Surface circulation in the Gulf of Thailand from remotely sensed observations: seasonal and interannual timescales

. The Gulf of Thailand (GoT), a shallow semi-enclosed basin located in the western equatorial Paciﬁc, undergoes high wind variabilities on both seasonal and interannual timescales . The study reveals that the surface current at different regions responds to the wind variability differently (cid:58)(cid:58)(cid:58) that (cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58) produce (cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58) complex (cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58) surface (cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58) circulation. The local Ekman pumping modiﬁes sea level in the northern GoT, while remote wind forcing inﬂuences sea level variability at the GoT western boundary (cid:58) , potentially through the coastal trapped Kelvin waves. The importance of wind-driven (cid:58)(cid:58) the (cid:58) Ekman current on the ageostrophic current is 5 also important; higher correlation between the ageostrophic and (cid:58)(cid:58) the (cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58) higher (cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58) inﬂuence

Previous observational and numerical studies show that circulation in the Gulf of Thailand varies seasonally (e.g., Yanagi et al., 2001;Buranapratheprat et al., 2008;Saramul, 2017;Buranapratheprat et al., 2002;Aschariyaphotha et al., 2008;Saramul and Ezer, 2014).A series of 6 hydrographic cruises over October 2003 -July 2005 in the upper GoT (uGoT; north of 12.5°N) suggests the overall cyclonic circulation during the northeast monsoon (November -February; Buranapratheprat et al., 2008) ::::::::::::::::::::::::: (Buranapratheprat et al., 2008) in agreement with numerical studies that accounts :::::: account : for tidal forcing, bottom friction, and river runoffs (Buranapratheprat et al., 2002;Saramul and Ezer, 2014).The numerical studies also suggest the dominance of contrasting anticyclonic circulation in the uGoT during the southwest monsoon(May -August).Still, both numerical simulations are forced by spatially uniform reanalysis wind products which is likely not representing ::::: likely ::: do ::: not :::::::: represent the actual wind field over the region (Yanagi and Takao, 1998).A study based on numerical simulation forced by spatially varying wind (Buranapratheprat et al., 2006) emphasizes the importance of both zonal and meridional wind gradient ::::::: gradients : on the circulation over the uGoT.For example, the development of an anticyclonic circulation to the north of 13°N during the southwest monsoon, observed by Buranapratheprat et al. (2002) and Saramul and Ezer (2014), is highly dependent on the intensity of wind at the south or east of the uGoT.Results from hydrographic surveys in May 2004 and July 2005 do not show a clear dominant circulation pattern during this period (Buranapratheprat et al., 2008).High spatial-resolution coastal radar of the monthly-mean surface current during both of the southwest (June 2015) and northeast monsoon (February 2015) reveals : a complex circulation pattern in the uGoT, although the circulations during the two seasons are not distinctly different (Saramul, 2017).The circulation based on the coastal radar suggests an overall cyclonic circulation in the northern part of the uGoT (north of 12.8°-12.9°N)and an anticyclonic circulation in the southern part (south of 12.8°-12.9°N)during both monsoon seasons.
South of 12.5°N, various observational and numerical studies were conducted (Wyrtki, 1961;Yanagi and Takao, 1998;Aschariyaphotha et al., 2008;Sojisuporn et al., 2010); however, the findings are not quite consistent due to the different studied periods and spatial resolution being considered.During the southwest monsoon, altimetry-based observation over the 1995 -2001 period (Sojisuporn et al., 2010) shows that the circulation intensifies at the rim of the GoT with : a : strong southward current within 1°of the GoT western boundary and westward current to the south of the uGoT yielding a cyclonic circulation.
At the southeastern entrance, satellite altimetry indicates an outflow into the SCS (Sojisuporn et al., 2010).Numerical simulation assimilating measurements from the NAGA expedition in 1959-1960 (Yanagi and Takao, 1998) shows similar results, except the presence of strong northwestward flow in the mid-basin which yields a cyclonic circulation to its west and an anticyclonic circulation to its east.In contrast, findings from Princeton Ocean Model (Aschariyaphotha et al., 2008) indicate : a : strong southeastward flow in the mid-basin and the dominance of an anticyclonic circulation over the GoT with an outflow at the southeastern entrance during the southwest monsoon.Note that the study by Aschariyaphotha et al. (2008) allows inflow and outflow from the lateral boundaries while that by Yanagi and Takao (1998) does not which could contribute to the discrepancy.
Still, the circulation pattern found by Aschariyaphotha et al. (2008) resembles the GoT surface velocity surveyed during the NAGA expedition (Wyrtki, 1961).
During the northeast monsoon, the altimetric observations and numerical simulations suggest the dominance of an anticyclonic circulation at the rim of the GoT and an inflow at the southeastern entrance (Yanagi and Takao, 1998;Aschariyaphotha et al., 2008;Sojisuporn et al., 2010).Circulation in the GoT interior is quite complex and findings from the studies do not necessarily agree.The NAGA expedition, however, shows that cyclonic circulation prevails during the northeast monsoon (Wyrtki, 1961).Still, an inflow is present at the southeastern entrance in agreement with the observational and numerical studies.The inflow is found to reach the bottom at the GoT western boundary during the spring monsoon transition as : a ::::::::::: hydrographic :::::: survey ::::: shows ::: the :::::::: presence :: of cold and saline water originated ::::::::: originating in the SCS is observed there by a hydrographic survey :: at :: the :::::: region : (Yanagi et al., 2001).
Although previous studies have recognized the role of monsoon winds on seasonal variability of the GoT circulation (e.g., Yanagi and Takao, 1998;Aschariyaphotha et al., 2008), the associated dynamics are not well understood.Therefore, this study aims to examine :: the : seasonal variability of surface circulation in the GoT and the associated dynamics by investigating the influence of geostrophic current and wind-driven Ekman current using remotely-sensed observations.The mechanisms that set up the geostrophic flow will also be discussed.In addition, interannual variability of the GoT circulation will be examined to understand the effect of ENSO and the IOD on the circulation pattern.

Datasets
To examine circulation pattern in the GoT, the gridded Ocean Surface Currents Analyses Real-time product (OSCAR; Bonjean and Lagerloef, 2002) between 8°and 14°N, 99°and 105°E is considered (Figure 1).The product is calculated based on satellite sea surface height, wind, and water temperature from both remotely-sensed and in-situ measurements, e.g.drifters, moored, and shipboard measurements : , : etc.The resulting current is an average in the upper 30 m of the water column.Therefore, OSCAR current represents the total current (sum of the geostrophic and ageostrophic currents) over the GoT.The gridded OSCAR product has a resolution of 1/3°(36-37 km in the GoT) with a temporal resolution of 5 days available from 1992 to 2020.To validate the OSCAR velocity over the GoT, the monthly average velocity maps in February 2015 and June 2015 are compared to tide-removed surface currents from high frequency :::::::::::: high-frequency : radar system shown in Saramul (2017).
Generally, OSCAR velocity exhibits : a similar circulation pattern to the coastal-radar velocity : , particularly in February 2015.
Still, : a : much more complex circulation is observed in the coastal-radar velocity due to its much higher spatial resolution.The difference between OSCAR velocity and high-frequency coastal-radar velocity is the largest in the uGoT; as the region is quite small and shallow (Figure 1), the spatial resolution provided by the OSCAR products might not be sufficient to resolve the circulation there.
To examine the effect of wind-driven Ekman current on the GoT circulation, the gridded surface vector winds Version 2 Cross-Calibrated Multi-Platform (CCMPv2) obtained from Remote Sensing Systems are used.The CCMPv2 wind product, available from July 1987 to December 2019, has a resolution of 1/4°with a temporal resolution of 6 hours (Wentz et al., 2015).
The Ekman current (u e and v e ) at each depth (z) is calculated following Alberty et al. (2019): where τ is wind stress, d is thickness of the surface Ekman layer defined as 2A |f | with A being a function of wind speed (|U|; A = 8 × 10 −5 |U| 2.2 ).In addition, wind stress curl (∇ × τ = ∂τ y ∂x − ∂τ x ∂y ) is also calculated.
The weekly sea surface temperature averaged over the Niño 3.4 box (hereafter referred to as Niño3.4)provided by the National Oceanic and Atmospheric Administration (NOAA) is used to indicates :::::: indicate ENSO conditions (Trenberth, 1997).
To assess the IOD conditions, the Dipole Mode Index (DMI) is used.The weekly DMI calculated from sea surface temperature in the tropical Indian Ocean is calculated and provided by the NOAA/ Earth System Research Laboratory (Saji et al., 1999;Black et al., 2003).

Methodology
3.1 ::::::: Complex ::::::::: empirical :::::::::: orthogonal ::::::: function To determine the dominant pattern and the associated temporal variation of the surface current in the GoT, the complex empirical orthogonal function (CEOF) is utilized.The CEOF is similar to the empirical orthogonal function (EOF) which is suitable for analysis of a dataset with both spatial and temporal variation (e.g., Weare et al., 1976;North et al., 1982).The EOF technique decomposes the data matrix that has its mean removed (X) into orthogonal EOF modes (U ) that display spatial patterns.
Each mode corresponds to a time series known as the principal component (PC) that demonstrates :: the : temporal variation of that EOF mode; the PC identifies when and how intense each EOF pattern occurs.The collection of the PCs forms an orthogonal matrix (V ).The matrix decomposition is done as follow :::::: follows: where the superscript T denotes the transposition of a matrix.Each EOF and PC explain different fractions of the dataset variance (variance of the i-th mode is calculated as ); the first EOF mode shows the most dominant pattern and the subsequent modes account for : a smaller fraction of the variance by the mathematical construction.When the technique is applied to vector quantities, e.g., velocity, the CEOF is often adopted, where each vector is transformed into a complex number (e.g., Kundu and Allen, 1976;Klinck, 1985).In this study, the velocity vector with the time-mean removed (u) is decomposed as where u is the zonal velocity, v is the meridional velocity, and i is √ −1.Applying the same EOF technique (Eq.5) to the complex number, the resultant PC is complex where its magnitude represents :: the : temporal fluctuation of the corresponding CEOF.The phase calculated as ::: the arctan of the imaginary part divided by the real part represents the direction that the CEOF mode has to rotate (positive clockwise).

Circulation in the Gulf of Thailand
The mean and variance of OSCAR surface velocity is ::: are calculated over the 2014 -2019 period (Figure 2).Strong mean flow is observed near the northern and western boundaries of the GoT and at the southeastern entrance.The mean current is northward along the western boundary to the south of 12°N and westward :::::::::::: southwestward : at the southeastern entranceproducing an outflow along the eastern boundary and an inflow along the western boundary.The mean circulation in the GoT interior consists of a few weak eddies.The mean circulation pattern from OSCAR products generally agrees with the satellite-derived geostrophic current (color contour in Figure 2a), except in the uGoT where OSCAR products are present at only six locations.
Also, as the uGoT is shallow and enclosed by land on the western, northern, and eastern sides (Figure 1), OSCAR products over the region could contain : a substantial error.Thus, discussion regarding OSCAR velocity over the uGoT will be omitted.High variance of the surface circulation is found along the western boundary of the GoT, approximately between 9.5°and 11.5°N, with most of the variance associated with meridional velocity (Figure 2b).Variance of the ADT is also high along the western boundary indicating the influence of geostrophic flow (Eq. 1 and 2), particularly between 9°and 10.5°N.At the southeastern entrance, high variance is also observed in both OSCAR velocity and ADT with the highest velocity variance observed at the southeastern part of the observing domain and highest ADT variance observe a bit farther north, at ∼9°N.
4.1 Seasonal circulation in the Gulf of Thailand

Overall description
As suggested by the dominating circulation pattern calculated based on :: the : CEOF, more than one quarter ::::::::: one-quarter : of the variance in the GoT current can be simply explained by an annually reversing circulation that follows the monsoon seasons (Figure 3a-c).Thus, :: the : monthly mean current over the 2014 -2019 period is calculated; the monthly current reveals : a : circulation pattern generally consistent to :::: with that shown by CEOF1 (Figure 4a, c, e, g), particularly in June (representing the southwest monsoon) and December (representing the northeast monsoon).The pattern describes the circulation with strong current at three main regions which are the western boundary of the GoT, the interior of the GoT, and the southeastern entrance of the GoT.In the interior of the GoT, an anticyclonic circulation is present (centered at ∼10.5°N, 101.5°E) during the southwest monsoon (Figure 4b, c).Strong : A :::::: strong : southward current is observed along the western boundary and : a : strong southeastward current is observed at the southeastern entrance suggesting an outflow into the SCS at the surface.The circulation pattern generally reverses its direction during the northeast monsoon, consistent with : a previous surface current observation (Saramul, 2017).The monthly mean surface current hints : at : the connection between the currents along the western boundary and that at the southeastern entrance (Figure 4a, c, e, g).Thus, complex correlation analysis (Kundu, 1976) between a :::::::: complex ::::::::: correlation :::: (Eq.:: 7) ::::::: between ::: the current along the western boundary and that over the GoT is performed using complex number constructed from velocity vector (u, Eq. 6) to understand the dynamics associated with the strong western boundary current particularly if it is related to the current at the southeastern entrance.The complex correlation coefficient (R) is computed as , where * denotes complex conjugate.The resultant R is a complex number where its magnitude describes how the magnitude of the two time series covary and its phase (arctan of the imaginary component divided by the real component) describes the angle between the two vector time series in order to achieve the highest correlation.On timescales longer than 30 days, currents along the southern boundary of the domain (south of 8.5°N) gives ::: give : a : higher correlation to the western boundary current (9.0°-11.5°N,99.5°-100.2°E)compared to the rest of the GoT (Figure 5); the correlation is higher than 0.25 with the highest value of 0.57 at the entrance.The correlation between the western boundary current and that along the southern boundary of the domain is significant with 95% confidence as determined by a non-parametric method (Sprent and Smeeton, 2007) where correlation coefficients are computed repeatedly (5000 times) using both of the time series that are randomly rearranged.The significant correlation strongly suggests : a : connection between the GoT western boundary current and the GoT inflow/outflow at the southeastern entrance through a passage to the south of ∼8.5°N.
During the spring monsoon transition (represented by March), the current resembles that during the northeast monsoon; a westward flow at the southeastern entrance is observed.The cyclonic circulation in the GoT interior is still present, but weak.However, the northward flow along the western boundary is stronger and wider compared to that during the northeast monsoon; width of the northward current reaches :::::: extends ::::: more :::: than :: 80 ::: km :::::::: offshore :: to ∼100°E (Figure 4a, d).Similarly, the GoT circulation in September, representing the fall monsoon transition, resembles that during the southwest monsoon despite the weak anticyclonic circulation in the interior.The circulation pattern during the monsoon transitions shows the dominant circulation pattern captured by both CEOF1 and CEOF2 (Figure 3) reflecting the influence of both monsoon winds and :: of the current connecting to the SCS (Figure 5).

Geostrophic and ageostrophic component
Satellite altimetry is used to estimate geostrophic components of the surface circulation over the GoT.Although the altimetry may include short-period contribution during the satellite over-pass, geostrophic velocity calculated from the altimetry is found to be reasonably close to the observation (Yu et al., 1995).In the Mediterranean Sea, the altimetry-derived geostrophic velocity is generally smaller than the drifter observations (Poulain et al., 2012); the error increases with the geostrophic velocity with an error of 7-17% at the velocity of 1.5 m s −1 (Kubryakov and Stanichny, 2013).
The satellite ADT is linearly interpolated onto the 1/3°OSCAR grid (Figure 4).Magnitude of ::: The ::::::::: magnitude :: of :::: the geostrophic current is generally comparable to that of the total current, although their directions are not perfectly aligned.Large geostrophic velocity is observed along the western boundary of the GoT and at the southeastern entrance.The geostrophic velocity is weak in the GoT interior.The root-mean-square (rms) difference between the total and the geostrophic current (i.e. the estimated ageostrophic current) ranges from 0.04 to 0.11 m s −1 with the largest difference observed along the western boundary of the GoT; however, the rms difference is reasonably proportional to the speed of the total current there.Complex :: A ::::::: complex correlation between the total and geostrophic current is calculated to determine the correlation and phase relationship between the two velocity fields (Figure 6a, b; :::: Eq. : 7).Over the entire basin, the rms correlation coefficient is 0.70 (Figure 6a).
The correlation between the total and geostrophic current is higher along the southern boundary of the observing domain, at 8°N . Along the northeastern boundary, geostrophic current only explains a small fraction of the total :::::: current variance (14 -34%).
The rms correlation coefficient over the interior is 0.71 and that over the western boundary region is 0.67.The strong correlation indicates the dominance of geostrophic circulation over the GoT.Phase relationship shows the direction that the geostrophic current has to rotate to align with the direction of the total current where positive denotes counterclockwise rotation.A negative phase relationship is found roughly between 10.5°and 12°N while a positive relationship is found to the south of 10.5°N (Figure 6b).With the presence of anticyclonic circulation centered at 10.5°-11°N during the southwest monsoon and fall monsoon transition, the phase relationship requires : a : southeastward ageostrophic flow.In contrast, the phase relationship implies : a northwestward ageostrophic current during the northeast monsoon and spring monsoon transition when cyclonic circulation dominates the GoT interior.As the prevailing monsoon wind is southwesterly during the southwest monsoon and northeasterly during the northeast monsoon, the resulting wind-driven Ekman current aligns with the direction of the ageostrophic flow in the respective seasons.Therefore, the Ekman current is calculated from the CCMPv2 wind (Eq. 1 and 2) and : 3 ::: and :: 4) :: to : compared with the ageostrophic current.
Wind-driven Ekman current averaged in the upper 30 m of the water column (or to the seafloor where the water column is shallower than 30 m) has a distinct seasonal cycle being :: the : strongest during the northeast monsoon and the weakest during the spring monsoon transitions (Figure 4b, d, f, h).During the northeast monsoon, :: the : speed of the Ekman current exceeds 0.06 m s −1 almost everywhere except between 10°-12°N along the eastern coast.The weak Ekman current is likely due to the presence of the Cardamom Mountains between 10°-13°N at the coast of Thailand and Cambodia that blocks the northeasterlies (Li et al., 2014).The strongest current is at the southeastern entrance with the speed of 0.1 m s −1 transporting water into the GoT (Figure 4h).The Ekman current at the southeastern entrance still transports water into the GoT during the spring monsoon transition, although the speed decreases due to the weakening of the monsoon wind (Figure 4b).During the southwest monsoon and fall monsoon transition, the Ekman current is quite uniform over the entire GoT; the flow is southeastward producing an outflow into the SCS (Figure 4d, f).Magnitude ::: The ::::::::: magnitude of the Ekman current is similar to the ageostrophic current in the GoT interior and the southeastern entrance, but smaller along the boundaries.
Complex correlation between the ageostrophic component and wind-driven Ekman current is calculated to examine the contribution of wind-driven current on the ageostrophic circulation (Figure 6c).Higher correlation is found over the southern part of the domain with the highest correlation coefficient of 0.54 reflecting that up to 29% of the ageostrophic circulation is winddriven.Over the western boundary region, the correlation coefficient between the wind-driven current and the ageostrophic current is 0.40.Similar :: A :::::: similar : value of correlation coefficient is observed over the interior of the GoT (R = 0.41).Phase relationship indicates the direction that the Ekman current has to rotate to align with the direction of the ageostrophic current; it is small overall with most values between − π 4 and π 4 (Figure 6d).When ageostrophic current is entirely driven by wind stress, the phase relationship is zero.Thus, the small phase relationship hints : at : the importance of forcings other than wind stress on ageostrophic current, e.g.counterflow produced by the bottom friction (the bottom Ekman layer), etc. Negative ::: The ::::::: negative phase is only found in a narrow band along the northwestern boundary indicating that :::::::: suggesting :::: the ::::::: direction ::: of :: the : wind-driven current ::: that is to the left of ageostrophic flow.The negative phase is clearly evident in June, September, and December, while a positive phased is suggested ::::: phase :: is ::::::: observed : during the spring monsoon transition (March) (Figure 4b, d, f, h).

The role of wind stress curl on the sea surface height
The impact of the wind stress curl on sea surface height is examined through a linear regression using daily measurements at the original spatial resolution of 1/4°.Both of the basin-averaged wind stress curl and ADT have distinct seasonal cycle; high ::::: cycles.::::: High (low) ADT is observed during the northeast (southwest) monsoon over the entire basin, while the wind stress curl exhibits the opposite pattern (Figure 7a).In addition, the intraseasonal signals of the basin-averaged ADT covary quite well with the basin-averaged wind stress curl.The anticorrelation yields a significant negative correlation of -0.84 indicating the dominance of wind stress curl on the ADT through local Ekman pumping; positive (negative) wind stress curl induces upward (downward) flow in the water column and depresses (raises) the ADT.In addition, the correlation suggests no time lag between the wind curl and the ADT reflecting the instantaneous adjustment of the ADT as being forced by the local wind stress curl.As Zhou et al. (2012) suggest a delayed response of the ADT to wind stress curl over the SCS (decaying timescale of ∼40 days), the result demonstrates the different wind-associated dynamic that underlies the GoT compared to the rest of the SCS.
As most energy of the GoT circulation locates along the western boundary of the GoT (Figure 2, 3, 4), the influence of wind stress curl on current variability at 9.6°N, 99.9°E representing high ADT variance to the south of the uGoT (purple cross in Figure 2b) and : at : 12.9°N, 100.1°E representing high variance along the western central GoT (green cross in Fig- ure 2b) are further investigated.Since an effect of winds on the ocean circulation is not necessarily local nor applied over a large scale (e.g.Meyers, 1996;Giddings and MacCready, 2017), ::::::::::::::::::::::::::::::::::::::::::: (e.g., Meyers, 1996;Giddings and MacCready, 2017) : , ::: the relationship between ADT at the selected locations and wind stress curl over the entire GoT is examined to identify the location of wind stress curl that influences the ADT.::: The : ADT with high variance :: to ::: the south of the uGoT correlates well with the nearby wind stress curl with a correlation coefficient of -0.80suggesting .the : high correlation between ADT at the western boundary and remote wind stress curl is still unclear and beyond the scope of this study.Still, the result suggests the importance of coastal trapped Kelvin waves which travel equatorward along the western boundary of the basin (Wang, 2002).Coastal trapped Kelvin waves are also commonly found in regions with shallow and complex bathymetry, e.g., the Indonesian Archipelago (Sprintall et al., 2000;Delman et al., 2018)and the SCS : , ::: the :::: SCS, : and the East China Sea (Wang et al., 2003;Yin et al., 2014;Liu et al., 2011).At the GoT interior, wind stress curl does not exhibit : a : seasonal cycle, and thus, the local wind stress curl does not locally influence the ADT there.

Interaction with the South China Sea
Since the GoT connects to the SCS, variability of the SCS circulation would provide a better understanding on : of : the GoT circulation as well as ::: and ::: the : origin of water masses transported into the basin.In the southern part of the SCS, the circulation is highly influenced by the monsoon winds (e.g.Hu et al., 2000;Gan et al., 2006).During the northeast monsoon when the inflow from the SCS to the GoT is observed (Figure 4g), a strong southwestward flow is present off the eastern coast of Vietnam; the current partly turns northwestward transporting water into the GoT (Hu et al., 2000;Gan et al., 2006;Liu et al., 2008).

Interannual variability of the circulation in the Gulf of Thailand
The influences of ENSO and IOD are examined to understand ::: the interannual variability of the GoT circulation.Low-frequency OSCAR velocity is calculated by removing the seasonal cycle, taken as a linear combination of :: the : annual and semiannual harmonics that best fit the 6-year observations, and signals with periods shorter than 90 days.A 90-day lowpass filter is also applied to Niño3.4 and DMI (see Datasets section).Complex correlation between the low-frequency indices, which have a phase of zero (±π) for a positive (negative) value, and the low-frequency currents is calculated over the entire GoT (Figure 8a, b).The correlation shows fascinating patterns revealing that the ENSO conditions highly influence the circulation over the central and the eastern parts of the GoT, while IOD conditions influence the current along the western boundary of the GoT and the southern boundary of the observing domain.
Along the western boundary of the GoT, high correlation between the low-frequency current and low-frequency DMI is found with a phase relationship of π 2 indicating northward (southward) current anomaly during a positive (negative) IOD condition (Figure 8b).The low-frequency meridional current averaged in 9.0°-11.5°N,99.5°-100.2°Eregion is used to represent the current along the western boundary; low-frequency DMI explains 45% of the variance of the alongshore meridional flow (Figure 9b).During the :::::::: southwest ::::::: monsoon :::: and ::: fall :::::::: monsoon :::::::: transition :: in : 2016 ::::: when : a : negative IOD event ::::: occurs, the southward western boundary current (Figure 4) intensifiesduring the southwest monsoon and fall monsoon transition.In contrast, the seasonal southward current significantly weakens during the southwest monsoon and fall monsoon transition of 2019 when a positive IOD occurs.In addition, DMI correlates with the OSCAR current along the southern boundary of the domain.
With the phase relationship of approximately ±π, the current is westward (eastward) along ∼8°N during a positive (negative) IOD event.The result suggests that IOD events does :: do : not only affect the current along the western boundary but they ::: also impact the continuous current from the southeastern entrance to the western boundary of the GoT (Figure 3-5).The results are opposite to ::::::::: contrasting :::: with : the finding by Higuchi et al. (2020) that suggests an anomalous outflow during the southwest monsoon season of a positive IOD event.To understand the dynamics associated with the low-frequency variability of the GoT circulation, correlations between low-frequency DMI and selected forcings, which are ADT, zonal wind stress, and wind stress curl, are calculated (Figure 8c, e, g).Similarly, those between Niño3.4 and the selected forcing are also computed (Figure 8d, f, h).Note that the meridional wind stress is also considered; however, its correlation with either of the indices does not exhibit a distinct variation pattern over the GoT.The low-frequency variability of the current along the western boundary during IOD events is likely associated with local zonal wind stress (Figure 8f).Low-frequency component of the zonal wind stress shows strong correlation with the DMI along the GoT western boundary with correlation coefficients of up to -0.75 and along the southern boundary of the domain with correlation coefficients of up to -0.58; the correlation pattern is similar to that between the OSCAR current and DMI (Figure 8b).The negative correlation suggests the westward (eastward) wind stress anomaly during a positive (negative) IOD event yielding a northwestward (southeastward) surface Ekman current anomaly along the western boundary, consistent with the low-frequency OSCAR current.Although the low-frequency wind stress curl and ADT also suggest northward flow anomaly along the western boundary, the influence is roughly the same along the entire western The ADT, zonal wind stress, and wind stress curl are also examined over the uGoT region to understand how these forcings, which potentially influence the uGoT interannual circulation, vary during the ENSO and IOD events (Figure 8).Positive correlation is found between the ADT over the uGoT and Niño3.4 indicating a tendency of an anomalously high (low) sea level, particularly along the eastern boundary during an El Niño (La Niña) event (Figure 8c).The pattern is consistent with that produced by the wind stress curl (Figure 8g).Therefore, the geostrophic current is likely anomalously northward (southward) along the western boundary of the uGoT ::::: during ::: an :: El ::::: Niño ::: (La ::::: Niña) ::::: event.In addition, a positive correlation between the zonal wind stress and Niño3.4 is present reflecting anomalously southward (northward) wind-driven Ekman current during an El Niño (La Niña).Correlations between the selected forcings and DMI are generally lower than when compared to ::::: those :::: with the Niño3.4(Figure 8c the ::::::: eastern :::::::: boundary : (Figure 8d).A positive correlation between DMI and wind stress curl is also found over the uGoT potentially contributing to a lower increase in the sea level compared to that in the rest of the GoT where negative correlation is present (Figure 8f).IOD events are overall not significantly correlate :::::::: correlated : to the zonal wind stress over the uGoT (Figure 8f).

Conclusions
This study exploits the synergy of :: the : available remotely-sensed observations to understand variability of the GoT circulation that reveals different responses to the different climate modes.The interannual current along the western boundary is more sensitive to IOD conditions, while that in the GoT interior is more sensitive to ENSO conditions (Figure 8,9).At the seasonal timescale, the observation reveals spontaneous adjustment of the basin-averaged ADT following :: the basin-averaged wind stress curl signal that is different from the rest of the SCS (Zhou et al., 2012).Still, the associated mechanisms are different :::: vary over different parts of the GoT (Figure 7).For example, the ADT at the southern part of the uGoT highly correlates with the local wind stress curl reflecting the influence of ::: the local Ekman pumping, while the ADT along the western boundary is highly related with :: to the wind stress curl to its north suggesting the influence of coastal trapped Kelvin waves on modifying the sea level along the western boundary.Similar to circulation in the SCS (Gan et al., 2006), current over the GoT is mostly geostrophic.
The OSCAR surface current product also demonstrates the seasonal reversing circulation pattern at the surface following the monsoon wind reversal that accounts for 28% of the total current variance over the 2014 -2019 period (Figure 3a-c).The seasonal pattern confirms the anticyclonic circulation in the GoT interior with an outflow at the southeastern entrance during the southwest monsoon (Figure 4) consistent with findings from Wyrtki (1961); Sojisuporn et al. (2010);and Aschariyaphotha et al. (2008).A cyclonic circulation along the western boundary of the GoT as suggested by a numerical simulation (Yanagi and Takao, 1998) is present but narrow, confined to the west of ∼100.5°E:::::: (Figure ::: 1a).Also, the western flank of the cyclonic circulation is stronger than the eastern flank consistent with a previous altimetric observation (Sojisuporn et al., 2010).During the northeast monsoon, OSCAR product shows strong northward flow along the western boundary and an inflow at the southern entrance consistent with observational (Sojisuporn et al., 2010) and numerical studies (Yanagi and Takao, 1998;Aschariyaphotha et al., 2008)that suggest anticyclonic circulation over the GoT.The 6-year averaged surface velocity displays the dominance of cyclonic circulation in the GoT interior (Figure 4), in agreement with that observed during the NAGA expedition (Wyrtki, 1961).

:
:::::: entrance.:::::: During ::: the :::::: spring :::::::: monsoon :::::::: transition, :: a strong northward flow along the western boundary that superimposes on the circulation pattern during the northeast monsoon is found during the spring monsoon transition (Figure 3-4).During the fall monsoon transition, :In :::::::: contrast, : a : southward flow is present along the western boundary and strong southeastward current is observed at the southeastern entrance :::::: during ::: the ::: fall :::::::: monsoon :::::::: transition.As the western boundary current is connected to that at the southeastern entrance (Figure5), the results highlight the connection between circulation in the GoT and the SCS that distinctly occurs during the monsoon transitions.Moreover, variability of the circulation during the transition seasons could largely impact the properties of water in the GoT.

Figure 2 .
Figure 2. :: (a) : The 2014-2019 :: (a) : mean ::: and :: (b) ::::::: variance :: of OSCAR current shown in : (maroon quivers ::::: quiver/ ::: line) : and the 2014-2019 mean ADT shown in : (color contour) : over the Gulf of Thailand (a) and the OSCAR current variance shown in maroon lines where the ::: GoT.:::: The zonal (meridional) component of the maroon lines : in ::: (b) indicates variance of the zonal (meridional) OSCAR currentwith ADT variance shown in color contour (b).Crosses in (b) mark locations with high ADT variance in the uGoT (green) and at the GoT western boundary (purple)and ; : the triangles with respective colors mark the locations of wind stress curls that correlate the best with the ADT shown in Figure 7.

Figure 5 .
Figure 5. :: (a) Complex correlation between OSCAR current along the western boundary (purple box) and OSCAR current over the entire Gulf of Thailand ::: GoT ::: and (a : b) and the corresponding phase(b).Black contour in (a) is plotted every 0.25 and that in (b) is plotted at 0 ::: zero and π ::: ±π.Crosses indicate the regions where the correlation is not significant with the 95% confidence.

Figure 7 .
Figure 7.Comparison between sea surface height anomaly (black) and wind stress curl (colors): :: (a) : both sea surface height anomaly and wind stress curl (orange) averaged over the entire Gulf of Thailand :::: GoT, (a : b) , sea surface height anomaly at the Gulf of Thailand :::: GoT western boundary shown as purple cross in Figure 2b and wind stress curl (purple) to the south of the upper Gulf of Thailand ::::: uGoT shown as purple triangle in Figure 2b(b), and :: (c) both sea surface height anomaly and wind stress curl (green) to the south of the upper Gulf of Thailand :::: uGoT : shown as green cross and triangle in Figure 2b(c).Correlation coefficient between each comparison is shown on the upper right corner of each subplot.Note the reversed y-axis for wind stress curl.