The circulation model is based on the Regional Ocean Modeling System (ROMS) . The model uses non-local K-profile parameterization (KPP) scheme for vertical mixing . Advection is calculated using a rotated-split, third-order upstream-biased scheme following , which has been shown to reduce numerical dispersion while maintaining low diffusion. Covering the Indian Ocean from 31.5° S to 31° N and 30 to 120° E, the model employs a horizontal resolution of 1/10° and 32σ-coordinate vertical layers, with enhanced resolution near the surface. Seafloor bathymetry is derived from the one arc-minute ETOPO1 dataset . Open lateral boundary conditions are specified along the southern and eastern boundaries of the domain. For depth-averaged velocity and sea surface height, radiation boundary conditions following and , respectively, are applied. Barotropic tidal forcing is imposed at the open boundaries using eight tidal constituents (K2, S2, M2, N2, K1, P1, O1, Q1) obtained from the TPXO8 global tidal model . Finally, river discharge from major rivers within the model domain, including the Shatt al-Arab in the northeastern Gulf, is imposed as point sources. Monthly climatological runoff from rivers discharging into the northern Indian Ocean, including the Arabian Sea and the Gulf, is incorporated using the global discharge dataset of . The hindcast simulation is forced by ECMWF ERA-Interim 6-hourly heat fluxes, air temperature, pressure, humidity, precipitation, and winds spanning January 1980 to December 2018. The initial and lateral boundary conditions for various parameters are derived from the ECMWF Ocean Reanalysis System 5 ORAS5; and World Ocean Atlas 2018 . Further details of the model setup are provided in . The performance of the model in reproducing key aspects of the hydrography of the region was extensively evaluated in . In general, the model captures the essential hydrographic features of the Gulf region, including the seasonal progression of temperature, salinity, and Gulf outflow, as well as the long-term trends in sea surface temperatures. A detailed description of the model evaluation is provided in .

In order to understand the drivers of SST variability in the Gulf, we consider the SST tendency equation: ∂T∂t=∇⋅K∇T︷Diffusion-u⋅∇hT︷Horizontal advection-w∂T∂z︷Vertical advection+Qnetρwcp,wh︷Atmospheric heat flux where T is the SST, K is the diffusivity tensor, and where ∇ and ∇h are the 3-D and horizontal gradient operators, respectively. The symbols u and w denote the horizontal and vertical velocities, respectively, while ρw, cp,w and h are the water density, specific heat capacity and thickness of the model's surface layer. Qnet is the atmospheric net surface heat flux.

The net heat flux, Qnet, at the ocean surface is the sum of four components: solar shortwave radiation (Qsw), longwave radiation (Qlw), latent heat flux (Qlat), and sensible heat flux (Qsen): 1Qnet=Qsw+Qlw+Qlat+Qsen

The longwave radiation, Qlw, is further decomposed into downward radiation (Qdlw) and upward radiation (Qulw): 2Qlw=Qdlw+Qulw

While the solar radiation Qsw and the downward longwave radiation Qdlw are prescribed from the atmospheric reanalysis used to force the model, the outgoing longwave radiation Qulw is estimated from the SST using the Stefan–Boltzmann law: 3Qulw=ϵσT4 where ϵ and σ are the emissivity of the ocean surface and the Stefan–Boltzmann constant, respectively, and T is the SST expressed in Kelvin.

The latent and sensible heat fluxes are parameterized using bulk formulae following : 4Qlat=-ρaceLUΔQ,with ΔQ=QS-Qair5Qsen=ρacpcsUΔT,with ΔT=Tair-SST where ρa and cp are the air density and the specific heat of air at constant pressure, respectively; ce and cs are the bulk transfer coefficients for latent and sensible heat; L is the latent heat of vaporization; U is the wind speed at 10 m; ΔQ is the humidity gradient; QS and Qair are the saturated specific humidity at the sea surface and the specific humidity at 10 m, respectively; ΔT is the air-sea temperature difference; and Tair and SST are the surface air and sea temperatures.

To characterize the relationship between anomalies in atmospheric heat fluxes and SST, we analyze the statistical correlations between interannual anomalies of Qnet and anomalies in the SST tendency term dSSTdt. Additionally, to quantify the drivers of Gulf SST variability, we regress SST on the following heat flux components: solar shortwave radiation (Qsw), downward longwave radiation (Qdlw), latent heat flux (Qlat), and sensible heat flux (Qsen).

ERA5 reanalysis data was retrieved from the Copernicus Climate Change Service (C3S) Climate Data Store (CDS) . Data are provided on a regular latitude–longitude grid with a spatial resolution of 0.25°. For the purpose of this study, the data were extracted for a domain encompassing the Arabian Peninsula, the Gulf, and the Arabian Sea, extending from the equator to 35° N and from 40 to 80° E. The reanalysis data, used for composite and regression analyses, covers the same period as the hindcast simulation, that is, 1980 to 2018. For the Self-Organizing Map (SOM) analysis, the data was extended to the entire 1950–2024 period to ensure robust training of the neural network, as described in Sect. 2.5. In addition to SST, the ERA5 dataset includes the following meteorological variables: 2 m air temperature; 10 m wind speed and its zonal and meridional components; sea level pressure; evaporation rate; total column water vapor; 2 m relative humidity; cloud cover; and, in the free atmosphere, geopotential height, specific humidity, and wind vectors at 850 hPa (lower troposphere) and 300 hPa (upper troposphere).

Finally, to assess the robustness of the key findings with respect to the choice of reanalysis product, we additionally use two independent atmospheric reanalysis datasets: the Japanese 55-year Reanalysis JRA-55; and the Modern-Era Retrospective Analysis for Research and Applications, Version 2 MERRA-2;. Data from these datasets were extracted over the same study period and include the following variables: sea level pressure, 10 m winds, 850 hPa winds and geopotential height, and total-column water vapor.

To explore the potential influence of large-scale natural climate variability on extreme temperatures in the Gulf, we consider four global and regional climate variability modes known to affect the climate of the Middle East and the Arabian Peninsula: ENSO, NAO, IOD, and the Indian Summer Monsoon (ISM). Other modes, such as the Arctic Oscillation, were also examined but showed very weak and statistically insignificant correlation with summer Gulf SST (Fig. S1 in the Supplement), and were therefore excluded from further analysis. The intensity and phase of each mode are characterized using the following state-of-the-art indices:

For ENSO, we use the Oceanic Niño Index (ONI), defined as the 3-month running mean of sea surface temperature anomalies in the Niño 3.4 region (5° N–5° S, 120–170° W), based on NOAA’s ERSST.v5 dataset . The anomalies are calculated relative to centered 30-year base periods, updated every 5 years.

To extract the interannual variability in Gulf temperature and its drivers, the data were deseasonalized by removing the monthly climatological means from the original time series, and detrended by subtracting the long-term linear trends over the study period. Specifically, for a variable A, we considered the interannual anomaly A′, defined as: 6A′=A-Aclim-Atrend where Aclim and Atrend represent the monthly climatology and the linear trend of the variable A, respectively. The analysis was further restricted to the summer months (July to September), when extreme temperatures are most likely to occur. We tested the normality of the data using the Shapiro–Wilk test (Table S1 in the Supplement) and by visually inspecting quantile–quantile (Q–Q) plots, which compare the data quantiles to the theoretical quantiles of a normal distribution (Fig. S2). The results of the test, along with the visual inspection, indicate that all variables are either normally distributed or approximately normal. Given that the data do not substantially deviate from normality, we chose to use parametric methods to analyze statistical correlations. We computed local Pearson correlation coefficients between SST anomalies and anomalies in potential drivers at each grid point, as well as correlations between Gulf-mean SST anomalies and regional or global climate variability modes. The statistical significance of the correlations was assessed using a Student’s t-test at a 95 % confidence level. Because these modes are not independent and exhibit mutual correlations, we further conducted a partial regression analysis to isolate the unique relationship between each mode and Gulf SST anomalies while statistically controlling for the influence of the others. This approach involves regressing the residuals of SST (after removing the effects of all other predictors) against the residuals of each individual climate mode (after removing the effects of the remaining modes).

To quantify the typical SST response to variations in surface heat fluxes, we perform a composite of SST anomalies corresponding to the difference between high (>+1 SD above the mean) and low (<-1 SD below the mean) values of atmospheric heat fluxes. To identify and interpret the atmospheric conditions associated with extreme summer SST anomalies in the Gulf – or those commonly observed under specific climate variability modes – we conduct a composite analysis, in which atmospheric fields are averaged during periods when Gulf SST anomalies exceed the 90th percentile, or when climate modes such as ENSO and NAO are in strong positive or negative phases. The most extreme summers are identified as summers where SST exceeds 2 standard deviations (98th percentile) over at least one-third of the Gulf.

To quantify the extent to which summer SST variability in the Gulf is collectively explained by the four climate modes, we performed a stepwise multiple linear regression analysis with forward variable selection. In this approach, NAO, ENSO, IOD, and ISM were successively introduced as predictors based on the incremental variance each contributed to the model. Finally, to partition the total explained variance into independent (unique) and joint (shared) components, we applied a hierarchical partitioning approach , which systematically evaluates all possible combinations of predictors to quantify how much variance each explains individually versus in combination with others.

To visualize the relationship between Gulf SST anomalies and the leading climate variability modes we apply a Self-Organizing Map (SOM) analysis. SOM is a nonlinear, unsupervised learning algorithm that projects high-dimensional data onto a lower-dimensional (typically 2D) grid while preserving topological relationships – such that similar patterns are mapped close to one another . Each neuron (or map unit) represents a typical or prototype pattern learned from input observations and has the same dimensionality as the input. In this study, each neuron corresponds to a specific Gulf SST anomaly and the concurrent intensity and phase of key climate variability modes, represented by their respective indices. To ensure robust training, the data coverage was extended to the full 1950–2024 period. Prior to training, the data were standardized to zero mean and unit variance. During training, input vectors are iteratively matched to their best matching unit (BMU), and neighboring neurons are updated to better represent the underlying data distribution . Upon convergence, the SOM provides a low-dimensional approximation of the probability density function of the input space. We use a 10×10 rectangular SOM (100 neurons), which offers a balance between pattern resolution and generalization. Neuron weights are initialized using the linear initialization method which ensures a faster convergence, and training is performed with a Gaussian neighborhood function, which minimizes topographic error . The initial learning rate is set to 0.2, and the initial neighborhood radius, defining the influence range of the BMU, is set to 5. Slight variations in map size and training hyperparameters yielded comparable results.

The summer SST in the Gulf exhibits pronounced spatial and temporal variability. During the study period (1980–2018), average summer SSTs ranged from below 28 °C to above 35 °C across the Gulf (Fig. 1A). This variability reflects interannual fluctuations superimposed on a long-term warming trend, further modulated by spatial heterogeneity. On average, SSTs are generally higher in the southern part of the Gulf compared to the northern region (Fig. 1B). When spatial variations are averaged out, the Gulf-wide mean summer SST increased at a rate of approximately 0.31 °C per decade over the study period (Fig. 1A). Superimposed on this long-term warming trend are substantial interannual anomalies, with the most extreme basin-scale positive and negative departures occurring in the summers of 1998 and 1991, respectively (Fig. 1A). The magnitude of these interannual fluctuations also displays strong spatial variability, with weaker anomalies near the Strait of Hormuz (SD ≈ 0.3 °C) and stronger anomalies in the northern Gulf (up to SD ≈ 0.7 °C; Fig. 1C).

A heat budget analysis of the surface layer over the study period reveals that anomalies in the SST tendency term dSSTdt primarily reflect the near-compensating effects of anomalies in atmospheric heat fluxes (r=0.88*) and vertical transport processes (mixing and advection; r=-0.87*) (Fig. 2A). In contrast, anomalies in Gulf-integrated lateral heat transport – associated with heat exchange with the Sea of Oman – exhibit only a weak correlation with the SST tendency (r=0.25; Fig. 2A). The strong anticorrelation between anomalies in atmospheric heat fluxes and vertical transport (r=-0.99*) further indicates that vertical transport largely acts as a response to surface forcing rather than an independent driver of SST variability (Fig. 2A). Overall, at the scale of the entire Gulf, variations in atmospheric heat fluxes emerge as the dominant control on SST variability. Spatially, this influence is strongest in the northern Gulf, where atmospheric heat fluxes account for 92 % of the SST variance (Fig. 2B), and remains dominant in the southern Gulf, accounting for about 83 % (Fig. 2C).

Warm summer SST anomalies are associated with positive anomalies in downward longwave radiation and, to a lesser extent, latent heat flux – particularly in the northern Gulf (Fig. 3). However, high summer SSTs are generally accompanied by weaker-than-average sensible heat flux and incoming solar (shortwave) radiation (Fig. 3). The reduced sensible heat flux from the atmosphere to the ocean surface during warm summers results from the combined effect of weaker winds and a smaller air–sea temperature difference (ΔT=Tair- SST; defined in Eq. 5), as air temperature does not increase as much as SST during these periods (Figs. 4A–B and S3). The weaker-than-average shortwave radiation is likely due to increased atmospheric moisture content during warm SST summers, since changes in cloud cover are not statistically significant (Fig. 4D–E). The contrasting impact of atmospheric moisture on surface shortwave and longwave radiation fluxes is further evidenced by the strong anticorrelation between anomalies in downward longwave and shortwave radiation (Fig. S4).

Finally, changes in evaporative cooling – which depend on both the air-sea humidity gradient (ΔQ=Qs-Qair; defined in Eq. 4) and wind speed – are governed by the sign and relative contribution of these two factors. The surface saturation specific humidity (Qs) and air specific humidity (Q^air) increase with rising SST and air temperature, respectively. However, because SST anomalies tend to be larger than air temperature (Tair) anomalies during warm Gulf SST summers, and because relative humidity typically decreases with increasing air temperatures, the humidity gradient (ΔQ) increases under these conditions (Fig. 4C).

In the southern Gulf and the Sea of Oman, where wind changes are weak or uncorrelated with SST anomalies, evaporation increases, leading to cooler SSTs (Fig. 4F). In contrast, in the northern Gulf, the pronounced weakening of the Shamal wind during warm SST summers offsets the modest increase in humidity gradient (ΔQ), resulting in reduced evaporation and, consequently, diminished evaporative cooling (Figs. 4F and S3).

In summary, higher summer SSTs in the Gulf appear to be associated with enhanced downward longwave radiation, and – locally in the northern Gulf – by reduced evaporative cooling. In the following section, we investigate the large-scale atmospheric circulation patterns that modulate these surface heat fluxes and contribute to the occurrence of extreme summer SSTs in the Gulf.

The mean atmospheric circulation near the surface and in the lower troposphere during a typical summer is characterized by a high-pressure system over East Africa, the western Arabian Peninsula, and the southern Arabian Sea, and a low-pressure system over the eastern Arabian Peninsula, the northern Arabian Sea, southern Iran, and Pakistan (Fig. 5). This configuration favors the Shamal winds, while the pressure gradient over the Arabian Sea drives strong southwesterly monsoon winds across the Arabian Sea (Fig. 5A–C). While the Shamal winds transport dry air from Iraq and the northern Arabian Peninsula into the Gulf, the monsoon-associated southwesterly winds bring moisture-rich air to the southeastern Arabian Peninsula and the Gulf (Fig. 5B). Finally, the low-pressure system over the Arabian Peninsula is associated with ascending air in the lower troposphere, while the upper troposphere is dominated by subsidence (Fig. 5D–F).

Extreme summer SSTs in the Gulf are associated with lower atmospheric pressure over the Arabian Peninsula and higher pressure over Iran and the northern Arabian Sea (Fig. 6A and E). This pressure configuration weakens the Shamal winds and strengthens the monsoon winds over the southern and western Arabian Sea, resulting in reduced evaporation across most of the Gulf and enhanced evaporation over much of the Arabian Sea, respectively (Fig. 6C). Increased moisture transport from the Arabian Sea into the Gulf region, combined with reduced moisture export from the Gulf due to weakened local Shamal winds, leads to the accumulation of atmospheric moisture over the region (Fig. 6D). This, in turn, results in higher downward longwave radiation and reduced evaporative cooling, both of which contribute to elevated Gulf SSTs (Figs. 3 and S5). In the upper troposphere (300 hPa), extreme Gulf SSTs are associated with increased convergence and enhanced subsidence over the region (Fig. 6G–H). The intensified upper-level subsidence contributes to heat trapping near the surface.

In summary, higher-than-average Gulf SSTs are associated with lower surface pressure over the Arabian Peninsula and higher pressure over Iran and Pakistan, leading to a weakening of the Shamal winds and a strengthening of the monsoon winds in the southern and western Arabian Sea (Fig. 7). These conditions result in: (i) a buildup of atmospheric moisture over the Gulf, (ii) reduced evaporative cooling, and (iii) enhanced subsidence in the upper troposphere. Together, these three mechanisms contribute to heat being trapped near the surface and suppress surface cooling, thereby leading to warmer SSTs. Next, we examine how strongly these atmospheric conditions are tied to large-scale climate teleconnections.

We examine how summer Gulf SSTs are linked to four major modes of interannual variability that are likely to influence atmospheric circulation in the region: ENSO, NAO, IOD, and ISM.

We find that high summer SST anomalies in the Gulf are negatively correlated with ENSO, NAO, and IOD, and positively correlated with ISM (Fig. 8). However, only the correlations with ENSO and NAO are statistically significant at the 95 % confidence level across most of the Gulf, whereas the correlations with IOD and ISM reach, at best, the 90 % confidence level over the western and central Gulf, respectively. Moreover, while ENSO exhibits a negative correlation with SST throughout most of the summer (July–September) and leads the SST signal by 1 to 2 months, NAO is negatively correlated with SST only during August–September, with no discernible lead time (Fig. 9A and B). Similar to ENSO, IOD shows a persistent negative correlation with a 1-month lead (Fig. 9C). In contrast, ISM displays a positive correlation with SST, peaking at a 2-month lead during most of the summer (Fig. 9D). Interestingly, with the exception of late summer (September), the zero-lag correlation between ISM and SST is negative or very low, suggesting that higher SSTs are favored by strong monsoon winds in late spring to early summer (May–June), while stronger monsoon circulation in the peak summer season (July–August) has either little impact or even a suppressing effect on Gulf SSTs (Fig. 9D).

Six out of the nine summers in which SSTs exceeded the 90th percentile (i.e., 1.28 standard deviations above the mean) over more than three-quarters of the Gulf occurred during La Niña conditions (1998, 1999, 2000, 2007, 2010, and 2016; Table 1). During these extreme summers, the NAO was either in a negative phase (1998, 2000, 2010, 2015, 2016, 2017), or in a neutral phase (1996, 1999, 2007) (Table 1). The IOD, on the other hand, was mostly in a neutral phase (6 out of 9 summers) or in a negative phase (1996, 1998, 2016). Similarly, ISM was generally near its climatological mean (6 out of 9 summers) or above-average monsoon strength in the early summer (1999, 2000, 2017). When considering more extreme events – defined as SSTs exceeding the 98th percentile (i.e., 2 standard deviations above the mean) over more than one-third of the Gulf – 2 out of the 3 such summers coincided with La Niña conditions (1998, 1999). Additionally, 1998, which recorded the highest interannual SST anomaly, also featured simultaneous negative phases of both NAO and IOD (Table 1). Other extreme SST summers that occurred in the absence of La Niña (2017, 1996, and 2015) were associated with strong negative IOD (1996), strong negative NAO (2015), or a combination of strong monsoon winds in early summer followed by weaker monsoon winds during peak summer along with a negative NAO in late summer (2017) (Table 1).

In summary, extreme summer SSTs in the Gulf are generally associated with La Niña conditions and a negative NAO phase, and to a lesser extent with negative IOD conditions and a monsoon pattern characterized by strong winds in early summer followed by weaker winds during peak summer. In the following section, we describe the mechanisms through which these teleconnections influence Gulf SSTs.

La Niña favors elevated summer SSTs primarily because it is associated with lower atmospheric pressure over the Arabian Peninsula at the surface and in the lower troposphere (Fig. 10B). This pressure anomaly weakens the Shamal winds over the Gulf and strengthens the monsoon winds over the southern and western Arabian Sea (Fig. 10D). The intensified winds over the Arabian Sea lead to enhanced evaporation (Fig. 10F). Furthermore, the intensified low-pressure system over the Arabian Peninsula facilitates the transport of excess air moisture from the Arabian Sea to the Gulf region, trapping heat in the lower troposphere (Fig. 10H). Additionally, the weakened Shamal winds over the Gulf further amplify this warming by reducing evaporative cooling – and by enhancing the accumulation of atmospheric moisture in the region.

Negative NAO favors elevated summer SSTs in the Gulf primarily because it is associated with lower atmospheric pressure over the Arabian Peninsula and higher pressure over Iran and the northern Arabian Sea (Fig. 11B). This pressure pattern significantly weakens the surface pressure gradient across the Gulf, thereby substantially weakening the Shamal winds and reducing evaporation within the Gulf, while exerting limited influence on monsoon winds and evaporation over the Arabian Sea (Fig. 11D and F). These changes contribute to higher Gulf SSTs by reducing evaporative cooling and enhancing the retention of atmospheric moisture over the region, which in turn traps heat near the surface (Fig. 11H).

Finally, the impacts of the ISM and IOD on summer Gulf SSTs are weaker and more complex. A strong monsoon circulation in early summer (i) enhances evaporation over the Arabian Sea, increasing atmospheric moisture in the region, and (ii) coincides with weaker Shamal winds over the Gulf during peak summer (July–August) (Fig. S6). Both factors favor warmer Gulf SSTs during the peak summer season. In contrast, strong monsoon winds during peak summer (i) suppress evaporation over the Arabian Sea due to upwelling-induced surface cooling, and (ii) tend to be associated with stronger Shamal winds over the Gulf (Fig. S6). Together, these effects limit Gulf warming. Negative IOD events appear to be associated with slightly weaker Shamal winds and increased atmospheric moisture over the Gulf (Fig. S7), both of which contribute to warmer SSTs. However, the magnitude of atmospheric anomalies associated with the IOD over the Gulf remains relatively weak compared to those linked to ENSO and NAO.

In summary, both La Niña and negative NAO phases favor elevated summer SSTs in the Gulf, primarily because they are associated with lower pressure over the Arabian Peninsula, which weakens the Shamal winds. However, La Niña is also linked to enhanced monsoon winds in the southern and western Arabian Sea, leading to greater moisture transport to – and accumulation in – the Gulf region. In contrast, a negative NAO is associated with a stronger weakening of the Shamal winds, resulting in a more pronounced reduction in evaporative cooling over the Gulf. A strong ISM in early summer followed by a weaker ISM during peak summer promotes increased moisture buildup in the Gulf, while a negative IOD contributes slightly to warmer Gulf SSTs due to marginally weaker Shamal winds.

The climate variability modes considered in this study are not independent but can interact, resulting in cumulative or nonlinear effects. In particular, ENSO, IOD, and the ISM exhibit strong interdependence . Although their relationship is more complex, ENSO and NAO can also interact and produce synergistic effects , albeit much more weakly during summer . Our cross-correlation analysis confirms these links, showing that ENSO is significantly correlated with both the IOD (r=0.36, p<0.01) and the ISM (r=0.38, p<0.01), while its correlation with the summer NAO is not statistically significant (r=0.09, p>0.05) (Table 2). This weak summer ENSO-NAO coupling is consistent with , who reported an asymmetric summer relationship, with La Niña episodes weakly associated with negative NAO phases, but no significant link between El Niño and positive NAO phases. To isolate the unique relationship between Gulf summer SST and each of the four climate modes while controlling for the influence of the others, we conducted a partial regression analysis (Fig. S8). This analysis shows that both NAO and ENSO are significantly and negatively correlated with summer Gulf SSTs, whereas IOD and ISM display only weak and statistically insignificant correlations.

To further quantify the collective influence of these modes, we applied a stepwise multiple linear regression analysis (Fig. 12). The results indicate that ENSO and NAO together explain more than 50 % of the total SST variability, while the addition of IOD and ISM increases the explained variance only marginally. This limited contribution reflects both their weaker individual correlations with SST and their strong association with ENSO, which reduces their independent explanatory power (Table 2). A decomposition of the total explained variance using hierarchical partitioning further confirms the dominance of NAO and ENSO, with the NAO showing a somewhat stronger unique contribution owing to its weaker correlations with the other three modes (Fig. S9). The analysis also reveals that approximately 20 % of the total variance is jointly explained by all four modes, highlighting the intertwined nature of these large-scale climate drivers.

To better visualize the impact of the interaction of the climate modes on SST, and how it manifested in specific years, we use the Self-Organizing Maps (SOM). Since ENSO and NAO together explain most of the variance in Gulf SST, we applied the SOM analysis to three variables: Gulf SST as the dependent variable, and ENSO and NAO as the independent variables or predictors (Fig. 13).

The distribution of SST anomalies and ENSO/NAO phases on the SOM reveals four Gulf SST regimes, corresponding to four quadrants on the SOM grid: (i) very warm summers where La Niña coincides with a negative NAO such as in 1998, 2000, 2010 (upper-right quadrant); (ii) very cool summers where El Niño coincides with a positive NAO such as in 1982, 1991, 2018 (lower-left quadrant); (iii) average to moderately warm summers where La Niña coincides with a positive NAO such as in 1981, 2020, 2022 (upper-left quadrant) and (iv) average to moderately warm summers where El Niño coincides with a negative NAO such as in 1986, 1987, 2015 (lower-right quadrant) (Fig. 13).

In summary, since ENSO and NAO are largely uncorrelated (i.e., orthogonal), their effects on Gulf SSTs appear to be predominantly additive, with each mode independently contributing to the overall SST anomaly.

ENSO, NAO, IOD, and ISM exert distinct influences on surface temperatures in the Gulf and the Arabian Sea (including the Sea of Oman) (Fig. 14). For example, while La Niña, negative NAO, negative IOD, and strong early-summer monsoon circulation are all associated with extreme summer temperatures in the Gulf, these same climate patterns tend to either favor cooling over much of the Arabian Sea – including the northern Arabian Sea and the Sea of Oman (in the case of the IOD and ISM) – or produce only limited warming, confined to specific areas, such as the northern Arabian Sea (La Niña) or the eastern Arabian Sea (negative NAO), with cooling over the remaining regions.

These contrasting responses between the Gulf and the Arabian Sea are noteworthy, as several previous studies e.g., have assumed that Gulf SSTs respond similarly to those of the adjacent Arabian Sea under large-scale teleconnections. In contrast, our results reveal an anti-correlated behavior between the two basins. For instance, while positive IOD and El Niño events are typically linked to warming across the Arabian Sea, they coincide with weak to moderate cooling in the Gulf.

These contrasts likely arise from two main factors: (i) differences in how these modes modulate local winds – positive IOD and El Niño events are associated with weaker winds over the Arabian Sea but a modest strengthening of the Shamal winds over the Gulf (Figs. 10 and S7); and (ii) stronger role of ocean circulation including upwelling and lateral heat advection in the Arabian Sea relative to the Gulf, due to the shallow and semi-enclosed nature of the latter, which limits the propagation of temperature anomalies from the Arabian Sea into the Gulf, especially in summer .

These opposing SST responses enhance the temperature gradient between the Gulf and the Arabian Sea during summer, potentially weakening the density contrast and, consequently, the water exchange between the two seas. Such a reduction in exchange could have important biogeochemical implications for both systems, affecting nutrient fluxes, oxygen concentrations, and ecosystem productivity .

Our study reveals that extreme summer SSTs in the Gulf are typically associated with both local climatic changes (e.g., weakening of the Shamal winds) and remote large-scale influences (e.g., modulation by climate modes such as ENSO and the intensity of the Indian monsoon circulation).

These findings align with previous research on MHWs, indicating that extreme SSTs often arise from a combination of local-scale processes and large-scale teleconnections . Regarding local-scale drivers, our results align with those from studies of other semi-enclosed seas, which highlight the role of anomalous atmospheric conditions in the development of MHWs , particularly during warmer months and MHW onset periods . Specifically, the weakening of winds and the resulting reduction in surface heat loss via evaporative cooling have been shown to contribute to MHWs in various oceanic settings e.g.,. However, many of these studies also link MHWs to increased shortwave solar radiation due to reduced cloud cover. In contrast, our findings suggest that Gulf SSTs increase in summer despite reduced shortwave radiation. This is due to enhanced atmospheric moisture, which increases haze and reduces incoming solar radiation, while simultaneously enhancing longwave radiation and trapping heat near the surface. On the large-scale teleconnection side, studies in tropical and subtropical regions have shown that ENSO and IOD strongly influence the occurrence of MHWs in the tropical Indian Ocean, while ENSO and the NAO exert strong influence over MHWs in the Pacific and Atlantic Oceans, respectively . However, in contrast to these findings where El Niño events are typically associated with increased MHW activity in the tropical Indian and Pacific Oceans, our study shows that La Niña conditions are more strongly associated with higher Gulf summer SSTs. Similarly, while increased MHW occurrences in the western Indian Ocean are often linked to positive IOD phases, we find that warm summer SSTs in the Gulf are generally associated with neutral or negative IOD phases.

To assess the sensitivity of our results to the choice of reanalysis product, we repeated the analyses using two additional atmospheric datasets: JRA-55 and MERRA-2. Consistent with the ERA5-based results, composites of atmospheric conditions corresponding to extreme summer SSTs (90th percentile) across most of the Gulf, derived from these alternative products, display similar large-scale patterns (Fig. S10). Specifically, both JRA-55 and MERRA-2 show that elevated Gulf SSTs are associated with lower atmospheric pressure over the Arabian Peninsula and higher pressure over Iran and the northern Arabian Sea, accompanied by a weakening of the Shamal winds, a strengthening of the monsoon flow over the southern and western Arabian Sea, and enhanced surface air temperature and total-column water vapor over the Gulf and the Arabian Peninsula (Fig. S10). Similarly, analyses of summer atmospheric conditions associated with major climate variability modes (ENSO and NAO) were repeated using JRA-55 and MERRA-2 (Figs. S11 and S12). The resulting patterns are consistent across all three reanalysis products. As with ERA5, La Niña (El Niño) conditions are associated with lower (higher) atmospheric pressure over the Arabian Peninsula (Fig. S11), while negative (positive) NAO phases correspond to lower (higher) pressure over the Arabian Peninsula and higher (lower) pressure over Iran and the northern Arabian Sea (Fig. S12).

Overall, these comparisons indicate that the large-scale atmospheric circulation patterns associated with extreme summer SSTs in the Gulf are robust across multiple reanalysis datasets. Nonetheless, all three reanalyses share a relatively coarse spatial resolution, which may limit their ability to capture small-scale features such as the effects of complex orography. These limitations are further compounded by the scarcity of long, continuous meteorological observations in the region, which constrains the validation of reanalysis-based fields at local scales.

While this study establishes robust statistical links between Gulf SST anomalies and major climate variability modes such as ENSO and the NAO, primarily through their associated modulation of surface pressure and wind patterns over the Arabian Peninsula and Iran, it does not explicitly examine the underlying physical mechanisms by which these remote modes influence regional atmospheric circulation. Understanding the dynamical pathways that connect Pacific and North Atlantic variability to the Arabian Peninsula’s atmospheric conditions remains a complex problem that extends beyond the scope of the present analysis, which focuses on characterizing SST variability within the Gulf itself. Several previous studies have reported that ENSO and NAO modulate pressure systems over the Arabian Peninsula and Iran without fully explaining the mechanisms responsible for this modulation e.g.,. Other studies e.g., have proposed potential teleconnection pathways involving large-scale Rossby wave trains or adjustments of the subtropical jet stream in the upper troposphere. Nevertheless, a comprehensive dynamical attribution has yet to be established. Future work combining observational analyses with targeted climate-model experiments will be required to elucidate these linkages in greater detail.

Finally, this study focuses on understanding extreme SSTs at the scale of the entire Gulf. Our large-scale analysis reveals that most of the interannual variability is driven by fluctuations in local atmospheric conditions, with only a limited contribution from remote oceanic influences. However, at more localized scales – particularly near the Strait of Hormuz – the influence of oceanic circulation and heat transport from the Sea of Oman, and thus changes within the Sea of Oman itself, may become more significant and warrant further detailed investigation.

We show that summer Gulf SSTs are strongly influenced by large-scale climate modes such as the NAO and ENSO. Consequently, the predictability of these teleconnections can enhance the predictability of summer SSTs in the Gulf.

While summer NAO predictability remains limited – owing to the weaker North Atlantic jet stream and reduced ocean–atmosphere coupling during summer – previous studies have demonstrated moderate skill using coupled models, particularly when initialized in late spring .

In contrast, ENSO generally exhibits higher and improving predictability. However, forecasts are challenged by the spring predictability barrier (March–May), a period marked by weakened ocean–atmosphere coupling . Forecast skill improves in late spring as the barrier subsides and model performance in tracking ENSO evolution through summer increases , particularly for strong events associated with pronounced subsurface heat content anomalies in the equatorial Pacific .

Therefore, subseasonal-to-seasonal forecasts that incorporate North Atlantic and equatorial Pacific precursors may provide skill in predicting summer Gulf SSTs with 2–3 month lead time when initialized in late spring (e.g., May). This predictive skill is likely enhanced during strong ENSO and NAO phases, offering potential for early warning of summer MHWs in the region.

As coral reefs represent the most biodiverse ecosystem in this arid region, with significant economic importance in supporting fisheries and ecotourism , the ability to predict anomalously hot summers weeks to months in advance would be of great value. Mass coral bleaching associated with recurrent marine heat waves over the past several decades have resulted in a loss of 40 % of coral across the Gulf since the mid-1990s, largely because there was no advanced warning to permit preventative measures . Having forecasts of imminent heat risks would allow regional reef managers and other practitioners to proactively deploy approaches to minimize coral bleaching. These include rapid interventions such as reef shading , pumping of deep, cool water onto reefs , or artificial mixing to decrease stratification , as well as using management efforts to reduce additive stressors that exacerbate heat stress e.g., by limiting dredging activities or reducing nutrient inputs during peak heat periods,. Thus, early warning forecasts provide the capacity for rapid adaptive management intervention in ways that were not previously available.