Relative contributions of heat flux and wind stress on the spatiotemporal upper-ocean variability in the tropical Indian Ocean

High-resolution ocean general circulation model (OGCM) simulations are employed to investigate interannual variability of the upper-ocean temperature in the tropical Indian Ocean (20°S–20°N). The seasonal cycle and interannual variability in the upper-ocean temperature in the tropical Indian Ocean in the forced ocean simulation are in good agreement with available observation and reanalysis products. Two further sensitivity OGCM simulations are used to separate the relative contributions of heat flux and wind stress. The comparison of the model simulations reveals the depth-dependent influences of heat flux and wind stress on the ocean temperature variability in the tropical Indian Ocean. Generally, heat flux dominates the temperature variability in the top 30 m, while wind stress contributes most strongly to the subsurface temperature variability below 30 m. This implies that a transition depth should exist at each location, where the dominant control of the ocean temperature variability switched from heat flux to wind stress. We define the depth of this transition point as the ‘crossing depth’ and make use of this concept to better understand the depth-dependent impacts of the heat flux and wind stress on the upper-ocean temperature variability in the tropical Indian Ocean. The crossing depth tends to be shallower in the southern tropical Indian Ocean (20°S-EQ), including the Seychelles-Chagos Thermocline Ridge (SCTR) and the eastern part of the Indian Ocean Dipole (IOD), suggesting the dominance of forcing due to wind stress and the resulting ocean dynamical processes in the temperature variability in those regions. The crossing depth also shows prominent seasonal variability in the southern tropical Indian Ocean. In the SCTR, the variability of the subsurface temperature forced by the wind stress dominates largely in boreal winter and spring, resulting in the shallow crossing depth in these seasons. In contrast, the intensified subsurface temperature variability with shallow crossing depth in the eastern part of the IOD is seen during boreal autumn. Overall, our results suggest that the two regions within the tropical Indian Ocean, the SCTR and the eastern part of the IOD, are the primary locations where the ocean dynamics due to wind-stress forcing control the upper-ocean temperature variability.


Introduction
Prominent warming has been observed throughout the global upper-ocean since the 1950 s (Levitus et al 2009(Levitus et al , 2012). Yet, the rate of warming in the tropical Indian Ocean far exceeds that in the tropical Pacific and Atlantic Oceans (Han et al 2014). Existing studies indicate that atmospheric circulation patterns are closely related to the upper-ocean temperature and sea surface temperature (SST) variability in the tropical Indian Ocean (e.g. Ashok et al 2004, Annamalai et al 2005, Trenary and Han 2012. In the tropical oceans, net heat flux and wind stress are the main drivers of the variability in SST and subsurface ocean temperature on seasonal to interannual timescales (e.g. Behera et al 2000, Rao and Sivakumar 2000, Schott et al 2009, Sayantani and Gnanaseelan 2015. The net heat flux, consisting of short and long-wave radiative fluxes and the latent and sensible heat fluxes, is the key term in the upper-ocean temperature equation (Moisan and Niiler 1998). Not only is the net heat flux the main driver for seasonal variability of upper-ocean temperature in most areas of the tropical Indian Ocean Sivakumar 2000, Cyriac et al 2019), but it also plays a vital role in controlling interannual SST variations associated with the El Niño-Southern Oscillation (ENSO) (Behera et al 2000). In particular, heat flux into the ocean in the tropical Indian Ocean is shown to be enhanced during the El Niño years in association with the induced atmospheric circulation changes (i.e. atmospheric bridge) (Klein et al 1999, Alexander et al 2002, Lau and Nath 2003, Liu and Alexander 2007, often inducing basin-wide warming during boreal winter (Shinoda et al 2004a, Zhong et al 2005. Behera et al (2000) pointed out that the interannual variability of upper-ocean temperature within the Seychelles-Chagos Thermocline Ridge (SCTR) region in the southwestern tropical Indian Ocean, cannot be fully explained by heat flux alone. The SCTR is unique in that the thermocline depth remains climatologically shallow (Schott et al 2009). As a result, a strong coupling between the surface and subsurface temperature fields is observed in the SCTR, enabling interannual variability of SST to be sensitive to the subsurface temperature variability driven by local and remote wind stress forcing (Xie et al 2002). Du et al (2009) suggested that interannual variability of SST in the SCTR is influenced by ocean dynamics within the tropical Indian Ocean and the Rossby wave forced by ENSO , Yu et al 2005, Zhou et al 2008. During El Niño years, SST warming in the SCTR is due to a thickened thermocline depth induced by the downwelling Rossby waves in boreal winter and spring (Xie et al 2002). In contrast, during La Nina years, SST cooling is related to the upwelling Rossby wave in boreal winter (Singh et al 2013).
Moreover, the interannual variability of SST in the eastern tropical Indian Ocean also cannot be solely explained by heat flux variation. Murtugudde and Busalacchi (1999) suggested that wind stress is an indispensable contributor to interannual warming in some regions within the tropical Indian Ocean, such as the Arabian Sea and the southern tropical Indian Ocean. In fact, the well-known Indian Ocean Dipole (IOD) mode arises from enhanced southeasterly monsoon winds along Java and Sumatra, resulting in pronounced interannual SST variability in the tropical Indian Ocean through the Bjerknes feedback (Saji et al 1999). For instance, during the positive IOD years co-occurring with ENSO, the surface cooling in the eastern tropical Indian Ocean results from the negative heat flux anomaly (Tanizaki et al 2017). However, during the IOD years independent of ENSO, the cooling is a response to wind-driven upwelling during boreal summer-autumn (Chen et al 2016, Delman et al 2016. Despite the apparent simultaneous effects of heat flux and wind stress, their relative contribution to interannual variability of upper-ocean temperature in the tropical Indian Ocean has not been systematically quantified. Here, we use a series of highresolution ocean general circulation model (OGCM) simulations to investigate the relative impacts of heat flux and wind stress on the upper-ocean temperature interannual variability in the tropical Indian Ocean, focusing on their spatial distribution, vertical structure, and seasonal variability. In this exploration, we will use the concept of a 'crossing depth' to tease apart and diagnose the relative contributions of buoyancy and wind stress forcing.
The remainder of the paper is structured as follows. In section 2, we introduce the OGCM and analysis methods. Section 3 details the relative contribution of buoyancy and wind stress forcing to the variability of the tropical Indian Ocean. In section 4, we provide a brief summary and discussion for our findings.

Ocean model simulations
Numerical experiments were performed with the Nucleus for European Modelling of the Ocean (NEMO, version 3.1.1; Madec 2008). We used an eddy-active global ocean-sea ice configuration with a spatial resolution of 0.25 • latitude/longitude, ORCA025. The ocean model is coupled to a dynamicthermodynamic sea-ice model, namely the Louvain-La-Neuve sea-ice Model version 2 (LIM2; Fichefet and Maqueda 1997). The model uses a tripolar grid that has a 21-28 km effective resolution in the tropical Indian Ocean. Its vertical coordinate is discretized with 46 height (z) levels, ranging from 6 m at the surface to 250 m for bottom layers. The lowest grid cells apply a partial step topography (Bernard et al 2006). The model has 15 layers in the top 250 m of relevance to the study here. However, it is noteworthy that the results could be dependent on the exact vertical structure compared to another ORCA configuration with 75 vertical levels. Vertical mixing and boundary layer mixing are parameterized by a turbulent kinetic energy scheme (Blanke and Delecluse 1993). Lateral diffusion is performed along isopycnal surfaces.
The initial conditions of the model for temperature and salinity are obtained from climatology (Levitus et al 1998). A 30-yr spin-up was forced with interannually varying atmospheric forcing over the period 1978-2007, followed by a series of OGCM simulations that cover the period 1948-2007. The first four years (1948)(1949)(1950)(1951) of the OGCM simulations might still contain adjustment to the initial spin-up period and thus are excluded from the subsequent analysis. The atmospheric forcing fields are adapted from the corrected global observational dataset based on the National Centers for Environmental Prediction (NCEP)-National Center for Atmospheric Research (NCAR) reanalysis products (Large and Yeager 2009). Specifically, atmospheric variables include 6-hourly wind, air temperature, and humidity; daily shortwave and longwave radiation; monthly precipitation and runoff. In addition to the interannually varying atmospheric forcing, the simulations also make use of the Coordinated Ocean Reference Experiments 2 (CORE2; Griffies et al 2009) 'normal year' forcing that is representative of a climatological year but still includes realistic synoptic variability.
Specifically, we use three simulations and analyze output for the period 1952-2007 to examine the role of wind stress and heat flux on interannual variability of upper-ocean temperature in the tropical Indian Ocean. The reference run is forced with full interannual forcing in both heat flux and wind stress and will be hereafter referred to as ALL. The second run is forced with interannually varying heat flux, but with wind stress from the 'normal year' forcing (i.e. climatology), thereby lacking interannual variability in the wind stress field. This simulation is henceforth called HF. Conversely, the third experiment is integrated with interannually varying wind stress at the surface, while heat flux is from the 'normal year' forcing. This simulation is henceforth called WS. To be able to correct for any remaining spurious model drift, a fourth simulation had been conducted forced with the 'normal year' forcing in both the heat flux and wind stress. Linear trends from this 'normal year' simulation for the period 1952-2007 were then used to remove that component of the linear trends in all three simulations (ALL, HF, and WS) that is due to spurious model drift (rather than trends inherent in the forcing fields). However, it should be noted that these linear trends in upper ocean temperatures in the 'normal year' simulation are very small (See more details in Ummenhofer et al 2013Ummenhofer et al , 2017.
There are a few caveats in our experimental set-up to note before proceeding. First, the experiments are based on an OGCM forced by atmospheric fields, and thus the set-up does not allow for feedback from the ocean to the atmosphere. More importantly, the heat fluxes in the simulations are calculated through bulk formulae using the prescribed surface air temperature/humidity and the simulated SST. This means that even in the WS run, where wind stress is the only forcing that varies interannually, actual heat flux can vary as it reflects the SST response to interannual wind stress forcing. However, since the use of bulk formula represents essentially a restoring of simulated SST toward the prescribed air-temperature on time scales of a month, this unwanted effect might be considered small. Details for the model configuration can be found in Behrens et al (2013). Comparable ORCA simulations at both 0.5 • and 0.25 • horizontal resolution were employed previously to study Indian Ocean variability on seasonal, interannual, to multi-decadal time scale and were found to adequately reproduce the upper-ocean thermal structure of the Indian Ocean (Schwarzkopf and Böning 2011, Ummenhofer et al 2013, Jin et al 2018. A more detailed comparison between the model and observations can be found in the supplementary material (available online at stacks.iop.org/ERL/15/084047/mmedia).

Methods
To investigate the spatiotemporal upper-ocean temperature variability of the tropical Indian Ocean, a concept of 'crossing depth' is introduced. We define it based on the standard deviation of interannual ocean temperature obtained from the three model outputs such that, where Z crossing is the crossing depth and Z argmin(Var(THF(z)−TWS(z))) indicates the depth at which the difference of interannual standard deviation between T HF and T WS reaches the minimum. T HF and T WS represent the oceanic temperature, forced by heat flux and wind stress alone, respectively. z represents the oceanic depth.
Monthly mean fields are averaged over different three-month periods for different seasons, such as December-January-February (DJF) for boreal winter, March-April-May (MAM) for boreal spring, June-July-August (JJA) for boreal summer and September-October-November (SON) for boreal autumn, the choice based on the wind changes of the South Asian monsoon and the seasonal change of heat flux (Yu et al 2007, Schott et al 2009. We confirm that the use of different combinations of months to define seasons, e.g. according to the seasonal variability of upper-ocean temperature variability or thermocline depth produces negligible impacts on our results (figures not shown). The thermocline depth is defined as the depth of the 20 • C isotherms. Mixed layer depth (MLD) is defined as the depth of the water with differences in potential density of less than 0.01 kg m −3 relative to the surface.

Depth-dependent effect of heat flux and wind stress forcing on the interannual variability of the upper-ocean temperature fields
Previous studies have shown that interannual variability of the upper-ocean temperature in the tropical Indian Ocean features a significant seasonal phaselocking related to ENSO and IOD (e.g. Behera et al 2000, Huang andKinter 2002). Thus, we compare the seasonal averages of interannual variability of ocean temperature in the top 180 m averaged over 20 • S-20 • N, 50 • E-100 • E (the whole tropical Indian Ocean) from ALL, HF, and WS. In all four seasons, the interannual variability of the whole tropical Indian Ocean from ALL (figures 1(a), (d), (g), (j)) is similar to that of HF (figures 1(b), (e), (h), (k)) in the top 30 m, while its variability is greatly reduced at increasing depth. On the other hand, WS underestimates the variability seen in ALL in the surface layer, only to become comparable to ALL in the deeper ocean. This highlights the depth-dependent role of wind stress and heat flux forcing in determining interannual variability of upper-ocean temperature in the tropical Indian Ocean. Taking the 28 • C isotherm as an example to demonstrate the response to the near-surface temperature, HF well captures the cold and warm events shown in ALL during all seasons, suggesting the yearround effect of heat flux on interannual variability of ocean temperature above the depth of the 28 • C isotherm. On the other hand, the interannual variability of ocean temperature below approximately 28 • C in WS is in good agreement with that in ALL for the four seasons, as shown by the thermocline depth (red line). Thus, the relative impact of heat flux and wind stress on interannual variability of upper-ocean temperature in the tropical Indian Ocean is depthdependent, and this depth-dependent feature can be found throughout different seasons of the year.
To further quantify the contributions of heat flux and wind stress to the ocean temperature variability, the spatial distributions of the standard deviation of SST and thermocline depth obtained from three simulations are shown in figure 2. The largest values of SST standard deviation near 30 • S, consistent with previous work obtained from SST observational data (Baquero-Bernal et al 2002), can be seen in all three simulations (figures 2(a), (b)). While the SST standard deviation near the equatorial tropical Indian Ocean (5 • S-5 • N, 70 • E-90 • E) in WS is relatively weaker compared to ALL and HF. At depth, in contrast, the standard deviations of thermocline depth in both ALL and WS indicate strong thermocline variability in the southwestern tropical Indian Ocean (5 • S-15 • S, 50 • E-70 • E) (figures 2(e), (f)), while the HF run fails to reproduce this thermocline signal.
We also calculate the probability distribution functions of the interannual variability of SST and thermocline depth within the tropical Indian Ocean to further illustrate the relative contributions of heat flux and wind stress on the interannual variability in the tropical Indian Ocean (figures 2(d), (h)). All three experiments exhibit a similar unimodal pattern of SST variability distribution in the tropical Indian Ocean. However, a closer examination shows that the SST variability distribution in WS is displaced towards smaller values by about 0.2 • C compared to that in ALL in the entire tropical Indian Ocean. On the other hand, the distribution of the SST variability in HF is nearly identical to that of ALL. This indicates that heat flux exerts a stronger control on the interannual SST variability in the tropical Indian Ocean. As for the thermocline depth, in contrast, wind stress is the main driver since the thermocline depth variability in both ALL and WS features the bimodal distribution in the tropical Indian Ocean, which is in stark contrast to a unimodal distribution seen in HF. From this analysis, we deduce clear that the influence of heat flux and wind stress on the upper-ocean temperature variability in the whole tropical Indian Ocean is distinct with depth and potentially separable.
We also compare the depth-time diagrams of the ocean temperature from ALL, HF, and WS within the tropical Indian Ocean (20 • S-20 • N) in figure 3. Again, taking the 28 • C isotherm (thick black line) as an example, HF captures the two cold events in the 1950 s and 1970 s when the isotherm above the 28 • C isotherm outcrops, while WS fails to reproduce such strong events in the tropical Indian Ocean. However, the variability of ocean temperature from WS is almost identical to ALL below the 28 • C isotherm, indicating that the wind stress is the main driver of the interannual temperature variability at the subsurface. This finding is consistent with previous studies showing that interannual variability of the thermocline depth (20 • C; thick red line in figure 3) was found to be entirely attributed to wind stress Behera 2005, Yu et al 2005). The pronounced spatial differences between SST and thermocline standard deviations in figure 2 also motivate us to examine the depth-dependent influence of heat flux and wind stress in sub-areas of the tropical Indian Ocean, including the SCTR (12 • S-5 • S, 50 • E-75 • E), the western tropical Indian Ocean (5 • S-5 • N, 50 • E-75 • E), and the eastern tropical Indian Ocean (10 • S-5 • N, 80 • E-100 • E). In the SCTR (figures 3(d)-(f)), interannual variability of ocean temperature above the 28 • C isotherm in ALL is better reproduced by HF, while the stronger interannual variability of the subsurface temperature is in good agreement between ALL and WS. In the western tropical Indian Ocean (figures 3(g)-(i)) and eastern tropical Indian Ocean (figure 3(j)-(l)), although the impact of heat flux on interannual variability of ocean temperature does not reach below the depth of the 28 • C isotherm, the simulated temperature between HF and ALL is reasonably consistent in the top 30 m. The interannual variability of ocean temperature below the depth of 28 • C isotherm in WS is consistent with that in ALL. Although there are different physical mechanisms between the northern (EQ-20 • N, 50 • E-100 • E) and the southern (20 • S-EQ, 50 • E-100 • E) tropical Indian Ocean, for example, there is a meridional SST gradient between north and south tropical Indian Ocean due to different wind direction induced by both ENSO and IOD (Wu et al 2008), the depthdependent feature can also be seen in these two areas (figures not shown). Thus, this depth-dependent feature of the impact of heat flux and wind stress on

Crossing depths: attribution of the interannual temperature variability in the tropical Indian Ocean
Such a depth-dependent effect of surface forcing implies that a transition depth would exist, demarcating the regime of heat flux controlled temperature variability near the surface from the deeper depth ranges where the forcing due to wind stress dominates the temperature variability. Figure 5 shows the distribution of the crossing depth in the tropical Indian Ocean. The shallower crossing depths are located in the southern tropical Indian Ocean associated with a shallower thermocline depth, which is consistent with previous studies (Yokoi et al 2008, Schott et al 2009. In other words, in the southern equatorial tropical Indian Ocean, heat flux dominates interannual variability of upper-ocean temperature only near the surface, while wind stress plays a more important role at the deeper depth. In the northern tropical Indian Ocean, the crossing depths are deeper than they are in the southern equatorial tropical Indian Ocean, indicating that both heat flux and wind stress contribute to interannual variability of upper-ocean temperature. However, the crossing depth becomes very deep south of 15 • S, where the wind stress impacts both the SST through wind-driven vertical mixing (Behera and Yamagata 2001). We note that the concept of a crossing depth is meaningful only in regions where the effects of wind stress and heat flux can be quasilinearly separable such as in the tropics and where the thermocline remains relatively shallow. We expect that it does not offer particularly useful information in regions where wind stress also drives significant irreversible mixing that affects the upper ocean temperature fields, such as a layer as in the higher latitudes. Thus, we focus our analysis based on the crossing-depth in two tropical regions, the SCTR and the eastern part of the IOD. These two regions are interesting because SCTR has the shallowest crossing depth, and the eastern part of the IOD supports the most significant interannual ocean temperature variability in the tropical Indian Ocean (Qu andMeyers 2005, Chen et al 2016). Figure 5(b) presents the annual cycle of the crossing depth in the SCTR and eastern part of the IOD. The crossing depth in the SCTR has a clear annual cycle, with the maximum in September at 58 m and a minimum of 23 m in March. This seasonal variation is very similar to the seasonal cycle of SST in the SCTR (Yuan et al 2018). From March to August, the crossing depth deepens in the SCTR. The shoaling time of the crossing depth in the SCTR starting from September is in good agreement with the seasonal phase locking impact of ENSO on the SCTR when the SST variability in the SCTR is affected by the ENSO induced anomalous heat flux (Shinoda et al 2004b, Nagura and Konda 2007, Burns and Subrahmanyam 2016. The seasonal variability of the crossing depth in the SCTR seems one month lagging than those of MLD. For example, the deepest MLD is in August, while the deepest crossing depth exists in September. The reason is that MLD is defined locally as functions of stratification and vertical shear, which is determined with the local conditions. When the heat flux is cooling or wind stress is strong, it deepens immediately. The crossing depth gets deeper but slowly in the SCTR due to the ocean dynamics (e.g. the downwelling Rossby waves). In contrast, the seasonal variation of the crossing depth in the eastern part of the IOD is relatively small, ranging between 30 m and 48 m.
An alternative way to examine the seasonal variability of the crossing depth would be to compare seasonally averaged vertical profiles of the standard deviations of interannual ocean temperature spatially averaged over the SCTR and the eastern part of the IOD (figures 5(c)-(j)). In the SCTR, the depth of the crossing point between HF and WS is in the top 30 m of the water column during boreal winter and spring while it drops to below 30 m during boreal summer and autumn. In the eastern part of the IOD, the transition depth is below 30 m during boreal winter, spring and summer but rises to above 30 m in autumn. Furthermore, corresponding to the shallower crossing depth in boreal winter and spring, interannual variability of the subsurface temperature affected by wind stress is prominent within the SCTR with its standard deviations exceeding 2 • C (figures 5(c), (d)). Similarly, the large interannual variability of the subsurface in the eastern part of the IOD is seen along with the shallower crossing depth in boreal winter and autumn, but its magnitude is weaker (1.5 • C standard deviation, figures 5(i), (j)). Therefore, the crossing depth can separate the relative role of heat flux from wind stress in ocean temperature variability, as well as potentially be an indicator of the intensity of the subsurface variability. Finally, the contributions of HF and WS to ALL in both the SCTR and the eastern part of the IOD during different seasons are shown in figure 6. The contribution of HF to ALL above the crossing depth almost reaches 100% in different seasons in both the SCTR and the eastern part of the IOD while WS contributes most to ALL below the crossing depth. It is noteworthy that wind stress almost entirely controls the interannual variability of upper-ocean temperature in the SCTR during boreal winter and spring, which is consistent with the subsurface-to-surface feedback (Xie et al 2002). Thicker thermocline depth in the SCTR is due to the downwelling Rossby wave and anomalous easterlies along the equatorial Indian Ocean induced by El Niño, which in turn, results in a warmer ocean surface (Xie et al 2002, Yokoi et al 2008. This strong control of wind stress leads to a shallower crossing depth in the SCTR during boreal winter and spring (figures 5(c), (d)). Shallower crossing depth can also be observed in the eastern part of the IOD during boreal winter and spring due to the dominant impact of wind stress on the interannual variability of upper-ocean temperature. But this wind stress is induced by the easterly wind anomalies (Sayantani and Gnanaseelan 2015).

Summary and discussion
In this study, we diagnose the relative contributions of heat flux and wind stress to the interannual upper-ocean temperature variability in the tropical Indian Ocean by analyzing three sets of highresolution OGCM simulations that differ only in their surface forcing. ALL, forced by both interannually varying heat flux and wind stress, reasonably well reproduces the observed mean and variability of the vertical structure of ocean temperature in the tropical  Indian Ocean when compared to the WOA climatology and SODA reanalysis data.
Comparison of the sensitivity simulations, HF and WS, against the ALL reveals the depth-dependent impacts of heat flux and wind stress on the ocean temperature variability in the tropical Indian Ocean. In general, heat flux dominates the variability of ocean temperature in the upper 30 m in much of the tropical Indian Ocean, while its impact is quickly subdued below 30 m. On the contrary, wind stress is the decisive driver of temperature variability below 30 m. This suggests that there is a transition depth at which the main driver of the upper-ocean interannual temperature variability shifts from heat flux to wind stress. We investigated this crossing depth to better understand the driving forces of the spatial pattern and seasonality of the upper-ocean interannual temperature variability.
Our results show that the variability of ocean temperature in the top of the crossing depth is deemed mainly a response to heat flux, while the one below is induced by wind stress either locally or remotely. The shallowest crossing depth is found in the southern tropical Indian Ocean, including the SCTR and eastern part of the IOD, indicating that these are the regions where wind-stress forcing exerts a strong control on upper-ocean temperature variability. The SCTR is characterized by a shallower crossing depth along with intensified variability of the subsurface temperatures driven by wind stress, because of the anomalous wind stress and Rossby wave-induced upwelling forced by ENSO (Xie et al 2002, Yokoi et al 2012, Burns and Subrahmanyam 2016 and the Interdecadal Pacific Oscillation (Jin et al 2018). On the other hand, the crossing depth in the eastern part of the IOD shoals during boreal autumn and winter while it deepens in late boreal spring and summer.
The crossing depth in the SCTR also exhibits a clear seasonal variability with shoaling during boreal winter and spring and deepening during boreal summer and autumn. The strong annual cycle in the crossing depth in the SCTR is in agreement with results from previous studies (Hermes and Reason 2008, Yokoi et al 2008, which associated the strong annual cycle in the upper-ocean temperature in the SCTR with upwelling due to wind stress forcing related to the trade winds and Rossby waves. Furthermore, recent high-resolution modeling studies highlight the role of ocean dynamics in setting the upper ocean temperature anomalies and variability in the tropical oceans (Roberts et al 2017, Small et al 2020. For example, Roberts et al (2017) adopted the Pearson correlation and a fraction of variance (FOV) skill score to diagnose the drivers of ocean variability; they showed that up to 80% of the total interannual variance of the mixed layer temperature is accounted for by the ocean heat transport convergence in the SCTR (their figure 8). We calculate the minimum vertical difference of the interannual ocean temperature standard deviation in the sensitivity experiments to distinguish the relative impacts of heat flux and wind stress. Our finding that the wind stress dominates the subsurface temperature variability in the SCTR through downwelling Rossby waves forced by wind stress, which contributes to the shoaling of the crossing depth, is consistent with their result.

Data statements
• The data that support the findings of this study are available from the corresponding author upon reasonable request.