthens the NADW cell, the SO salinity anomaly which weakens the NADW cell and the increase in wind stress in the Southern Hemisphere which strengthens the NADW cell. These processes act on different time 130 scales ranging from a few to thirty years (Swingedouw et al., 2009a).
Previous studies have shown that freshwater release from melting ice sheets can have a major impact on climate (Defrance et al., 2017(Defrance et al., , 2020. However, the response of global and regional monsoons to such rapid ice melting has not been investigated in great details. The main objective of this study is to determine the impact of melting ice sheets on global monsoon but 135 also on each regional monsoon using detailed analysis of mechanisms at play. In this study we highlight potential changes in rainfall seasonality and intensity using Hovmöller diagrams. We also investigate mechanisms at play at the ocean-atmospherecryosphere interface and assess relationship between rainfall changes and moist static energy using the IPSL-CM5A-LR model. To determine the impact of a significant release of freshwater at high/low latitudes on tropical monsoons, we used a simulation framework developed by Defrance et al. (2017). A release of freshwater is simulated in the North Atlantic (offshore 140 Antarctica) to simulate a partial melting of the Greenland (West Antarctica) ice-sheet. This freshwater release is added to the standard RCP8.5 scenario to simulate a break-up of the ice sheet using the Institute Pierre Simon Laplace Climate Model, version 5A (IPSL-CM5A).
This study aims to understand the impact of a rapid ice-sheet melting on monsoons and the physical mechanisms at play. 145 First, we focus on the impact of ice sheet melting on oceanic and atmospheric circulation (Sect. 3.1 and Sect.3.2 respectively).
In a second step, we will study more detailed impacts of such melting on monsoons, first globally and then regionally (Sect. 3.3 and Sect. 3.4 respectively). Finally, we will discuss these results and will provide final recommendations (Sect. 4).

Climate model 150
All experiments were conducted using the IPSL-CM5A model at Low spatial Resolution (IPSL-CM5A-LR, 3.75 • in longitude and 1.875 • in latitude), using the r1i1p1 simulation as described by described by Dufresne et al. (2013). IPSL-CM5A is one of the global climate model used for CMIP5 (Taylor et al., 2012), that feeds into the IPCC 5th assessment report. This model is a atmosphere ocean global climate model (AOGCM). This GCM includes an atmosphere-land surface model coupled to an ocean-sea ice model. This model is made up of physical and biogeochemistry models . The dynamical 155 atmospheric model is LMDz (Laboratoire de Météorologie Dynamique zoom) version 5A with 39 vertical levels, 15 levels of them are below 20 km (Hourdin et al., 2013). The ORCHIDEE (ORganizing Carbon and Hydrology In Dynamic EcosystEms) land surface model is included in IPSL-CM5A-LR (Krinner et al., 2005). NEMOv3.2 (for Nucleus for European Modelling of Ocean) is the ocean model included in IPSL-CM5A (Madec et al., 2017). The resolution is about 2 • (with a meridional increased resolution of 0.5 • near the equator) and with 31 vertical levels for the ocean . NEMOv3.2 inludes the 160 simulation of ocean dynamics with OPA (Océan PArallélisé), of biogeochemistry processes with PISCES (Pelagic Interaction Scheme for Carbon and Ecosystem Studies) (Aumont and Bopp, 2006) and sea ice processes with LIM2 (Louvain-la-Neuve Sea Ice Model, Version 2) (Fichefet and Maqueda, 1997). The OAsis (Ocean Atmosphere Sea Ice Soil) coupler allows the syn-chronization of all models and the exchange of energy and moisture fluxes between the different sub-climatic systems (Valcke, 2013). The biogeochemistry models are INCA (The INteraction with Chemistry and Aerosol) for tropospheric chemistry and 165 aerosols (Hauglustaine et al., 2004), the REPROBUS (Reactive Processes Ruling the Ozone Budget in the Stratosphere) module for stratospheric chemistry (Lefevre et al., 1994). The prescribed variables are CO 2 and other greenhouse gases emissions based on RCP scenarios (Moss et al., 2010), land use (Hurtt et al., 2011), solar irradiance (Lean et al., 2005), and volcanic aerosols .

Experimental design
The experimental design used in this study (Chemison et al., 2022) is based on Defrance et al. (2017). The RCP8.5 scenario (Moss et al., 2010) is used as our reference simulation. RCP8.5 is a worst-case scenario assuming the continuation of recent trends without mitigation during the 21st century and leading to an atmospheric radiative imbalance of 8.5 W.m −2 by 2100 (Moss et al., 2010). To simulate the ice-sheet melting, two sets of simulations were used. The first one corresponds to a partial 175 melting of the Greenland Ice-Sheet (GrIS scenarios) and the second one to a melting of the West Antarctica Ice Sheet (WAIS scenarios). Freshwater fluxes (FWFs) of 0.22 and 0.68 Sv (where 1 Sv = 10 6 m 3 s −1 ) for the GrIS scenarios, and 0.68 Sv for the WAIS scenario, were introduced from 2020 to 2070 using the common RCP8.5 radiative forcing scenario, leading to +1 and +3 m additional SLR respectively. These simulations are hereafter referred as to GrIS1m, GrIS3m and WAIS3m. The annual rate of freshwater release (0.68 Sv or 0.22 Sv depending on the simulation) is constant over 2020-2070. For the GrIS scenarios, the 180 freshwater is added in the North Atlantic (45 • N-65 • N, 45 • W-5 • E) where deep water is formed. For WAIS3m, freshwater is added into the western Antarctic Ocean, off the coasts of Southern America (Defrance et al., 2020). The choice to introduce large amounts of freswater aims to magnify potential impacts of rapid ice sheet melting on monsoon systems, despite known low sensitivity of current climate models to the amount of freshwater release (Swingedouw et al., 2013;Hansen et al., 2016). 185 We compare a mid-century period (2041-2070) during freshwater release, with the historical period  for RCP8.5, GrIS and WAIS scenarios. For all experiment, only one simulation was used (see Appendix A for further details about model internal variability). To quantify the potential impact of melting ice sheet on ocean circulation, we study the evolution of the AMOC, which is derived from the maximum annual mean stream function at 30 • N based on the criterion by Cheng et al. (2013). 190

The monsoon domains
Monsoon areas are defined based on the criterion by Lee and Wang (2014). A monsoon area is characterised by an annual difference in precipitation between local summer and winter exceeding a threshold of 2.5 mm day −1 and the summer precipitation must exceed 55 % of the annual total. To compare observed and simulated monsoon areas, observed rainfall data is based on the Global Precipitation Climatology Project (GPCP), Monthly V2.3 available from the NOAA PSL, Boulder, Colorado, USA 195 (Adler et al., 2018). This data was averaged between 1979 and 2005 and compared with the historical simulation from 1976 to 2005.
Each regional monsoon was defined following fifth IPCC report naming convention (Figure 14.3 in Christensen et al. (2013)): the North America Monsoon System (NAMS), South America Monsoon System (SAMS), North Africa (NAF), Southern 200 Africa (SAF), Eastern Asia (EAS), South Asia (SAS) and Australia and Maritime Continent (AUSMC), see Table 1 and Fig.   1. The equator separates the Northern Hemisphere monsoons whose summer wet season lasts from May until September (MJ-JAS), and the Southern Hemisphere monsoons where the summer period lasts from November to March (NDJFM) following Wang and Ding (2008), Wang et al. (2011) and Lee and Wang (2014).

205
Monsoon areas consist of any land grid point corresponding, in at least one of the simulations, to the aforementioned criterion (Lee and Wang, 2014). Thus, the selected monsoon areas include monsoon regions for historical and scenario simulations (RCP8.5, GrIS1m, GrIS3m, WAIS3m). All land grid points per monsoon region were retained to derive spatial averages, except for the AUSMC box for which one outlier (southernmost point) was removed. Only land data was considered. To better understand simulated precipitations changes, we calculated Moist Static Energy (MSE) in J.kg −1 as follows: is the the specific heat at constant pressure, T (K) is the layer temperature, g (m.s − 2) is the gravity constant, Z (m) is the geopotential height, L (m 2 .s −2 ) is the latent heat of evaporation and q (kg.kg −1 ) is the specific humidity.
Hovmöller diagrams were derived for each land monsoon region. Rainfall was averaged longitudinally. They represent the average monthly precipitation over our 30-year study period. Then, the difference between the future period (2041-2070) 215 and the historical period  was calculated. The statistical significance of this difference was evaluated using the Wilcoxon-Mann-Whitney test for p-values greater than 0.05 (Seneviratne et al., 2013).
For each land monsoon area, ∆ MSE, the difference between the MSE at 200 hPa and 850 hPa, was calculated following Seth et al. (2013).
Then, the ∆ MSE difference between the future and historical period is calculated for each future simulation and for each region (∆ MSE anomaly hereafter). ∆ MSE anomaly was averaged longitudinally and monthly and overlaid on the Hovmöller precipitation diagrams.
Hovmöller diagrams are shown for NAMS, SAMS, NAF, SAF and SAS. Diagrams for AUSMC and EAS are presented 225 in Appendix C. It is noteworthy that changes for AUSMC and EAS are mostly non-significant and too few land pixels were available for the AUSMC region.

Characterisation of monsoons
Six indices, defined by the Expert Team on Climate Change Detection and Indices (ETCCDI), were used to determine changes in daily rainfall extremes and statistics per land monsoon region (Sillmann et al., 2013): the average precipitation (Pav), the 230 number of rainy days (R1mm), simple precipitation daily intensity index (SDII), seasonal maximum 5 days precipitation total (RX5day), seasonal maximum consecutive dry days (CDD) and seasonal maximum consecutive wet days (CWD). The indices are calculated annually for the May to September period over the Northern Hemisphere and for the November to March period for the Southern Hemisphere. Calculation details for each index are provided in Table 2.
These indicators can be used to study the impacts of changes in rainfall. They can then be used for population adaptation.

235
In order to have the most accurate indices possible, we have chosen to apply a bias-correction on rainfall by the Cumulative Distribution Function transform (CDF-t) method that was developed by Michelangeli et al. (2009). For this purpose, by a mathematical transfer function (Vrac and Friederichs, 2015) , the CDF of the precipitation variable simulated by the  (Famien et al., 2018). This dataset is derived from the ERA-Interim (Dee et al., 2011) reanalyses. It 240 extends from 1 January 1979 to 31 December 2013 and has a horizontal resolution of 0.5°x0.5°. Thus, before being able to set up the CDF matching, the precipitation data from the GCM were spatially interpolated on the same 0.5°x0.5°grid. Due to the importance of seasonality for the monsoons, the CDF-t method is applied monthly. This method preserves the long-term trends but moments or quantiles are not conserved (Vrac and Friederichs, 2015).

245
The validation of these indicators for our simulations is presented in supplementary materials by comparing the inter-annual variability of each index for each monsoon between the historical simulation and the EartH2Observe, WFDEI and ERA-Interim data Merged and Bias-corrected data for ISIMIP (EWEMBI) (Lange, 2016). To study the impact of each simulation on these monsoon indices, the difference in inter-annual variability for each simulation between the period 2041-2070 and the historical simulation 1976-2005 is then calculated as the difference between a year and the historical mean divided by the historical mean 250 and converted to a percentage: 100.(year i − mean hist )/mean hist These results are presented in the form of whisker boxes for each region and each simulation.

Ocean Dynamics
The most direct impact of the addition of freshwater, resulting from an ice melting, occurs in the ocean. Future changes for the 255 RCP8.5 simulation correspond to a -4 Sv decrease of the stream function between 500 and 2500 m depth in the North Atlantic ( Fig. 2a). This difference is slightly reduced for the WAIS3m scenario (Fig. 2b) denoting a moderate impact of Antarctica ice melting on the North Atlantic Ocean circulation. This moderate impact may be related to the presence of the circumpolar current around the Antarctica continent which tends to dilute the FWF disturbance. Conversely, the addition of FWF in the North Atlantic associated with the melting of the Greenland ice sheet strongly amplifies the simulated decrease in stream function 260 (about -6 Sv between 500-2500 m). The larger the amount of freshwater added, the greater the decrease in simulated oceanic stream function ( Fig. 2c and Fig. 2d). The addition of freshwater in the Labrador Sea changes the water density resulting in changes in oceanic currents. The seasonal signal is weak although there are slightly stronger differences simulated in boreal summer with respect to winter for the GrIS scenarios (see Appendix A Figure B1 and Figure B2).

265
These results are consistent with the simulated evolution of the AMOC during the 21 st century (Fig. 3). In 2020, before the simulated release of freshwater, all simulations are extremely similar. Simulated AMOC ranges between 7 and 8 Sv, and small differences across simulations might be related to internal model variability (Fig. 3). Between 2025 and 2050, the AMOC simulated for the RCP8.5 scenario decreases to 6 Sv and then increases to 8 Sv while the AMOC for the WAIS3m scenario remains relatively constant. After 2050, the AMOC slowly decreases to 4 Sv for WAIS3m and 5 Sv for RCP8.5. The AMOC

Atmosphere Dynamics
Differences between future scenarios and the historical period are shown for several atmospheric variables (temperature, rain-280 fall, sea level pressure and winds at 850 hPa) during boreal summer on Figure 4, and austral summer on Figure 5. A large increase in temperature is shown at global scale for the RCP8.5 scenario, with simulated temperature differences of about 10 • C at the North Pole during boreal winter (Fig. 5a). This high latitude increase in temperature is slightly amplified in the WAIS3m scenario (Fig. 5c). Conversely, the effect of the RCP8.5 radiative forcing on temperature is strongly buffered by the melting of the Greenland ice sheet (GrIS1m and GrIS3m). Simulated temperature increases for GrIS1m ( Fig  in purple. The AMOC index is derived from the maximum annual mean stream function at 30 • N from Cheng et al. (2013) and GrIS3m ( Fig. 4g and Fig. 5g) are much smaller all year round. The freshwater released into the Labrador Sea cools down this region; this local cooling extends to the western part of the Northern Atlantic Ocean for GrIS3m ( Fig. 4.e, Fig. 5.e). The induced cooling of the North Atlantic is more pronounced during boreal winter (Fig. 5e). For the WAIS3m simulation, the cooling due to the melting of the Western Antarctica is shown during austral summer (Fig. 5c) and is greatly amplified during austral winter (Fig. 4c). WAIS3m shows a regional cooling along the Antarctica coast which is linked to the circumpolar cur-290 rent ( Fig. 4c and Fig. 5c).
All simulations show a decrease in sea level pressure (SLP) at both poles in boreal summer and winter ( Fig. 4 and Fig. 5).
During boreal summer (MJJAS), there is a decrease in SLP over the Northern Hemisphere, no change in simulated rainfall between 10 • N and 30 • S and an increase in SLP in the Southern Hemisphere between 30 • S and 60 • S for the RCP8.5 (Fig.   295 4b) and WAIS3m (Fig. 4d) simulations. Over the Eastern Pacific and Tropical Atlantic Oceans, the RCP8.5 simulation shows a southward shift of the ITCZ (Fig. 4b). These changes are not simulated in WAIS3m, a slight increase in rainfall is simulated over this region with more westerly winds (Fig. 4d). In the GrIS1m and GrIS3m simulations, SLP increases near the coasts of Central America and over the southern USA during boreal summer ( Fig.4f and Fig. 4h). A decrease in SLP is simulated between 30 • S and 60 • S for GrIS1m (Fig. 4f) and GrIS3m (Fig. 4h) during boreal summer. The resulting inter-hemispheric 300 SLP gradient causes a southward shift of the rain belt ( Fig. 4f and Fig 4h). The SLP gradient between the Southern and Northern Hemispheres increases with the addition of freshwater in the North Atlantic Ocean. This gradient is consistent with an increased southward pressure force that pushes the ITCZ southward over the Atlantic, leading to a significant decrease (increase) in rainfall north (south) of the Equator. A similar mechanism is highlighted during boreal winter (NDJFM) with an increase in precipitation simulated further south, over Brazil and southern Africa ( Fig. 5f and Fig. 5h).

Impacts on global monsoon
All model experiments tend to simulate a double Intertropical Convergence Zone (ITCZ), highlighted by the presence of two distinct rain bands, a classical drawback in state-of-the-art GCMs (Fig. 6). Despite this standard bias, the IPSL-CM5 model tends to reproduce the main tropical monsoon areas (shaded areas on Fig. 6) with respect to observed estimates (black contours on Fig. 6). However, the West African and Indian monsoons are simulated too far south and the model also underestimates 310 rainfall over Central America and Northern Australia (Fig. 6). Over the African continent, simulated future changes are moderate over the Sahel for the RCP8.5 (Fig 6a) and WAIS3m scenario (Fig 6b). A southward shift of the ITCZ is simulated over the Tropical Atlantic region in the GrIS1m (Fig. 6c) and GrIS3m (Fig. 6d) experiments. In GrIS3m, the future rain belt extends further south over southern Africa and south-western Brazil (Fig. 6d). Most experiments tend to suggest that south-eastern US states might become Tropical monsoon regions in future. A northward shift of monsoon regions over Northern China is also 315 depicted.

Impacts on regional monsoons
For the NAMS, rainfall is large from May to September between the equator and 10 • N (Fig. 7a). Future RCP8.5 changes reveal a dipole pattern with a decrease in rainfall simulated at the beginning of the rainy season and an increase towards the end (Fig. 7c). The Greenland ice sheet melting scenarios amplify these differences ( Fig. 7.g-i). For GrIS3m, rainfall decreases 320 during the rainy season at 10 • N (Fig. 7i). For WAIS3m, a slight increase in rainfall is simulated at the beginning of the wet season (Fig. 7e).  GrIS3m.
For the SAMS, the simulated rainy season lasts from October to April, with large rainfall occurring between 5 • S and 15 • S (Fig 7b). For GrIS1m and GrIS3m changes, a rainfall dipole is shown between the equator and 15 • S (Fig 7h-j). Between 325 January and March rainfall intensifies in the southern part. Conversely, a drying signal is simulated over the northern part in future. In addition, future rainfall significantly increases at the beginning of the rainy season (Oct-Dec). These changes are more pronounced in the GrIS3m experiment (Fig. 7j). For the RCP8.5 and WAIS3m scenarios, future rainfall changes are much weaker ( Fig. 7d-f). The RCP8.5 scenario tends to simulate slightly wetter conditions in future (Fig. 7d). The WAIS3m scenario seems to show an opposite pattern with a slight increase in precipitation near the equator and a decrease at 10 • S (Fig. 7f).

330
Over the NAF region, two rainy seasons occur over the Gulf of Guinea (April-May and Oct-Nov), one rainy season occurs over the Sahel (July-Sep) and two rainy seasons occur over East Africa (short rains during Oct-Nov-Dec and long rains in period. Light blue shading shows the monsoon domains that spatially intersect for both periods. Mar-Apr-May). The ITCZ first reaches the Guinean coast in April-May, moves northward to reach the Sahel during boreal summer and then quickly retreats southward to reach the Guinean coast in Sep-Oct (Fig. 8a). The RCP8.5 and WAIS3m sce-335 narios simulate a slight rainfall increase in April-May nearby the coast (0-5 • N), and a larger increase from Sep to Dec over 0-10 • N ( Fig. 8c and Fig. 8e). WAIS3m (Fig. 8e) simulates a larger increase from Sep to Dec with respect to RCP8.5 (Fig. 8c), and the RCP8.5 scenario simulates moderate drier than average conditions at 7 • N during the West African monsoon season (July to Sep). The GrIS1m and GrIS3m scenarios simulate a significant decrease in precipitation over the whole region ( Fig.   8g and Fig. 8i). This decrease is larger for the GrIS3m simulation during the West African monsoon season (July-Sep) at 6-10 340 • N (Fig. 8i), and denotes a southward shift of the ITCZ over NAF.
For the SAF domain, the rainy season extends from October to March and peaks between 10 • S and 15 • S (Fig. 8b). All simulations show an increase in future rainfall between December and May between 15 • S and 5 • S ( Fig. 8-d-f-h-j). RCP8.5 ( Fig. 8d) and GrIS scenarios ( Fig. 8h-j) simulate a rainfall increase at 15 • S-5 • S. The RCP8.5 and WAIS3m scenario simulate 345 drier than average conditions during the onset of the rainy season over Southern Africa (Fig. 8d-f). There is a clear southward shift of the ITCZ over the SAF region in GrIS1m and GrIS3m (Fig. 8h-j).
The SAS domain includes both the Indian and South-east Asian monsoons. Depending on the latitude, rainfall is bimodal in the southern part with peaks simulated in March and October-November or uni-modal in the northern part with a peak in 350 August (Fig. 9a). The patterns are very similar between the RCP8.5 (Fig. 9b) and WAIS3m (Fig. 9c) scenarios, with an increase in future precipitation simulated from July to December. The melting of the Greenland ice sheet tends to buffer this increase in rainfall over SAF (Fig. 9d-e).
For the EAS region, although few points show significant differences, there is little difference between the scenarios (Fig.   C1). For the AUSMC region, all scenarios simulate an increase in precipitation in the north. This increase extends southwards  with the GrIS3m scenario (Fig. C1). See Appendix C for more information. All other diagrams correspond to the precipitation difference (colors) and the MSE difference (black contours, dashed lines for negative values) between the future scenario (2041-2070) with c,d) RCP8.5, e,f) WAIS3m, g,h) GrIS1m and i,j) GrIS3m and the historical period , in mm month −1 . Please note that the colour bar values do not correspond to the same values from one column to another.
Significant differences at the 95 % confidence interval are depicted by red dots according to the Wilcoxon-Mann-Whitney test (see Sect. 2.3).
As a summary: Greenland ice melt has a strong impact on rainfall seasonality for the NAF and NAMS regions, thus for the monsoons bordering the North Atlantic. The WAIS3m scenario mainly impacts the seasonality of the North American region  and otherwise follows the trends simulated by the RCP8.5 scenario. Changes in rainfall seasonality are linked to changes in 360 the ∆ MSE anomaly. When ∆ MSE anomaly increases, atmospheric stability increases and precipitation decrease. Conversely, when the ∆ MSE anomaly decreases, atmospheric destabilization and precipitation increase. This relationship between precipitation and MSE is shown for the American (Fig. 7), Southern African (Fig. 8 d-f-h-j), Southern Asian (at high latitudes, see Fig. 9) and South-east Asian monsoons (Fig. C1 c-e-g-i). In some regions, an increase in MSE occurs without an associated decrease in precipitation. This is related to the fact that MSE increases during the dry season, when precipitation is already 365 close to zero, as is the case for Southern Africa (Fig. 8 d-f-h-j), for the North American monsoons at high latitudes (Fig. 7 c-e-g-i) and for the South American monsoon for the RCP8.5 and WAIS3m scenarios (Fig. 7 d-f).
Our scenarios suggest that a melting of the ice sheet impacts the quantity, the geographical distribution and the seasonality of the precipitation supplied to each monsoon system. However, the studied indicators were calculated at a monthly time step, hence not providing detailed information about frequency and magnitude of potential rainfall extremes and dry spells. In the 370 following we focus on monsoon indicators calculated at a daily time step (see Table 2).
The CDD indicator, representing the length of drought episodes, shows a very large interannual variability for all scenarios and regions (Fig. 10). Nevertheless, a clear increase in the duration of droughts is shown for the GrIS scenarios over the NAMS region (Fig. 10a). Decreases in the duration of the wet season, the number of rainy days per year and total precipitation 375 (CDW, R1mm, Pav) are also shown over NAMS by the GrIS scenarios. Changes in intensity of wet events, characterised by the RX5day and SDII indicators, are moderate over this region (Fig. 10a).
For the SAMS region, changes in CDD are less marked for the GrIS scenarios with respect to the WAIS3m scenario which simulates an increase in cumulative dry days (Fig. 10b). An increase in the intensity of precipitation events is shown over SAMS (RX5day, SDII), for the RCP8.5 and GrIS scenarios. Interestingly, the WAIS scenario tends to simulate a decrease in 380 RX5day with respect to the other scenarios. An increase in total annual rainfall (Pav) is also shown over SAMS (Fig. 10b).
Although the interannual variability is very large over the NAF region, the melting of the Greenland ice sheet induce an increase in the number of the CDD and a decrease in the number of the CWD for GrIS3m leading to an overall decrease in annual precipitation (Pav) for the GrIS3m scenario (Fig. 10c). Conversely, an increase in extreme precipitation events (RX5day, 385 SDII) and annual precipitation (Pav) is shown for the WAIS3m scenario. More moderate changes are simulated by the RCP8.5 scenario (Fig. 10c).
In the SAF region, there is an intensification of large precipitation events (RX5day, SDII) and a decrease in the number of cumulative wet days and rainy days (CDW, R1mm) for all simulations (Fig. 10d). Nevertheless, total annual rainfall (Pav) slightly increases for all scenarios (Fig. 10d) In SAS region, there is an increase in annual precipitation (Pav) for all scenarios linked to an intensification of heavy precipitation events (RX5day, SDII) (Fig. 10e). These changes are largest for the WAIS3M simulation (Fig. 10e). We investigated the impact of a partial melting of the ice sheet on monsoons and the associated physical mechanisms using the IPSL-CM5 GCM. This model was forced by three hosing scenarios all considering the RCP8.5 scenario as a common radiative 395 forcing. The WAIS3m experiment simulates a melting of the West Antarctica ice-sheet and the GrIS experiments simulate a partial melting of the Greenland ice sheet equivalent to an additional 1 m and 3 m sea level rise. The Greenland ice sheet, due to its sub-polar position, releases freshwater directly into the North Atlantic, where cold, dense water sinks to form deep currents.
This leads to the slowing of the AMOC, as demonstrated in several studies (Stouffer et al., 2006;Swingedouw et al., 2007;Kageyama et al., 2013;Marzin et al., 2013b). The induced collapse of the AMOC, which we consider to be a major disruption 400 of the thermohaline circulation of the oceans (Vellinga and Wood, 2002), does not occur with the RCP8.5 scenario alone. As shown by Weaver et al. (2012), it is unlikely that the AMOC will undergo an abrupt change when only considering standard RCPs scenarios.
The method for water hosing simulations varies greatly from study to study: the amount of water input and salinity can vary, 405 simulations can be carried out at different time-scales for different climatic contexts (paleoclimates, pre-industrial conditions, future GHGs scenarios). The cooling of the North Atlantic induced by the slowing of the AMOC, and the subsequent southward shift of the Atlantic ITCZ are robust results, shown by many climate models (Vellinga and Wood, 2002;Stouffer et al., 2006;Kageyama et al., 2013;Jackson et al., 2015;Liu et al., 2020) and corroborated by climate proxies (Clement and Peterson, 2008;Mulitza et al., 2008;Marzin et al., 2013b).

410
Previous studies that do not consider a dynamical evolution of GHGs show a cooling of the Northern Hemisphere, and in some cases a warming of the Southern Hemisphere (Vellinga and Wood, 2002;Jackson et al., 2015). In our study, the increase in GHGs leads to a standard global warming signal that is buffered by the slowdown of the AMOC. A rapid melting of the West Antarctica ice sheet tends to increase future temperature further at high latitudes in the Northern Hemisphere during boreal 415 winter. Studies that consider changes in GHGs, suggest that a shut-down of the AMOC might lead to a return to pre-industrial climate conditions (Vellinga and Wood, 2008), which is not the case in the present study. Liu et al. (2020) also use the baseline RCP8.5 scenario to produce climate sensitivity experiments for which they weaken the AMOC. Their findings are consistent with ours, with a lower temperature increase simulated over the Northern Hemisphere when the AMOC is weakened, a southward shift of the ITCZ and no significant change in the Pacific Ocean. The choice of scenarios and climate models has a strong 420 impact on the robustness of the results as demonstrated by Stouffer et al. (2006). Stouffer et al. (2006) conducted a multi-model analysis in pre-industrial climatic conditions using freshwater inputs of 0.1 Sv and 1 Sv (to compare to 0.22 Sv and 0.68 Sv in our simulations). All climate models simulate a temperature decrease over the Northern Atlantic and Greenland, consistently with our findings, but the responses between climate models vary significantly over other regions of the Northern Hemisphere.
A southward shift of the ITCZ is also simulated by other climate models and this change was robust with the addition of +1 Sv. confirmed by the proxy studies carried out by Mulitza et al. (2008); Marzin et al. (2013b) and Liu et al. (2020). The addition of freshwater to a hemisphere leads to cooling of that hemisphere and a southward shift of the ITCZ is simulated in response to freshwater inputs in the North Atlantic, regardless of the the selected climate model. Regarding the choice of scenario, the northern or southern location of freshwater input plays a crucial role in the see-saw effect. The amount of added water also 430 impacts the intensity of changes. For example, the spatial and temporal trends between the GrIS1m and GrIS3m scenarios are similar, but much more pronounced with the latter for which adds a larger amount of freshwater is released. For the WAIS3m scenario, the freshwater disturbance tends to be diluted by circumpolar currents in our simulations.
This latitudinal shift leads to significant changes in African climate, with a drying simulated over the Sahel and increased 435 rainfall simulated over the central and southern part of Africa. These results are consistent with the studies by Mulitza et al. (2008); Kageyama et al. (2013) and Liu et al. (2020). A few paleoclimate studies also show an impact of freshwater inputs on the Indian monsoon (Kageyama et al., 2013;Marzin et al., 2013a), this impact was not evident in our experiments and in results shown by Marzin et al. (2013a). Our study also reveals a southward shift of the American monsoon, with a simulated drying over Central and the Northern part of South America and a large increase in precipitation over the Amazon basin. Jackson year for 10 years. This input in freshwater is much larger than the one used in our study, but it shows similar trends for global temperature changes and a southward shift of the ITCZ which also impacts the Indo-Pacific basin. Jackson et al. (2015) also emphasize a drying out of the Amazon, while a strong increase in precipitation is simulated over this region in GrIS experiments, especially during the wet season. 445 Swingedouw et al. (2008) studied the impact of the Antarctica melting over a longer time period (3000 years). They showed that the melting of Antarctica counteracts climatic changes induced by a melting of Northern Hemisphere ice-sheet due to a see-saw mechanism, consistently with former findings by Stocker (1998). In our study, freshwater input in the Southern Hemisphere is diluted by the circumpolar current and has varying impacts in different parts of the world. The impacts of north-450 ern/southern freshwater inputs have different impacts on the regional monsoon systems.
The spatial resolution of the IPSL-CM5A-LR climate model is coarse. The slowing trend of the AMOC, the associated decrease in temperature and the southward shift of the rain belt in response to the addition of fresh water into the North Atlantic Ocean are robust results. These changes were found in other studies and they are related to large-scale parameters 455 (Mulitza et al., 2008;Kageyama et al., 2013;Jackson et al., 2015;Liu et al., 2020). Concerning freshwater inputs in West Antarctica, a higher resolution ocean model (at 0.1°spatial resolution) could represent oceanic eddies that are not represented in our low-resolution simulations (Kirtman et al., 2012). These eddies occur in the Southern Ocean around Antarctica. These eddies are important sinks of energy between the ocean and the atmosphere, and this sink is more important the further east the wind is (Jullien et al., 2020). Thus, the representation of realistic surface winds in the atmospheric model is also an important effect is significant but moderate compared to the temperature changes simulated in our ice-melt simulations (global cooling between RCP8.5 and GrIS3m of about 0.6 • C on average, and may reach a maximum of 1.15 • C over the period 2041-2070). Jackson et al (2020) also show that finer spatial resolution can have an impact on the AMOC weakening. Both high and lowresolution models have significant biases (Chassignet et al., 2020;Jackson et al., 2020). Additional simulations on the impact 465 of horizontal model resolution and bias improvement will be very valuable and useful for improving future climate projections.
The impact of the spatial resolution is more important at regional scale. Therefore, for the Asian and AUSMC regions it is difficult to simulate reliable trends. The presence of the Himalayas, which has a strong impact on the Asian monsoon (Boos and Kuang, 2010), is poorly represented in our model simulations due to a too coarse resolution and very few island grid boxes are available in the AUSMC region to obtain reliable results. Our findings based on low spatial resolution model still simulate 470 realistic monsoon dynamics and simulated future changes are in agreement with other published studies (Mulitza et al., 2008;Kageyama et al., 2013;Jackson et al., 2015;Liu et al., 2020).
The IPSL-CM5 model belongs to the CMIP5 intercomparison exercise whose parameterization has evolved for the CMIP6 framework (Boucher et al., 2020). The transition from CMIP5 to CMIP6 has led to changes in many GCMs with reduced biases and uncertainties in many regions of the world. The presence of a double ITCZ is a persistent bias in the different in-475 tercomparison exercises, however this bias has reduced in CMIP6 (Tian and Dong, 2020). The representation of monsoons by GCMs has improved between CMIP5 and CMIP6 for China (Xin et al., 2020), India (Gusain et al., 2020), East Asia (Xin et al., 2020, and East Africa, although there are still significant errors (Ayugi et al., 2021). For the Sahel, future rainfall uncertainties remain large despite the evolution of GCMs (Monerie et al., 2020). For Central America, and South America, there is a large dispersion across the different climate models utilized in CMIP5 and CMIP6, with however a slight improvement in CMIP6 480 GCMs (Ortega et al., 2021). It is noteworthy that the multi-model average is often closer to reality than each model considered independently (Ayugi et al., 2021). Significant changes in rainfall seasonality are simulated by the IPSL-CM5 model as well as in the CMIP5 and CMIP6 intercomparison exercises (Wainwright et al., 2021). In our methodological framework, freshwater input is added continuously between 2020 and 2070. However, the melting and release of freshwater might be highly non-linear (tipping point) and may vary seasonally. Sensitivity climate experiments that consider different values of freshwater 485 input depending on the season would allow a more realistic simulation. Although the transition from CMIP5 to CMIP6 has led to improvements in climate simulation in many regions of the world, the models still have systematic errors in the simulation of precipitation. The IPSL-CM5 model, like other GCMs, shows a significant bias with the presence of a double ITCZ.
The most extreme emission scenario, RCP8.5, was used in this study. The choice of RCP8.5 as a credible scenario for the 490 21 st century has been recently criticized by Hausfather and Peters (2020). Hausfather and Peters (2020) argue that the real word GHGs emissions, in order to reach the RCP8.5 scenario, would require an increase in coal use beyond recoverable and available reserves. They also note that even without climate policies, as assumed in RCP8.5, clean energy costs tend to decrease over time. Consequently, they advise that the RCP8.5 scenario should not be used as a most likely future scenario, but only as an extreme one (Hausfather and Peters, 2020). The potential large number of retroactive events in snow-covered regions resulting 495 from rising temperatures (Fettweis et al., 2012) suggest that tipping points can be very rapid and sudden. Recent trends are worrying: the recent melting of the A68 iceberg was estimated to have released about 152 Gt of freshwater and nutrients near South Georgia (Braakmann-Folgmann et al., 2022). In addition, nearly twice as much lightning was detected north of 80°N in 2021, than in the previous nine years combined, denoting an increase in liquid precipitation at high latitudes (Network, 2022).
We only investigated the impact of ice-sheet melting, but a melting of the permafrost, which would potentially release a large 500 amount of methane in the atmosphere, is worth investigating (Stendel and Christensen, 2002). As perspectives, multi-model studies of tipping point scenarios (ice sheet melting, melting of the permafrost, dieback of the Amazon...) and their potential impacts on societies, should be encouraged (Lenton et al., 2019). Rising sea levels and induced climatic impacts could have primordial consequences for human societies, health (Chemison et al., 2021), agriculture (Defrance et al., 2017) and the global economy (Kuhlbrodt et al., 2009).

Appendix A: Ensemble experiments
All experiments are based on the first ensemble member r1i1p1. Only one simulation was carried out for each ice melt experiment because of the magnitude of the added freshwater signal. To compare simulations, only a single ensemble member was used to avoid smoothing out internal variability of the model. In the following figures we compare global mean temperature and rainfall for all ensemble members available for the historical (Fig. A1) and RCP8.5 (Fig. A2) scenario carried out with 510 the IPSL-CM5A-LR model. All historical and RCP8.5 scenario experiments simulate a warming-wetting trend. The spread between ensemble members is relatively moderate and the ice experiments (GrIS1m, GrIS3m and WAIS3m, with colder and drier conditions) clearly stand out from the internal model variability derived for all RCP8.5 simulations (Fig. A2). Mean r1i1p1 r2i1p1 r3i1p1 r4i1p1 1 9 7 6 1 9 7 8 1 9 8 0 1 9 8 2 1 9 8 4 1 9 8 6 1 9 8 8 1 9 9 0 1 9 9 2 1 9 9 4 1 9 9 6 1 9 9 8 2 0  The differences in sea ice fraction (%) between our different future scenarios (2041-2070) and the historical simulation  for boreal winter is shown on Figure B3 and austral winter on Figure B4. The temporal evolution of sea ice fraction and temperature for our different simulations for the northern is presented on Figure B5 and for the southern on Figure   B6. To calculate indices, the sea ice area was defined as follows: any grid point with a 30-year median  ≥ 15% sea ice. This analysis was done by seasons (MJJAS and NDJFM) and for each hemisphere. The largest ice melting is simulated 525 for the RCP8.5 and WAIS simulations in the northern hemisphere during boreal winter (Fig. B3a-b and Figure B5a-c). The addition of freshwater in GrIS experiments tend to limit future ice-melting around Greenland (Fig. B3c-d and Fig. B5a-c).
Most future experiments tend to simulate ice-melting over western Antarctica, while more ice is simulated over south-eastern Antarctica (Fig. B4a-c-d and Fig. B6a-c). These findings are consistent with results from the IPCC AR6 report (Fox-Kemper et al., 2021). Conversely, the WAIS experiments simulate more ice over the western part of Antarctica and a decreased sea ice 530 extent over north-eastern Antarctica (Fig. B4b). The addition of freshwater in the northern Atlantic leads to colder temperatures ( Fig. B5b-d), a decreased AMOC, and a more moderate sea-ice melting (Fig. B5a-c). A similar relationship is shown over the southern hemisphere for the WAIS3m experiment (Fig. B6). There is no clear lag between simulated temperatures and sea-ice extent so we assume that this process is related to the coupling between the atmosphere and the ocean. The study by Merino et al. (2018) also establishes for the Antarctic region that both the atmosphere and the freshwater input, in varying proportions, 535 lead to sea ice anomalies.  Appendix C: Impact on regional monsoons For the EAS region, the wet season extends from May to August with maximum rainfall simulated over the south in July (Fig.   C1a). Although only a few points show significant differences, there is still little difference between the scenarios when comparing future and historical periods. All experiments simulate a decrease in precipitation during boreal winter. During summer, 540 there is an increase in precipitation in the north. In the south, there is a decrease in precipitation during winter but also at the beginning of the wet season, whereas there is an increase in precipitation at the end of the wet season. (Fig. C1c-e-g-i). To validate the use of the different monsoon indicators for each study region, indices for the historical period are compared with those obtained for the EWEMBII data. For CDD, the values obtained are close between the two datasets, although the 545 values from the historical simulations have a slight negative bias for the East Asian, North American and South African regions (Fig. C2a). For the CDW, all regions have a slight negative bias with the historical simulation (Fig. C2b). This bias is also shown for the RX5day indicator ( Fig. C2e) but is not found with the Pav, R1mm, SDII indicators which have very similar values between the two data sets (Fig. C2c-d-f). Despite a slight negative bias for some indicators in some regions, the indicators show the same trends between the two datasets (Fig. C2).

550
For the EAS region, there are almost no changes in the different monsoon indexes for all scenarios, only a slight increase in total precipitation (Pav) due to an intensification of precipitation events (RX5day, SDII) (Fig. C3a).
For AUSMC region, a very high inter-annual variability is shown (Fig. C3b). Total precipitation (Pav) increases slightly for the RCP8.5, GrIS1m and GrIS3m scenarios, which may be related to a decrease in the duration of the dry season (CDD) and an