Impacts of a weakened AMOC on precipitation over the Euro-Atlantic region in the EC-Earth3 climate model

47 Given paleoclimatic evidence that the Atlantic Meridional Overturning Circulation (AMOC) may 48 affect the global climate system, we conduct model experiments with EC-Earth3, a state-of-the-art 49 GCM, to specifically investigate, for the first time, mechanisms of precipitation change over the 50 Euro-Atlantic sector induced by a weakened AMOC. We artificially weaken the strength of the 51 AMOC in the model through the release of a freshwater anomaly into the Northern Hemisphere 52 high latitude ocean, thereby obtaining a ~57% weaker AMOC with respect to its preindustrial 53 strength for 60 model years. Similar to prior studies, we find that Northern Hemisphere 54 precipitation decreases in response to a weakened AMOC. However, we also find that the 55 frequency of wet days increases in some regions. By computing the atmospheric moisture budget, 56 we find that intensified but drier storms cause less precipitation over land. Nevertheless, changes in 57 the jet stream tend to enhance precipitation over northwestern Europe. We further investigate the 58 association of precipitation anomalies with large-scale atmospheric circulations by computing 59 weather regimes through clustering of geopotential height daily anomalies. We find an increase in 60 the frequency of the positive phase of the North Atlantic Oscillation (NAO+), which is associated 61 with an increase in the occurrence of wet days over northern Europe and drier conditions over 62 southern Europe. Since a ~57% reduction in the AMOC strength is within the inter-model range of 63 projected AMOC declines by the end of the 21 st century, our results have implications for 64 understanding the role of AMOC in future hydrological changes. 65


Introduction
The Atlantic Meridional Overturning Circulation (AMOC) plays a crucial role in the climate system by redistributing heat, carbon and salinity globally (e.g., Buckley and Marshall 2016, Weijer et al. 2019, Zhang et al. 2019, Srokosz et al. 2021, Jackson et al. 2022).In response to past changes in net radiation, it has been shown that the Earth's climate experienced periods of rapid cooling that in several cases led to partial or total glaciation (e.g., Rahmstorf 2002, Clement andPeterson, 2008).
Abrupt changes in AMOC strength are believed to have substantially amplified these global temperature anomalies, if not led to them (Broecker 1997, Broecker 2003, Moffa-Sanchez et al. 2019).Idealized box models also support the hypothesis that the AMOC may exist in multiple states of equilibrium, specifically there would exist a state in which the AMOC is strong, similar to today's climate, but also states in which the AMOC is much weaker or even reversed (Lenton et al. 2013, Johnson et al. 2019).Although multiple equilibria in the AMOC have not been directly observed, it is thought that in past climates the AMOC existed in a much weaker state, or even completely shut down, such as during Dansgaard-Oeschger events (e.g., Dansgaard et al. 1993, Burckel et al. 2015, Henry et al. 2016).Further, proxy records corroborate that global patterns of abrupt past climate change reflect variability in the AMOC, especially for the Younger Dryas and many of the Heinrich events (see Lynch-Steglitz et al. 2017 for a complete overview, and references therein).
In response to increasing concentrations of greenhouse gases, climate models simulate a steady decline in the strength of the AMOC; however, the amount, the rate and the effects of this decline are highly uncertain across models (Collins et al. 2013, Weijer et al. 2020, Bellomo et al. 2021, Lee et al. 2021, Fox-Kemper et al. 2021, Masson-Delmotte et al. 2022).Although simulations archived in the Coupled Model Intercomparison Project (CMIP) do not in general predict a shutdown of the AMOC, even when forced with quadrupled concentration of CO2 with respect to the preindustrial climate (Bellomo et al. 2021), it has been hypothesized that models may overlook this possibility (Liu et al. 2017).In fact, some recent studies have shown that AMOC's strength may be at its weakest in the last millennium (Rahmstorf et al. 2015, Caesar et al 2021, Thornalley et al., 2018), while numerous research efforts investigate potential precursors of an AMOC shutdown (Caesar et al. 2018, Jackson et al. 2020) and early warning signals (Boers et al. 2021).Direct observations of AMOC strength are too short to corroborate an hypothetical AMOC shutdown, in fact some recent work question a declining trend (e.g., Worthington et al., 2021).The collapse of the AMOC is nevertheless considered a 'low-likelihood, high-impact' outcome (e.g., Arias et al. 2021), thus understanding its impacts relative to those of global warming is of utmost importance.
To investigate the potential impacts of an AMOC shutdown, previous studies have performed coupled climate model simulations in which the AMOC is artificially weakened through the release of freshwater into the North Atlantic or Arctic Oceans.These experiments are typically referred to as 'water hosing' simulations, and have been carried out both in preindustrial background conditions, or in conjunction with rising concentrations of greenhouse gases (e.g., Manabe and Stouffer 1997, Zhang and Delworth 2005, Dahl et al. 2005, Jacob et al. 2005, Levermann et al. 2005, Stouffer et al. 2006, Vellinga and Wood 2008, Brayshaw et al. 2009, Kuhlbrodt et al. 2009, Woollings et al. 2012a, Woollings et al. 2012b, Jackson et al. 2015, Liu et al. 2020).These studies have mainly focused on the seasonal or annual mean climate response, finding similar results in independent models.In all previous studies, following collapse of the AMOC or a significant weakening, the models simulate a widespread cooling in the Northern Hemisphere, which is only slightly compensated by a moderate warming in the Southern Hemisphere, especially in the South Atlantic.The cooling at the Earth's surface is accompanied by precipitation decrease in the Northern Hemisphere, especially in the Atlantic sector and over Europe (Jackson et al., 2015).
Previous modeling experiments, showed that a weakened AMOC causes a reduction in northward ocean heat transport (OHT), which is compensated by an increase in northward atmospheric heat transport (AHT).The increase in AHT is accomplished through an intensification of the northern branch of the Hadley cell.As a consequence, models also simulate a southward shift in the annual mean ITCZ, which migrates towards the Southern Hemisphere (e.g., Kang et al. 2008, Vellinga and Wood 2008, Frierson et al. 2013, Marshall et al. 2014).Prior model experiments also showed a strengthening and eastward extension of the Atlantic eddy-driven mid-latitude jet stream, which corresponds to an increased storm activity over the European continent (Brayshaw et al. 2009, Woollings et al. 2012a).These studies found that enhanced and eastwardly displaced storm tracks are due to an increased meridional temperature gradient south of the British Isles, which leads to increased baroclinic instability to fuel storm development.In addition, in water hosing experiments the subtropical high strengthens, whereas no significant changes in the Icelandic low occurs, thus suggesting an increase in North Atlantic Oscillation positive phase (NAO+) events (Jackson et al. 2015), although a previous study by Brayshaw et al. (2009) found no statistically significant changes in the frequency of the NAO.
While previous studies focused on the mean climate response to an AMOC collapse, the impacts on daily timescales remain unknown.To address this question, we perform water hosing simulations with EC-Earth3, a state-of-the-art climate model participating in the Coupled Model Intercomparison Project phase 6 (CMIP6; Eyring et al. 2016).Here we investigate mechanisms of precipitation change in the model at daily timescales in boreal winter (DJFM) and over the Euro-Atlantic sector through an analysis of the atmospheric moisture budget and of weather regimes.

Model and experimental design
We carry out model experiments with EC-Earth3, a state-of-the-art GCM participating in the CMIP6 archive and developed by a consortium of European research institutions (Döscher et al., 2022).EC-Earth3 includes the ECMWF IFS cy36r4 atmospheric model, the land-surface scheme H-TESSEL (Balsamo et al., 2009), the NEMO 3.6 ocean model (Madec, 2015) and the LIM3 seaice component (Rousset et al., 2015).The OASIS3-MCT coupler version 3.0 exchanges fields between the model components (Craig et al., 2017).We use the standard resolutions of TL255L91 for the atmosphere and ORCA1L75 for the ocean (same as in CMIP6).These settings correspond to an atmospheric horizontal resolution of ~80 km and 91 vertical levels, and an oceanic horizontal resolution of ~100 km and 75 vertical levels.
We run an experiment under preindustrial control conditions (hereafter referred to as 'control') for 150 years.In this experiment the external radiative forcing is fixed at the level of representative year 1850, thus all model simulated climate variability is internally driven.We compare the control simulation with an experiment in which a freshwater anomaly of 0.3 Sv is uniformly added poleward of 50°N in the Atlantic and the Arctic oceans for 140 years (see fig. 1).
After this time, the hosing is halted and the model is left to freely evolve for an additional 70 years, for a total experiment length of 210 years.Hereafter, this experiment will be referred to as 'water hosing'.We obtain the freshwater anomaly by applying a virtual salinity flux: where S0 is the local salinity in the upper layer, dz0 is the upper layer thickness, and h is the water hosing field (ℎ =

𝐻𝐻 𝐴𝐴 𝑅𝑅
), where the denominator (  ) is the area of the region in which the water hosing is applied (the North Atlantic and the Arctic in our case), while the numerator (H) is the strength of the freshwater flux anomaly (0.3 Sv = 0.3•10 6 m 3 s -1 ).Then, the following water hosing correction is applied to conserve the total amount of salt throughout the rest of the ocean to the 3D salinity field: which represents the total added flux divided by the total ocean volume (  ).
We note that the strength of freshwater flux used in this study is deemed unrealistic, however since current climate models are unable to reproduce AMOC bi-stability and might overestimate AMOC stability (Liu et al. 2017), similar to prior studies we apply a rather large freshwater hosing to force a much weaker AMOC and explore its impacts on the climate system.
Note that this forcing has been tested in previous studies (Jackson and Wood 2018)

Response of the AMOC
Fig. 2a shows the AMOC index in the experiments, defined as the annual mean maximum in AMOC strength between 25.5°N and 27.5°N, and below 500 meters.The dark purple curve is the AMOC index in the control simulation.The horizontal dark purple curve represents the climatological mean value of the AMOC index, which over the span of the control simulation corresponds to 17.66 Sv.The light purple curve represents the AMOC index in the water hosing experiment.The horizontal light purple line represents the value corresponding to a 50% reduction in strength compared to the control climate (8.83 Sv).Also plotted in fig. 2 is the ocean meridional mass stream-function in the Atlantic/Arctic sector of the control experiment (fig.2b), and the average of the years 100 -159 of the water hosing experiment (fig.2c).Years 100-159 in the water hosing experiment (shaded in fig.2a) are those in which the low-pass filtered (10 years running mean) AMOC index is more than 50% weaker than the control simulation.As expected, the streamfunction associated with the AMOC circulation is much reduced in the water hosing experiment (fig.2c).
We note that similar to some other GCMs, EC-Earth3 does not simulate a steady decline in high-latitude surface temperature as the AMOC weakens due to hosing, but rather shows periods in which the Arctic warms in the first 50 years.We advance an hypothesis for a possible physical mechanism responsible for this in the Supplemental Information (SI).In this study, since the focus is to investigate the effects of a weakened AMOC, for the water hosing experiment we analyze the years 100-159 (shaded in fig.2a) in which the AMOC is more than half weaker (< 50%) of its original mean strength.By making this choice, we eliminate from our analysis the possible effects of the temporary warm anomalies and sea-ice decrease (fig.S1), which we see in the AMOC transient response, on large-scale atmospheric circulation and precipitation patterns (see SI).This leaves us with an adequate sample size (60 years) to compute robust statistical significance against the control experiment at monthly and daily timescales.

Statistical significance
For all diagnostics, the anomalies are computed with respect to the mean control climate.We retain all 150 years of the control simulation for our calculations.We note that the AMOC (fig.2a) as well as global mean surface temperature exhibits centennial timescale fluctuations in EC-Earth3.Meccia et al. (2022) attribute the oscillatory behavior of the AMOC in the EC-Earth3 model to a selfsustained ocean variability.Enhanced OHT, and consequently increased sea-ice melting, are linked to a strong AMOC phase, leading to a build-up of freshwater anomalies in the central Arctic.This anomaly is then released by liquid freshwater transport through the Arctic boundaries, yielding to a freshening of the North Atlantic and a decrease in deep water formation and AMOC strength.A similar mechanism was found by Jiang et al. (2021) for the IPSL-CM6A-LR model, which shares the same ocean model (NEMO) with EC-Earth3, while self-sustained AMOC fluctuations also appear in intermediate complexity models (Mehling et al., 2022).
Centennial AMOC variability in these studies is shown to have a large influence on subpolar temperature and sea ice anomalies.In our analysis, we average the control simulation over a period of 150 years, which covers a full AMOC oscillation.Hence, the influence of the centennial fluctuations on the computation of climatological means from the control run is limited.Moreover, the departure of the AMOC in the water hosing experiment from the mean control climate is much larger than the amplitude of the AMOC fluctuations (fig.2a).In fact, Meccia et al. (2022) in a 2000 year long preindustrial control run of EC-Earth3 estimated the average amplitude of centennial fluctuations to be ~3 Sv, while the departure of the hosed AMOC compared to the control AMOC strength in this study is more than 8.83 Sv.We further account for the possible influence of the centennial AMOC fluctuations by performing statistical tests.To assess the statistical significance of the averages of the changes in the gridded anomalies (i.e., temperature, precipitation and moisture fields) in the years 100-159 of the water hosing experiment against the control mean climate, we compute averages of 30 years length taken from the control simulation, starting each one of them 5 years apart.Thus, we obtain for each grid point a distribution of 30 averages, which represent the estimated internal variability in the control run.We deem the anomaly in the water hosing experiment statistically significant if it exceeds 2 standard deviations of the distribution of the 30 averages computed from the control simulation.In the figures, we use hashes to indicate where changes are statistically significant according to this test.We tested whether significance is sensitive on the choice of period lengths of 30 years and starting 5 years apart, with different periods ranging from 20 years to 60 years and starting them 2, 5, or 10 years apart.We found that our conclusions were not affected.We note that the initial AMOC strength from which the water hosing experiment is started from, may influence the time required by the water hosing to force a weakened AMOC state: however, here we do not focus on mechanisms and timescales of AMOC response to the water hosing forcing (c.f.Jackson et al. 2022), but on its impacts.Finally, although the water hosing anomalies are much larger than the model's AMOC internal variability, we note that more ensemble members could help corroborate our findings.

Moisture budget
To investigate mechanisms of precipitation change, we compute the atmospheric moisture budget following D'Agostino and Lionello (2020).A similar moisture budget formulation can also be found in Seager et al. (2010), while a thorough derivation can be found in Trenberth and Guillemot (1995).The moisture budget (precipitation minus evaporation) for the control mean climate can be written as: where P and E are the precipitation and evaporation respectively, u is the wind vector, q is specific humidity, p is pressure,   is the density of water, and g is the standard gravity.Overbars indicate monthly means, while primes indicate sub-monthly variability.Subscript s indicates values at the surface.For all variables, we use daily data output.The first integral on the RHS of eq. ( 3) is the convergence of moisture due to the mean flow, while the second integral on the RHS is the convergence of moisture due to the turbulent flow (transient eddies).We refer to the last term on the RHS as S: this term involves surface quantities representing the deformation of the surface moisture transport by the surface pressure gradient.By computing the moisture budget for both experiments using eq.( 3), the change in moisture budget in the water hosing experiment can be approximated as: Where  indicates the difference between the water hosing and the control experiment. is the term involving surface quantities, and tends to be significant in the presence of steep orography.In this analysis, all terms except  are calculated explicitly. is computed as the residual of the other terms.
In the first integral on the RHS of eq. ( 4), the contributions of thermodynamics can be separated from those of dynamics: all terms involving a difference in q but not in u are due to thermodynamic processes, while the terms involving a difference in u but not in q are due to dynamic processes.We can rewrite the moisture budget equation as: Where: and  represent the thermodynamic and dynamic contributions to the moisture budget, respectively.The transient eddies term is computed as a covariance, thus it is not straightforward to split this contribution into thermodynamic and dynamic terms.As in previous studies, we leave it as is (c.f.Seager et al. 2010).In this study, because the nonlinear term ( � �) is small, we incorporate it with the term due to transient eddies.This term contains both thermodynamic and dynamic processes as well.
The term  (eq.6) represents the change in atmospheric moisture budget that is due to ), and will not be discussed here.A review of possible sources of errors in the computation of the moisture budget can be found in Seager and Henderson (2013).Finally, we note that the moisture budget framework does not take into account precipitation recycling (Brubaker et al. 1993).Precipitation recycling contributes to the moisture budget over land areas, and tends to be less important when considering large-scale processes such in the present study: it would be however possible in a future work to further disentangle mechanisms of precipitation changes over land using a moisture tracking formulation.

Weather regimes
To further characterize the mechanisms of precipitation change, we investigate the association of precipitation anomalies with wintertime (DJFM) Euro-Atlantic weather regimes (WRs) (e.g., Hannachi et al. 2017, Straus et al. 2017, Dawson et al. 2012).In our study, WRs are computed using clustering as described by Fabiano et al. (2020).Following their methods, we calculate WRs from daily geopotential height at 500hPa (hereafter z500) in the Euro-Atlantic sector (30°N-90°N, 80°W-40°E).We obtain z500 daily anomalies (z500') by removing the daily mean seasonal cycle smoothed with a 20-day running mean.Since WRs are related to large-scale atmospheric circulation, to reduce complexity arising from variability at small scales, we first compute the first 4 EOFs of the z500', which explain ~55% variance in the control simulation.
We then apply a K-means clustering on the principal components (PCs) associated with these EOFs in the reduced dimensionality space to derive 4 cluster centroids.Each centroid corresponds to a WR.Each day in the control simulation is assigned to the closest centroid based on the minimum Euclidean distance.We then obtain patterns of z500' associated with each WR by compositing over all days in each cluster.In agreement with a large number of previous studies (e.g., Vautard 1990, Michelangeli et al. 1995, Cassou 2008, Dawson et al. 2012, Strommen et al. 2019, Fabiano et al. 2020), we set the a priori number of centroids (i.e., WRs) to be 4, corresponding to the positive phase of the North Atlantic Oscillation (NAO+), Scandinavian Blocking (SBL), negative phase of the North Atlantic Oscillation (NAO-), and Atlantic Ridge (AR).
The z500' composites associated with each WRs in the control simulation are shown and discussed in Section 3.4.Fabiano et al. (2020) found that all climate models show biases in the simulation of WRs in the historical period, nevertheless WRs are quite consistent with reanalyses.EC-Earth3 was found to lie well within the inter-model spread (Fabiano et al., 2021).
We compute WRs in the water hosing experiment assigning each day to one of the 4 centroids obtained from the control simulation.The z500' in the water hosing experiment are obtained with respect to the control climatology.To avoid the chance that changes in the WRs in the water hosing simulation are caused by a change in z500, rather than changes in variability, we linearly detrend the northern hemisphere z500 at each grid point (c.f.fig.S3), although we note that detrending does not alter our results.The z500' anomalies are then projected onto the 4 EOFs computed from the control simulation, thereby producing 4 pseudo-PCs.Each day in the water hosing experiment is then assigned to a centroid based on the minimum Euclidean distance between the 4 pseudo-PCs and the control centroids.Statistical significance of the difference in the number of days and persistence in each WRs with respect to the control simulation is computed using the 99% probability threshold of the Welch's t-test.
We note that to test the robustness of our choices, we tried setting the variance explained by the EOFs to be 80% in the control experiment to obtain the reduced dimensionality space, instead of choosing the a priori number of EOFs to be 4.With this choice, we obtain a total number of 12 EOFs, but we find that our results are not affected.Here, WRs in the water hosing experiment are computed using the control climatology as reference to compare the changes in atmospheric circulation patterns with respect to the control experiment.Computing WRs in the water hosing experiment using the water hosing climatology itself as reference, leads to largely similar WRs patterns.However, the question of how the spatial patterns may change in the new water hosed climate is not specifically addressed here.

Analysis of precipitation anomalies
Fig. 4 shows the change in winter precipitation (fig.4a), evaporation (fig.4b) and net precipitation (fig.4c), here defined as precipitation minus evaporation.Given the large reduction in atmospheric moisture (fig.4d), we would expect a decrease in net precipitation.Instead, because the decrease in evaporation (fig.4b) is much larger than the decrease in precipitation over the mid-latitude North Atlantic Ocean, the net precipitation locally increases.We note that despite the overall drying, there is a small region off the coasts of western Europe where the precipitation anomaly is slightly positive (fig.4a).Net precipitation is also slightly positive over the UK, Ireland and some regions over continental western Europe (fig 4c).This suggests that in some areas there is an increase in precipitation anomalies, albeit very small.In fig.4c we plot 3 boxed areas, which indicate representative regions that we will examine in more detail.We name these regions: 'NWEU' (Northwestern Europe), 'NEEU' (Northeastern Europe), and 'SWEU' (Southwestern Europe).We choose these boxes over the more common regions used in the IPCC reports (Iturbide et al. 2020) because they encompass uniform changes in precipitation anomalies, and we will show that precipitation change in each region is associated with the same drivers.
To characterize the precipitation increase that we notice in some regions in the Euro-Atlantic sector at daily timescales, we compute the change in the number of wet days exceeding the 75 th percentile of daily precipitation between the water hosing experiment and the control simulation in DJFM.First, we compute for each grid point the 75 th percentile of the precipitation anomalies in the control simulation (fig.5a).Then, we count the number of days exceeding the 75 th percentile threshold at each grid point in the water hosing simulation and show in fig.5b the change in the number of days between the water hosing experiment and the control climate.Although there is a widespread moisture deficit in the water hosing experiment (fig.4d), fig.5b shows that the number of wet days increases in some regions, particularly off the coasts of northwestern Europe.
We note that the number of wet days also increases over northern Africa and on the central and eastern side of the continental U.S, although we do not examine these further.

Mechanisms of precipitation change
The amount of moisture in the atmosphere is thermodynamically constrained by the Clausius-Clapeyron relation (e.g., Held and Soden, 2006).Hence, from thermodynamic considerations alone the colder air temperature caused by the decline in the AMOC would be responsible for a decrease in the overall precipitation, as previously hypothesized (e.g., Jackson et al. 2015).However, precipitation exhibits a more complex pattern of change.Specifically, there exist regions that experience an increase in anomalously wet conditions during winter despite the overall mean drying, as shown in fig. 5.This indicates that there is an interplay of thermodynamic and dynamic processes responsible for precipitation changes.Here we investigate the relative role of each of these mechanisms using the moisture budget framework (as described in detail in the Methods section).
Fig. 6 shows the relative contributions of each term in the moisture budget to the change in net precipitation δ(P-E) (fig.4c).The influence of the thermodynamic term δTH (fig.6a) is scattered over the European continent: it contributes to wetter conditions over the Iberian peninsula, southern Europe and the northern coasts of Scandinavia.Over the rest of the European continent, the contribution of δTH is less clear.Over the ocean, δTH is more uniformly distributed, showing a zonal dipole of wet/dry change in the west/east mid-latitude North Atlantic Ocean.δTH contributes to dry also some regions located off the coasts of France, the British Isles and Scandinavia.
Recalling eq. ( 6), we plot the individual contributions of δ in fig.7a and 7b.The δ  term (fig.7b) shows a clear increase in the moisture budget by the mean divergent flow in the subtropics.The divergent flow in the control climate tends to dry the subtropics, but since the moisture deficit is larger upstream of the atmospheric flow in the eastern side of the North Atlantic when the AMOC is weakened (fig.4d), in the presence of this gradient the overall effect downstream towards the regions of convergence (i.e., ITCZ) is an enhancement of the moisture budget.δ  also explains the drying off the coasts of France, the British Isles and Scandinavia.
The wet/dry contrasts between these coasts and the adjacent seas seem to be related to a change in land-sea breeze.Above 45°N over the oceans, δ  dominates the sign of δ because the moisture field shows a larger decrease in the eastern side of the North Atlantic basin compared to the western side (fig.4d), hence the advection is reduced towards the eastern side, and an increase in the moisture budget is favored on the western side of the ocean basin.Over the European continent both δ  and δ  contribute to the overall scattered pattern, with δ  generally favoring wetter conditions due to the change in the moisture gradient (fig.4d) and the land/sea contrast with the Mediterranean Sea.
The dynamic term δDY contributes to a moistening of central and northern Europe, moderate drying of southern Europe, and moistening of the southeastern United States.From eq. ( 7), the dynamic term can also be split into advective (δ  ) and divergent (δ  ) terms, which are shown in fig.7c and 7d, respectively.Both δ  and δ  contribute to moister conditions over northern Europe, however, the drying over southern Europe is due mainly to the divergent contribution, indicating a possible connection with the southward shift of the ITCZ.We note that the combination of the dynamic and thermodynamic terms fails to fully account for the precipitation deficit over Europe.Instead, interestingly, fig.6c shows that the transient eddies term (δTE+δNL) provides a drying contribution throughout the European continent, and a moistening offshore, between the coasts of northern Europe and Greenland.
In This helps us estimate uncertainty in the computation of the moisture budget, although internal variability is large at decadal timescales and longer integrations would provide us with better uncertainty estimates.Finally, we note that the surface term δS (fig.6d) reflects the effect of the wind circulation around boundary conditions, such as orography over land, while over the ocean it retains the sign of the largest term in our calculations (i.e., δTE).While this term contributes to the overall budget, it is of difficult interpretation since it is calculated as a residual (see eq. 3).
In fig. 9 we show changes in dynamic atmospheric variables to help elucidate the role of δDY and δTE in driving moisture budget changes.Enhanced precipitation over northern Europe is consistent with an eastward extension of the jet-stream and enhanced wind speed at upper-levels (fig.9a), which explain the term δDY (see also fig.S6).We see an increase in the frequency of storm tracks (fig.9b), which also exhibit a slight southeastward extension towards the European continent.The enhancement of the storm tracks is in apparent contradiction with the sign of δTE, considering that more storms would imply larger precipitation anomalies over land, whereas the contribution of δTE is to dry the continent (c.f.fig.6c).The opposite effect of storm tracks over Europe, where they dry, and over the ocean, where they enhance moisture, is due to the pattern of evaporation change and moisture transport (fig.4).Because over the ocean evaporation exceeds precipitation, storms are more intense and enhance the moisture budget.Instead, over land the precipitation deficit is larger than evaporation, hence storms are more intense but drier.
To conclude, enhanced wind speed together with a moisture deficit explain the pattern of net precipitation changes over Europe.Although the effect of δTH per se is scattered over the European continent, its indirect effect on storm activity causes the decrease in precipitation over land.The increase in the number of wet days in some regions (above the 75 th percentile, fig.5) is explained by the enhanced storm track activity and wind speed (δDY).Over the ocean, the δTE is the dominating factor in explaining the overall δ(P-E), whereas over the European continent δDY explains the wetter conditions over the northern parts of western Europe, while both δDY and δTE explain the drier conditions over southern Europe.We now further investigate how these precipitation mechanisms may be related to changes in atmospheric circulation patterns at daily timescales.

Changes in weather regimes
From the moisture budget analysis, we find that the dynamic contributions (i.e., δDY and δTE) to precipitation change play a crucial role in determining δ(P-E).Hence to better understand the origin of the dynamic terms, we now investigate changes in daily weather regimes (WRs) and their association with precipitation patterns.We use the methodology developed by Fabiano et al. (2020) and The average persistence of days within each WRs is plotted in fig.12 for the control simulation and the water hosing experiment.In our calculation, persistence is defined as the number of consecutive days in which z500' are clustered within the same regime.The anomaly needs to persist for at least two consecutive days in another regime to be reassigned to a new regime.The average persistence of NAO+ changes from 5.0 days to 6.8 days, with the inter-quartile range changes from 2-6 days to 2-9 days.The 90 th percentile changes from 11 to 15 days, suggesting an increase in extreme events associated with NAO+.The average persistence in the other WRs decreases.We note that the change in the persistence is statistically significant for all WRs, except again for AR.Table S2 reports statistics associated with the boxplots in fig.12. Notably, all mean values related to NAO+ increase, while they decrease for the other WRs.
Because the change in NAO+ occurrence is much larger than the other WRs, and the NAO also is the regime that explains most of the wintertime precipitation anomalies in reanalyses (Seager et al. 2020), in fig.13 we show composites of precipitation anomalies associated with NAO+ in the control (fig.13a) and water hosing (fig.13b) experiments.The precipitation composites relative to the other WRs are shown in supplemental fig.S7.The composite of precipitation anomalies during days classified as NAO+ in the control simulation (fig.13a) resembles the well-known pattern of precipitation anomalies that we expect from a reinforcement of the Icelandic low pressure and Azores high pressure systems, which push the mid-latitude jet stream towards higher latitudes, thereby promoting wetter conditions in northern Europe and drier conditions over southern Europe.
If we compare this composite with the composite obtained from the water hosing experiment for NAO+ (fig.13b), we see an overall drying associated with the mean temperature decrease.However, we can clearly see increased precipitation located downstream of the enhanced mid-latitude jet stream (fig.9a), which promotes wet conditions over northwestern Europe.The precipitation anomalies associated with NAO+ in the water hosing experiment are remarkably consistent with the anomalies associated with the dynamic term in moisture budget (fig.6b) and with the regions where we find an increase (or zero change) in the number of wet days (fig.5b).
This indicates that in the water hosing experiment a more prevalent NAO+ WR in winter, both in terms of frequency of occurrence and persistence, contributes to increase the precipitation anomalies associated with the dynamic term in the moisture budget.In contrast, the reduction in NAO-days is associated with a reduction in wet anomalies over southern Europe and of dry anomalies over northern Europe (fig.S7), also consistent with δDY.

Discussion
The impacts of a weakened AMOC in the annual mean climate found in this study largely resemble those in previous studies using different climate models and with various degrees of AMOC strength declines.For example, one of the most recent studies (Jackson et al. 2015) used HadGEM3 with a higher resolution than in our study (~60 km in the atmosphere) and obtained a ~2.5 °C global mean cooling with a much weakened AMOC strength (~2/3 Sv) at 26.5°N.For comparison, we obtain a reduction in global mean temperature of -1.09°C, but with a weaker AMOC decline (average of 7.55 Sv for 60 years).Despite the difference in the AMOC weakening, in both studies the North Atlantic is the region exhibiting the strongest cooling and there is a global southward shift in the ITCZ.The jet speed is enhanced and the jet is found to extend eastward towards Europe.
Also, the storm tracks are enhanced in both studies (also in Brayshaw et al. 2009).Woollings et al. (2012a) analyzed a water hosing experiment together with simulations from the CMIP archives finding that the storm track response to the AMOC decline is consistent across models and related to the AMOC strength, in support of the present study.Bellomo et al. 2021 also found a strong correlation between the amount of AMOC decline and the mid-latitude upper level jet stream, which is closely linked to changes in the storm tracks at mid-latitudes.Jacob et al. (2005) examined ECHAM5 coupled to MPI-OM together with a regional climate model to specifically investigate climate impacts over Europe.They found that in response to an AMOC weakening, the maritime influence on the European climate is larger compared to today's climate, but also brings cooler conditions because the sea surface temperature cools.In agreement with Jacob et al. (2005), we find an enhanced influence of the ocean temperature on the European continent driven by changes in the jet-stream and storm tracks (also supported by the findings in Brayshaw et al. 2009).Overall, we can conclude that for all of the above mentioned aspects, the results obtained with the EC-Earth3 model lie well within the simulated response to an AMOC suppression found in previous studies.
Our results are also consistent with the models analyzed in Stouffer et al. (2006), who presented an inter-model comparison of water hosing experiments carried out by both GCMs and earth system models of intermediate complexity (EMICs).As mentioned earlier, in some of the models analyzed by Stouffer et al. (2006) there is a partial recovery of the AMOC and northward migration of the deep water formation sites, as in EC-Earth3.A study by van der Berk et al. (2021) using EC-Earth2.2also finds that enhanced convection in the Nordic Seas can inhibit the effect of the water hosing on the AMOC strength.
Ours is the first study that explicitly calculates changes in WRs.Jackson et al. (2015) examined mean SLP change following a collapse of the AMOC in the HadGEM3 model.They argued that winter SLP changes and the strengthening of the storm tracks suggest a mean state that looks more like the positive NAO, although they did not calculate a NAO index explicitly.An earlier study by Brayshaw et al. (2009) investigated the impacts of an AMOC collapse in another version of the Met Office unified model, HADCM3.Using EOF analysis and stationary box indices methods they found no significant changes in the variability of the NAO, although they found that changes in the mean flow project onto the NAO+ phase, similar to our results and Jackson et al. (2015).Brayshaw et al. (2009) found an eastward extension of the NAO pattern in the AMOC collapse run, which is consistent with our analysis of SLP change (fig.S5).We must take into consideration the fact that using WRs as in the present study is quite different than isolating the NAO using the methods used in Brayshaw et al. (2009).In fact, computing the NAO by using EOFs the response of WRs in future scenarios could be influenced by the decline of the AMOC in addition to feedbacks related to the increase in greenhouse gases.We suggest that this aspect should be further investigated.
Finally, we note that horizontal resolution may influence the results presented herein.
Fabiano et al. (2020) compared standard and higher resolution of the models participating in PRIMAVERA, including EC-Earth3.While higher resolutions generally lead to improved representation of WRs, they point out that other biases, such as mean SSTs and geopotential fields, may also contribute to the models' ability of reproducing realistic WRs.The impact of the ocean model resolution may also affect the pathways and adjustment processes in the transient response of the AMOC to the water hosing (e.g., Jackson et al. 2020, Mecking et al. 2016, Swingedouw et al. 2022).Higher resolutions in both the atmosphere and the ocean influence the air-sea interactions, ultimately affecting precipitation impacts, especially over coastal areas or complex topography (e.g., Tsartsali et al. 2022, Bellucci et al. 2021, Haarsma et al. 2019).Reproducing these experiments with higher model resolution and in different models could thus help us gain further insights and advance the results of this study.

Conclusions
Previous studies investigated the global and regional impacts of a possible AMOC collapse by artificially weakening the AMOC strength through the release of a freshwater anomaly into the North Atlantic and Arctic (e.g., Jackson et al. 2015).Most of these studies were carried out using single model experiments, but inter-model comparisons have also been conducted (Stouffer et al. 2006).We perform experiments with EC-Earth3, forcing a 57% reduction of the AMOC strength for 60 model years.We find that the changes in mean surface temperature and precipitation exhibit similar patterns and amplitude compared to previous studies, but for the first time we conduct an indepth analysis of precipitation anomalies at daily timescales in the Euro-Atlantic sector in winter (DJFM).
Our results show that when we artificially weaken the AMOC compared to the preindustrial climate, the global temperature cools especially in the Northern Hemisphere.Net precipitation decreases over most of the European continent, although in some regions there is a slight net precipitation increase.We further characterize precipitation change by examining the number of days exceeding the 75 th percentile of precipitation anomalies computed from the control climate.
This analysis indicates while over most of the Euro-Atlantic sector precipitation decreases, in some regions the number of wet days increases following a decline of the AMOC, especially over northwestern Europe.
We apply the moisture budget framework to investigate mechanisms of precipitation change.We find that changes in the dynamics of the atmosphere, in particular the enhancement of the mid-latitude jet stream and its eastward extension towards northwestern Europe explains the increase in wet days anomalies.On the other hand, it is not simply the deficit in moisture that explains the widespread drying over Europe, but the transient eddies term.Transient eddies are responsible for the drying: in fact, despite storm tracks are enhanced, the moisture they carry is reduced, resulting in overall drier storms, even if more intense.
Finally, to further characterize precipitation anomalies and their connection with the largescale atmospheric circulation, we analyze changes in weather regimes.We find a large increase in the persistence and frequency of occurrence in NAO+ events.Through composite analysis of precipitation anomalies over all NAO+ days, we find that precipitation associated with NAO+ exhibits a pattern of enhanced precipitation over northern and western Europe, and reduced precipitation over southern Europe, consistent with the dynamic contributions in the moisture budget analysis and the enhancement of the jet stream.
To summarize, our study demonstrates that in the event of an AMOC weakening, widespread drying over the European continent is to be expected mainly from intensified but drier storms, and regions of wet anomalies arise from the change in the location and intensity of the midlatitude jet in association with an increase in NAO+ days.Because a ~57% reduction in AMOC strength is within the projected range in AMOC decline in scenarios from the CMIP6 archive, our results have implications for understanding the role of AMOC decline in future climate change (e.g., Bellomo et al. 2021, Liu et al. 2020).
, and water hosing model simulations of EC-Earth3 forced with a 0.3 Sv of freshwater forcing over the region shown in fig. 1 are also part of a larger inter-model comparison, the North Atlantic Hosing Model Intercomparison Project (NAHosMIP, Jackson et al. 2022).
the change in vertically integrated mean humidity.It consists of two components: the first one is the advection of moisture   = − ∫ (  � control •  �)   ��� 0 , and is related to the change in the humidity gradient (e.g., land-sea contrast) along the mean flow.The second component is defined as   = − ∫ (  �  •  � control )   ��� 0 , and is related to the change in mean humidity in areas of mean flow convergence (upward air motion) or divergence (downward air motion).The term  (eq.7) represents the change in atmospheric moisture budget that is due to the change in the mean flow.Similar to , it consists of two components: the first advective component   = − ∫ ( � •  control ���������)   ��� 0  is related to the change in mean wind flow in the presence of spatial gradients of humidity.The second component   = − ∫ ( control ���������  •  �)   ��� 0  is the convergence or divergence of moisture due to the change in the mean flow.The terms of the moisture budget computed for the control climate of EC-Earth3 from eq. (3) are consistent with previously published results, including residuals (e.g., Seager et al. 2010, D'Agostino and Lionello 2020

Fig. 3
Fig. 3 depicts an overview of annual mean changes in the global climate associated with the artificially weakened AMOC strength in the EC-Earth3 model.Due to the AMOC weakening, the global near-surface air temperature cools by an average of -1.09 °C.Most of the cooling occurs in the Northern Hemisphere, which experiences an average decrease in temperature of -2.09 °C .Regional warming in the Southern Hemisphere, such as in the South Atlantic Ocean, partially compensates, resulting in a weaker cooling of only -0.10 °C.Changes are largest in the Northern Fig.8, we compute averages of each moisture budget term in the 3 representative regions of precipitation change (boxed in fig.4c), considering only grid points over land.Over Southwestern Europe (SWEU) we see a reduction in δ(P-E), which is contributed by the dynamic and transient eddies term, while the thermodynamic and surface terms oppose the change.Over Northwestern Europe (NWEU), the slight increase in δ(P-E) is largely due to the dynamic term, while the thermodynamic and transient eddies would favor drier conditions.In Northeastern Europe (NEEU), the reduction in δ(P-E) comes from the thermodynamic and transient eddies term, while the dynamic term would again favor increased net precipitation.In all cases, δTE is responsible for drier continental conditions, while δDY contributes to increased moisture budget over northern Europe.Fig.8also shows error bars estimated by calculating the standard deviation of the moisture budget computed separately for 6 consecutive and non-overlapping 10 years long period of the water hosing experiment over the years 100-159 (same period analyzed throughout the analysis).
Fig.11shows the WR frequency in DJFM in the two experiments, where the boxplots span 25 th to the 75 th percentiles and the whiskers the 10 th to the 90 th percentiles of the frequency distributions.The colored dots and horizontal lines in each boxplot are the mean and median, respectively.Here the mean control values (blue dots) correspond to the average percent frequency reported in the captions of fig.10.While the mean frequency of the other WRs regimes decrease in the water hosing experiment, the average frequency of NAO+ grows from 26.66% to 42.85%.TableS1reports the change in the mean frequency for all WRs, and the values associated with the boxplots in fig.11.Except for AR, the changes in the frequency of occurrence of the other WRs are statistically significant.
isolates the mode of variability that explains most of the variance on inter-annual timescales, which includes coupling with the ocean.The use of WRs instead better isolates atmospheric patterns at daily timescales, hence the two methods are not directly comparable.For what concerns the impacts of NAO on precipitation,Seager et al. 2020  analyzed the ERA-Interim reanalysis over the years 1979-2017 to investigate mechanisms of precipitation variability in winter over Europe.They found that the positive phase of the NAO index is associated with wet anomalies over the British Isles and Scandinavia, and dry anomalies over southern Europe.They applied a moisture budget decomposition, and found that precipitation anomalies associated with the NAO are mainly driven by the mean flow moisture convergence anomalies, and offset by transient eddies moisture fluxes.Although in our context we specifically investigate the impacts of an AMOC weakening, we find an increase in the NAO+ events together with an enhancement of wet anomalies over the British Isles and Scandinavia.In our case, the terms δDY explained the increase in wet anomalies, while δTE opposed it over those regions.Therefore, our results are consistent with the mechanisms found bySeager et al. (2020)  for interannual variability.An interesting parallel with the results obtained here could be made with studies investigating the response of the models to future climate change.Fabiano et al. (2021) examined WRs in historical simulations and future projections in the CMIP5 and CMIP6 archives, showing that models project an increase in NAO+ frequency and persistence.They found a decrease in the SBL and AR regimes, and a slight increase in the NAO-phase.This indicates that except for the NAO-, the influence of future climate change on WRs has the same sign as the influence of an AMOC decline.Fabiano et al. (2021) further note that the increase in NAO+ is consistent with a contraction of the mid-latitude jet towards the central position (see alsoPeings et al. 2018, Oudar et    al. 2020).Moreover,Bellomo et al. (2021) andWoollings et al. (2012a) found a strong influence of the AMOC on the position of the jet in projections of future climate change.Hence, we argue that

Fig. 3 :
Fig. 3: Annual mean changes (water hosing minus control).(a): change in near-surface air temperature, hashes indicate statistical significance; (c) same as (a) but for precipitation; (b) change in atmospheric zonal mean meridional stream-function.In (b) the control climatology is superimposed in contours; (d) zonal mean precipitation change: in black the control climatology, in blue the water hosing anomaly.