Contribution of buoyancy fluxes to tropical Pacific sea level variability

Abstract. Regional anomalies of steric sea level are either due to redistribution of heat and freshwater anomalies or due to ocean–atmosphere buoyancy fluxes. Interannual to decadal variability in sea level across the tropical Pacific is mainly due to steric variations driven by wind stress anomalies. The importance of air–sea buoyancy fluxes is less clear. We use a global, eddy-permitting ocean model and a series of sensitivity experiments with quasi-climatological momentum and buoyancy fluxes to identify the contribution of buoyancy fluxes for interannual to decadal sea level variability in the tropical Pacific. We find their contribution on interannual timescales to be strongest in the central tropical Pacific at around a 10∘ latitude in both hemispheres and also relevant in the very east of the tropical domain. Buoyancy-flux-forced anomalies are correlated with variations driven by wind stress changes, but their effect on the prevailing anomalies and the importance of heat and freshwater fluxes vary locally. In the eastern tropical basin, interannual sea level variability is amplified by anomalous heat fluxes, while the importance of freshwater fluxes is small, and neither has any impact on decadal timescales. In the western tropical Pacific, the variability on interannual and decadal timescales is dampened by both heat and freshwater fluxes. The mechanism involves westward-propagating Rossby waves that are triggered during El Niño–Southern Oscillation (ENSO) events by anomalous buoyancy fluxes in the central tropical Pacific and counteract the prevailing sea level anomalies once they reach the western part of the basin.


the eastern basin saw no or even negative trends. These anomalies have been attributed to gradually increasing trade winds since the early 1990s (Merrifield and Maltrud, 2011). As such they are part of a multidecadal pattern of variability, that can be related to climate modes such as PDO, ENSO and SOI (Merrifield et al., 2012).
While it seems clear that the better part of sea level variability in the tropical Pacific is due to adiabatic processes and can 30 be understood in terms of redistribution of heat driven by wind stress changes (Timmermann et al., 2010;Piecuch and Ponte, 2011;Merrifield, 2011;Merrifield and Maltrud, 2011;McGregor et al., 2012;Merrifield et al., 2012;Moon and Tony Song, 2013; Moon et al., 2013;Qiu and Chen, 2012), the importance of buoyancy fluxes is still under debate. Only a few studies have addressed the importance of diabatic processes for SLC in the region. Piecuch and Ponte (2011) assessed the contribution of surface buoyancy fluxes to steric SLC and identified a few regions where the contribution of local buoyancy fluxes is not neg-35 ligible. One of them is the warm pool region in the western tropical Pacific. Other studies identified the central tropical Pacific as another region where local buoyancy fluxes contribute to interannual sea level variability (Piecuch and Ponte, 2012;Forget and Ponte, 2015;Meyssignac et al., 2017). Piecuch et al. (2019) argue that local, latent heatfluxes are in particular relevant on decadal timescales and contributed to the recent reversal of sea level trends in the tropical Pacific since 2012. All these studies used either local budget calculations or ocean model sensitivity simulations to estimate the contribution of buoyancy fluxes. 40 Fukumori and Wang (2013) chose a different approach in a semi-Lagrangian model study and found buoyancy fluxes to be the dominant driver of sea level trends between 1993 and 2004 in the western Pacific warm pool.
A detailed assessment of the relative contribution of buoyancy flux anomalies to sea level variability in the region is important for projecting future sea level trends in this region, since adiabatic and diabatic forcing mechanism might evolve rather 45 differently. However, in particular for the pre-altimetry era and on decadal timescales, we currently lack a detailed picture on how buoyancy fluxes affect sea level in the tropical Pacific and which mechanisms are involved. By means of eddy-permitting ocean model experiments that allows us to individually apply buoyancy and momentum flux forcing to the underlying ocean and by decomposing the steric sea level component into thermosteric and halosteric contributions, we assess the importance of buoyancy fluxes on interannual to decadal SLC over the last 6 decades, analyze their interplay with the wind stress driven 50 variability and determine the importance of heat vs. fresh water fluxes.

Model experiments
We use a global ocean circulation model (OGCM) configuration of the "Nucleus for European Modelling of the Ocean" (NEMO) code version 3.6 (Madec and NEMO-team, 2016). The model uses a global tri-polar ORCA grid at 1/4 • horizontal resolution. The vertical grid consists of 46 z-levels with varying layer thickness from 6 m at the surface to 250 m in the deepest 55 levels. Bottom topography is interpolated from 2-Minute Gridded Global Relief Data ETOPO2v2 1 and represented by partial steps (Barnier et al., 2006). The model is forced with the JRA55-do forcing (Tsujino et al., 2018) which builds on the JRA55 reanalysis product but is adjusted relative to observational datasets.
Laplacian and bilaplacian operators are used to parameterize horizontal diffusion of tracer and momentum respectively. The ocean model is coupled to the Louvain-La-Neuve sea-ice Model version 2 (LIM2-VP, (Fichefet and Maqueda, 1997).

60
To avoid spurious drifts of global freshwater content, all models use a sea surface salinity restoring with Piston velocities of 137 mm per day, which corresponds to a relaxation timescales of one year for the upper 50 m, and a freshwater budget correction, that sets changes in the global budget to zero at each model timestep.
In addition to a hindcast simulation (O025-HC) that uses realistic, interannual forcing and is integrated from 1958 to 2016, 65 two sensitivity experiments with interannual variability only in the momentum (O025-W90) or buoyancy fluxes (O025-B90) were conducted. As a means to construct a "quasi-climatological" forcing, we followed the approach of a "Repeated Year Forcing" (Stewart et al., 2020) to construct quasi-climatological atmospheric fields in order to compute buoyancy and momentum fluxes for the two sensitivity experiments. Following the recommendation of Stewart et al. (2020), we used the period from Mai 1990 to April 1991 repeatedly for 59 cycles to match the length of the hindcast. The three experiments have been used 70 previously for studies of Indian Ocean heat content (Ummenhofer et al., 2020) and marine heatwaves (Ryan et al., 2021).
Interannual sea surface height 2 (SSH) variability from the model hindcast compares well with observations. Figure 1 shows interannual SSH variability as observed by satellite altimetry 3 and simulated by the OGCM. Because the boussinesq model used here does not capture the GMSL signal and the freshwater budget correction forces the GMSL anomaly to zero, the global 75 mean signal has been removed from the altimetry data. The domain is characterized by the well-known meridional dipole with its eastern pole centered on the equator in the central to eastern Pacific. The western pole shows two maxima located around 10 • S and 10 • N in the western part of the basin. Observations show amplitudes of up to 10 cm in the west and around 6 cm in the east (Fig. 1a). Overall, this structure is well captured by the model experiment, although the overall strength of the dipole is slightly underestimated with amplitudes of 8 cm and 5 cm in the western and eastern part respectively (Fig. 1b). SSH anoma-80 lies, averaged over three representative boxes in the northwestern (0 tropical Pacific (shown as red boxes 1-3 on the maps in Fig. 1), from observations and model simulation confirm the reduced amplitude of interannual variability by about 20%. However, the phase of the interannual variability is reproduced well, with correlation coefficients above 0.95 for all three boxes.

Results
Regional SSH variability is predominantly due to changes in the density of the water column ( Fig. 2), with contributions from changes in the mass distribution negligible everywhere, except in shallow coastal regions (not shown). Fig. 2 shows the total SSH signal and its decomposition into steric, thermosteric and halosteric contributions for all three experiments. The steric signal in O025-HC is to a large extent due to changes in heat content (compare Fig. 2d and g). Compared to the total steric signal, 90 the thermosteric variability shows slightly higher amplitudes in particular in the South Pacific Convergence Zone (SPCZ). The same area is characterized by elevated values of halosteric variability (Fig. 2j), reflecting a compensating effect of thermosteric and halosteric changes in the region.
The momentum flux experiments yield a very similar result. Here, total SSH variability is mostly due to heat content changes, 95 with small contributions from halosteric changes that tend to dampen the thermosteric signal (compare Fig. 2b,e,h,k). Overall, these results concur with earlier accounts of the dominant role of thermosteric changes due to wind stress variability for SSH variability in the tropical Pacific.
However, although interannual SSH variability driven by surface buoyancy fluxes is small compared to the momentum flux 100 driven variability, it is not negligible. Its spatial pattern differs from the wind stress driven component (Fig. 2c). The buoyancy flux contribution is most pronounced in the southern part of the study domain with its maximum in the central Pacific around 10 • S and is lowest on the equator. The buoyancy flux driven signal is again steric in nature but in contrast to the wind stress driven signal the halosteric contribution is comparable in magnitude to the thermosteric part (compare Fig. 2 f, i and l).

105
In order to assess the interplay between momentum and buoyancy fluxes, we analyze the change in variability between the hindcast experiment O025-HC and the momentum flux experiment O025-W90. This allows us to identify regions where momentum and buoyancy fluxes are in phase and to asses whether they tend to amplify or dampen one another. Figure 3 shows the absolute change of interannual SSH variability in O025-W90 compared to O025-HC in terms of standard deviation (SD).
Red/blue areas indicate regions where the variability increases/decreases when the interannual variability is removed from the 110 buoyancy flux forcing, i.e. where buoyancy fluxes dampen/amplify wind stress driven SSH variability. The change in total SSH shows a pattern that is almost symmetrical about the equator (Fig. 3a). Negative values prevail around the equator and east of 225 • E. A horseshoe shaped minimum is located at the eastern boundary. Positive values are centered at 175 • E to both sides of the equator at 10 • latitude. The same pattern can be seen for steric SSH (Fig. 3b) and, to a slightly lesser extent, for thermosteric SSH (Fig. 3c). The change in halosteric SSH is similar in magnitude but shows a different pattern. Changes are mainly 115 limited to western and southwestern tropical Pacific where the halosteric SSH variability decreases in O025-W90. Buoyancy fluxes seem to have a small effect on halosteric SSH in the eastern to northeastern tropical Pacific. Notably, regions where halosteric SSH variability is increased coincide with regions where total steric variability is damped. This indicates again the compensating effect of halosteric SSH changes. In short, buoyancy fluxes tend to dampen SSH variability in the eastern part of the domain and amplify it in the central and western part of the region. 120 We take a closer look at two regions where buoyancy fluxes have a strong, but opposite effect and analyze spatial averages over the two boxes marked in Fig. 3. In the southwestern tropical Pacific, SSH changes driven by momentum and buoyancy fluxes are clearly in phase but anticorrelated (p=-0.6) on interannual to decadal timescales. This negative correlation corresponds to the damping effect of buoyancy fluxes identified before. The relative importance of buoyancy fluxes is strongest on 125 decadal timescales, where it dampens the SSH variability by 20 % (Fig. 4a). Steric SSH variability in the eastern box shows similar amplitudes of interannual and decadal variability with the corresponding signals again being in phase, but this time positively correlated (p=0.7). This is in line with the amplifying effect of buoyancy fluxes described above. In contrast to the western box, buoyancy fluxes have little effect on decadal SSH variability in the eastern part of the basin (Fig. 4b).
130 Figure 2 already indicates that both heat and freshwater fluxes drive SSH variability in particular in the region of the SPCZ.
We investigate this further by decomposing SSH from O025-B90 into their thermosteric and halosteric parts again. Figure 5 shows this decomposition for SSH anomalies in the southwestern and eastern tropical Pacific. In the southwestern domain ( Fig.   5a) steric SSH is influenced in equal parts by halosteric and thermosteric contributions from the beginning of the simulation until approximately 1990, after which the variability of thermosteric component decreases by more than 20% (SD drops from 135 75 cm to 59 cm) and steric SSH is subsequently primarily controlled by halosteric SSH (SD of 0.76 cm throughout the integration period). Steric SSH in the eastern equatorial Pacific is almost exclusively governed by thermosteric SSH. While its variability is also smaller during the later part of the simulation, the decrease is not as drastic as in the western box (SD reduces from 71 cm to 63 cm). Halosteric SSH shows only minor contributions, in particular during El Niño events (e.g. the 1997/98 El Niño; SD of halosteric SSH increases from 20 cm to 25 cm between the two periods, which is due to this strong El Niño 140 event).
The effect of buoyancy fluxes on SSH variability is uniform across the zonal extent of the tropical Pacific. This is in particular true south of the equator and explains the opposite effect of buoyancy fluxes (i.e. damping in the west and amplification in the east) on the wind stress driven zonal dipole structure. Figure 6 shows the temporal evolution of SSH along a zonal section 145 at 10 • S. The dipole structure in thermosteric SSH is clearly visible in O025-HC and O025-W90 (Fig. 6a,b,d,e) whereas the effect of halosteric SSH in the same experiments is limited to the central and western part of the region (Fig. 6c,f). Its compensating effect, related to the adiabatic advection of warm and saline surface waters, is clearly visible. Negative SSH anomalies dominate in the western part of the area during the negative PDO phase in the 1980s and 90s and are caused by thermosteric anomalies. Positive halosteric anomalies act to increase SSH. SSH anomalies in O025-B90 show a zonally uniform structure 150 across the basin (Fig.g-j), i.e their effect on the western and eastern poles of the wind stress driven, thermosteric dipole is exactly opposite. Maximum anomalies are located between 200 • E and 220 • E and are more pronounced in the western than in the eastern part of the area. The relative importance of buoyancy fluxes is highest for halosteric SSH, where amplitudes driven by momentum and buoyancy fluxes are comparable in magnitude (compare Fig 6f and i).

155
Our set of experiments specifically suggests that buoyancy-forced SSH variability in the tropical Pacific is relevant on interannual to decadal timescales. Since the early 1990s buoyancy flux driven SSH in the southwestern tropical Pacific is mostly controlled by halosteric effects with the strongest signals emerging during El Niño or La Niña events. In order to obtain insight 5 https://doi.org/10.5194/os-2021-31 Preprint. Discussion started: 12 April 2021 c Author(s) 2021. CC BY 4.0 License. into the mechanism by which halosteric SSH anomalies are triggered by buoyancy fluxes, how they spread zonally and how they interact with the wind driven SSH anomalies during ENSO events, we examine the particularly strong El Niño event in 160 1997/98.

Case Study: 1997/98 El Niño
SSH anomalies from O025-HC during the El Niño event show the characteristic zonal dipole (Fig. 7a). In contrast to that, O025-B90 shows positive SSH anomalies centered in the central basin at the beginning of 1998. These anomalies propagate 165 westward with phase speeds of about 17 cm s −1 , which is in accordance with phase speeds of Rossby waves derived from linear theory (Killworth et al., 1997). These anomalies counteract the negative wind stress driven anomalies once they reach the western part of the domain by mid-1998 (Fig. 7b). These anomalies are mostly halosteric in nature (Fig. 7c). Areas of high precipitation start to propagate eastward at the beginning of 1997 and reach their easternmost position by the end of the same year where they trigger the SSH anomalies (Fig. 7d). There are only small precipitation anomalies east of 240 • E, which 170 explains the weak halosteric SSH anomalies in the eastern part of the basin described above. We note that the freshwater flux is controlled by precipitation at this latitude and evaporation is negligible (not shown). Thermosteric sea level anomalies are smaller than their halosteric counterpart (Fig. 7e) and heat flux anomalies do not show a coherent structure (Fig. 7f).
A different picture emerges along the equator. The zonal SSH dipole in O025-HC (Fig. 8a) is related to positive anomalies in 175 the eastern basin, which are amplified by buoyancy fluxes (Fig. 8c). The steric anomalies are dominated by their thermosteric component (Fig. 8b,e), triggered by positive heatflux anomalies that evolve throughout the zonal extent of the basin after mid-1997 and early 1998 (Fig. 8f). Precipitation anomalies between mid-1997 and mid-1998 (Fig. 8d) are even stronger at the equator than at 10 • S but steric SSH anomalies in the central basin are weaker and we do not find a pronounced zonal propagation of these anomalies. A possible reason is that sea level and thermocline depth expressions of Rossby waves are stronger 180 off the equator, rendering signals of zonal propagation weaker on the equator. Another possible reason concerns the monthly resolution of the model output which is insufficient to show the adjustment process associated with the eastward propagation of Kelvin waves. Given phase speeds of 2.8 m s −1 for the first baroclinic mode (Gill, 1982), Kelvin waves would take about 45 days to cross the distance between the dateline and the American west coast.

185
In summary, positive SSH anomalies appear to be induced by buoyancy flux anomalies in the tropical Pacific during the peak of the El Niño event. Precipitation anomalies and collocated halosteric SSH anomalies suggest that fresh water fluxes play an important part in this mechanism. Off-equatorial SSH anomalies propagate westward as Rossby waves and counteract the negative SSH anomalies in the western tropical Pacific.
We used a global ocean model to run a set of sensitivity experiments to determine the impact of ocean-atmosphere buoyancy fluxes on interannual to decadal SSH variability in the tropical Pacific. As expected, wind stress variability, associated with the major basin-scale climate modes, is the most important driver of SSH variability in the region, causing the well known zonal dipole with opposing tendencies of SLC in the western and eastern parts of the tropical Pacific. We find that SSH variability due to buoyancy fluxes is generally small but not negligible, with strongest contributions around 10 • N/S and a maximum in Previous studies based on sensitivity experiments (Piecuch and Ponte, 2012;Forget and Ponte, 2015;Meyssignac et al., 2017) tend to agree with the results presented here in that they find a maximum of buoyancy flux forced interannual SSH variability on the order of 2-3 cm in the central tropical Pacific south of the equator. Studies using different methods, including 200 budget analyses of ocean state estimates, identified an impact of buoyancy flux forcing mostly in the western tropical Pacific (Piecuch and Ponte, 2011;Fukumori and Wang, 2013), with amplitudes of the related sea level variability of up to 10 cm.
However, all previous studies were limited to a period of 2-3 decades, i.e., the altimetric period since 1992. By building on model simulations over the last six decades, the present study provides an extended perspective on the vari-205 ability on interannual time scales and also allows a view on the decadal-scale changes. While previous studies mostly focus on heat fluxes we also analysed the importance of fresh water fluxes for steric sea level anomalies.
The results of the sensitivity experiments presented here suggest that, in contrast to the prevailing wind stress driven zonal dipole, buoyancy flux driven sea level variability is zonally uniform across the extent of the basin but varies in phase with the 210 wind stress driven part. More specifically, buoyancy fluxes tend to dampen SSH variability in the western part of the basin but amplify it in the east. The eastern part of the domain is mostly dominated by interannual heat flux variability that increases during El Niño events, thereby increasing the thermosteric SSH and amplifying positive sea level anomalies in the region. Here, buoyancy forcing has no impact on the variability on decadal timescales. Halosteric sea level variability driven by buoyancy fluxes is negligible in this area since its variability is largely confined to the central and western parts of the domain. In contrast, 215 steric SSH variability due to surface buoyancy fluxes in the central and western Pacific is equally affected by thermosteric and halosteric contributions, with an increasing, relative importance of halosteric SSH in the last three decades due to a reduction of thermosteric SSH variability by more than 20%. Overall, buoyancy fluxes contribute to SSH variability on both interannual and decadal timescales, with a higher relative importance for the latter where they dampen the wind stress driven signal by 20%.  Exact areas are shown as boxes 1 and 2 respectively in Fig. 3. Standard deviations for thermosteric and halosteric SSH anomalies are given for the periods from 1965-1990 and 1990-2015. All time series are smoothed with a 12-year running mean.