Atmospheric impacts of local sea surface temperatures versus remote drivers during strong South China Sea winter cold tongue events

Sea surface temperatures (SSTs) in the western part of the South China Sea (SCS) are cooler than in the eastern part in boreal winter, owing to a winter climatological cold tongue (CT). In this study, using a regional atmospheric model configured for the Maritime Continent, we assess the atmospheric impacts of local (or SCS) SSTs versus those from remote drivers (e.g., western tropical Pacific SSTs) during strong CT events with anomalously cool SSTs. In the local run, more rainfall is observed over the eastern SCS, but no significant atmospheric impacts are found over the CT region when SSTs associated with strong CT events are imposed within the SCS while climatological conditions are imposed elsewhere. SCS SST anomalies during strong CT events do not significantly modify the regional wind circulation. The lack of atmospheric response to SSTs over the CT region may be explained by the wintertime mean SSTs (i.e., <27–28°C) over the CT region that are inadequate to trigger deep atmospheric convection, while eastern SCS SSTs are high enough. The increase of anomalous positive moist static energy (MSE) near the sea level over the eastern SCS indicates underlying warm eastern SCS SST anomalies could be influencing positive rainfall anomalies. In the remote run, imposing climatological SCS SSTs but remote SSTs and lateral boundary conditions linked to strong CT events results in cyclonic wind and positive rainfall anomalies over the eastern SCS and Philippines, which are a Matsuno–Gill response to the diabatic heating anomalies over the warm western tropical Pacific SST anomalies. Positive rainfall and cloud cover anomalies associated with the cyclonic wind anomalies are due to the anomalous positive MSE import into the eastern SCS by horizontal advection.


INTRODUCTION
The South China Sea (SCS) is a marginal tropical sea that is part of the Indo-Pacific warm pool. During the boreal winter, the SCS experiences climatologically warm sea surface temperatures (SSTs) that are strongly regulated by surface heat fluxes linked to the prevailing northeast monsoon (Liu et al., 2004;Koseki et al., 2013;Wu and Chen, 2015;Thompson et al., 2016;Xiao et al., 2018;Seow and Tozuka, 2019). An important winter climatological feature in the SCS is an elongated cool SST pool extending from the southern China, Vietnam coast to Peninsular Malaysia (darker gray shades <27 • C, Figure 1a). As such, the western part of SCS is cooler than the eastern part of SCS, and a zonal SST gradient exists across the SCS basin. The cooler SST region is known as the "cold tongue (CT)" (Liu et al., 2004;Varikoden et al., 2010;Koseki et al., 2013;Thompson et al., 2016). Most areas within the SCS with high winter SST interannual variability are located in the CT region and CT SST variations are the key mode of winter variability for the SCS, which implies that the CT is important in regulating the SCS winter climate variability (Seow and Tozuka, 2019;Seow et al., 2021). An extensive number of studies (e.g., Liu et al., 2004;Thompson et al., 2016;Xiao et al., 2020aXiao et al., , 2020bSeow et al., 2021;Wang and Wu, 2021) have discussed the atmospheric and oceanic forcings behind the CT formation and its interannual variability. From November to February, the winter monsoon results in latent heat loss and southward advection of cold water over the western SCS, which are responsible for the CT formation (Liu et al., 2004;Seow and Tozuka, 2019). The northern CT (i.e., 10-20 • N) variability is mainly controlled by latent heat flux, while the southern CT (i.e., 2-10 • N) variability is controlled by horizontal advection (Seow and Tozuka, 2019). Seow et al. (2021) noted that the CT interannual variability is forced by cyclonic wind anomalies aloft ( Figure 1b). Furthermore, the atmosphere over the SCS is strongly influenced by tropical remote forcings, such as the tropical Pacific and Indian Oceans air-sea interactions (Xiao et al., 2020a(Xiao et al., , 2020bSeow et al., 2021).
While there are many studies covering the atmospheric influences on the CT formation and variability, the current understanding of atmospheric impacts from CT SSTs remains sparse. Only Varikoden et al. (2010) and Koseki et al. (2013) have investigated atmospheric impacts of the CT. There are several questions that their studies have yet to address. Varikoden et al. (2010), by observation and reanalysis data, postulated that the zonal SST gradient induced by the CT variability results in less (more) rainfall over the CT (non-CT) region during strong CT events. This is because SST gradients in tropical regions are known to control the rainfall and/or wind circulation changes according to previous studies (e.g., Lindzen and Nigam, 1987;Joseph et al., 2005;Chung and Ramanathan, 2006;Good et al., 2008;Shi et al., 2022). However, observation and reanalysis data analyses alone are unable to confirm if the SST gradient is indeed the cause, as non-SST-related factors like winter monsoon variability might instead be responsible for the zonal contrast in rainfall anomaly. Using regional atmospheric model experiments, Koseki et al. (2013) concluded that the climatological CT SSTs reduce convection and rainfall over the southern CT region and strengthen the monsoon over Java. However, firstly, their model domain covers only the southern part of the CT while excluding its northern part, which covers the latitudinal region of 10-20 • N. Also, their model experiments are evaluating atmospheric impacts of the climatological CT instead of interannually varying CT SSTs. Moreover, both studies have yet to investigate the role of SST threshold in affecting the atmosphere over the CT region. Earlier studies (e.g., Graham and Barnett, 1987;Zhang, 1993;Lau et al., 1997;Neelin et al., 2009;Johnson and Xie, 2010) noted that deep atmospheric convection development probably becomes significant when SSTs surpass a threshold (in the vicinity of 26-28 • C across tropical oceans in the world). Neelin et al. (2009) argued that warmer SSTs result in more surface energy flux, low-level convergence, tropospheric column water vapor and higher chance of intense precipitation, which are all accompanied by deep atmospheric convection. Given that the SCS lies in the tropical region, it makes studying the role of winter SST threshold more relevant in order to explain the atmospheric impacts over the SCS.
Motivated by the above, this study will evaluate the atmospheric impacts from local or SCS SSTs and remote drivers related to strong interannual CT events. A regional atmospheric model is used to achieve the research goals. Descriptions of the data, method to construct strong CT event composites and model experiment designs are detailed in Sections 2, 3 and 4 respectively. Section 5 describes the model performance and atmospheric patterns between different model runs. Sections 6 and 7 discuss the results before concluding in Section 8.

DATA
We use the following reanalysis and observation data to force the atmospheric model and to  (Reynolds et al., 2002). For the outgoing long-wave radiation (OLR) data, the NOAA/Earth System Research Laboratory (ESRL) monthly mean interpolated dataset (Liebmann and Smith, 1996) with a 2.5 • × 2.5 • horizontal resolution from 1982 to 2018 is used. The monthly mean rainfall data with a 2.5 • × 2.5 • horizontal resolution is obtained from the Global Precipitation Climatology Project (GPCP) version 2.3 dataset, which is produced by combining satellite observations over land and oceans, and those of land rain gauges and soundings (Adler et al., 2003). Three-hourly mean single-level variables (i.e., OLR, skin temperature, soil temperature, volumetric soil water layers and snow depth) and interpolated multilevel atmospheric variables (i.e., meridional and zonal winds, geopotential, specific humidity, temperature and vertical velocity) from the ERA5 reanalysis data of European Centre for Medium-Range Weather Forecasts (ECMWF) are used either to validate the model performance or to drive the model. The ERA5 reanalysis has a horizontal resolution of 31 km with 137 hybrid vertical levels up to 0.01 hPa (Hersbach et al., 2020).

STRONG CT EVENT DEFINITION
The CT index is defined in the same manner as in Seow et al. (2021) via the leading mode of multivariate empirical orthogonal function (MEOF) analysis, which captures the CT variability structure well. Mathematical details on the MEOF procedures are explained in Seow et al. (2021). The input data is November-January (NDJ) mean wind and December-February (DJF) mean SST over the SCS (0-20 • N, 100-120 • E). The wind data over land areas are omitted to focus on the air-sea interaction. The long-term linear trend is removed via a least-square fit from all input data prior to the MEOF analysis. The first mode of MEOF captures a strong CT associated with cyclonic wind anomalies at the center of the SCS (Figure 1b). The eigenvalues of the first two modes are well separated according to the standard error criterion by North et al. (1982). Information on the second mode is presented in Figure S1. Hence, we use the first principal component (PC) time series as the CT index in this study (Figure 1d).
A strong CT year is defined when the PC value lies above the 0.5 standard-deviation threshold. A total of 15 out of 37 years are identified as strong CT years. After identifying the strong CT years, composites of model forcing data are prepared, as discussed in the next section.

MODEL AND EXPERIMENT DESIGN
We adopt the Singapore Variable-Resolution Model (SINGV) developed from the United Kingdom Met Office Unified Model Dipankar et al., 2020;Heng et al., 2020). It is based on the Even Newer Dynamics for General Atmospheric Modelling of the Environment (ENDGAME) dynamical core, which uses a semi-Lagrangian advection scheme and a semi-implicit scheme for the temporal discretization of non-hydrostatic, deep-atmosphere equations of motion (Wood et al., 2014). It uses the Arakawa C-grid staggering in the horizontal and a terrain-following hybrid height coordinate in the vertical with a Charney-Phillips staggering. The global correction scheme of Zerroukat (2010) is used to maintain both dry and moist mass conservation. The physics schemes used in the SINGV include: (1) the cloud microphysics scheme of Wilson and Ballard (1999); (2) the prognostic cloud scheme of Wilson et al. (2008); (3) the radiation scheme of Edwards and Slingo (1996); (4) the boundary layer parameterization scheme of Boutle et al. (2014), which blends the one-dimensional planetary boundary layer scheme of Lock et al. (2000) and the three-dimensional Smagorinsky-Lilly scheme (Lilly, 1962). The community land surface model JULES (Best et al., 2011) is incorporated to simulate the exchanges of mass, momentum and energy between the atmosphere, land and sea surfaces. The SINGV has been used in past studies (e.g., Dipankar et al., 2019) to study the mesoscale circulation in the Maritime Continent, and it is currently used as a convective-scale numerical weather prediction system for the western Maritime Continent by the Meteorological Service Singapore Dipankar et al., 2020;Heng et al., 2020).
The model domain covers the entire Maritime Continent, southern China, eastern Indian Ocean, western tropical Pacific and northern Australia, as illustrated in Figure 2. The simulations are free-running except that the following forcing data and initial conditions are prescribed. The SSTs were continuously forced using the daily mean NOAA OISST; soil moisture and atmospheric conditions were initialized using the ERA5 reanalysis; the lateral boundary conditions (LBCs) are imposed using three-hourly mean ERA5 reanalysis wind, geopotential, specific humidity, temperature and vertical velocity data. In this study, the model resolution is set to 8 × 8 km.
We perform three experiments to compare the atmospheric impacts from local SSTs with remote drivers over the SCS during strong CT events. Each experiment is integrated from November 21 to February 28 of the following year. The first 10 days (i.e., November 21-30) are treated as the spin-up period. In the control run, the daily mean climatological SSTs are used to force the model, with the LBCs provided by the three-hourly mean climatology of the ERA5 reanalysis data. The second experiment, called the local run, evaluates the atmospheric impact of SCS SSTs during strong CT events. In the local run, daily mean SSTs of the strong CT event composite are imposed within the SCS but climatological SSTs are used elsewhere to force the model. Like the control run, LBCs are provided by the three-hourly mean climatology of the ERA5 reanalysis data. The third experiment, called the remote run, evaluates the remote atmospheric impact of SSTs outside of  the SCS during strong CT events. In the remote run, daily mean SSTs of the strong CT event composite everywhere within the model domain, but climatological SSTs within the SCS, are used to force the model. The LBCs in the remote run are provided by three-hourly mean strong CT event composites of ERA5 reanalysis data. The experiment specifications are summarized in Table 1, with the DJF mean input SSTs for the local and remote runs illustrated in Figure 2b, c. Figure 3 compares the DJF mean climatology of reanalysis and observation data with model outputs of the control run. The model can realistically simulate the wind climatology for the control run across 200 hPa and 850 hPa levels. Also, the model can simulate the pronounced equatorial band of cloud cover and rainfall associated with the intertropical convergence zone over the Indonesian islands and Borneo. Before analyzing the model results, we first analyze the possible impacts that SSTs have on the atmosphere. We refer to Figure 1b, c that show the SCS key spatial winter variability patterns based on the MEOF analysis. The MEOF result in Figure 1c is expressed as covariances between the PC time series of the first mode with SST and 850 hPa zonal and meridional winds, with the MEOF input data being DJF mean SST and 850 hPa wind over the SCS (0-20 • N, 100-120 • E). The MEOF input data for Figure 1b are similar to those for Figure 1c, except that NDJ mean 850 hPa wind is used. According to Seow et al. (2021), given that the SST data used lag the wind data by one month, Figure 1b shows the atmospheric influences on the SSTs, where the CT variability is controlled by variations in cyclonic wind anomalies. The cyclonic wind anomalies are controlled by tropical remote forcings, one of which is related to the western tropical Pacific SSTs (Seow et al., 2021). In contrast, Figure 1c shows the contemporaneous SST impacts on the atmosphere if there are

F I G U R E 3
December-February (DJF) mean (a) 200 hPa wind and outgoing long-wave radiation (OLR) and (b) 850 hPa wind and rainfall. The wind, rainfall and OLR data are from the ERA5 reanalysis, GPCP rainfall and NOAA OLR respectively. (c, d) As in (a, b) but for SINGV model outputs [Colour figure can be viewed at wileyonlinelibrary.com] any, given that the SST and wind data are of the same period in the MEOF analysis. Despite cooler CT SSTs, Figure 1c depicts cyclonic wind anomalies existing over the SCS, similar to Figure 1b. Thus, we hypothesize that CT SST variations do not significantly modify the wind pattern (i.e. weakening of cyclonic wind anomalies or formation of anticyclonic wind anomalies) over the SCS compared to remote drivers.
Next, the atmospheric impacts from the imposed SSTs and LBCs in the local and remote runs are examined by taking the difference between the respective responses with the control run. Figure 4 shows the 200 hPa, 500 hPa and 850 hPa wind differences, as well as rainfall and OLR differences between the local and control runs. Over most of the Maritime Continent, wind anomalies at the various levels are weak, especially at the mid-and lower-level troposphere ( Figure 4). Within the SCS, while OLR anomalies are generally negligible, positive rainfall anomalies are seen over the eastern SCS ( Figure 4). The lack of wind anomalies over the CT region shows that cool CT SST anomalies do not trigger any regional circulation impacts. In contrast, the positive rainfall anomalies that are seen without significant OLR anomalies over the eastern SCS might be linked to warm eastern SCS SST anomalies and the SCS basin zonal SST anomaly gradient (Figures 2b and 4). Also, positive rainfall and negative OLR anomalies are seen over the entire region south of the equatorial line and over the Philippines, where no SST anomalies are imposed (Figures 2b and 4). These anomalies are likely caused by atmospheric internal variability (e.g., Madden-Julian Oscillation [MJO]-induced variabilities), which are active over the eastern Indian Ocean and northwestern Australia basin during winter (Vialard et al., 2013). Hence, we will verify if the eastern SCS rainfall anomalies could be partly forced by underlying eastern SCS SST anomalies in Section 6. Nonetheless, the model results show that SCS SST anomalies linked to strong CT events do not modify the wind pattern over the SCS and rest of the Maritime Continent, thereby confirming our hypothesis that CT SST variations do not significantly affect the regional wind circulation (cf. Figures 1b, c).
In contrast, a comparison between the remote and control runs reveals cyclonic wind anomalies centered around the Philippines with northeasterly wind anomalies over the SCS north of 10 • N at 500 hPa and northerly anomalies at 850 hPa (Figure 5b, c, e, f). At the 200 hPa level, divergent wind anomalies emanating from the western tropical Pacific and easterly wind anomalies over the SCS are seen (Figure 5a, d). These wind anomalies are accompanied by positive rainfall and negative OLR anomalies over the Philippines and the eastern SCS (Figure 5a, d). The anomalous cyclone, which can exist even when CT SST anomalies are present in observation (Figure 1c), is a Matsuno-Gill response to diabatic heating anomalies over the prescribed warm SST anomalies in the western tropical Pacific (Figure 2c; Matsuno, 1966;Gill, 1980;Wang et al., 2000;Wang and Zhang, 2002;Seow et al., 2021). As LBCs of the strong CT event composite are prescribed, the anomalous cyclone may also be an atmospheric Kelvin wave response to diabatic cooling anomalies over the eastern Indian Ocean associated with strong CT events (Seow et al., 2021), according to atmospheric model experiments by Annamalai et al. (2005). Nevertheless, it is beyond the scope of this study to compare the atmospheric impacts of different remote drivers over the SCS.

CONVECTION TYPE AND MECHANISMS OF RAINFALL CHANGES OVER THE EASTERN SCS
To analyze the convection type associated with rainfall increases over the eastern SCS between the local and remote runs with the control run and whether they forced by underlying eastern SCS SST anomalies, we study the moist static energy (MSE) term over the eastern SCS. The MSE is a thermodynamic potential energy term conserved under moist adiabatic processes and hydrostatic balance, making the MSE a useful quantity to understand energetic processes behind moist convection (Yano and Ambaum, 2017). The MSE budget has been widely used as a framework to analyze tropical convective rainfall under cases like MJO (e.g., Maloney, 2009), convective coupled equatorial waves (e.g., Sumi and Masunaga, 2016) and tropical climate phenomen such as Ningaloo Niño (Zheng et al., 2020). An increase of MSE in an area implies an import of MSE from the surrounding environment, which potentially destabilizes the atmospheric column in that area via heating and moistening processes (i.e., condensation and freezing) that in turn promotes deep convective rainfall. The MSE (h) can be calculated via where c p is the specific heat of air at constant pressure, T is the air temperature, g is the gravitational constant, z is geopotential height, L v is the latent heat of vaporization of water and q v is specific humidity. The time rate of column-integrated MSE, where curly brackets {⋅} denote column-mass integrals, v is the horizontal velocity made up of both zonal and meridional winds, is the vertical velocity, p is the pressure, Q R is the radiative heating rate and R is the residual term representing various processes like subgrid-scale, eddy and ice processes. We only evaluate the SCS area-averaged vertical profiles of horizontal (−v ⋅ ∇ p h) and vertical (− h p ) MSE advection terms of the MSE budget to deduce the convection type associated with rainfall changes over the eastern SCS. Following Hill et al. (2017), we decompose the advection response terms into their dynamic, thermodynamic and covarying components to identify the dominant component that controls the advection response: ⏟⏞⏞⏞⏞⏞⏞⏟⏞⏞⏞⏞⏞⏞⏟ covarying (3b) The covarying components are omitted from the analysis as their magnitudes are negligible. Figure 6 shows the time-mean vertical profiles of MSE advection terms, vertical velocity and moist static stability (or h p ) averaged over the eastern SCS (3-15 • N, 112-120 • E). Climatologically, over the eastern SCS, the horizontal MSE advection below 950 hPa is positive (Figure 6a) due to the moist lower tropospheric winds from the western tropical Pacific converging toward the SCS (Figures 3b, d; 7c). The vertical advection exports the MSE out of the eastern SCS throughout the mid-and upper-level troposphere but imports the MSE into the lower troposphere. The shape of the vertical velocity profile and negative (positive) h p explain the export (import) of MSE at the mid-and upper-level (lower) troposphere that is consistent with the circulation associated with deep convection (Figure 6b-d; Nasuno and Satoh, 2011).
For the local run, the increase of MSE within the eastern SCS is introduced by horizontal and vertical advections at the lower troposphere (i.e., 950-850 hPa; Figure 6e, f). However, both horizontal and vertical advections remove the MSE at the mid-and upper-level troposphere (i.e., 700-200 hPa; Figure 6e, f). From Figure 6e, the horizontal MSE advection change across the whole tropospheric column is mainly dominated by the thermodynamic component. For the lower troposphere, the dominant role of thermodynamics is clearly supported by Figure 7f that shows an increasing west-east MSE anomaly gradient along with climatological easterlies directing MSE into the eastern SCS at the 850 hPa level. In Figures 7d-f and S2a, we note significant negative MSE anomalies over the CT region only at the lower-but not mid-and upper-level troposphere. For the vertical MSE advection response, the dynamic component is dominant across the whole tropospheric column (Figure 6f). The increased time-mean ascent in the local run over the whole troposphere (i.e., more negative vertical velocity) indicates a slight deepening of the convective mode relative to the control run (Figure 6c, g). At the mid-and upper-level troposphere, MSE is removed by the negative horizontal and vertical advection anomaly, leading to small negative changes in h p (Figure 6h) at the mid-and upper-level troposphere (Figure 6d, where negative h p implies that the MSE is increasing with height). At the same time, the positive horizontal advection anomaly is responsible for the concomitant increase in the MSE import in the lower troposphere (Figure 6f). Based on the lower tropospheric negative advection changes, positive rainfall anomalies over the eastern SCS are likely caused by shallow cumulus and congestus precipitating systems (Figure 6e, f). Existing literature (e.g., Ghate et al., 2011;Nasuno and Satoh, 2011;de Roode et al., 2012;Wang and Geerts, 2013;Nuijens et al., 2017;Klingebiel et al., 2021) on shallow cumulus systems support the above hypothesis. According to Nuijens et al. (2017), shallow cumulus and congestus have tops up to 850 and 650 hPa levels respectively, whose altitudes of peak ascent lie in the 850-700 hPa range, as calculated from a radiative-convective equilibrium model. Composite observational vertical velocities of shallow cumulus clouds show peak ascent (approximate height is 1-1.5 km or 900-800 hPa) within the core of cumulus clouds; meanwhile the magnitude of ascent decreases from the shallow cloud base to sea level, where surface subsidence is instead detected in some cases (Ghate et al., 2011;Wang and Geerts, 2013;Klingebiel et al., 2021). Modeling studies (e.g., Nasuno and Satoh, 2011;de Roode et al., 2012) support the above observations, as well as highlight the characteristic decrease of ascent above the top of the cumulus clouds. The vertical velocity change time-mean profile in Figure 6g shows peak ascent anomalies at the 850-800 hPa range, which are consistent with the characteristic vertical velocity profile in cumulus clouds as explained above. Moreover, the height of peak ascent anomalies coincides with maximum positive horizontal MSE advection change (Figure 6e, g), while, by 2D continuity, the peak ascent in cumulus cloud cores is consistent with maximum horizontal convergence (Wang and Geerts, 2013). Nevertheless, the positive h p change in the lower troposphere implies an increase of MSE toward the sea level between local and control runs (Figures 6h and S2b). It can be inferred that positive MSE anomalies are drawn upwards from the sea level (as seen from the ascent anomalies in Figure 6g) and fed into the shallow clouds. All of these suggest rainfall changes over the eastern SCS between the local and control runs are possibly due to a combination of shallow cumulus and congestus clouds, where most of the anomalous MSEs driving the growth of precipitating clouds are forced locally by the warm SST anomalies.
While the MSE over the CT region is not the focus in this section, we briefly describe the causes behind the negative lower tropospheric MSE anomalies (Figure 7f). They are due to the negative lower tropospheric horizontal MSE advection change forced thermodynamically over the CT region ( Figure S3e). Climatological lower tropospheric southeasterlies over the CT region that turn into southwesterlies south of the equator might be drawing MSE out of the CT region toward the anomalous convective-active northwestern Australia, resulting in the observed thermodynamically forced negative horizontal MSE advection change over the CT region (Figures 7c, f, and S3e). Another possible reason is due to the lack of MSE input from the underlying cool CT SST anomalies, as confirmed by Figures S2a and S3h.
For the remote run, the time-mean vertical velocity profile reflects the characteristic top-heavy deep mode of convection (Chen and Yu, 2021; Figure 6c). The profile is also similar to the characteristic composite profile of deep convective clouds generated by models (Nasuno and Satoh, 2011). The increase of MSE within the eastern SCS is introduced by horizontal advection over the entire troposphere and is dominated by the thermodynamic component, particularly in the lower-(i.e., below 900 hPa level), and mid-and upper-level troposphere (i.e., 700-300 hPa; Figure 6i). This is in stark contrast to the horizontal advection features of the local run. In turn, the deep layer of moisture import would promote and support the development of more vigorous deep convection in the eastern SCS (Figure 6k) despite the climatological SSTs being imposed within the SCS domain in the remote run (cf. Figure 2c). Thus, compared to the local run case, there is stronger MSE export (import) in the mid-to-upper-(lower-)level troposphere by vertical advection that is dominated by the dynamic component (Figure 6j). The moisture layer is consistent with Figure 7g-i showing an increasing west-east MSE anomaly gradient along with climatological easterlies directing MSE into the eastern SCS at the 200, 500 and 850 hPa levels. From the positive horizontal MSE advection change across the whole tropospheric column, along with the stronger MSE export by vertical advection and top-heavy vertical velocity change structure, we can infer that rainfall changes over the eastern SCS are likely caused by deep-cloud-precipitating systems (Figure 6i, j).
In contrast to the local run, we note the negative h p change at the lower troposphere (i.e. 950-750 hPa) in Figure 6l. This implies a decrease of MSE toward sea level between remote and control runs (Figures 6l and S2c). This means most of the anomalous MSE behind rainfall changes over the eastern SCS between the remote and control runs do not originate from the sea level. Rather, the anomalous MSE is drawn in by horizontal advection from the surroundings (i.e., western tropical Pacific; Figures 6i and 7g-i).

ROLE OF MEAN SSTS IN REGULATING CONVECTIVE CHANGES
Next, we discuss the reasons for the lack of atmospheric response to the CT SSTs. For this purpose, we analyze the DJF mean SST with OLR and vertical velocity averaged over 200-850 hPa in observational and reanalysis datasets to identify the SST-atmosphere relationship (Figure 8). The SST and OLR are positively correlated with each other over the central SCS between 0 • and 12 • N where most of the CT is situated, but negatively correlated over the western tropical Pacific (Figure 8a), which is consistent with findings of Wu and Kirtman (2007). Also, the vertical velocity and SST are not significantly correlated with each other over the SCS but negatively correlated over the western tropical Pacific (Figure 8b). These suggest that the SSTs (atmosphere) over the SCS CT region (western tropical Pacific) are forced by atmospheric (SST) changes, thereby implying the lack of atmospheric impacts by CT SST changes.
The passive role of SSTs over the CT region is likely due to its low mean climatological SSTs, compared to those over the eastern SCS (i.e., non-CT region) and western tropical Pacific (Figure 9a, b). Such a convection-SST relationship over the SCS and western tropical Pacific is confirmed by the collocated SST and tropospheric mean vertical velocity plots over the SCS and western tropical Pacific (Figure 8c, d). Over the western tropical Pacific, for example, we arbitrarily select the threshold SST of 28 • C. This SST also corresponds to the critical SST where the shape of the probability density function of tropospheric column water vapor (a measure of convection) over the western tropical Pacific transforms from positive-skewed to negative-skewed, as noted by Neelin et al. (2009). From Figure 8d, an SST of 28 • C corresponds to a mean vertical velocity of −0.015 Pa s −1 , which vertical velocity we define as the critical value of deep atmospheric convection. This is consistent with the entire SCS basin, where a mean vertical velocity of −0.015 Pa s −1 corresponds to around 28 • C (Figure 8c). For the western (eastern) SCS, a mean vertical velocity of −0.015 Pa s −1 corresponds to around 27 • C (28 • C; figure not shown). Hence, slight warming in the western tropical Pacific SSTs is adequate in generating anomalous lower tropospheric convergence and more rainfall as seen in our remote-run results, since most of their mean SSTs are high, above 28 • C (Wang et al., 2000;Figure 5). Also, during strong CT events, SSTs over certain parts of the eastern SCS increase above 28 • C, which corroborates the fact that warm eastern SCS SST anomalies Despite that, there are other factors, like large-scale surface convergence induced by the SST gradient, remote impacts of monsoon circulation, tropospheric temperature and moisture availability, besides SSTs, that control the occurrence of deep atmospheric convection (Lindzen and Nigam, 1987;Zhang, 1993;Lau et al., 1997;Neelin et al., 2009). Furthermore, Neelin et al. (2009) argued that higher SSTs only increase the frequency of tropospheric column water vapor greater than the critical vapor threshold, which depends only on tropospheric temperature but not SST, before precipitation and deep atmospheric convection occurs. In other words, tropospheric temperature is another factor other than SST in controlling convection. This explains why SST thresholds for significant deep atmospheric convection vary across different tropical oceans around the world (Zhang, 1993). Also, collocated SST and vertical velocity plots in Figure 8c, d show varying distributions of vertical velocities for each SST value, where we see positive (negative) vertical velocities despite SSTs above (below) the 27-28 • C threshold. Nevertheless, the lack of atmospheric impacts by cool CT SST anomalies and Figure 8 still show that high mean SSTs (i.e. >27-28 • C) are very critical for an increased chance of deep atmospheric convection for the SCS and tropical western Pacific, as supported by Graham and Barnett (1987), Zhang (1993) and Neelin et al. (2009).
We now compare our results with those by Varikoden et al. (2010) and Koseki et al. (2013). Varikoden et al. (2010) had associated the observed zonal contrast in rainfall anomaly with the SCS basin zonal SST gradient with increased rainfall over the non-CT region during strong CT events. While the local run results support observations by Varikoden et al. (2010), we should not neglect the role of the Borneo vortex over the non-CT region that may contribute to the zonal contrast in rainfall anomaly in observations. During strong CT events, the Borneo vortex tends to strengthen, as forced by strengthened winter monsoon winds that transport absolute vorticity and moisture, which can increase the rainfall over the southeastern SCS (Koseki et al., 2014). The strengthened monsoon winds are related to cold surges from high-latitude Siberia that intermittently penetrate into tropical Southeast Asia during boreal winter (Chang et al., 1979;Shoji et al., 2014;Abdillah et al., 2021). These intraseasonal surges are known to increase the northerly wind flow and precipitation over the SCS (Abdillah et al., 2021). Nonetheless, since LBCs of the local run correspond to the climatology, remote influences like the strengthened winter monsoon during strong CT events are eliminated, and we are unlikely to observe any strengthening of the Borneo vortex that affects the non-CT region rainfall. Koseki et al. (2013) conducted two numerical experiments with a regional atmospheric model for the Maritime Continent: one is the control run and the other is the no-CT run, to study the atmospheric impacts of the climatological CT. In their control run, the model was forced with daily mean SSTs from November 2007 to March 2008. In their no-CT run, it is similar to their control run except that the cold tongue is removed throughout the integration by keeping SSTs within the SCS to the monthly mean value of October 2007. Taking the difference between their control and no-CT runs, they found that the lack of CT or warmer SSTs results in cyclonic wind anomalies and increased rainfall over the southern CT region. The cyclonic wind anomalies weaken the northerly climatological monsoon wind over Java. However, the key F I G U R E 10 The 2007/08 December-February (DJF) mean sea surface temperature (SST) input used by Koseki et al. (2013) in running the atmospheric model, which shows the 2007 October monthly mean SSTs imposed in the South China Sea (SCS) (marked by black box) to remove the cold tongue (CT) but daily climatological varying SST imposed elsewhere. 28 • C isotherm is marked by the red contour [Colour figure can be viewed at wileyonlinelibrary.com] reason is because they imposed the October 2007 monthly mean SSTs over the SCS mostly above 29 • C that exceed the SST threshold, as shown by the 2007/08 DJF mean SST input in Figure 10. Such high SSTs are optimal to trigger deep atmospheric convection, leading to significant tropospheric wind and rainfall changes. Thus, when designing atmospheric model experiments, it is important to consider whether the imposed SSTs exceed the threshold required for deep atmospheric convection.
The likely atmospheric impacts of CT SSTs during weak CT events are briefly discussed here, although such an experiment is not performed. The DJF mean SCS SSTs of the weak CT event composite and weakest CT year (i.e., 1997/1998 winter) with those of strong CT event composite are shown in Figure 9b-d. In both strong and weak CT event composites, the CT SSTs are mostly below the SST threshold of 28 • C (Figure 9b, c). However, for the 1997/98 winter CT event, SSTs over the southern CT region are above 28 • C and the SST threshold ( Figure 9d). Hence, while the CT SST increase in the weak CT event composite is still unlikely to exert atmospheric changes, a weak event whose intensity is similar to the 1997/98 winter may instead trigger atmospheric changes. It will be illuminating to conduct an experiment in the future to confirm any atmospheric changes exerted by the CT SST increase during various weak CT events.

CONCLUSIONS
This study evaluates the atmospheric impacts from local (or SCS) SSTs during strong CT events and those by remote drivers on the SCS using a regional atmospheric model. The results are schematically summarized in Figure 11. The model results of the local run show that during strong CT events, negative CT SST anomalies are unable to induce any circulation and rainfall changes but positive SST anomalies over the eastern SCS induce positive rainfall anomalies (Figures 4 and 11a). The anomalous MSE import into the eastern SCS at the lower troposphere by both horizontal and vertical advections, dominated by the thermodynamic and dynamic terms respectively, suggest precipitating shallow cumulus and congestus systems are responsible for the positive eastern SCS rainfall anomalies. Positive h p at the lower troposphere suggests warm eastern SCS SST anomalies are likely to contribute to the rainfall increase (Figures 6h and S2b). Nevertheless, SCS SST anomalies during strong CT events do not significantly modify the regional wind circulation.
In contrast, results of our remote run confirm that remote drivers excite cyclonic wind anomalies and increased rainfall over the Philippines and eastern SCS. Warm SST anomalies leading to diabatic heating anomalies over the western tropical Pacific contribute to the anomalous cyclone via the Matsuno-Gill mechanism. The anomalous positive MSE import from the western tropical Pacific into the eastern SCS across the whole tropospheric column by horizontal advection is responsible for the increase of rainfall and cloud cover associated with the cyclonic wind anomalies by deep-cloud precipitating systems (Figure 11b). The thermodynamic component is dominant in increasing the horizontal MSE advection. Negative h p change at the lower troposphere confirms that the underlying eastern SCS SSTs are not responsible for atmospheric differences over the eastern SCS between the remote and control runs (Figures 6l and S2c).
The lack of atmospheric impacts of negative CT SST anomalies associated with strong CT events in the local run can be explained by the low mean SSTs (i.e., <27-28 • C) over the CT, which are inadequate to trigger deep atmospheric convection. In contrast, SSTs over the eastern SCS in the local run and over the western tropical Pacific in the remote run are high enough (i.e., >28 • C) to trigger atmospheric changes.
While studies (e.g., Joseph et al., 2005;Chung and Ramanathan, 2006;Good et al., 2008) demonstrated that rainfall and monsoon changes in some tropical regions (e.g., India and the Amazon) are driven by the SST gradient, we show that the SST gradient induced by strong CT events in the SCS during winter only exerts more rainfall over the eastern SCS but no rainfall changes over the CT region and wind changes across the SCS. Thus, the lack of atmospheric changes over the CT region (especially expected negative rainfall anomalies) despite cooler SSTs emphasizes that mean SSTs besides the SST gradient are important in controlling atmospheric impacts of SSTs. It is important to consider the SST threshold required for deep atmospheric convection when designing model experiments when comparing our results with those by Koseki et al. (2013).
There are several outstanding issues that should be addressed in the future. Since the regional atmospheric model is forced using a SST composite of strong CT events, the discontinuity in data input may influence the model output. Future studies can conduct each set of experiments repeatedly forced using daily SST means of various strong CT years instead of using a SST composite to confirm the results in this study. Also, the winter atmosphere over the SCS is likely subject to high intraseasonal atmospheric internal variability from the winter monsoon (i.e., cold surges) and MJO, which can affect the model output (Wu and Chen, 2015;Lim et al., 2017;Jiao et al., 2019). Hence, it is worth conducting SST perturbation simulations in future to overcome the variability issue, where each experiment is repeated several times with initial conditions that are slightly perturbed with a normally distributed random noise, like the method used by Seo et al. (2017). A comparison of the atmospheric impacts of different remote drivers (i.e., remote SSTs and LBCs associated with strong CT events) on the anomalous cyclonic winds should be investigated in future, as different LBCs can affect large-scale features generated in regional atmospheric models (Diaconescu et al., 2007;Køltzow et al., 2011). While the key goal of this study is to evaluate the atmospheric response to SSTs linked to strong CT events, future studies may compare the relative roles of the zonal SST gradient and SST threshold in influencing atmospheric changes. One way to compare is through conducting an additional experiment that imposes only CT SST anomalies linked to strong CT events but climatological SSTs over the eastern SCS and comparing with results of the local run. Lastly, experiments forced with SSTs associated with weak CT events should be carried out in future to check for any non-linear responses.

AUTHOR CONTRIBUTIONS Marvin Xiang Ce Seow:
Conceptualization; data curation; formal analysis; funding acquisition; investigation; methodology; project administration; writing -original draft; writing -review and editing. Muhammad Eeqmal Eesfansyah Hassim: Conceptualization; investigation; methodology; project administration; resources; software; supervision; validation; writing -review and editing. Prasanna Venkatraman: Conceptualization; project administration; resources; software; supervision. Tomoki Tozuka: Conceptualization; formal analysis; investigation; methodology; supervision; writing -review and editing. data/gridded/data.gpcp.html. The NOAA/ESRL interpolated OLR dataset was downloaded from https://psl.noaa. gov/data/gridded/data.interp_OLR.html. The ERA5 data were downloaded from the Climate Data Store via Python according to the instructions in https://cds.climate. copernicus.eu/api-how-to. The first author is financially supported by the Research Fellowship of Japan Society for the Promotion of Science (JSPS) under Grant-in-Aid for JSPS Fellows, Grant number 19J20585.