Projected sea level rise in Thailand: Regional effects of climate change and solar radiation modification based on observations and the GeoMIP6 Open Access

Sea level rise (SLR) is a pressing issue for coastal cities in Thailand. It leads to significant changes in the coastline, disrupt ecosystems, and force people to alter land use patterns. SLR can also exacerbate saltwater intrusions into river mouths, contaminating the rivers when the flows are weak. For example, this contaminates freshwater resources used for topwater production in Bangkok, Thailand’s capital, and the vicinity especially in the dry seasons.

Understanding SLR involves measuring both absolute and relative sea levels. Absolute sea level rise (aSLR) refers to the rate of changes in sea-surface height with respect to the geoid or the reference ellipsoid of the Earth. Relative sea level rise (rSLR), on the other hand, measures the rise or fall in sea-surface height relative to the land at a particular location. Relative sea level (RSL) can be measured using in-situ observations such as tide gauges, while absolute sea level (ASL) can be measured using remote sensing such as satellite altimetry. Alternatively, the aSLR can be deduced from tide gauge data by accounting for the vertical land motion (VLM), which can be measured using the Global Navigation Satellite Systems (GNSS). The relationship between rSLR and aSLR is as follows:

SLR is influenced by thermal expansion of the oceans and dynamic factors such as near-surface wind and surface pressure. Thermal expansion is mostly influenced by the long-term climate factor. Anthropogenic warming has caused a gradual rise in sea levels over the past few decades and will worsen if not addressed (Bouttes et al., 2013; Brown et al., 2021; Zhou et al., 2022). Near-surface wind, varying seasonally, contributes to the dynamic sea level (DSL) via a process called the Ekman transport, which causes a 90-degree net water movement from the wind direction due to the Coriolis effect (Price et al., 1987; Wenegrat and Thomas, 2017). As the wind stress (τ) is exerted on the ocean surface, it creates the rotational component referred to as the wind stress curl:

which in turn leads to divergence or convergence at the surface. Positive wind stress curl (counterclockwise rotation in the Northern Hemisphere) leads to surface water divergence, hence lowering local sea level, while negative curl (clockwise rotation) causes convergence and rising local sea level (Talley et al., 2011). Wind stress curl also impacts current-setup sea levels by driving large-scale wind patterns that push water toward or away from the coast, causing temporary rises (set-up) or falls (set-down) in sea levels, respectively (Siedler et al., 2013). To put this in a local perspective, near-surface wind over the Andaman Sea, west of Thailand, is influenced by a strong southwesterly monsoon wind in the boreal summer (June through August), which is the rainy season in this region, leading to a seasonal set-up of the sea levels. Conversely, a northeasterly wind in the Gulf of Thailand (GOT), on the east, is stronger in the boreal winter (December through February) due to the high-pressure systems over China, leading to a seasonal set-up in the GOT (Higuchi et al., 2020; Sripoonpan and Saramul, 2021).

Sea-surface height relative to land, or RSL, is influenced further by vertical movements of the land. VLM accounts for tectonic activities, glacial isostatic adjustment, ground deformation, and local anthropogenic factors such as groundwater extraction (Wöppelmann and Marcos, 2016). Tectonic activities and groundwater extraction have affected past VLM in Thailand (Trisirisatayawong et al., 2011; Saramul Ezer, 2014; Jaroenongard et al., 2021). Excessive groundwater extraction in the past, particularly in central Thailand, has led to significant land subsidence in Bangkok (Phien-wej et al., 2006). While the Groundwater Act, B.E. 2520, was developed in 1977, it took decades for the regulation of groundwater licensing and utilization to be strictly enforced, and years after that for the effects to be observed (Lorphensri et al., 2016). Overall, this helped alleviate the land subsidence issue starting from the early 2000s, as shown in a recent study focusing on the data from an in-situ observation station in the Inner GOT (Jaroenongard et al., 2021). It also indicates that when reporting local rSLR rates, it is important to specify the period being considered, as the rate is not necessarily linear.

One consequence of land subsidence in coastal cities is saltwater intrusion. Due to subsiding land at the delta, high tides, and occasional surges from strong monsoon winds, saltwater intrusion has become a significant issue for areas that rely on the Chao Phraya River, including Bangkok (Changklom et al., 2022; Pokavanich and Guo, 2024; Tomkratoke et al., 2025). Previous studies have estimated rSLR along the coasts of Thailand but the range of the estimated rSLR varies widely. For example, using the Permanent Service for Mean Sea Level (PSMSL) data, Yanagi and Akaki (1994) estimated rSLR during 1951–1991 of 16.4 ± 0.85 mm/yr at Fort Phrachul, a tide gauge station at the Chao Phraya River delta in the Upper GOT, and 2.3 ± 1.1 mm/yr at Ko Tapao Noi in the Andaman Sea. Trisirisatayawong et al. (2011) estimated the rSLR of 3.0–5.5 mm/yr in the Upper GOT, which is also significantly faster than the global SLR of approximately 2 mm/yr. Saramul and Ezer (2014) reported estimated rSLR rates of 6 mm/yr from 1940 to 2014 along both the Andaman Sea and GOT coasts. In short, station-based rSLR rates reported for this region between 1940 and 2006 vary across different periods, with some even showing opposite trends, despite the stations being only a few hundred kilometers apart (Yanagi and Akaki, 1994; Vongvisessomjai, 2006, 2010; Siwapornanan et al., 2011). This variability highlights the influence of VLM on rSLR measurements and underscores the need for more robust methods to reduce uncertainty.

One approach to address this uncertainty is to incorporate remote-sensing data, such as satellite altimetry, which provides extensive spatial and temporal coverage and is unaffected by land motion. However, while satellite altimetry offers accurate aSLR measurements over the open ocean, its accuracy drops within 20 km of coastlines mostly due to land contamination (Cazenave et al., 2022). Advanced processing techniques dedicated for addressing this challenge, such as the X-TRACK Coastal Altimetry data however, have improved coastal altimetry, and studies increasingly demonstrate consistent results between satellite altimetry and tide gauge data combined with GNSS land motion corrections (Cazenave et al., 2022; Zhou et al., 2022; Erkoç and Doğan, 2023; Leclercq et al., 2025). Nonetheless, as with any other remote-sensing measurements, satellite altimetry requires consistent calibration using in-situ observations such as tide gauges, and calibration errors can introduce biases and uncertainties in the data as well. In-situ observational data is so far the best approach for estimating long-term sea levels (Adebisi et al., 2021).

Another approach to estimate SLR is to use tide gauge data coupled with numerical simulations, ranging from regional models to global climate models (GCMs) (Bâki Iz et al., 2012; Oliver, 2014; Adebisi et al., 2021). For a station in Bangkok, Thailand, for example, Jaroenongard et al. (2021) used the tide gauge data and outputs from 35 GCMs from the Climate Model Intercomparison Project Phase 5 (CMIP5) under 2 greenhouse gas concentration scenarios: Representative Concentration Pathway 4.5 (RCP4.5) and RCP8.5. They estimated that at Fort Phrachul, the observed rSLR during 1940–2018 based on the tide gauge data was 14.98 mm/yr. However, after the regulation of groundwater extraction, recent land subsidence is found to be small. Assuming that the future land subsidence will remain small based on the tide gauge observations, coupled with a future RCP4.5 scenario projection from GCMs, they projected a long-term rSLR at Fort Phrachul to be 0.94–1.05 mm/yr for 2021–2099. For the RCP8.5 scenario, they projected the long-term rSLR rates of 1.07–1.18 mm/yr, also accounting for varying levels of ongoing land subsidence based on recent tide gauge data. At the other two stations in the Upper GOT, Jaroenongard et al. (2021) reported a much lower rSLR rates in the past, 0.75 mm/yr during 1940–2002 for Ko Sichang (eastern Upper GOT) and 0.48 mm/yr during 1940–2012 for Hua Hin (southwestern Upper GOT), and projected long-term rSLR of 0.94–0.95 mm/yr for RCP4.5 and 1.07–1.08 mm/yr for RCP8.5 for these 2 stations.

Recently, solar geoengineering or solar radiation modification (SRM) has been proposed to reflect more sunlight back to the space, aiming to temporarily lower global surface temperature while carbon reduction methods are implemented. These sunlight-reflecting methods may have the potential to prevent further global SLR by preventing ocean warming and glacial melting. Narenpitak et al. (2024) reported using GCMs in the Geoengineering Model Intercomparison Project Phase 6 (GeoMIP6) that in the G6Solar and G6Sulfur SRM simulations, sunlight dimming and stratospheric aerosol injection (SAI) could lower surface temperature in Thailand and nearby countries. But it will likely reduce precipitation after the mid-century compared to the medium-emission baseline scenario, the Shared Socioeconomic Pathways 2-4.5 (or SSP2-4.5). The effects of SRM on SLR in Thailand is yet to be examined. Therefore, it is also of interest to investigate further how SRM in these simulations might affect the future sea level projections, both in terms of the thermal expansion and dynamic factors.

As a result, the purpose of this article is to estimate the past, present, and future SLR rates along the west and east coasts of Thailand using various methods. In-situ observational data from tide gauges and the GNSS stations, remote-sensing data from the X-TRACK Coastal Altimetry, and GCM outputs are compared with one another, shown in Section 2. It also explores how climate change and SRM impact SLR in Thailand using the GeoMIP6 simulations, shown in Section 3. Given that saltwater intrusion is a prominent problem in December, January, and February, when surface wind in the GOT is high and precipitation is low, in Section 4, we also explore how mechanisms affecting seasonal sea levels such as the Ekman transport and the wind stress curl varies among different future climate scenarios. Finally, Section 5 provides the conclusions.

We use in-situ observational data from tide gauges and the GNSS stations, remote-sensing data from the X-TRACK Coastal Altimetry, and simulation outputs from GCMs to estimate SLR along the western and eastern coasts of Thailand. Data from 9 tide gauges are obtained from the PSMSL (Holgate et al., 2013; PSMSL, 2025) and used to estimate the rSLR (Table 1; Figure 1a). For the GNSS measurements, data from 10 stations are obtained from the Nevada Geodetic Laboratory (NGL), University of Nevada, Reno (Blewitt et al., 2018; NGL, 2025), and 1 from the Système d'Observation du Niveau des Eaux Littorales (SONEL) (Dow et al., 2005; Dow et al., 2009; SONEL, 2025) as the SONEL time series at this station is longer (Table 1; Figure 2a). Coastal altimetry data from 13 virtual stations, reprocessed from along-track altimetry data from the Jason-1, -2, and -3 satellites, which are obtained from the European Space Agency (ESA) Climate Change Initiative (CCI) (Cazenave et al., 2022; ESA, 2025), are juxtaposed with rSLR from the nearby tide gauge stations (Figure 4) (see Section 2.2 for details on how these rates are computed.)

Simulation outputs from 3 GCMs from the Climate Model Intercomparison Project Phase 6 (CMIP6), namely CNRM-ESM2-1 (Séférian et al., 2019), IPSL-CM6A-LR (Boucher et al., 2020; Lurton et al., 2020), and UKESM1-0-LL (Tang et al., 2019, ESGF), are used to estimate the aSLR (Table 3). These models provide variables for thermosteric and DSLs above the geoid (ZOSTOGA and ZOS, respectively). The 10-m (near-surface) zonal wind (u_10m) and meridional wind (v_10m) are also used. The monthly variables are downloaded from the U.S. Department of Energy Lawrence Livermore National Laboratory (LLNL)’s World Climate Research Programme’s website (LLNL, n.d.) All GCM outputs used in this study are cropped to the study domain, 0°N–17.5°N and 90°E–120°E (Figure 4a), and regridded to 0.5° × 0.5° resolution using nearest neighbor interpolation for 2-dimensional data (scipy.interpolate.griddata) based on Scientific Python library version 1.14.0 (SciPy). The SLR rates for the past are computed using outputs of the Historical simulations, available from January 1850 to December 2014 (Eyring et al., 2016), whereas the projected SLR rates are computed for the climate change SSP2-4.5 and SSP5-8.5 scenarios (O’Neill et al., 2016; Riahi et al., 2017), and the climate engineering simulations G6Solar and G6Sulfur (Kravitz et al., 2015; Visioni et al., 2021).

The SSP5-8.5 simulations represent a climate change scenario with extremely high emissions that triple the present-day level by 2075, as the societies still depend largely on fossil fuel (the 5^th SSP). The top-of-atmosphere (TOA) radiative forcing in SSP5-8.5 is expected to reach 8.5 W/m² in 2099. The SSP2-4.5 simulations represent a climate change scenario with intermediate emissions (the 2^nd SSP), which plateau around 2050 and then decline but do not reach net zero in 2099. In SSP2-4.5, the TOA radiative forcing is expected to reach 4.5 W/m² in 2099. In the G6Solar simulations, the emissions remain high as in SSP5-8.5 but the solar constant is reduced so the radiative forcing follows that of the SSP2-4.5 and thus the surface warming is reduced. The G6Sulfur simulations use injected sulfate aerosols into the stratosphere to reduce the surface warming from SSP5-8.5 to SSP2-4.5. In particular, the solar reduction in G6Solar is spatially uniform but gradual, increasing up to 2% in the end of the century. The injection sites of sulfate aerosols in G6Sulfur vary from model to model (Kravitz et al., 2015; Visioni et al., 2021). For instance, the IPSL-CM6A-LR and UKESM1-0-LL models inject sulfate aerosols between 10°S and 10°N, at 18–20 km altitude, at a single longitude band (0°). The CNRM-ESM2-1 model prescribes the aerosol optical depth (AOD) distribution derived by previous simulations of the same models. On average among these 3 models, the global solar irradiance reduction is 1.77% (AOD of 0.35) to offset about 2.3°C global surface warming (Visioni et al., 2021). The uniformity of the reduced solar constant in G6Solar and the focused injected aerosols in 2 of the GCMs in G6Sulfur may play a role on how much the tropical oceans warm or cool. There are 3 other models participating in the GeoMIP6, but they are not used for this analysis because they do not provide all the variables needed for this study or the coastlines in the study domain differ from the reality significantly, making comparisons with the observations challenging.

The rSLR is computed using monthly and annual time series from 9 PSMSL tide gauges. The available period and the station locations are shown in Figure 1 and summarized in Table 1. The second-order Butterworth low-pass filter is applied to the monthly time series with a cutoff period of 20 years using SciPy’s function scipy.signal.filtfilt(). The 20-year cutoff period is chosen to remove the decadal variability in the rSLR, and the lengths of the time series are sufficient for this approach. The Gustafsson’s method is used for handling the data at the edges of the time series. This is applied for two reasons. First, low-pass filtering often introduces artifacts near the edges of the time series due to padding, especially if the first or last data points differ greatly from the adjacent points. The Gustafsson’s method addresses this issue by using initial conditions so the forward and backward passes produces the same result as the forward-backward filter (Gustafsson, 1996). Second, given recent changes in the VLM and hence rSLR trends in Thailand following the groundwater extraction (Lorphensri et al., 2016; Jaroenongard et al., 2021), it is crucial to incorporate the most recent data to estimate current rSLR rates. Figure 1b shows that the filtered time series (thick solid lines) align with the monthly time series (thin lines) well. For the same reasons, the Gustafsson’s method is applied to other time series that undergo low-pass filtering in this study, although the cutoff periods may differ and will be specified.

The monthly RSL time series establish variabilities up to ±0.20 m at certain locations, such as Ko Lak (orange lines in Figure 1b), at which the tide gauge is in an open area of the Upper GOT, and Danang (green lines), along the coast of Vietnam connecting to the South China Sea (SCS). Other tide gauges establish smaller monthly variabilities. For instance, the monthly RSL variability at Fort Phrachul (blue lines), located at the Chao Phraya River delta in the Upper GOT, is approximately ±0.15 m, and that at Ko Tapao Noi (pink lines), facing the Andaman Sea, is approximately ±0.10 m. There are also annual and seasonal RSL variabilities. These variabilities are removed once the time series are low-pass filtered (thick solid lines in Figure 1b), so a long-term trend can be estimated based on the low-pass filtered data.

Since 1940, the RSL measured from most PSMSL stations shown have increased steadily at the rates of approximately 1–3 mm/yr, with a few exceptions. Ko Lak (facing the GOT) and Ko Tapao Noi (facing the Andaman Sea) establish more rapid changes in RSL especially after 2005. The RSL at Fort Phrachul has increased by 1 m since 1940, with the steepest rate of 25 mm/yr during 1970–1980 and slowly decreasing to 22 mm/yr (1980–1990), 18 mm/yr (1990–1999), and 8 mm/yr (2000–2009) during the following 3 decades. This is largely contributed by groundwater extraction in Bangkok that leads to land subsidence in the area (Jaroenongard et al., 2021). Since the groundwater extraction has been regulated, the land subsidence improves and changes in RSL at Fort Phrachul reduces to 3.32 mm/yr from 2010 onward (except with an indent in 2015 suggesting some interannual variability). The rSLR rate obtained postregulation is more consistent with the rate obtained from nearby areas, that is, 2.92 mm/yr during 1992–2002 at Ko Sichang, which is 66 km away (yellow lines, Figure 2b) on the east side of the Upper GOT and was not affected by groundwater extraction. The rSLR rates after 2010 or in the last 10 years of the available data are shown in Figure 1 and Table 2.

The VLM can be obtained directly from the GNSS stations. The daily time series from 11 stations are shown in Figure 2 and summarized in Table 1. The VLM establishes annual variabilities up to ±0.02 m, a factor of 10 smaller than the RSL variabilities. To compute a long-term trend, a second-order Butterworth low-pass filter is applied to the daily time series with a 2-year cutoff period, following the same approach as in the PSMSL time series. As with the tide gauge data, the Gustafsson’s method is applied to handle the artifacts at the edges of the time series. A difference here is that a 2-year cutoff period is used for the VLM time series. This is mainly because the VLM data establishes only a strong annual variability but slowly varies on a longer time scale. Furthermore, the lengths of the available GNSS time series are not sufficient for a longer cutoff period to be applied. For the chosen cutoff period and the application of Gustafsson’s method, the filtered time series also align well with the daily time series (Figure 2b).

Because of the sparse data at Fort Phrachul (station IDs: NIM1 and SPK3), the VLM from Pathum Wan (CUSV, cyan lines in Figure 2b) is also shown. Since 2008, there is a negative trend (indicating land subsidence) for the VLM at CUSV for 0.04 m excluding the annual variability, but the rate flattens after 2022. Most of the VLM time series establish such patterns, with a negative VLM trend initially, indicating that VLM may have temporarily stabilized recently. Thus, the last 2 years of the available low-pass filtered GNSS data are used to calculate the VLM from these locations. The VLM at Fort Phrachul at stations NIM1 and SPK3 are −2.42 and −1.83 mm/yr, respectively, with the slower land subsidence rate being more recent as shown in Table 2. The VLM trends in this study are thus computed using the last 2 years of the available data for most stations. An except is at Malaca (JUML), where the first 2 years of the time series are used since after that the time series is still declining and only starts to stabilize within the last available year. The VLM rates are shown in Figure 2 and Table 2.

Additionally, we use reprocessed satellite altimetry data to estimate the rSLR. The first dataset is the L3 X-TRACK/ALES Altimeter Sea Level Trends from the ESA CCI. This dataset is derived from altimetry data from the Jason-1, -2, and -3 mission, which underwent special reprocessing to enhance accuracy in the coastal regions (Cazenave et al., 2022; ESA, 2025). It provides the rSLR trends, the errors, and the time series of sea level anomalies relative to the rSLR trends at virtual altimetry stations from 2002 to 2019. The time series from 12 selected virtual stations are shown in Figure S1 of the Supplemental Material, along with the trends for 2002–2019 (provided in the original dataset) and 2010–2019 (computed using the low-pass filtered time series with a 4-year cutoff period and with the Gustafsson’s method being applied). The trends do not vary significantly over the period. While the X-TRACK dataset provides rSLR trends between 0 and 20 km from the coastlines, only data between 5 and 20 km are used in this analysis, as the data within 5 km of the coast exhibit a different trend due to land contamination (not shown).

The X-TRACK altimetry data are also compared with the Global Ocean Grid L4 Sea Surface Heights “Two-Sat” merged dataset (Figure 3). The gridded sea-surface height dataset combines measurements from 2 satellites operating in tandem mode for improved spatial and temporal coverage, and is obtained from the Copernicus Climate Service through the Marine Data Store (CMEMS, 2024). To compute the rSLR, the sea-surface height data are low-pass filtered using an 8-year cutoff period with the Gustafsson’s method applied (color maps). As previously noted, the gridded altimetry data struggle near shorelines but provides great coverage in the open ocean (Adebisi et al., 2021).

Figure 3b shows that the rSLR rates from the X-TRACK altimetry and tide gauge data, from or near the stations shown in Figure 1, are quite consistent. Exceptions are at Koh Sichang (KSC) and Koh Tapao Noi (KTN), where discrepancies arise. The causes for these discrepancies warrant further investigation but are beyond the scope of this study. Given the overall consistency between in-situ tide gauge observations and coastal altimetry data, and considering that tide gauges remain the most reliable tool for estimating the long-term sea trends especially near the coastlines (Adebisi et al., 2021), we focus on using the local tide gauge data in the rest of this study.

As previously noted, the aSLR can be estimated from the observations by adding the rSLR to the VLM rates (Table 2) (Adebisi et al., 2021). The caveat is that the periods during which the rSLR and VLM are calculated in this study do not overlap at some stations because of the sparse data.

The aSLR can also be computed from GCMs. The thermosteric and dynamic sea-surface heights above the geoid (ZOSTOGA and ZOS, respectively) are used. The ZOSTOGA variable is a global-mean thermosteric sea level (GTSL) change, an output from GCMs that is calculated such that it represents global SLR due to thermal expansion (Griffies et al., 2016; Gregory et al., 2019). The ZOS variable is grid-by-grid and varies locally due to horizontal gradients of surface wind and ocean circulation. By definition, the global-mean values of ZOS are zero (Griffies et al., 2016). Hence, for GCMs, the absolute sea level for the area of interest is defined as:

The ASL from 1850 to 2014 are obtained from the Historical simulations from 3 GCMs, as shown in Table 3 and Figure S2. The monthly data are low-pass filtered using a second-order Butterworth filter with a 20-year cutoff period and the Gustafsson’s method is applied, as in the PSMSL time series. For the GCM outputs, aSLR rates during 1990–1999 and 2000–2009 are computed (Figure 4a and b).

The low-pass filtered ASL data establish more rapid changes in 2000–2009 (Figure 4b, Figure S2d–f) than in 1990–1999 (Figure 4a, Figure S2a–c). This is due to the contribution from the ZOSTOGA variable in the GCMs, which represents the global thermosteric sea level, which increases in the recent decades with the anthropogenic warming. Among all 3 GCMs, CNRM-ESM2-1 has coastlines whose latitudes and longitudes match with the reality (gray lines, plotted using Python’s geopandas) (Figure S2a and d). However, IPSL-CM6A-LR and UKESM1-0-LL show narrower Borneo and Sulawesi (Celebes) islands and the Philippines island group, so the aSLR east of the Philippines from these 2 models can be observed (Figure S2b and c and S2e and f). The aSLR also has a steeper rate in the SCS than in the GOT, where the basin is shallower, especially in 2000–2009 in CNRM-ESM2-1 (Figure S2d) and UKESM1-0-LL (Figure S2f). The aSLR east of the Philippines also increase more rapidly than the west (Figure S2b, c, and e). These local variations are consistent with previous studies (Zhou et al., 2022, and the references therein).

The GCM-mean aSLR at the 9 tide gauge stations are shown in Table 2, juxtaposed with the observation-based aSLR estimated from the tide gauges and the GNSS data. The time series of changes in GCM-mean ASL are also shown in Figure 4c and d. Similar to the observed RSL from the PSMSL, the ASL monthly variabilities range from ±0.2 m in the GOT (Figure 4c) and ±0.1m in the SCS and Andaman Sea (Figure 4d). The estimated aSLR from the observations are also shown in Figure 4b. In general, the observed and simulated rates at Fort Phrachul (FPC), Geting (GT), Danang (DN), Tanjung Kling (TJK), and Bintulu (BTL) are consistent with one another (within 0.5 mm/yr). The rates at Ko Sichang (KSC), Ko Lak (KL), Ko Tapao Noi (KTN), and Penang (PN) are steeper from the observations than the GCM-mean values. This is likely because the GNSS stations and the tide gauges are not at the same location, and local factors could contribute to uncertainties when the VLM is measured. SLR also tends to be steeper after 2,000, but the VLM is stabilized, and the different calculation periods could contribute to such discrepancies. Furthermore, we note that different GCMs project different aSLR rates, that is, the UKESM1-0-LL estimates a steeper aSLR. Intermodel spreads and other factors could all lead to such uncertainties.

Changes in sea level felt by the locals are considered relative to vertical land movements. While GCMs offer both spatial and temporal coverages for GTSL and DSL, VLM is not yet taken into account. The ASL from GCMs should therefore be adjusted using VLM rates to obtain the RSL (see Equation 1). On the other hand, tide gauges provide local RSL but observations are sparse both temporally and spatially, but they are by far the best method to estimate the long-term sea level trends if the data are available (Adebisi et al., 2021). Previous studies have shown that the SLR trends in the western Pacific Ocean are often contaminated by the quasi-decadal and interannual variability, including those contributed by the El Niño Southern Oscillation (Zhang and Church, 2012; Lyu et al., 2017; Lyu et al., 2020). This makes it hard to compute the SLR trends solely from the observations which are often 20 years or shorter. Therefore, it is of interest to compare the RSL estimated from both the models and the observations when and where the data are available.

Figure 5a compares the rSLR rates deduced from GCMs with those measured from the tide gauges after 2005 (from Figure 1a). The aSLR from GCMs during 2000–2009 (Figure 4b) and the VLM rates from nearby GNSS stations (from Figure 2a) are also shown in the parentheses. The opaque and transparent markers display the locations of the tide gauges and the collocated GNSS stations, respectively. As discussed in the previous section, land subsidence in Bangkok has improved in recent years after groundwater extractions have been regulated. Therefore, the periods during which the VLM rates are calculated are chosen after the VLM time series stabilize. In general, the estimated rSLR from GCMs range between 1 and 4 mm/yr and tend to be slower than those measured from the tide gauges after 2005. But for the stations where land subsidence is no longer an issue, such as Fort Phrachul, the rSLR rates estimated from both methods are consistent. Other reasons that could lead to such discrepancies include, but are not limited to, internal variabilities that cannot be detected and removed using short time series, different measurement and calculation periods, different GNSS station and tide gauge locations.

Figure 4b and c also compare the observed rSLR rates from tide gauges (dotted bar graphs) with the calculated rSLR rates from GCMs (bar graphs with horizontal hatch lines) and the VLM rates (forward hatches). The observed and calculated rSLR rates are consistent, differing no more than 0.5 mm/yr, at Fort Phrachul, Kuala Terengganu or Geting, Danang, Malaca or Tanjung Kling, and Bintulu (Figure 5b). However, there are 4 stations at which the GNSS and tide gauge locations are not exactly the same, and the calculated rSLR rates differ from the observed rSLR rates more than 0.5 mm/yr: Laem Chabang or Ko Sichang, Chumphon or Ko Lak, Phuket or Ko Tapao Noi, and Penang. (As previously noted in Figure 3b, the rSLR trends derived from tide gauge and altimetry datasets are also inconsistent for 3 of these 4 stations: Chumphon [or Ko Lak, KL], Phuket [or Ko Tapao Noi, KTN], and Penang [PN]. This indicates that further investigations are needed for these locations in the future.) For these, new VLM rates are calculated based on the observed rSLR and the simulated aSLR from GCMs (Figure 5c) and used for projecting future rSLR. This points to uncertainties in estimating the VLM due to different measurements.

Changes in relative sea level from GCMs since 1990, deduced from the aSLR rates shown in Figure 5a, are displayed as time series in Figure 5b. The aSLR rates between 2000 and 2009, computed from low-pass filtered ASL from the Historical simulations (transparent lines), are used to project the future ASL until 2025 (dash lines). This assumes that the contributions from oceanic thermal expansion remain constant until then. This method suggests that, if land subsidence is not an issue, the changes in RSL along the GOT and Andaman Sea are up to 0.04–0.06 m between 1990 and 2025, or approximately 1.14–1.71 mm/yr. For the 4 stations where new VLM rates are calculated, the projected rSLR rates are much steeper than the aSLR, suggesting that accurate VLM measurements are needed to reduce uncertainties in assessing and projecting RSL.

Figure 6 shows the projected long-term SLR rates, computed from low-pass filtered multimodel mean ASL from GCMs along the western and eastern coasts of Thailand, as well as for the global sea level. The aSLR rates from the SSP2-4.5 (yellow), SSP5-8.5 (magenta), G6Solar (light blue), and G6Sulfur (deep blue) simulations are shown during the mid-century (2040–2069) and at the end of the 21st century (2070–2099). Overall, the ASL continues to rise throughout the 21st century, at a higher rate in the end of the century than the mid-century, for each simulation considered. During the end of the century, the aSLR rates are twice faster in SSP5-8.5 than SSP2-4.5. Solar dimming in the G6Solar simulations is able to keep the aSLR at the same rates as in SSP2-4.5. Tropical SAI in the G6Sulfur simulations results in significantly lower aSLR rates than in SSP2-4.5, as the differences between SSP2-4.5 and G6Sulfur are greater than the sum of one standard deviation from each of the simulations.

While GCMs predict consistent aSLR rates along different oceanic basins in the study domain, the rSLR rates are subject to differ depending on the VLM, which currently is measured from sparse observations and has large uncertainties (Figure 5). It should not be assumed that the VLM will remain the same throughout the century either, as the contributing factors including the tectonic movements will continue to change with time.

The changes in the aSLR rates are robust across all models considered along both coasts of Thailand (Figure 6d and e). The differences in ASL changes between each set of simulations after 2060 are also greater than the sum of one standard deviation of ASL changes from each simulation, indicating that the differences are statistically significant. The local trends in SSP2-4.5 and SSP5-8.5 follow the GTSL from both simulations quite well (Figure 6f). The GTSL in G6Solar and G6Sulfur are similar to that in SSP2-4.5, as the GTSL is generally proportional to the integral of radiative forcing (Bouttes et al., 2013). The ASLs along the western and eastern coasts of Thailand, on the other hand, are lower in G6Sulfur than G6Solar after the mid-century. A possible reason for this is due to differences in ice sheet melting in the G6Solar and G6Sulfur simulations (Moore et al., 2024), as discussed further in detail below. Local ASLs could also be contributed by other factors, such as local sea-surface temperature, which is cooled more in the tropics in G6Sulfur compared to G6Solar (Visioni et al., 2021; Narenpitak et al., 2024). To understand this, it is helpful to compare the spatial changes in ASL in these simulations.

Spatial changes in the ASL during 2070–2099 are shown in Figure 7. Among the 3 models, UKESM1-0-LL projects the greatest aSLR in SSP2-4.5 since 2020, while IPSL-CM6A-LR projects the smallest over the entire study domain (Figure 7a–d). This is generally because of the higher GTSL in UKESM1-0-LL than the other two simulations. The 3 models consistently project greater aSLR in the Bay of Bengal (along 90°W), where the bathymetry floor is deeper, than in the Andaman Sea and in the GOT. However, only UKESM1-0-LL projects greater aSLR in the SCS, near the coast of Vietnam, compared to the other basins. CNRM-ESM2-1 suggests similar aSLR and IPSL-CM6A-LR suggests the smallest aSLR in the SCS than the other areas.

The ASL changes are greater in SSP5-8.5 than in SSP2-4.5 during 2070–2099 (Figure 7e–h). This is expected as the global surface temperature rise leads to thermal expansion of the oceans. However, there are spatial differences among the 3 models: IPSL-CM6A-LR projects insignificant changes in aSLR along the Andaman coast and in the northern SCS, due to negative local DSL changes that offset the positive GTSL (Figure 7g). CNRM-ESM2-1 and UKESM1-0-LL project smaller aSLR in the GOT than the Andaman Sea (lower local ASL than the GTSL) and greater aSLR in the Bay of Bengal (higher local ASL than the GTSL) (Figure 7f and h).

These differences are examples of internal variabilities among the CMIP6 models, which make climate projections difficult especially when smaller numbers of models are used. Another example includes the finding from Jin et al. (2024) who considered the MPI-ESM-ER model. They found that the local ASL in the Andaman Sea is lower than the GTSL while the local ASL in the GOT is higher than the GTSL when compared between SSP5-8.5 and SSP2-4.5, a result similar to the IPSL-CM6A-LR model. They further found that in MPI-ESM-ER, the easterly zonal surface current from the Pacific Ocean to the Maritime Continent shifts northward in SSP5-8.5, relative to SSP2-4.5, leading to a weaker Indonesian through flow and a stronger SCS through flow, which in turn changes the water fluxes and affects the DSL in this region.

Another example is from Lyu et al. (2020), which showed that most GCMs in CMIP6, including the 3 models used in this study, establish a cold-tongue bias in the tropical Pacific. This means that the simulated sea-surface temperature in the equatorial Pacific is colder than the observations and the cold-tongue region extends farther west than the observed. In the future climate, these models also project more warming in the cold-tongue region, an El-Niño like pattern extending further west in the equatorial Pacific, accompanied by weaker trade wind and lower DSL in the West Pacific. Because of the cold-tongue bias, the DSL projected in SSP5-8.5 may increase up to 0.01–0.02 m in the future (Lyu et al., 2020). However, it is worth noting that the DSL differences due to intermodel variability within CMIP6 is greater than that from cold-tongue bias. So future projections from more GCMs will be helpful in lowering the uncertainty.

The solar dimming in G6Solar leads to a slight reduction in the ASL compared to SSP2-4.5 during 2070–2099 in all models, except in the northern region of the Bay of Bengal in CNRM-ESM2-1 and UKESM1-0-LL (Figure 7i–l). The ASL in G6Sulfur during 2070–2099 is even lower than G6Solar and SSP2-4.5. This is because both G6Solar and G6Sulfur cool the tropical surface temperature relative to SSP2-4.5, with stronger cooling in G6Sulfur (figure 7 of Visioni et al., 2021). This is especially true for the UKESM1-0-LL model which exhibits a stronger cooling in the Andaman Sea and GOT in G6Sulfur than in G6Solar (figure 2 of Narenpitak et al., 2024). As a result, the lower surface temperature leads to less thermal expansion of the oceans in G6Sulfur compared to G6Solar and SSP2-4.5, respectively.

Additionally, SLR is also influenced by ice sheet melting. While most GCMs include ice sheet schemes, their resolutions are too coarse to accurately resolve ice sheet mass balance, even for the Greenland Ice Sheet. Nonetheless, the projected SLR in our study (Figures 5 and 6) is consistent with previous findings, which used physics-based and regional climate models to downscale the ice sheet processes in Greenland. For instance, Moore et al. (2019) found that by 2080, ice sheet melting in the CMIP5 high-emission RCP8.5 scenario is about 40% greater than in the medium-emission RCP4.5 scenario, while melting in the GeoMIP G4 (corresponding to equatorial SAI at a rate of 5 Tg/year) is 20% less than in RCP4.5. Fettweis et al. (2021) projected approximately 50% more melting in SSP5-8.5 than in SSP2-4.5 after 2080, and 14% less melting in G6Solar than in SSP2-4.5, based on outputs from the CNRM-ESM2-1 model (Séférian et al., 2019). Another reason is that, in G6Sulfur, less warm water flows beneath the Western Antarctic ice sheet compared to G6Solar, and the winds blowing onto the ice sheet are also weaker, as found by Moore et al. (2024). These result in reduced ice sheet melting in G6Sulfur compared to G6Solar, which consequently contributes to less SLR in G6Sulfur. If ice sheet mass balance were better resolved in GCMs, the projected ASL would be greater in SSP5-8.5 than currently projected, and lower in G6Solar. Since G6Solar cools higher latitudes less than G6Sulfur, relative to SSP2-4.5, the projected ASL due to ice sheet melting would likely differ if it were better resolved in G6Sulfur.

Overall, this suggests that the latitudes where SRM is implemented will likely affect SLR patterns. This depends both on thermal expansion, which is directly related to the SRM location, and ice sheet mass balance, which is influenced on the cooling patterns after SRM. Heating and cooling patterns from SRM also contribute to changes in wind circulations, which in turn affect to the dynamic component of sea level. As wind patterns vary seasonally, the effects of seasonal wind changes on seasonal DSL or the Ekman transport, along with how this process may change in the future climate, will be discussed in the following section.

The article so far focused on the ASL trends due to climate change and SRM. However, the annual and seasonal variations in ASL can be as large as the differences between different climate scenarios.

Figures 8 and 9 show the monthly sea level averaged every decade from the coasts of Andaman Sea and GOT, respectively (see Figure S3 for the coastlines). Overlaid in the background is the monthly 10-m zonal wind averaged along the coasts. Along the Andaman coast, the ASL peaks in May–June, which is the beginning of the monsoon season when the westerly zonal wind approaches its maximum. The westerly zonal wind remains strong in this region until September, but the ASL begins to drop in June–July, suggesting that there are other factors that contribute to the annual variation of ASL in the western Thailand. The annual ASL variation ranges within a year (δ_y) are approximately 0.19 m in the Historical and the other 3 future simulations: SSP2-4.5, SSP5-8.5, and G6Solar. The δ_y ranges in G6Sulfur are slightly wider, 0.22 m, because of a wider δ_y range from the IPSL-CM6A-LR model. The annual variation in ASL is comparable to the long-term ASL changes before the end of the century in all simulations.

The annual variations in near-surface zonal wind along the GOT are similar to that along the Andaman Sea: strong westerly zonal wind in June through August (JJA), and strong easterly wind in December through February (DJF). However, the annual variations in ASL in the GOT have an opposite phase from those in the Andaman Sea. The δ_y range along the GOT is also wider, 0.41 m in the Historical and SSP2-4.5 simulations, 0.42 in SSP5-8.5 and G6Solar, and 0.44 m in G6Sulfur. The wider δ_y range in G6Sulfur is once again due to a contribution from IPSL-CM6A-LR. Throughout the 21st century, the annual ASL rise is approximately 0.30 m in SSP5-8.5, and between 0.15 and 0.20 in SSP2-4.5, G6Solar, and G6Sulfur. This is clear evidence that although long-term SLR is inevitable, the annual variation is in itself significant and can cause additional local impacts even within the near future.

Figures 10 and 11 show the monthly anomalies from the annual mean of the sea level and VLMs from the observations and GCMs. The RSL and ASL anomalies (RSL’ and ASL’) from tide gauges and GCM Historical simulations, as well as the VLM anomalies (VLM’), are shown. The VLM’ are an order of magnitude smaller than the RSL’ and ASL’.

While in general, the RSL’ patterns reach their annual maxima during DJF in the GOT, and JJA in the Andaman Sea, they also vary by locations. Stations that receive direct wind in the GOT have wider RSL’ ranges than those located inside the shelf, that is, the range from Geting is greater than Ko Lak, than Ko Sichang and Fort Phrachul, respectively. The range at Danang, which is located at the coast of SCS, is also greater than that at Bintulu, which is farther south and receives less influence from the northeasterly wind in DJF. Likewise, along the Andaman coast, Ko Tapao Noi receives the monsoon wind directly compared to Penang and Tanjung Kling, the latter which are behind the Sumatra Island. Since the monsoon wind is stronger in JJA, RSL’ along the western Andaman coast reach their maxima then.

GCMs in general are able to represent the ASL’ patterns that differ between the two coasts, but they are unable to capture such detailed differences that tide gauge data show. This is likely because of the coarse resolutions in GCMs, making them unable to capture the detailed topography in this region. In the future climate, GCMs do not project any changes in the ASL’ patterns when averaged during 2070–2099 in these future simulations. However, these factors may still change as the coastlines differ from the present day due to SLR and coastal erosions.

Ekman transport, driven by wind stress and the Coriolis effect, modulates local sea-surface height and its variability (Price et al., 1987; Talley et al., 2011; Wenegrat and Thomas, 2017). Seasonal shifts in monsoonal wind over the Andaman Sea and GOT subsequently affect the Ekman transport, leading to near-surface divergence and convergence, which can cause local DSL to rise or fall. How these factors may change in the future climates also need to be explored. Figures 12 and 13 show seasonal DSL averages (colors) in JJA and DJF, during which the GOT and Andaman Sea, respectively, experience higher DSL. The changes in DSL are with respect to the Historical simulations. Overlaid on the figures are the influencing factors to DSL: near-surface wind patterns (arrows) and the curl of surface wind stress (contours). Similar plots for the DSL during MAM and SON are shown in Figures S4 and S5, and similar plots for the changes in DSL with respect to the SSP2-4.5 simulations are shown in Figures S6–S9.

Southwesterly monsoon wind is dominant over the study region in JJA (Figure 12a–d). The wind blows onto the Andaman shore west of Myanmar and Thailand, resulting in local and seasonal DSL highs during the rainy monsoon season. This is consistent with the negative wind stress curl, downwelling, and hence near-surface convergence of the sea water (Talley et al., 2011). As the Andaman Sea is bounded by the coastline on the east, the southwesterly monsoon wind pushes water toward the shore, resulting in water set-up and hence local DSL highs (Siedler et al., 2013). These are seen in all 3 models, showing a negative wind stress curl along the Andaman shore, but UKESM1-0-LL shows a much more strongly negative curl than the rest.

In the GOT and SCS, the wind stress curl is weaker in CNRM-ESM2-1 and IPSL-CM6A-LR than UKESM1-0-LL as well, but all models agree that there is a positive curl in the western region of the Upper GOT and in the SCS along the coast of Vietnam, and a negative curl in the eastern Upper GOT and in the SCS west of Borneo Island. The negative wind stress curl on the east of the basin is associated with locally higher DSL in the eastern SCS, while the positive curl in the GOT and the western SCS is associated with water being pushed away from the western boundary, and thus lower DSL.

All 3 models agree that there is locally higher DSL in the Bay of Bengal (near 90°E and 10°N) in SSP2-4.5 compared to Historical (Figure 12e–h). However, 2 out of 3 models (CNRM-ESM2-1 and IPSL-CM6A-LR) project higher DSL, while the other (UKESM-1-0-LL) projects lower DSL in the Upper GOT. On the Andaman side, all models agree that the westerly component of the near-surface wind is weaker in the future climate, resulting in less negative wind stress curl and hence slightly lower DSL in the future.

Patterns of DSL changes (ΔDSL) in SSP5-8.5 (Figure 12i–l), G6Solar (Figure 12m–p), and G6Sulfur relative to Historical (Figure 12q–t) are consistent with those in SSP2-4.5 (Figure 12e–h), but stronger in magnitude. In G6Sulfur, UKESM1-0-LL projects a much lower DSL in the GOT and SCS than the rest of the models, and this strong negative ΔDSL dominates the multimodel averages. All 3 models agree that in the future climate simulations, the local DSL along the Andaman coast is projected to lower slightly relative to the Historical simulations during JJA. This is the case when comparing SSP5-8.5, G6Solar, and G6Sulfur with SSP2-4.5 as well, as shown in Figure S6.

Northeasterly trade wind is dominant over the study region in DJF (Figure 13a–d). As the wind blows onto the shore in the Thai and Malaysian peninsula, it creates negative wind stress curl (associated with anticyclonic conditions) in the GOT and part of the SCS along the coast of Vietnam, and positive wind stress curl (associated with cyclonic conditions) west of the Borneo Island and in the Andaman Sea. Because of the Ekman transport, the negative wind stress curl leads to downwelling and hence near-surface convergence of the water, resulting in higher DSL in the GOT. On the other hand, the positive wind stress curl is associated with upwelling and near-surface divergence, leading to lower DSL in the SCS than the GOT and in the Andaman Sea. These are all consistent among all 3 models.

For the future climate, in SSP2-4.5 (Figure 13e–h), ΔDSL is smallest compared to the other simulations: ΔDSL is negative in the Upper SCS (lower DSL in SSP2-4.5 compared to Historical), around 110°E–120°E and 10°N–17.5°N, and positive in the Upper GOT (higher DSL in SSP2-4.5 compared to Historical). This is as expected because the northeasterly wind becomes more positive in the future climate, exerting wind stress onto the water surface near the shore east of the peninsula and within the Upper GOT.

In the Andaman Sea, ΔDSL varies depending on the models, being positive in CNRM-ESM2-1, but mostly negative in IPSL-CM6A-LR and UKESM1-0-LL. The ΔDSL patterns in the Andaman Sea follows the spatial patterns of the positive wind stress curl over the ocean, that is, lower DSL in SSP2-4.5 than Historical where the wind stress curl is positive and strong.

The ΔDSL patterns in SSP5-8.5 (Figure 13i–l), G6Solar (Figure 13m–p), and G6Sulfur (Figure 13q–t) compared to Historical are similar to those in SSP2-4.5 (Figure 13e–h), except with larger magnitudes. These are as expected because the changes in wind speed and hence wind stress curl are stronger in these simulations. In G6Sulfur when compared to Historical, the multimodel averaged ΔDSL is more negative than in SSP5-8.5 and G6Solar. By definition of the DSL variable, the negative ΔDSL in the study domain is offset by positive ΔDSL in other regions, but that is outside the scope of this study.

Since local sea level in the Upper GOT reaches its seasonal high in DJF, during which the precipitation is sparse, not only are coastal cities affected by higher surges but some cities along the Chao Phraya River Delta may also be impacted by saltwater intrusion. Higher DSL in the SSP5-8.5 simulations will likely exacerbate this issue, especially because the dry seasons in Thailand are likely to be prolonged as emissions increase (Narenpitak et al., 2024).

But will SRM help alleviate saltwater intrusion? The answer to this question varies, as there are multiple factors that contribute to such issues. First, although the multimodel mean ΔDSL is negative in the Upper GOT in G6Sulfur relative to Historical, 2 out of 3 models (CNRM-ESM2-1 and IPSL-CM6A-LR) show slightly positive ΔDSL while the other model (UKESM1-0-LL) shows stronger negative ΔDSL. This multimodel variability suggests that more models are needed to lower the uncertainty. Furthermore, the severity of saltwater intrusion also depends on precipitation: Lower precipitation in G6Solar and G6Sulfur may reduce freshwater runoff and reservoir levels, decreasing overall availability for the Chao Phraya River, compared to the Historical simulations. Impact models that incorporate both precipitation and sea level data for saltwater intrusion prediction will be helpful.

Nonetheless, all 3 models agree that in G6Solar and G6Sulfur, the DSL along the coasts of Upper GOT and Andaman Sea will be lower than in SSP2-4.5 during DJF, while the DSL is higher in SSP5-8.5 than SSP2-4.5 (Figure S8). This suggests that SRM, both by the idealized solar constant reduction in G6Solar and in the form of SAI in G6Sulfur, has potentials to help prevent SLR along the coasts of Thailand and nearby countries, including Myanmar, Cambodia, Vietnam, Malaysia, and Indonesia.

Sea-surface height modulates daily, monthly, and annually due to surface wind variabilities and other factors. As the global surface temperature warms, the oceanic water expands, sea ice and glaciers melt, all leading to the rise of sea level. The thermal expansion and dynamic effects combined lead to excessive sea level in vulnerable regions along the coasts. Sea level rise is an emerging problem for Thailand, whose southern coasts boarder the Andaman Sea on the west and the GOT on the east. Sea level along the Andaman coast reaches its annual high around June, when the southwesterly monsoon wind is strongest. On the other hand, sea level along the GOT reaches its maxima during December through February, when the northeasterly wind is prominent as the high-pressure system from China pushes the cold and dry air southward. High sea level due to storm surges and tides during the dry and windy seasons often result in saltwater intrusion along the Upper GOT. This could be exacerbated if the long-term SLR increases significantly or the seasonal variability in sea level or near-surface wind speed becomes more intense.

Anthropogenic warming of the global surface temperature leads to continued SLR. Solar radiation modification, a novel approach proposed to slow down the warming by reflecting more sunlight into the space, can prevent the local surface warming in Thailand and the neighboring countries as intended but affects the precipitation patterns and reduces total rainfall (Narenpitak et al., 2024). In this study, we explore how climate change and SRM affect sea level and near-surface wind along the coasts of Thailand. Using in-situ observational data, satellite altimetry, and simulation outputs from GCMs, we estimate the rates of rSLR in this region. We then project future SLR given the past VLM, assuming that land subsidence has stabilized as groundwater extractions in Thailand have been regulated. Future SLR due to thermal expansion and dynamic changes caused by variations in near-surface wind are analyzed for the medium-emission and high-emission scenarios—SSP2-4.5 and SSP5-8.5—and the two SRM simulations—G6Solar and G6Sulfur.

To estimate the past SLR, we compare the RSL data from tide gauges (Figure 1) with the VLM data (Figure 2), the RSL data derived from coastal altimetry (Figure 3), and the ASL from GCMs (Figures 4 and 5). The tide gauge data are obtained from the PSMSL (PSMSL, 2025). The VLM data are measured using the GNSS, provided by the Nevada Geodetic Laboratory (NGL, 2025), University of Nevada Reno and SONEL (SONEL, 2025). The X-TRACK Coastal Altimetry data, reprocessed from along-track altimetry data from the Jason-1, -2, and -3 satellites (Cazenave et al., 2022; ESA, 2025), are obtained from the ESA CCI. Outputs from 3 GCMs in the GeoMIP6 are used: CNRM-ESM2-1, IPSL-CM6A-LR, and UKESM1-0-LL. Sea level from the GCM outputs comprises of the global thermosteric component (a globally average time series from the model outputs, variable name: ZOSTOGA) and the DSL (variable name: ZOS). A second-order Butterworth low-pass filter is applied to obtain long-term SLR from the analyzed data, using various cutoff periods for different datasets (e.g., a 20-year cutoff period for the tide gauge data and GCM outputs, and a 2-year cutoff period for the GNSS VLM data) and incorporating Gustafsson’s method to minimize edge effects.

By definition, RSL is the sea level measured relative to the land motion while ASL is the distance between the sea surface and the geoid; hence, rSLR equals aSLR subtracted by the VLM. Based on the tide gauge data, Fort Phrachul, a station located at the Chao Phraya River delta, has experienced approximately 1 m of rSLR since 1940. The rapid rSLR between 1970 and 2000 is due to land subsidence caused by groundwater extractions in Bangkok and the vicinity (Phien-wej et al., 2006; Lorphensri et al., 2016). Such activities have been regulated and land subsidence has stabilized in the recent decades (Figure 1b, Jaroenongard et al., 2021). The VLM data at Fort Phrachul, however, does not cover a long enough duration for this to be observed. The VLM data at Pathum Wan, a nearby station in the city center of Bangkok, is available since 2008. It shows that land subsidence in Bangkok has stabilized and the VLM rates in Bangkok and nearby areas are small (Figure 2b). Coastal altimetry data from the X-TRACK dataset also indicate a consistent SLR trend in the Upper GOT, compared to the tide gauge data, although there is no virtual station at the location Fort Phrachul (Figure 3).

Using our methods to estimate long-term sea level changes, rSLR measured by the tide gauge and rSLR calculated from GCMs and the VLM data at Fort Phrachul are consistent, being approximately 3.32–3.45 mm/yr (Figure 5). The estimated rSLR rates from both methods are also consistent, differing less than 0.5 mm/yr, at 4 other stations: Danang (off the coast of Vietnam by the SCS), Geting (off the coast of the Peninsula Malaysia by the Lower GOT), Malacca (along the western coast of Malaysia near the Strait of Malacca), and Bintulu (along the northern coast of Borneo Island bordering the southern basin of the SCS). The local variations observed at Ko Tapao Noi, a small island near Phuket located west of Thailand bordering the Andaman Sea, indicate a rapid rSLR. However, the tide gauge data shows significant fluctuations, particularly after the Indian Ocean earthquake and tsunami on December 26, 2004. Due to the insufficient length of VLM data, there are uncertainties in estimating the RSL at this location. Nevertheless, the calculated subsidence rates, ranging from 5 to 10 mm/yr following the 2004 earthquake, are consistent with the findings of Naeije et al. (2022) and Simons et al. (2019). The comparison between the tide gauge and coastal altimetry data between 2002 and 2019 also suggests inconsistent rSLR rates (Figure 3). While this inconsistency should be investigated further, it is beyond the scope of our study.

Future projections show that the sea level continues to rise throughout the 21st century. Medium emissions in the SSP2-4.5 simulations lead to faster SLR than in the Historical period based on GCM outputs (Figure 6). High emissions in the SSP5-8.5 simulations result in SLR at a rate almost twice faster than that in SSP2-4.5. This is expected, as the global sea level is generally proportional to the integral of the radiative forcing (Bouttes et al., 2013). Thus, more stringent mitigation means slower SLR. Our result also suggests that at this stage, there is a large uncertainty in projecting rSLR, as it relies not only on the accuracy of GCMs in projecting the aSLR but also on accurate measurements of VLM. However, past VLM data are sparse temporally and spatially, and land motions are subject to change in the future because of local factors, such as tectonic activities and ground deformation.

A reduction in the solar constant in the G6Solar, to maintain the TOA radiative forcing at 4.5 W/m² as that of the SSP2-4.5 simulations despite high emissions as in SSP5-8.5, leads to slightly slower SLR than SSP2-4.5. Finally, SAI in the tropics to keep the global radiative forcing at 4.5 W/m² in G6Sulfur results in the slowest SLR among the simulations compared. This is likely because the local surface temperature over the oceans surrounding Thailand are lower in G6Sulfur than in G6Solar, which is lower than in SSP2-4.5 (figure 6 of Visioni et al., 2021; figures 2 and 3 of Narenpitak et al., 2024). Lower oceanic temperature in G6Sulfur leads to less thermal expansion of the oceans, and stronger land-sea temperature contrast that could lead to bigger changes in local wind patterns. Models that resolve ice sheet and glacier melting have projected further SLR due to additional ice sheet melting in high-emission simulations and less melting in simulations with SRM, particularly with less ice sheet melting (Moore et al., 2024) and thus slower SLR in G6Sulfur than in G6Solar after the mid-century. If GCMs could resolve these processes, the actual SLR rates are subject to vary in the future climate scenarios. Nevertheless, since sea level will continue to rise even with SRM, adaptation strategies still need to be in place.

GCMs are able to capture the annual variations in sea level anomalies from the annual means quite well compared to the observations (Figures 10 and 11). They also predict that, along the coasts of Andaman Sea and GOT, annual variations of the sea level remain unchanged in the future climate, compared to the Historical simulations (Figures 8 and 9). But further into the oceans, where the seabeds are deeper, DSL will likely change as a result of changing near-surface wind patterns and the consequence of the Ekman transport. During the summer monsoon season (JJA), the southwesterly wind is strongest, exerting wind stress onto the ocean surface and more water converging along the Andaman shore, causing the Andaman Sea level to reach its seasonal high (Figure 12). In the future climate, especially with higher emissions in SSP5-8.5 and in the G6Sulfur simulations, the southwesterly wind during JJA weakens, resulting in less wind stress and less water converging along the Andaman shore but more in the Bay of Bengal. This causes a local reduction in the sea level along the Andaman coast in JJA when considered relative to the global mean sea level. However, the background sea level matters in terms of how much coastal communities feel the RSL. The more rapid long-term SLR in a world with little-to-no mitigation (as in SSP5-8.5) will still cause more damages than in a world with some mitigation (as in SSP2-4.5) or with SRM (G6Sulfur).

Over the GOT and SCS, near-surface northeasterly wind dominates in the boreal winter (DJF), exerting wind stress onto the eastern coast of Thailand and the Peninsula Malaysia, resulting in water converging along the coasts and locally higher sea level than the rest of the oceanic basin (Figure 13). In the future climate, the northerly wind component is projected to strengthen during DJF, especially in the SSP5-8.5 and G6Sulfur simulations. Such changes lead to more water converging along the coastline and locally higher sea level in the Upper GOT, on top of the rising background sea level which is much more rapid in SSP5-8.5 than in SSP2-4.5, G6Solar, and G6Sulfur. Storm surges and high tides, occurring on a shorter timescale than 1 month, may occasionally result in higher sea levels in the Upper GOT.

In a world with high emissions (as in SSP5-8.5), while the precipitation will become more intense, the dry periods will likely to be prolonged (Narenpitak et al., 2024). Could the prolonged dry period coupled with increased long-term sea level and high tides worsen saltwater intrusion in the Chao Phraya River and nearby deltas in the Upper GOT? In a world with tropical injection of sulfate aerosol (as in G6Sulfur), will the slower SLR help prevent saltwater intrusion? To precisely answer these questions, a regional impact model using sufficiently high resolution to capture the complex topography in the Upper GOT are needed (Tomkratoke et al., 2015). The model needs to have sufficient temporal resolution because saltwater intrusions and storm surges occur on a time scale shorter than a week. Daily GCM outputs and observational data are helpful for setting up such a model. More GCMs will be helpful in reducing intermodel variabilities especially when considering the changes in near-surface wind and sea level patterns. Using regional coastal models, along with data assimilation from the observations including satellite altimetry, is also a promising approach to downscale the GCM outputs especially in areas with intricate coastlines.

The release of tide gauge data by the Permanent Service for Mean Sea Level (PSMSL; Holgate et al., 2013; PSMSL, 2025) and the Global Navigation Satellite Systems (GNSS) land motion data by the Nevada Geodetic Laboratory, University of Nevada, Reno (Blewitt et al., 2018; NGL, 2025), and the Système d'Observation du Niveau des Eaux Littorales (SONEL) (Dow et al., 2005; Dow et al., 2009; SONEL, 2025) was essential to complete this work and is highly acknowledged by the authors. The L3 X-TRACK Coastal Altimetry data (Cazenave et al., 2022; ESA, 2025) were obtained from the European Space Agency (ESA) Climate Change Initiative (CCI) and the Global Ocean Gridded L4 Sea Surface Heights and Derived Variables Reprocessed Copernicus Climate Service from the Marine Data Store (CMEMS, 2024). We acknowledge the World Climate Research Programme (LLNL, n.d.), which, through its Working Group on Coupled Modelling, coordinated and promoted CMIP6. We thank the climate modeling groups for producing and making available their model outputs, the Earth System Grid Federation (ESGF) for archiving the data and providing access, and the multiple funding agencies who support CMIP6 and ESGF. No new datasets were generated in writing this article.

The supplemental files for this article can be found as follows:

Figures S1–S9.

We would like to thank Ben Kravitz from Indiana University and Peter Irvine from University College London for providing their scientific feedback. Conversations with John C. Moore from the University of Lapland, Cecilia Bitz from the University of Washington, Abolfazl Rezaei from the Institute for Advanced Studies in Basic Sciences (IASBS), Iran, and Sorja Koesuma from Sebelas Maret University (UNS), Indonesia, helped improve the discussion of our manuscript. We also thank two anonymous reviewers and the editors Detlev Helmig and Paul Palmer for their comments. Their feedback has helped improve this manuscript greatly.

This study is supported by the Degrees Initiative through the Degrees Modelling Fund (Grant Number: RGA-DMF23THA).

The authors declare no competing interests.

Contributed to conception and design: PN, ST, SS.

Contributed to acquisition of data: SK.

Contributed to analysis and interpretation of data: PN, ST, SS, SK.

Drafted and/or revised the article: PN, ST, SS, SK.

Approved the submitted version for publication: PN, ST, SS, SK.

How to cite this article: Narenpitak, P, Tomkratoke, S, Sirisup, S, Kongkulsiri, S. 2025. Projected sea level rise in Thailand: Regional effects of climate change and solar radiation modification based on observations and the GeoMIP6. Elementa: Science of the Anthropocene 13(1). DOI: https://doi.org/10.1525/elementa.2024.00069

Domain Editor-in-Chief: Detlev Helmig, Boulder AIR LLC, Boulder, CO, USA

Associate Editor: Paul Palmer, School of GeoSciences, The University of Edinburgh, Edinburgh, UK

Knowledge Domain: Atmospheric Science