Rainfall and Salinity Effects on Future Pacific Climate Change

Most climate models simulate an intensified eastern equatorial Pacific warming in response to increasing greenhouse gas concentrations. So far, the proposed mechanisms have focused on thermodynamic aspects such as the spatial inhomogeneity of evaporative feedbacks, and future increases in upper ocean thermal stratification, partly intensified by projected Walker circulation weakening. Here we show, using earth system model freshwater perturbation experiments, that the simulated future rainfall intensification along the equator plays an important role in tropical climate change. Associated negative equatorial salinity anomalies of about −0.6 permil, which are very similar to the projected late 21st century salinity anomalies in CMIP6 SSP585 scenario, strengthen upper ocean stratification in the Pacific which leads to a flattening of the thermocline, shoaling of the equatorial undercurrent, and major shifts in tropical ocean dynamics. The resulting eastern basin equatorial warming also contributes to a marked weakening of the Pacific Walker circulation. Our analysis illustrates the importance of rainfall and salinity changes in the tropical climate system with relevance for understanding both, patterns of future climate change as well as the El Niño Southern Oscillation phenomenon.

So far, the role of salinity in tropical climate dynamics has been studied mostly in the context of the El Niño-Southern Ocean phenomenon (ENSO) (Ballabrera-Poy et al., 2002;Maes & Belamari, 2011;Zheng & Zhang, 2015;Zhu et al., 2014). Some of these studies emphasize the role of the rainfall-driven barrier layer in the western tropical Pacific which reduces vertical mixing (Maes et al., 2002), and shields the surface layers from subsurface temperature anomalies. This in turn can impact the surface temperature sensitivity to wind forcings. This effect is relevant over the warm pool region during the onset of an El Niño event (Maes et al., 2002(Maes et al., , 2005. Other studies suggest that haline processes can strengthen ENSO SST asymmetry by reducing vertical mixing and entrainment (Guan et al., 2022). Motivated by these recent findings for ENSO and the fact that rainfall is expected to intensify considerably over the equatorial Pacific in response to greenhouse warming, our study addresses whether the projected massive enhancement of tropical rainfall-one of the most pronounced features of projected hydroclimate change in climate model simulations-may contribute to the establishment of the long-term future warming pattern in the tropical Pacific and the projected slowdown of the Walker circulation.

Model Experiments on iCESM
The freshwater perturbation experiment was conducted using the isotope-enabled Community Earth System Model (iCESM) version 1.2.2 (Brady et al., 2019) (available at GitHub http://cesm.ucar.edu). The ocean model is Parallel Ocean Program version 2 (POP2) in Greenland dipole grid system with ∼1° resolution and 60 vertical levels. The atmospheric model is the Community Atmosphere Model version 5 (CAM5) which has a 1.9° × 2.5° horizontal resolution and 30 vertical levels. We conducted a 1,000-year-long spin-up simulation to fully equilibrate for the present-day climate conditions (model component set of BC5CN). From this equilibrated state, we ran a control simulation and a freshwater perturbation (hosing) experiment (Pauling et al., 2017). To study the longer-term behavior, both the control and freshwater perturbation experiments were run for 200 years (results are shown in Figures S4 and S5 in Supporting Information S1).
It should be noted here that the 2081-2100 CMIP6 rainfall anomalies are applied in our model experiment as a continuous forcing. This means that the salinity anomalies at the end of the simulation exceed those of the transient CMIP6 ensemble projections. Moreover, the salinity anomalies in the CMIP6 experiment include the effect of the warming-induced evaporation changes, which have not been accounted for in the freshwater perturbation

Writing -original draft: Hyuna Kim
Writing -review & editing: Axel Timmermann, Sun-Seon Lee, Fabian Schloesser experiment, which only focuses on the role of projected precipitation changes. To account for these differences between transient greenhouse warming simulations (CMIP6) and our perturbation experiment, we apply two different methods: (a) we identified in the iCESM hosing simulation the time when the upper ocean salinity changes match the 2081-2100 projections. This occurs for model years 15-26 and we have used the corresponding changes in ocean conditions relative to the long-term mean control simulation as a first approximation of the salinity effect in CMIP6 (Figure 1d-1f); (b) we have scaled the long-term oceanic mean response in the equilibrium freshwater simulation by a factor of 0.65 (Figures 1g-1i). This factor was obtained from a linear spatial regression between the projected upper ocean (200 m) CMIP6 salinity changes and the long-term mean changes in the freshwater perturbation run. Applying this statistical scaling, we can then compare the oceanic response to similar salinity changes in the greenhouse warming ensemble and the hosing experiment. Method (a) has the disadvantage that the short averaging period may lead to a biased estimate due to internal ENSO variability. Method (b), on the other hand, assumes a linear response, which may lead to biases in areas where the nonlinearity of the equation of state (e.g., cabbeling effects) plays a role.

Data Processing of Model Outputs
We mainly used Python coding for data processing except for remapping the raw data of the iCESM model output. Due to the irregular grid spacing of iCESM, we remapped the raw data to the 1° × 1° spatial resolution using Climate Data Operators (CDO, https://code.mpimet.mpg.de).  (1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010) in longitude-latitude (surface) and longitude-depth plane (5°S-5°N average). Color shadings represent anomalies, and a dashed line denotes a zero boundary. The orange line in (b) is a SST anomaly (5°S-5°N). Panels (d-f) are the same as the left panels, but show the average differences between freshwater perturbation (years 15-26) and control (years 1-100) (Method a). This time span is optimal because its corresponding freshwater-driven simulated tropical Pacific salinity anomalies match those in the CMIP6 multi-model ensemble, but the short averaging time may lead to a biased signal due to interannual variability; (g-i) alternative estimate of salinity effect from scaling equilibrated ocean changes from freshwater (years 71-100) relative to control run by factor 0.65 (Method b). The scaling factor was obtained from a linear spatial regression between upper salinity changes in the upper 200 m simulated by equilibrated freshwater perturbation experiment and CMIP6 multi-model ensemble mean. This statistical scaling accounts for the fact that the CMIP6 simulations are forcing transient experiments, whereas the freshwater perturbation experiments apply 2081-2100 CMIP6 freshwater forcing continuously. Furthermore, the scaling also captures the lack of evaporative forcing in the freshwater perturbation experiment.

Temperature Tendency Equation
Our in-depth analysis is partly based on the temperature tendency equation which can be written as: where u, v, and w are zonal, meridional, and vertical velocity, and T and t are potential temperature and time. The overbar and prime denote the long-term mean of the control simulation and anomalies of the freshwater perturbation during the time slice of 15-26 model years, respectively. R stands for the other residual processes, including anomalous heat fluxes and mixing. Since the main components explaining the reduced zonal SST gradient in the equatorial Pacific are zonal and vertical terms, we exclude meridional and other negligible terms (combinations between anomalous advection and temperature gradient). Also, the effect of nonlinear zonal and vertical advection on the overall heat budget is small (not shown) and shall not be considered further. Therefore, our analysis focuses on a simplified temperature tendency equation:

CMIP6 Projections of Equatorial Pacific Climate
We study the overall linkage between EER, rainfall, salinity and future ocean circulation changes by first analyzing the multi-model ensemble mean of Coupled Model Intercomparison Project Phase 6 (CMIP6) simulations conducted under the Shared Socioeconomic Pathway (SSP) 585 scenario. The equatorial Pacific difference pattern for salinity between the time periods 2081-2100 and 1991-2010 (Figures 1a-1c) shows a pronounced freshening in the western equatorial Pacific (WEP) which can be partly explained by the rainfall intensification ( Figure S1 in Supporting Information S1) and anomalous zonal advection (Figure 1c). The freshening extends to ∼300 m depth in the WEP; in contrast to the EEP, where the signal is more surface-trapped. During the 21st Century projected tropical SST increases by 3.8°C on average (Figure 1b), exhibiting the typical EER pattern. Furthermore, the greenhouse warm ing response shows enhanced warming in the EEP (4.3°C, between 120°W and 90°W) relative to the WEP (3.4°C, between 110°E and 140°E) (Figure 1b, orange). We also find a pronounced subsurface "cold wedge" (Figure 1b)-a robust feature, which was identified in climate model simulations already in the late 1990s (Timmermann et al., 1999), and typically extends from the western to the central Pacific (Cai et al., 2018;Thual et al., 2013). The cold wedge, which can develop in response to the increased surface buoyancy, can dampen the enhanced EEP warming, because the colder waters get transported eastward by the mean equatorial undercurrent (EUC). CMIP6 models also simulate a positive zonal velocity anomaly in the thermocline, which is consistent with a shoaling of the EUC ( Figure 1c) and a weakening of the South Equatorial Current. Moreover, the commonly simulated slowdown of the Walker circulation (Soden & He, 2015;Vecchi & Soden, 2007) leads to weakened equatorial winds, a reduction of the zonal equatorial SST gradient, and an increase in the vertical shear of zonal velocities (Bjerknes, 1969).
One of the key features of greenhouse warming is the reduction of upper ocean salinity and an increase in surface temperatures (Figures 1a and 1b), which both contribute to a substantial decrease in upper ocean density and an intensified stratification, as documented by the projected vertical gradients (bottom to up) of the thermal and haline density anomalies (Δρ T , Δρ S ) (Figures 2a and 2b). Whereas the haline effect is strongest in the western Pacific warm pool (Figure 2a), the thermal stratification is dominant across the upper thermocline, extending also into the eastern tropical Pacific. The projected density changes are, in fact, qualitatively similar to the observed trends since the 1960s (Li et al., 2019(Li et al., , 2020Yamaguchi & Suga, 2019). This similarity suggests that anthropogenic climate change may have already contributed to a shoaling of the mixed layer in the equatorial Pacific (Richards et al., 2009), but also natural factors, such as shifts in El Niño frequency and the Interdecadal Pacific Oscillation need to be considered. The CMIP6 simulations exhibit increases of the squared-buoyancy frequencies over the upper ocean by 41% (Figure 2c, on average between 120°E-80°W for upper 200 m). The larger buoyancy frequencies are also associated with a larger Richardson number (Ri), which is calculated from the ratio of buoyancy frequency squared (N 2 ) and the square of the vertical shear of horizontal velocities (Ri = N 2 /(ΔU) 2 ). The greater Ri in the stabilized upper layer implies reduced vertical mixing of heat and momentum which in turn affects the mixed layer heat budget associated with oceanic ventilation and atmospheric forcing (Qiu et al., 2004).

Oceanic Response to Projected Changes in Rainfall
To quantify the effects of projected equatorial Pacific rainfall changes which can exceed regionally over 100% ( Figure S1b in Supporting Information S1) on upper ocean dynamics and thermodynamics, we further analyzed our idealized freshwater forcing experiment (Methods) which applies the projected multi-model mean CMIP6 precipitation changes as anomalous perpetual freshwater forcing to the iCESM (CESM1.2) under present-day climate conditions. To compare the spatial pattern in response to SSP585 greenhouse gas emitting scenario and freshwater perturbation, we apply methods (a) and (b) described in the Materials and Methods section. Considering the absence of explicit thermal forcing in the CESM1.2 experiment, the freshwater forcing is the main driver for the resulting temperature (Figures 1e and 1h), density (Figures 2d and 2g), and circulation (Figures 1f and 1i) changes. One of the key features of our freshwater forcing experiment is that it simulates a decrease in the zonal SST gradient along the equator (∼0.38°C for Method a and ∼0.32°C for Method b), which accounts for about 30%-40% of the CMIP6 projections (∼1.0°C), thereby suggesting that this common feature of greenhouse warming simulations may be caused in part by the rainfall enhancement along the equator and the resulting negative salinity anomalies. Moreover, the freshwater perturbation experiment also captures a pronounced subsurface cold wedge (Figures 1e and 1h) and a shoaling of the EUC (Figures 1f and 1i), in qualitative agreement with the CMIP6 simulations, albeit weaker. These responses can be explained by the simulated freshwater-driven changes in upper ocean buoyancy (Figures 2f and 2i) which include contributions both from salinity (Figures 2d and 2g) and temperature (Figures 2e and 2h).
The haline forcing influences upper ocean density, leading to an increase in stability (Figures 2f and 2i), which is similar in pattern to the one simulated by CMIP6 models in response to increasing greenhouse gas concentrations (Figure 2c). The increased buoyancy frequency N also increases the Richardson number Ri. A large Richardson number in a stratified ocean reduces the vertical mixing of heat ( Figure S2b in Supporting Information S1) between the upper layer and the subsurface (Richards et al., 2009). The reduced vertical mixing of heat shoals the mixed layer depth over the equatorial Pacific ( Figure S3a in Supporting Information S1), which in turn enhances upper ocean warming by distributing the same atmospheric heating over a shallower surface layer (Figures S2a and S2c in Supporting Information S1). The zonal mean mixed layer shoaling in the freshwater forcing experiment is quite substantial, amounting for Method (a) or (b) to ∼8 and ∼7 m (between 140°E-90°W), respectively. For comparison, the typical western to central Pacific mixed layer depth shoaling during observed El Niño events attains values of 10-15 m (Carton et al., 2008). The freshwater-forced reduction of vertical heat and salt exchange ( Figure S2b in Supporting Information S1) further strengthens the vertical density gradient by increasing the residence time of the negative salinity anomalies. These one-dimensional buoyancy-driven processes set the stage for a suite of other three-dimensional dynamical and thermodynamical adjustments.
In a simple shallow water reduced gravity model framework, the increased stratification (Figures 2c, 2f and 2i) can be expressed as a positive anomaly of the reduced gravity (g′). Based on the simplified zonal steady state equatorial momentum balance ( h x ′ ∼ ) the same long-term wind-stress τ x applied to a more stratified system will reduce the zonal thermocline gradient ( h x ). This process, which is apparent in the freshwater-perturbation experiment (Figure 1e,1h, Figure S3c in Supporting Information S1) as well as in the CMIP6 simulations (Figure 1b), leads to the formation of the western to central Pacific subsurface cold wedge and an associated shoaling and flattening of the EUC and thermocline. The establishment of the cold wedge can cause a near-surface cooling tendency (Figure 3b) due to the fact that the mean upwelling (Figure 3b contours) can bring anomalously cold subsurface waters into the mixed layer, as expressed by the heat budget term − ′ (Figure 3b, see Methods for mathematical expressions). Given the robustness of the overall results with respect to the choice of either Method (a) or (b) (Figures 1 and 2), we will from now on only illustrate the heat budget contributions for Method a. The thermocline cooling tendency is, however, compensated for by a reduction of the upwelling feedback −w ′ (Figures 3d and 4d), generating a positive temperature tendency, and by the anomalous zonal advective term −u ′ (Figures 3c and 4b, which is associated with a shoaling of the easterly transport of warm pool waters  Figure S4 in Supporting Information S1) is further spread out by the mean westward advection of the anomalous zonal temperature gradient − ′ (Figure 3a). The stratification-induced thermocline flattening, and the resulting thermodynamic effects are even more pronounced in our extended experiment which exhibits a very pronounced weakening of the EUC and a corresponding strengthened EEP warming ( Figure S5 in Supporting Information S1).
The interplay between forced thermocline and mixed layer changes controls the coupling between adiabatic (e.g., thermocline flattening) and diabatic (mixing) processes, leading eventually to the surface response, and contributing to the enhanced EEP warming. However, the oceanic response alone is not sufficient to explain the overall large-scale features. We need to also consider the atmospheric response and the associated coupled air-sea feedbacks.

Atmospheric Response to Projected Changes in Rainfall and Feedbacks
To examine changes in the coupled atmosphere-ocean system in our idealized Pacific freshwater perturbation experiment, we explore the zonal and vertical current velocities of the ocean and the zonal and vertical wind velocity of the atmosphere. The simulated EER weakens the subsidence over the eastern Pacific by about 46% for Method a (180°-90°W, 5°S-5°N, 1,000-400 hPa box average) (Figure 4c, blue indicates upward direction) and 20% for Method (b) (not shown). Furthermore, we observe a slowdown of the equatorial trade winds (Figure 4a) by about 2% (180°-90°W, 5°S-5°N, 1,000-700 hPa box average) in Method and 3% in Method (b) (not shown) and an overall weakening of the Walker circulation. The weakened easterlies also decrease Ekman divergence and  (Figure 4d), which contributes to further warming of the EEP and slowdown of the Walker circulation. The role of simulated atmospheric circulation changes in driving the ocean temperature response is reflected by the fact that the anomalous vertical and zonal advection of mean temperatures are more effective in warming the upper ocean than the mean advection of anomalous temperature gradients (Figure 3). By connecting oceanic and atmospheric feedbacks, the Walker circulation serves as an important amplifier in the freshwater-mediated EEP warming. The eastward shift of the Walker circulation toward the end of the idealized freshwater perturbation experiment increases precipitation in the EEP by about 13% for method a and 7% for Method (b), relative to the control simulation, whereas the evaporation increases by only about 0.5% ( Figure S6 in Supporting Information S1). This strong feedback on rainfall and eventually ocean salinity will further contribute to the final response to the imposed forcings through the mechanisms detailed above.
With the resulting changes in the mean state (Figures 1d-1i), there is the possibility that also ENSO variability might change in response to the increased haline stratification (Thual et al., 2011) and the emergence of the cold wedge. In our CESM1.2 simulations, however, the background state changes (Figures 1d-1i) are not sufficient to change the statistics of ENSO considerably ( Figure S7 in Supporting Information S1), except for a slightly increased frequency of El Niño events in the 200-year-long extended simulation (62 cases), compared to 53 cases in the control simulation and an enhancement of La Niña conditions following major El Niño events ( Figure  S7c in Supporting Information S1).

Summary and Discussion
Our CESM freshwater perturbation experiment shows that future rainfall changes and the corresponding surface salinity changes can contribute considerably to a reduction of the equatorial Pacific zonal SST gradient (40% of the projected CMIP6 value) and a slowdown of the Walker circulation via a combination of coupled feedback processes ( Figure 5). Decreased surface salinity (Figure 1) sharpens the vertical density contrast (Δ ′ > 0) , which reduces the zonal equatorial thermocline tilt due to the steady-state zonal pressure balance. An increase of the Richardson number (ΔRi > 0) reduces the vertical exchange of heat causing mixed layer shoaling (Δh M < 0), Figure 5. Freshening-driven air-sea coupling feedbacks. A schematic summarizing the main feedback processes. Arrows denote the direction of flow, and black circles indicate the physical process in each step. Red (blue) circles relate to warming (cooling) tendencies over the EEP.
10.1029/2022EF003457 9 of 10 which can promote surface warming by concentrating heat fluxes to a shallower depth. For the same mean wind-stress, an increased stratification (Δ ′ > 0) will also generate a flattening of the thermocline ( h x < 0 ). The thermocline shoaling is particularly well pronounced over the central and western Pacific, where we observe the largest decrease in salinity (Figure 1d). The thermocline anomalies and the resulting shifts in zonal pressure shoal the EUC, which in turn implies that warmer waters from higher up in the thermocline can be advected eastwards into the main upwelling region, leading to a warming tendency. Combined thermodynamical and dynamical responses generate an initial EEP warming, which is further intensified by anomalous westerly equatorial trade winds (Δτ x > 0) and the upwelling feedback. Our analysis reveals that haline and thermal contributions to future stratification changes can have comparable magnitudes (Figure 2), which further highlights the role of rainfall and salinity in tropical climate dynamics.
There is a possibility that the increased tropical rainfall caused by the freshwater perturbation could be partially counteracted by evaporative effects. Our analysis, which focuses on the effect of rainfall only, did not account for this effect, but the CMIP6 multi-model mean changes in evaporation and precipitation between the SSP585 and historical datasets suggest that the projected changes in evaporation, averaged over the region between 140°E-280°E and 10°S-10°N, between 1991-2010 and 2081-2100, are approximately 0.4 mm/day, whereas the absolute changes in precipitation attain values of approximately 1.2 mm/day (see Figure S8 in Supporting Information S1). This implies that the greenhouse warming-induced evaporative effect in the equatorial Pacific could cancel out one-third of the freshening effect resulting from increased tropical rainfall. However, as our study focused on the isolated role of precipitation changes rather than their combined effect, further investigation is necessary to examine the coupling between haline and thermal effects through the latent heat flux.

Data Availability Statement
The data and code used in this paper are available at https://climatedata.ibs.re.kr/data/papers/kim-et-al-2022-esf. The data set includes CMIP6 data to analyze future tropical climate responses, as well as the tropical responses of the ocean and atmosphere to freshwater perturbation in the CESM model.