Reversal of Projected European Summer Precipitation Decline in a Stabilizing Climate

Precipitation projections in transient climate change scenarios have been extensively studied over multiple climate model generations. Although these simulations have also been used to make projections at specific Global Warming Levels (GWLs), dedicated simulations are more appropriate to study changes in a stabilizing climate. Here, we analyze precipitation projections in six multi‐century experiments with fixed atmospheric concentrations of greenhouse gases, conducted with the UK Earth System Model and which span a range of GWLs between 1.5 and 5°C of warming. Regions are identified where the sign of precipitation trends in high‐emission transient projections is reversed in the stabilization experiments. For example, stabilization reverses a summertime precipitation decline across Europe. This precipitation recovery occurs concurrently with changes in the pattern of Atlantic sea surface temperature trends due to a slow recovery of the Atlantic Meridional Overturning Circulation in the stabilization experiments, along with changes in humidity and atmospheric circulation.


Introduction
The 2015 Paris Agreement defines legally binding temperature limits for the 196 signatory parties, with an implied aim to remain below defined temperature thresholds in the long-term.A key question for climate adaptation is understanding what the climate will look like in a net-zero emissions world where global temperatures have roughly stabilized.Many studies have investigated projected climate change at specified Global Warming Levels (GWLs), but a large majority of these projections are extracted from transient simulations with time-varying radiative forcings (e.g., Swaminathan et al. (2022) for projections using the UK Earth System Model (UKESM1)).While climate projections under different radiative forcing scenarios have been extensively studied across multiple model generations, there are fewer studies focusing on understanding the regional climate impacts of emission reductions or even cessation.
To date, a few climate model experimental designs to study net-zero climates exist.The Zero Emissions Commitment Model Intercomparison Project (ZECMIP, Jones et al. (2019)) was designed to quantify the amount of committed additional warming following a cessation of CO 2 emissions.By design, this results in a multi-model ensemble stabilizing at different GWLs largely depending on the different model climate sensitivities.Other experimental designs have been proposed which focus on sampling specific target GWLs rather than prescribing emission pathways (King, Sniderman, et al., 2021;Sigmond et al., 2020).Fixed concentration experiments, such as the millennial-length simulations contributing to LongRunMIP, have also been used to study climate differences between transient and equilibrated climate states (e.g., Callahan et al., 2021;King, Borowiak, et al., 2021;Rugenstein et al., 2019).Fixed-concentration experiments such as these continue to warm over time, as the ocean continues to take up heat while atmospheric greenhouse gas (GHG) concentrations remain constant (MacDougall et al., 2020).In contrast, in simulations where GHG emissions, rather than concentrations, are prescribed, atmospheric GHG concentrations decrease over time due to uptake from the ocean and terrestrial carbon sinks, more closely simulating a net-zero climate (MacDougall et al., 2020).While emission-driven experiments are well-suited to study the global mean temperature commitment after reaching net-zero emissions of GHGs, fixed concentration experiments are well-suited to study slow changes in the climate system as it slowly transitions to a new equilibrium state on millennial timescales.Other approaches to investigate temperature stabilization include using a simple climate model to predict time-varying concentration pathways consistent with stabilization at global warming level (GWL) targets (Sanderson et al., 2017) or the recently developed adaptive emission reduction approach (Terhaar et al., 2022).
In this paper, we seek to investigate how regional climates are not uniquely determined by their associated GWL.We explore how the rate of climate change is an important factor in shaping regional patterns of climate change.To do this, we exploit an ensemble of six multi-centennial simulations conducted under fixed atmospheric concentrations of GHGs with UKESM1, specifically designed to span a range of GWLs between 1.5 and 5°C above preindustrial by branching-off from ScenarioMIP simulations (O'Neill et al., 2016) at different times.Despite the slow residual warming, the experiments used are referred to as stabilization experiments hereafter for simplicity.We identify regions where precipitation projections at different GWLs in the stabilization experiments differ in sign from their rapidly warming transient counterparts, focusing on Europe in particular.

Climate Model Ensemble Design
Six 500-year long climate model simulations were performed with the UKESM1-0-LL Earth System Model (ESM) under constant external forcing conditions.This model has an atmospheric resolution of 1.25°× 1.875°on 85 vertical levels and a 1°ocean, and includes the following earth system components: a terrestrial carbon and nitrogen cycle; ocean biogeochemistry (BGC) and a unified troposphere-stratosphere chemistry model, tightly coupled to a multi-species modal aerosol scheme (Sellar et al., 2019).The terrestrial BGC includes dynamic vegetation and a representation of agricultural land use change.Its climate sensitivity is 5.4 K (Sellar et al., 2019).The simulations were designed to target a range of GWLs to explore projected climate change between 1.5 and 5°C above pre-industrial levels in a fully coupled, state-of-art ESM.The external forcings for each simulation are held at the level they were at in a parent Scenar-ioMIP simulation at the time the simulations are branched off.In the case of GHGs, that corresponds to constant concentrations and for anthropogenic aerosols it corresponds to constant emissions.The vegetation is dynamic, but each grid cell has a prescribed crop and pasture fraction based on the conditions of their parent simulations at the point of branching-off.Solar and volcanic forcing are set to the forcings used in the pre-industrial control simulations.The six experiments are branched from the historical and ScenarioMIP experiments as follows.
• 2014 in historical • 2020 in SSP3-7.0 • 2025 in SSP3-7.0 • 2025 in SSP2-4.5 • 2030 in SSP1-1.9 • 2040 in SSP3-7.0All simulations are branched from the first initial condition realisation of the parent ScenarioMIP experiment (r1i1p1f2 in the nomenclature of the Coupled Model Intercomparison Project Phase 6 (CMIP6, Eyring et al., 2016).The importance of branching from different scenarios and therefore having differences in the spatial pattern of anthropogenic aerosols and land-use change will be assessed in separate work.The results presented in this paper are not sensitive to the parent scenario the simulations were branched from.
The climate response in the six fixed concentration experiments (500 years each) is compared to the UKESM1-0-LL CMIP6 ScenarioMIP SSP3-7.0 experiments (86 years each), of which there are 16 initial condition members.Mitigation scenarios are not considered in this study due to fewer (5) ensemble members being available for this model and to avoid conflating responses due to other factors such as rapid changes in aerosols in scenarios like SSP1-1.9, in contrast to relatively stable aerosol emissions in SSP3-7.0 (Wilcox et al., 2020).

Global Overview of Simulated Climate Characteristics
As expected for this experimental setup, the fixed concentration experiments show ongoing global warming over the multi-centennial length of the experiments (Figure 1a; solid lines).In these experiments, the top-ofatmosphere (TOA) radiation imbalance is declining but remains positive even after 500 years, explaining the continuing slow warming in these simulations (Figure S2 in Supporting Information S1).The amplitude and timescales of the reduction in TOA radiation imbalance in our fixed concentration experiments is consistent with results from LongRunMIP (Rugenstein et al., 2019).However, the climate warms at a much slower rate in the fixed concentration simulations than the SSP3-7.0scenario (dashed purple line, Figure 1a).We can therefore use the fixed concentration experiments to investigate how a stabilizing climate differs from a rapidly warming one.Existing studies have shown differences in the warming pattern for the same GWLs between transient and stabilization experiments (e.g., King, Borowiak et al., 2021), with relatively reduced warming over land and enhanced warming over the oceans in the stabilization experiments relative to the typical transient experiments.A similar pattern is also found in our experiments (Figure S1 in Supporting Information S1), resulting in a slowly declining land/sea warming ratio in the stabilization experiments (Figure 1b).This slow decline in the land/sea warming ratio is in contrast to the increase in the land/sea warming ratio seen in SSP3-7.0,reflecting a strong amplification of the warming over land.Stabilization is also expected to change sea-surface temperature patterns and affect ocean circulation.The Atlantic Meridional Overturning Circulation (AMOC) is shown for approximately 40°N as obtained from the Overturning Mass Streamfunction (CMIP6 variable "msftyz").The maximum value across all depth ranges is used (i.e., the depth corresponding to the streamfunction maximum is time-varying).In transient experiments such as SSP3-7.0, the AMOC rapidly declines, consistent with previous studies (Figure 1d and e.g., Weijer et al., 2020).In the stabilization experiments, the AMOC initially declines for several decades, but eventually begins to recover and subsequently increases under constant forcing conditions (Figure 1d).This is comparable to findings from Sigmond et al. (2020), who found an AMOC recovery in their emission-driven GWL experiments, but contrasts with the persistent weakening of the AMOC in the overshoot experiments of Delworth et al. (2022), where global temperatures peak in the 2050s and decline thereafter.In our fixed concentration experiments, September Arctic sea-ice rapidly declines, approaching ice-free conditions within the first 100 years for the warmer experiments.The AMOC recovery could be tied to timescales of when Arctic sea-ice loss stabilizes (Nobre et al., 2023) but we do not investigate the mechanisms driving the AMOC recovery further here.
Globally averaged land precipitation is increasing in both the transient and the stabilization experiments, but at different rates, approximately following the different rates of global warming (Figure 1c).The global land hydrological sensitivity spans a comparable range for both the transient and fixed concentration experiments, but both vary in time (Figure 1e).However, regional-scale precipitation projections over land are more complicated due to the importance of different mechanisms, including atmospheric circulation changes and land-surface feedbacks (e.g., Byrne & O'Gorman, 2015;Seager et al., 2014).In the following section, we analyze regionalscale precipitation projections in the transient and stabilization experiments for the 46 land regions used in the IPCC 6th Assessment Report (Iturbide et al., 2020).All regionally averaged results are presented for data that has been masked to include land points only (Figures 2 and 3).

Global and Seasonal Precipitation Projections
When considering the regional differences between transient and stabilization simulations it is clear that precipitation trends show some striking differences.Figure 2 summarizes projected precipitation trends in all IPCC AR6 land regions for each season (Iturbide et al., 2020).The colors in Figure 2 indicate the sign of the trends in each region for the transient experiments (left rectangle), and the fixed concentration runs (right rectangle): blue for increasing trends, red for decreasing trends, and gray for lack of agreement across ensemble members.For boreal summer (June-July-August, JJA) there is a striking feature across Europe where summer drying in the transient simulations becomes wettening in the stablisation simulations; note this is largely a summer phenomenon, as precipitation increases in the transient simulations in the other seasons of the year.The Mediterranean trend reversal occurs in both March-April-May (MAM) and JJA, with a precipitation increase also seen in December-January-February (DJF) in the fixed concentration experiments.No agreement on the sign of the trend across ensemble members in the fixed concentration experiments in September-October-November (SON) indicates a possible stabilization.We consider these European changes further below, noting that UKESM1-0 achieved a "satisfactory" performance classification for precipitation in all European regions (Palmer et al., 2023).
Elsewhere, a reversal of a transient precipitation decline occurs in many regions across the globe: the Caribbean in MAM and JJA; North-East (DJF), Northern (MAM), South-West (all seasons except DJF) South America; Western Africa in JJA.A possible stabilization (gray) of a transient precipitation decline also occurs in several other Central and South American regions, Eastern North America (JJA), Australia (MAM, JJA and SON) and Western Southern-Africa (SON).
Overall, reversals of drying trends occur in multiple regions and seasons, while there are no robust reversals of wettening trends, although changes from a wettening trend to a stabilization (gray) do occur in some instances.However, we do not investigate these instances further here.The sensitivity to the time period over which the trends are calculated is illustrated in Figures S4 and S5 in Supporting Information S1.

European Summer Precipitation Projections and Associated Patterns of Change
In boreal summer, transient climate projections have long shown a robust decrease in precipitation over the Mediterranean region (e.g., Giorgi & Lionello, 2008) and the region is considered a climate change "hot-spot" (Giorgi, 2006).Northern European projections are more uncertain due to different model responses in atmospheric circulation changes but also tend to project a precipitation decline (e.g., Brunner et al., 2019).The UKESM1 SSP3-7.0 transient projections are in agreement with this finding (Figure 3; red).In the Mediterranean, the projected precipitation decline under SSP3-7.0 is approximately 48% by the end of the century relative to 1850-1900, compared to 7% for Northern Europe (Figure 3).However, regardless of which scenario and year the fixed concentration experiments were branched from, the stabilization experiments show a wettening trend instead of a drying, reversing the precipitation decline seen in the transient projections.This is particularly striking when precipitation change is viewed as a function of global mean warming (Figure 3, left): in the transient projections, precipitation decreases approximately linearly with additional warming.In contrast, in the stabilization experiments, precipitation increases with further warming within a single simulation.Across multiple simulations, the warming level does not uniquely define a precipitation level, suggesting that the timescale since branching-off from the parent scenario simulation is an important control on the projected precipitation amplitude.We note that for this model in SSP1-1.9there is also a reversal in the Mediterranean drying trend as global temperatures cool after 2050 although we do not consider those experiments further here.
By the end of the stabilization simulations, Mediterranean precipitation is still below the 1850-1900 baseline.In contrast, Northern Europe precipitation overshoots the reference baseline by the 2nd century of the simulations.Expressed as a percentage change relative to 1850-1900, the precipitation deficit is much larger in the Mediterranean than in Northern Europe.Perhaps unsurprisingly, internal variability is larger in the Northern Europe region, with a separation between the six stabilization runs not clearly discernible by eye.In contrast, in the Mediterranean, the separation between the different stabilization experiments is clear: the lowest warming runs, starting off with a smaller precipitation deficit, consistently have a smaller precipitation deficit than the higher warming simulations.This clear separation between the different simulations in the Mediterranean persisting to the end of the 500-year simulations implies that while the precipitation deficit might be partially reversible over long timescales, the recovery timescale is so long that the initial deficit still matters centuries later.
The mechanisms explaining this reversal of summertime drying across Europe warrant some discussion.To compare trend patterns between the transient and stabilization experiments, Figure 4 shows the linear regression patterns in the variable of interest divided by the trend in global average surface temperature.This ensures that the scaled patterns are of comparable magnitude between both sets of experiments.It also ensures that the patterns can be averaged across the different stabilization simulations.It is however important to note here that the magnitude of the absolute trends is substantially larger in the transient simulations than the stabilization experiments (see Figure 1).As a measure of robustness of the trends, lack of stippling indicates that >80% of available ensemble members per experiment agree on the sign of the trend.
The pattern of warming across the North Atlantic differs substantially between the transient and stabilization experiments (Figure 4; top row).The strong warming in the subpolar gyre region in the stabilization experiments is consistent with a strengthened northward heat transport due to a partial recovery in the AMOC (Figure S3 in Supporting Information S1), in contrast to the AMOC decline in the transient experiments (Figure 1d).While the jet is shifting polewards in the transient runs, the jet latitude in the stabilization experiments is roughly stabilized at the level of branching-off from the parent transient run, with perhaps an indication of a weak equatorward trend in some of the runs (Figure 3).This is potentially consistent with a reduced meridional temperature gradient induced by the subpolar gyre warming.
The specific humidity changes in the fixed concentration experiments over the North Atlantic closely mirror the pattern of warming, with increased specific humidity in the fixed concentration runs in the subpolar gyre and along the eastern coast of the North Atlantic.Relative humidity and precipitable water content in the stabilization experiments also increase (not shown).These results suggest that concurrent humidity changes are an important factor explaining the precipitation increases in the stabilization runs.There are also indications that changing SST patterns could lead to atmospheric circulation changes in the North Atlantic.A low-pressure anomaly southwest of the UK and Ireland and changes in the zonal wind speed consistent with a southward shift of the jet are both robust across ensemble members in the JJA mean.These atmospheric circulation changes could contribute to the positive precipitation trends in Northern Europe, but their exact role is difficult to quantify.We further note that the strongest surface circulation changes are found for MAM (see Figure S6 in Supporting Information S1) which also shows a precipitation reversal in the Mediterranean region and so precursor wetter conditions may also play a role for the reversed trends in summer (e.g., Leutwyler et al., 2021).
Summer precipitation in Europe mostly occurs in transient weather systems, including intense convective events.The moisture source for this rainfall either comes from storms as they pass over the ocean or is evaporated from the land surface.Without a detailed analysis of moisture sources, and dynamical and thermodynamical components of precipitation (e.g., Chadwick et al., 2016), it is not possible to quantitatively attribute changes in the different experiments to specific physical processes.Nevertheless, this would be a suitable subject for future research, particularly in the context of understanding projected changes in mean and extreme precipitation events.

Discussion and Conclusions
The projected summertime (JJA) precipitation decline across different European regions in transient scenarios is well studied (e.g., Tuel & Eltahir, 2021).Studies have shown that the assumption of a linear relationship between projected local precipitation change and increasing GWLs is well justified in most cases for transient climate change simulations and observations (e.g., Seneviratne & Hauser, 2020).Our results show that this relationship holds in the UKESM1 transient projections (brown dots and lines, Figure 3) but this assumption no longer applies in the fixed concentration simulations we have used to study climate stabilization (blue dots and lines; Figure 3).Other stabilization experimental designs have also identified instances where precipitation recovers from a decline in rapidly warming transient projections in both hemispheres (e.g., Sniderman et al., 2019;Grose & King, 2023 for the Southern Hemisphere).For Europe, Delworth et al. (2022) found a recovery of summertime precipitation in the Mediterranean, and identified a dependence of the winter precipitation trends in the Mediterranean upon the state of the AMOC in their simulations.Globally, Zappa et al. (2020) identified three Mediterranean-like regions where the time evolution of the local hydro-climate response did not scale linearly with global mean warming under stabilization scenarios.They linked these deviations from a simple pattern scaling to different timescales of the response: rapid adjustments, fast changes in sea-surface temperature patterns and slow changes in sea-surface temperature patterns.Our results are consistent with this perspective, in that slow changes in sea surface temperature play a key role in changing the characteristics of regional precipitation in Europe.Our results also highlight further that the regional climates are not uniquely determined by GWL but also depend on the regional pattern of warming over the Atlantic associated with slow changes in the ocean circulation.For the Mediterranean, the rates of recovery in precipitation are comparable across the different ensemble members, which implies that the point at which concentrations are fixed (i.e., the beginning of "stabilization") is an important factor that could have impacts on precipitation levels even centuries later.
The results presented here and elsewhere highlight an important finding that in many regions and seasons, a precipitation decline under transient warming may be reversible to some extent, if global temperatures stabilize.The consistency of our Mediterranean results with previous studies suggests that these findings are likely to be robust across different climate models.For Northern Europe, we identify a precipitation increase in summer in the stabilization experiments, overshooting the pre-industrial baseline by the 2nd century of the simulations.Atmospheric circulation changes associated with a change in the surface temperature pattern over the North Atlantic, consistent with a slow recovery of the AMOC, are also found.Future studies with other models or different experimental designs will help quantify the sensitivity of the Atlantic circulation changes and northern European wettening to model structural uncertainty and experimental design choices.The precipitation response in mitigation scenarios where future temperature increases are limited also warrant further investigation.
In summary, our results show that climate projections under stabilized warming can substantially deviate from expectations derived from transient climate change projections.Reversals in the sign of precipitation projections would have important implications for adaptation and planning decisions.In particular, the prospect of halting an ongoing precipitation decline would have important implications for water resources management in so-called climate change "hotspots" such as the Mediterranean.More research is needed to understand how these differences affect changes in the frequency and intensity of extreme precipitation events in a stabilized world.
Supporting Information may be found in the online version of this article.

Figure 1 .
Figure 1.16-year running means of global mean surface air temperature (GSAT) (a), the land-sea warming ratio (b), global land precipitation (c), the Atlantic Meridional Overturning Circulation at 40°N (d), hydrological sensitivity defined as global land precipitation divided by GSAT (e), and September Arctic sea-ice area (f) in the various simulations.The reference time period for anomalies (panels a-c) is 1850-1900.The SSP3-7.0 transient projections are a 16-member ensemble mean and are not temporally smoothed.

Figure 2 .
Figure 2. Overview of precipitation trends in SSP3-7.0 (left rectangle) and the fixed concentration experiments (right rectangle) in all seasons for different land regions.Blue indicates a positive precipitation trend and red is used for negative precipitation trends.These colors are shown where > 80% of ensemble members agree on the sign of the trend.Gray rectangles are used where < 80% of ensemble members agree on sign, and therefore represents lack of agreement across the ensemble.The trends are calculated over the full duration of the simulations (86 and 500 years respectively).

Figure 3 .
Figure 3. June-July-August precipitation projections in SSP3-7.0 (red) and the stabilization experiments (blues) for (a) Northern Europe and (b) the Mediterranean.Panel (c) shows the summer Atlantic jet latitude following the definition of Ceppi et al. (2018).The left panels show the projected changes as a function of GWLs, and the right panels as a function of time.All stabilization experiments are smoothed by a 16-year running mean, while the transient experiments are a 16-member ensemble average.A single SSP3-7.0simulation is also shown (thin brown line), representing the parent simulation for those stabilization experiments that are branched from SSP3-7.0.Only every 16th point is shown on the scatterplot for illustration purposes, but the regression in the left panel is performed over all data points.A quadratic least squares fit was performed for the precipitation data and a linear fit was used for the jet latitude index (left panels).

Figure 4 .
Figure 4. June-July-August trends in air temperatures, sea level pressures, 850 hPa zonal winds, precipitation and surface specific humidity in the 6-member ensemble mean of the fixed concentration runs (left) and the transient SSP-3-7.0 16-member ensemble mean (middle).The trends were computed over the full time period, 86 and 500 years respectively.The difference in the trend patterns between fixed concentrations minus transient runs is shown on the right.All trend patterns have been scaled (i.e., divided by) the trend in global mean surface temperatures, ensuring that the patterns across all simulations are comparable.Unscaled, the transient trends are much larger in magnitude than the trends in the fixed concentration experiments.Lack of stippling indicates that >80% of ensemble members agree on the sign of the trend (at least 5/6 in the left column, and 13/16 in the middle column).