Links between large-scale Atlantic Ocean variability and coastal sea level (CSL) along the U.S. East Coast have been suggested long time ago from early observations (Montgomery 1938; Blaha 1984; Maul et al. 1985) and early models (Sturges and Hong 2001; Ezer 2001). Numerous studies suggested different ways in which changes over the Atlantic Ocean can affect the coast, for example, through variations in the Gulf Stream (GS), variations in the Atlantic Meridional Overturning Circulation (AMOC), variations in the North Atlantic Oscillation (NAO), or due to other ocean dynamic factors (Leverman et al. 2005; Sallenger et al. 2012; Ezer et al. 2013, 2015, 2025a, 2025b; Gawarkiewicz et al. 2012; Chen et al. 2014; Piecuch et al. 2016; Little et al. 2019; Dangendorf et al. 2021, 2023; Volkov et al. 2019, 2023; Ezer and Dangendorf 2020; Ezer and Updyke 2024). The way that offshore ocean dynamics can affect CSL variability often involves sea level signals that propagate toward the coast by mechanisms such as Rossby waves and barotropic waves, while coastal-trapped waves (CTW) spread signals along the coast (Huthnance 2004; Hughes and Meredith 2006). The clearer connector between offshore dynamics and CSL is the GS. The GS flows from the Florida Straits (where it is referred to as the Florida Current, FC) along the coast of the South Atlantic Bight (SAB), then it separates from the coast near Cape Hatteras, North Carolina, and turnes toward the northeast, offshore of the Mid-Atlantic Bight (MAB). Since the GS flow speed is proportional to the sea level slope across the stream (due to the geostrophic balance), variations in the position and/or strength of the GS were found in many studies to be linked with CSL (i.e., weakening GS is correlated with increased sea level near the coast and decreased sea level offshore east of the stream; see for example on the Gulf Stream’s induced sea level in Ezer et al. 2013). This GS-CSL connection is important for decadal variations, as well as for the potential of climate related AMOC slowdown (Smeed et al. 2014, 2018; Caesar et al. 2018; Dong et al. 2019; Volkov et al. 2023) that may contribute to future sea level rise and increased flooding (Sweet et al. 2009; Ezer and Atkinson 2014; Sweet and Park 2014; Goddard et al. 2015; Wdowinski et al. 2016).

Unlike the great interest in remote influence on the coast from long-term variability (time scales of interannual to decadal and longer), less is known about high frequency remote forcing of coastal variability. HFV in coastal dynamics is, in most cases, linked to local drivers such as daily weather events, seasonal wind pattern, river discharges, as well as extreme events like hurricanes, tropical storms and winter storms (Lee and Williams 1988; Kohut et al. 2006; Ezer et al. 2017; Ezer 2018; Park et al. 2022, 2024; Todd et al. 2018). However, some studies show that high frequency variations in the GS can influence coastal sea level (Ezer 2016) and coastal currents (Ezer 2025a) in a similar manner as long-term variabilities do (i.e., weakening GS is linked with rising CSL). Observations and model simulations indeed show significant anticorrelation between HFV in the GS and in CSL (Ezer 2016). The latter study also demonstrates that GS-driven CSL variations are quite different than wind-driven CSL variations in that the coastal response to the GS is more coherent along the entire U.S. coast than the response to wind which is affected by local land and coastal topography. One important topographic feature that influences the CSL-GS relations is Cape Hatteras, which separates between the SAB in the south, where the GS is flowing closer to the coast, and the MAB in the north, where the GS is flowing away from the coast in deep waters. Several studies thus found different sea level response to forcing in the SAB and the MAB (Piecuch et al. 2016; Valle-Levinson et al. 2017; Domingues et al. 2018; Ezer 2019). These differences between the SAB and the MAB will be assessed here by comparing the coastal response in locations north and south of Cape Hatteras.

Since the goal here is to assess the impact on the coast from HFV in the GS, this forcing must be isolated from other forcing- in particular, eliminating variations in the wind. To do that, the study follows on the footsteps of Ezer (2016), which used regional ocean circulation model with constant surface forcing (heat and wind), but with time-dependent oscillation in the GS transport (which is applied through the lateral open boundary conditions of the model). Unlike Ezer (2016), who conducted short simulations of 60 days, each one with one forced frequency, here longer simulations of one year were conducted with oscillations of multiple frequencies. The goal was to see if the response at the coast includes only the forced oscillations, or that multiple forced oscillations can produce a spectrum of CSL oscillations like those seen in observations. Since the GS produces natural mesoscale variability from ts instability, experiments with and without forced oscillations can tell us about the interaction of forced oscillations with the natural variability of the GS. These simulations over 360 days are long enough to capture many cycles for conducting spectral analysis. However, one may acknowledge that these idealized simulations cannot last much longer without more realistic surface conditions that include for example the seasonal surface heat flux and wind. The model results are also compared with observations of the FC transport (Baringer and Larsen 2001; Meinen et al. 2010) and CSL from tide gauges, to see if the model relations between the GS and CSL resemble the observed relations.

The study is organized as follows: first, the observations are described and analyzed in section 2, then the model setting and the experiments conducted are described in section 3, the results are described in section 4, and finally a summery and conclusions are offered in section 5.

Daily Florida Current (FC) transport from a cable across the Florida Strait (at ~ 27°N) has been recorded by NOAA/Atlantic Oceanographic and Meteorological Laboratory since the 1980s (Baringer and Larsen 2001; Meinen et al. 2010; www.aoml.noaa.gov/phod/floridacurrent/). However, there are some gaps in the data during 1998–2000 and more recently since 2024. Examples of the time evolution and the power spectra for each year between 2021 and 2023 are shown in Fig. 1. There are great differences between the three years with no clear seasonal cycle that is evident in all years. The focus here will be on the existence of intraseasonal high-frequency oscillations with periods ranging from ~ 1 week to ~ 2 months. The most energetic peaks change from year to year, they are: 15, 22, 36 and 61 days in 2021, 26, 46 and 73 days in 2022, and 28, 33, 47 and 73 days in 2023. The source of these oscillations can be local wind variations and mesoscale dynamics (Lee and Williams 1988; Meinen et al. 2010; Frajka-Williams et al. 2013). The goal of this study is to assess if these HFV in the FC can contribute to the variability of CSL.

Hourly water level records from 2 locations were obtained from NOAA tide gauges (https://tidesandcurrents.noaa.gov/), one location in the SAB (Fernandina, Florida; near 31°N) and one in the MAB (Norfolk, Virginia; near 37°N). The latter location was the subject of many sea level studies (Ezer et al. 2013; Ezer, 2019, 2020, 2022) because of the impact of sea level rise on flooding in this region, and past studies that link CSL there to variations in the GS. The CSL of the two locations (Fig. 2 and Fig. 3) show considerable HFV ranging from ~ 1 week to ~ 2 months, but not necessarily at the same periods as in the FC (Fig. 1); correlations between the GS and CSL will be discussed later. The variability shows significant differences between the two sites and from year to year. For example, during 2021 and 2022 sea level in Norfolk had more energetic HFV than Fernandina did, but not in 2023. Some periods tend to repeat in multiple years at the two locations, such as 14–15 days, 23–24 days, 33 days and 52 days. Also, in 2022 both locations had a peak more energetic than the HFV with a period of 122 days (not completely shown). A common oscillation with the same period at the two locations suggests that a large scale change beyond the coast can affect the entire region, such as variations in AMOC or NAO. For example, between April to July 2022 there is an unusual increase in the FC transport (Fig. 1c) accompanied by a decreased sea level in both Fernandina (Fig. 2c) and Norfolk (Fig. 3c); during the same period the NAO index shifts from negative to positive (not shown). This is consistent with studies that show that sea level rises along the U.S. east coast during periods of low NAO index (Ezer 2015; Goddard et al. 2015).

To evaluate more quantitatively the relations between the GS (as measured by the FC transport) and CSL (as measured by tide gauges), linear correlation coefficients between the two are calculated from daily values and shown in Fig. 4a (another year, 2020, was added to the previous analysis). While the correlations indicate that less than 30% of the total CSL variability is directly linked to the GS (maximum R² ~ 0.3), the correlations are statistically significant at over the 95% level (P < 0.05) for all cases except at Norfolk in 2020. In Fernandina the significance level of the correlation is as high as 99.99%. Correlations are higher in the SAB (Fernandina) near the FC than they are farther downstream the GS in the MAB (Norfolk). An interesting result is that correlations increase with decreasing mean transports from 2020 to 2023. Figure 4b shows that when the variability of the FC increases (as in 2022), the CSL variability increases as well, further supporting the hypothesis that the two are connected. However, statistical correlation does not necessarily mean causation, for example, storm events passing the U.S. east coast can affect both the GS and CSL. To assess if variations in the GS are in fact causing variations in CSL, controlled experiments with a numerical ocean circulation model will be used (see next section).

Experiment #1: a control run with fixed inflows and outflows as described above, so the only variability is the natural mesoscale variability due to the GS meandering instability.

Experiment #2: High-Frequency Experiment (HFE) with oscillations of ± 5 Sv at periods of 7 days and 12 days imposed on the FC inflow (and the GS outflow to conserve the volume).

Experiment #3: Low-Frequency Experiment (LFE), same as experiment #2, but with oscillations at periods of 33 days and 46 days.

Figure 6 shows the imposed FC transport and the power spectra in the HFE and LFE cases. These frequencies and their amplitudes were chosen to roughly represent the typical high-frequency oscillations of the observed FC (Fig. 1) while ignoring low-frequency variabilities with periods longer than about 1.5 months. The goal of these experiments was to see what frequencies of CSL variability are generated when the GS includes only distinct known cycles.

Figure 7 shows the mean model surface elevation in HFE (which is almost identical to the other experiments, so not all are shown) and the variability (Standard Deviation, SD, which is the Root Mean Square, RMS, of the surface elevation anomaly) of the three experiments. The area of the largest variability is in the GS extension after it has separated from the coast with variability associated with the mesoscale variability of the meandering stream, eddies, and recirculation gyres (Andres et al. 2020). The maximin variability of ~ 25–30 cm is only slightly less than the observed variability from altimeter data and ocean models of similar resolution which is around 30–40 cm (Chassignet and Xu 2017). The extent of the high variability area here is somewhat smaller than observations, but this is expected since surface wind and heat fluxes as well as interannual variations are neglected in the idealized model. All three experiments have similar maximum variability but very different spatial patterns, showing that the variability away from the coast is mostly driven by internal variability of the GS, and not directly influenced by the imposed HFV in the FC. In the HFE case (Fig. 7d) there is increased variability farther downstream the GS near the MAB – later analysis will further show that indeed, the HFE has larger impact on the coastal MAB than on the SAB.

Figure 8 shows the correlation between the model surface velocity of the FC near 27°N and sea level over the model domain. In both experiments HFE (Fig. 8a) and LFE (Fig. 8b) the entire coast from Florida to Canada has negative correlations (blue). Positive correlations (red) are found only in a few locations east of the GS. This result is consistent with past studies that indicate CSL rise when the GS is weakening. While the pattern of correlations is similar in both cases, the LFE case shows larger absolute correlations (both positive and negative) than the HFE, indicating a large coastal response to GS oscillations with periods of about 1-1.5 months.

To look closely at the type of variations induced by the GS, Fig. 9 shows the time series of CSL at two locations, one in the SAB around 31°N and one in the MAB around 38°N. Because of the coherent correlations across the SAB and MAB (Fig. 8), choosing slightly different locations did not make any significant difference. Qualitative assessment of the CSL clearly show the opposite change in FC velocity and CSL, for example, in the LFE case (Fig. 9b) when the FC was very weak around day 30 CSL in SAB was extremely high, and around day 240 when the FC was stronger, CSL in SAB was lower. However, there are clearly CSL variability that is unrelated to the GS, like the low CSL from days 140–180. The variations of CSL in the SAB, close to the FC, are much larger than the variations in the MAB farther downstream the GS path, when the GS is also farther away from the coast. Compared to the mostly wind-driven observed CSL variability (Fig. 2 and Fig. 3) of ~ 40 cm (excluding big storm surges), the GS-induced variability is only up to ~ 20 cm in the SAB and ~ 5 cm in the MAB.

Power spectra of the CSL in the three experiments are shown in Fig. 10 – they are clearly very different than the GS forcing in the model (Fig. 6). The control experiment with no time-dependent forcing (Fig. 10ef) produces CSL oscillations at periods ranging from about 2 weeks to 1.5 months – these oscillations, while very small in amplitude, represent the natural variability of the GS system due to instability of the GS, meanders, and mesoscale eddies. The inclusion of even small FC oscillations in HFE and LFE increases the CSL energy by 5–10 times and produces unpredictable frequencies beyond the forcing frequencies; period of peaks ranging from 1 week to 90 days (Fig. 10a-d). There are, however, significant differences between the SAB and MAB (left and right panels of Fig. 10, respectively), and between the HFE and LFE (top and middle panels, respectively). In the MAB, oscillations at periods ~ 33–34 days and ~ 43–45 days appear in all 3 cases, even though only in the LFE case (Fig. 10d) these frequencies were forced on the FC. This indicates some intrinsic modes of the GS system in the MAB. On the one hand, only the CSL in the MAB shows high energy at the 7 and 12 days periods (Fig. 10b) when these frequencies were forced in the HFE case. In the SAB on the other hand, low frequency CSL oscillations dominate, with energetic peaks at periods of 60, 72, and 90 days. While the idealized model results are not intended to reproduce the observations, which are changing from year to year (Fig. 2 and Fig. 3), some general characteristics of the CSL variability in the model are consistent with the observations. For example, both the model experiments and the observations show higher energy at high frequencies in the MAB (Norfolk), compared with the SAB (Fernandina). Also, a specific frequency of variability in the FC does not necessarily produces the same frequency in CSL, as the model experiments demonstrate.

Both the idealized model experiments and the observations indicate large differences in the response of CSL in the MAB and in the SAB to offshore variations in the GS. This result is consistent with past studies that show that the coastal areas north and south of Cape Hatters respond differently to remote forcing due to topography and the distance of the GS from the coast (Domingues et al. 2018; Ezer 2016, 2019; Valle-Levinson et al. 2017). Figure 11 shows the correlation between CSL north and south of Cape Hatteras (Fig. 11a) and the correlation between CSL and the GS (Fig. 11b); model correlations (in green) were also compared with the mean observed correlations (in blue). Surprisingly, the correlation between CSL in the MAB and SAB are completely different in the 3 experiments (Fig. 11a), with negative correlations in the control and HFE, and positive correlations in LFE and the observations. This demonstrates that oscillations with periods around 1–2 months can affect larger extent along the coast, while natural GS variability in the MAB or higher frequency oscillations in the GS have difficulty crossing the Cape Hatteras topography separating the MAB from the SAB. The correlation between the GS (i.e., the flow of the FC) and CSL is negative in all the forced model experiments and in the observations with statistically significance of at least 95% (Fig. 11b), indicating that sea level rises when the GS weakens, as expected from theory and past studies. However, the GS may explain only about 3–30% of the total CSL variability (based on R²) depending on the case and location. Correlations in the LFE case are higher than in the HFE. However, one discrepancy between the model and observations is that in the model correlations are higher in the MAB than in the SAB, while observations show the opposite; this may be due to the neglection of wind variability in the model (Ezer 2016 shows that the response of CSL to variations in the wind is very different than the response to GS variations). In summary, while the model and the observations agree with past studies of the relations between CSL and the GS, the model results demonstrate unexpected sensitivity of CSL to the particular frequencies in the GS variability, as well as great spatial variations in the CSL response.

The motivation for this study comes from past and recent research that found numerous processes that link coastal processes and sea level variability along the U.S. east coast with remote influence from the Atlantic Ocean. Most of these studies focus on large-scale or long-term open ocean variability including variations in AMOC, the GS, NAO, Rossby Waves, and the subtropical gyre (Leverman et al. 2005; Ezer et al. 2001, 2015, 2013, 2025b; Gawarkiewicz et al. 2012; Chen et al. 2014; Piecuch et al. 2016; Little et al. 2019; Dangendorf et al. 2021, 2023; Volkov et al. 2019, 2023; Ezer and Updyke 2024). Much less is known about how high frequency variability (HFV) in offshore dynamics, such as those associated with the meandering GS, may affect the coast. These HFV are often related to atmospheric variations such as wind and air pressure changes due to daily and seasonal weather, or extreme events such as hurricanes that cause storm surges (Lee and Williams 1988; Kohut et al. 2006; Ezer 2018; Park et al. 2022, 2024). One exception was the study of Ezer (2016), who used a simple GS model to show that high frequency oscillations (periods of 2–10 days) in the GS can produce coherent CSL variations along the coast. The transfer of the offshore signal to the coast involved fast moving barotropic waves and coastal trapped waves that spread the signal along the coast (Huthnance 2004; Hughes and Meredith 2006). However, the experiments in Ezer (2016) involved GS forcing with only one frequency at a time over a short period of only 60 days, thus analysis of the spectrum of frequencies in the observations and the model were not previously possible. The current study follows on the footsteps of the earlier study using the same numerical model but combining several forcing frequencies and conducting longer simulations of 360 days each with high frequency output interval of 3 hours (2880 data points for each case). Time series of the daily Florida Current transport and hourly tide gauge sea level data for several years were also analyzed using power spectra to assess the high frequency variability that is found in the observations. Another goal was to see if there is a difference in the response of the coast between the SAB when the GS is close to the coast, and the MAB after the GS separated from the coast; this was done by looking at one location north of Cape Hatteras and one location south of Cape Hatters (both in the model and in the observations). Choosing different locations would not make much of a difference in the main results because of the coherence signal found over large coastal area along the coast.

The main findings are summarized as follows:

In summary, the analysis of the observations and the model’s simulations demonstrate the important, but complicated, role of high frequency GS oscillations in contributing to CSL variability. This result makes prediction of sea level variability, sea level rise, and unexpected coastal flooding more difficult to predict. It should be acknowledged though, that even though there is a clear link between a weakening GS and rising CSL with correlations that are statistically significant at 95-99.9%, the high frequency GS variability found here is only responsible for ~ 3%-30% of the total CSL variability, and this link can change dramatically from year to year and from place to place (e.g., between the SAB and MAB). In comparison, on decadal time scales Ezer et al. (2013) found correlations between temporal trends in the GS flow and CSL in the MAB as large as R=-0.85 (i.e., GS may be responsible for ~ 72% of the decadal variability; R² = 0.72). Another resent study of surface currents in the MAB found that the GS may contribute ~ 10%-30% of the coastal variability (Ezer 2025a). In conclusion, while the high frequency GS variability cannot be neglected, wind variability (including storms) should still be recognized as the major driver of coastal dynamics.