Volume 129, Issue 1 e2023JC019873
Research Article
Open Access

What Forcing Mechanisms Affect the Interannual Sea Level Co-Variability Between the Northeast and Southeast Coasts of the United States?

Ou Wang

Corresponding Author

Ou Wang

Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA, USA

Correspondence to:

O. Wang,

[email protected]

Contribution: Conceptualization, Methodology, Software, Validation, Formal analysis, ​Investigation, Data curation, Writing - original draft, Writing - review & editing, Visualization

Search for more papers by this author
Tong Lee

Tong Lee

Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA, USA

Contribution: Conceptualization, Methodology, Formal analysis, ​Investigation, Resources, Writing - original draft, Writing - review & editing, Supervision, Project administration, Funding acquisition

Search for more papers by this author
Thomas Frederikse

Thomas Frederikse

Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA, USA

Joint Institute for Regional Earth System Science and Engineering, University of California, Los Angeles, Los Angeles, CA, USA

Contribution: Conceptualization, Methodology, Formal analysis, ​Investigation, Writing - review & editing

Search for more papers by this author
Rui M. Ponte

Rui M. Ponte

Atmospheric and Environmental Research Inc., Lexington, MA, USA

Contribution: ​Investigation, Writing - review & editing

Search for more papers by this author
Ian Fenty

Ian Fenty

Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA, USA

Contribution: ​Investigation, Writing - review & editing

Search for more papers by this author
Ichiro Fukumori

Ichiro Fukumori

Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA, USA

Contribution: Conceptualization, Methodology, ​Investigation, Writing - review & editing

Search for more papers by this author
Benjamin D. Hamlington

Benjamin D. Hamlington

Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA, USA

Contribution: ​Investigation, Writing - review & editing

Search for more papers by this author
First published: 25 January 2024

Abstract

Interannual sea-level variations between the United States (U.S.) Northeast and Southeast Coasts separated by Cape Hatteras are significantly less correlated than those within their respective sectors, but the cause is poorly understood. Here we investigate atmospheric forcing mechanisms that affect the interannual sea-level co-variability between these two sectors using an adjoint reconstruction and decomposition approach in the framework of Estimating the Circulation and Climate of the Ocean (ECCO) ocean state estimate. We compare modeled and observed sea-level changes at representative locations in each sector: Nantucket Island, Massachusetts for the Northeast and Charleston, South Carolina for the Southeast. The adjoint reconstruction and decomposition approach used in this work allows for identification and quantification of the causal mechanisms responsible for observed coastal sea-level variability. Coherent sea-level variations in Nantucket and Charleston arise from nearshore wind stress anomalies north of Cape Hatteras and buoyancy forcing, especially from the subpolar North Atlantic, while offshore wind stress anomalies, in contrast, reduce co-variability. Offshore wind stress contributes much more to interannual sea-level variation at Charleston than at Nantucket, causing incoherent sea level variations between the two locations. Buoyancy forcing anomalies south of Charleston, including over the Florida shelf, the Gulf of Mexico, and the Caribbean Sea, also reduce co-variability because they induce sea-level responses at Charleston but not Nantucket. However, the relative impact of buoyancy forcing on interannual sea-level co-variability between the two sectors is much smaller than that of offshore wind stress.

Key Points

  • Nearshore winds north of Cape Hatteras and buoyancy forcing cause coherent interannual sea-level variations between Nantucket and Charleston

  • Offshore winds contribute much more to interannual sea-level variation at Charleston than to that at Nantucket

  • Offshore winds are the major factor causing incoherent interannual sea-level variations between Nantucket and Charleston

Plain Language Summary

The magnitude of year-to-year sea-level variations along the East Coast of the United States (U.S.) can be comparable to that of global mean sea-level rise over a few decades. These year-to-year sea-level variations contribute to more frequent nuisance floods that affect coastal communities. Year-to-year sea-level variations between the U.S. Northeast and Southeast Coasts separated by Cape Hatteras are significantly less correlated than those within the respective sectors. Understanding causal mechanisms affecting the coherence of year-to-year sea-level variations in the two sectors would help improve sea-level predictions. Our study attributes year-to-year sea-level variations at Charleston and Nantucket, proxy locations for the U.S. Southeast and Northeast Coasts, to wind and buoyancy forcing (air-sea heat exchange and precipitation/evaporation at sea surface that both affect seawater buoyancy). We find that nearshore winds north of Cape Hatteras and buoyancy forcing both cause coherent year-to-year sea-level variations between Nantucket and Charleston. Offshore winds affect much more year-to-year sea-level variation at Charleston than that at Nantucket. Offshore winds are the major factor causing less correlated year-to-year sea-level variations between Nantucket and Charleston. Warming/cooling of water south of Charleston affects sea-level variations at Charleston, but not at Nantucket. However, this effect is smaller than that of the offshore wind.

1 Introduction

Interannual sea-level variations along the East Coast of the United States (U.S.) have been a subject of many studies, because sea-level rise (SLR) rate in that region exceeded the global mean SLR rate, posing a serious threat to the densely populated and economically important coastal communities (Andres et al., 2013; Little et al., 20172021; Piecuch et al., 2016; Sallenger et al., 2012; Thompson, 1986; Valle-Levinson et al., 2017; Wang et al., 2022; Woodworth et al., 201420172019; Yin & Goddard, 2013). On interannual through decadal timescales, sea-level variations along the U.S. Northeast and Southeast Coasts, separated by Cape Hatteras, are distinctively more coherent within each sector than between the two sectors (Calafat et al., 2018; Diabaté et al., 2021; Ezer, 2019; Piecuch et al., 2016; Thompson, 1986). Probably because of this reason, many studies of sea-level variations along the U.S. East Coast focused on one sector, or even subregions within a particular sector.

The incoherence for sea-level variations between the U.S. Northeast and Southeast Coasts are time- and frequency-dependent (Little et al., 2021). There is weaker correlation than previous decades for sea-level variations between the U.S. Northeast and Southeast Coasts since around 1992 (e.g., Andres et al., 2013). Roughly from 1990s through 2010, the rate of sea-level rise along the U.S. Northeast Coast has accelerated, while along the U.S. Southeast Coast no acceleration was observed (Andres et al., 2013; Boon, 2012; Sallenger et al., 2012). From 2010 to 2015, sea-level rise has accelerated south of Cape Hatteras and decelerated north of Cape Hatteras (Valle-Levinson et al., 2017).

A natural question is what forcing mechanisms cause sea-level variations between the two sectors to be less coherent than sea-level variations within each sector. Previous studies using simple models and correlation-based analysis suggest different factors related to sea-level variations between the U.S. Northeast and Southeast Coasts. Along the U.S. Northeast Coast, most previous studies suggest interannual sea-level variations are mainly associated with local wind stress on the continental shelf (Andres et al., 2013; Piecuch et al., 20162019; Wang et al., 2022; Woodworth et al., 20142017). Frederikse et al. (2017) found sea-level variations along the U.S. Northeast Coast and steric height in the subpolar North Atlantic are well correlated on decadal time scale. Wang et al. (2022) found that buoyancy forcing overall has a smaller contribution than wind stress to the U.S. Northeast Coast sea-level variations. However, remote buoyancy forcing from the subpolar North Atlantic does play a role on sea-level variations along the U.S. Northeast Coast and can have contributions comparable to local wind stress in some years.

In contrast, sea-level variations along the U.S. Southeast Coast are affected by the Florida Current as well as the Loop Current and the Caribbean Current upstream of the Florida Current (Domingues et al., 2018). Local “nuisance flooding” events along the U.S. Southeast Coast, minor flooding caused by sea-level rising above a certain relatively low threshold, often coincide with low transport of the Florida Current (Sweet et al., 2016). This coincidence is likely associated with the geostrophic balance of the Florida Current (Domingues et al., 2018). As the Florida Current becomes the Gulf Stream and flows offshore south of Cape Hatteras, it has less impact on sea-level variations along the U.S. Northeast Coast north of Cape Hatteras than the Southeast Coast. In the North Atlantic, there is an out-of-phase variability between the subtropical and subpolar gyres, which is part of the so called North Atlantic sea surface height (SSH)/sea surface temperature (SST) tripole (e.g., Volkov et al., 201920222023). Similar to Frederikse et al. (2017) for the U.S. Northeast Coast, Volkov et al. (2019) showed that sea level along the U.S. Southeast and Gulf Coasts is correlated with steric sea level in the North Atlantic subtropical gyre.

The temperature change of the Florida Current is also found to be associated with significant sea-level change along the U.S. Southeast Coast. Domingues et al. (2018) attributed the accelerated SLR south of Cape Hatteras between 2010 and 2015 to a warming Florida Current. In a recent modeling study, Ezer and Dangendorf (2022) found a ±2°C temperature anomaly of the inflow of the Florida Current can cause 5–12 cm of coastal sea-level change.

Wind stress on the continental shelf contributes more to interannual sea-level variations along the U.S. Northeast Coast than wind stress in the open ocean (Wang et al., 2022). Along the U.S. Southeast Coast, however, past studies suggest that open-ocean wind contributes much more to interannual and decadal sea-level variations than along the U.S. Northeast Coast (Bingham & Hughes, 2012; Piecuch et al., 2016; Woodworth et al., 20142017). Wind stress curl in the interior of the Atlantic Ocean induces westward-propagating Rossby waves (Calafat et al., 2018; Dangendorf et al., 2023; Domingues et al., 2016). Once they reach the coast, these waves transform into coastal trapped waves that propagate southward, affecting coastal sea level along their path. Wind-driven gyre-scale heat convergence/divergence has also been found to be correlated with coastal sea-level variations (Volkov et al., 20192023).

In the context of comparing sea-level variations between the two sectors across Cape Hatteras, Yin and Goddard (2013) suggested baroclinic and barotropic processes dominate decadal and multidecadal sea-level changes north and south of Cape Hatteras, respectively. However, their finding of the dominance of baroclinic processes north of Cape Hatteras contrasts with Piecuch et al. (2016), probably because the former study focuses on longer time scales over which baroclinic processes are more important.

Ezer (2019) contended that the strength and location of the Gulf Stream is related to different SLR between the two sectors: weakening and southward shifting (strengthening and northward shifting) of the Gulf Stream is associated with accelerated SLR along the Northeast (Southeast) Coast. Valle-Levinson et al. (2017) argued that the pulses of accelerated SLR events along the U.S. East Coast are related to the cumulative effect of atmosphere forcing anomalies associated with the North Atlantic Oscillation (NAO) and El Niño–Southern Oscillation (ENSO). NAO forcing determines the latitudinal locations of these SLR events.

In addition to the incoherent variability, there is also some coherent variability based on tide gauge sea-level time series between the U.S. Northeast and Southeast Coasts over 1950–2009 (Thompson & Mitchum, 2014). They attributed coherent coastal sea-level variations to variances in divergence of Sverdrup transport off the coastal region. Note, however, that some subset of their time series, for example, after mid 1990s still shows sea-level variations between the two sectors are either uncorrelated or even anticorrelated (see their Figure 4). Valle-Levinson et al. (2017) found that a coherent pattern in 7-year filtered cumulative sea-level variations is anticorrelated with a similarly processed index of ENSO. During El Niño events, sea level drops as weakened northeasterlies and amplified westerlies cause net divergence of the western region of the Atlantic. Supporting the influence of El Niño, Dong et al. (2022) found that warming/cooling conditions in the eastern equatorial Pacific correlate with the Florida Current transport on interannual time scales, which in turn is linked to the coastal sea level.

Correlation-based studies have identified many potential factors that could contribute to sea-level variations along the U.S. East Coast. These factors include local and remote wind and buoyancy forcing, the Gulf Stream and its precedent currents, the Atlantic Meridional Overturning Circulation (AMOC), the Atlantic Multidecadal Oscillation (AMO), NAO and other climate modes (Dangendorf et al., 2023; Ezer, 2019; Ezer & Dangendorf, 2022; Hameed et al., 2021; Kopp, 2013; Little et al., 2019; McCarthy et al., 2015; Valle-Levinson et al., 2017; Volkov et al., 20192023). Some of these factors are not independent of each other, making it difficult to separate one effect from another and even harder to identify the causality of sea-level variations along the U.S. East Coast.

Recently, Wang et al. (2022) used a more rigorous method based on adjoint decomposition to find the causal mechanisms of interannual sea-level variations at Nantucket, Massachusetts (MA) on the U.S. Northeast Coast. In particular, they found that local wind stress is the main contributor and remote buoyancy forcing from the subpolar North Atlantic can have contributions comparable to wind stress in some years.

Here, we apply the same adjoint-based method to determine the causal mechanisms for dynamic sea-level variations at Charleston, South Carolina (SC), south of Cape Hatteras. We then compare contributions from different forcing types and their spatial distribution to shed light on the forcing mechanisms reducing the interannual sea-level co-variability between the U.S. Northeast and Southeast Coasts. We briefly describe the model and the adjoint-based methodology in Section 2. The results are presented in Section 3, in particular about how causal forcing to sea-level variations at Charleston differs from that at Nantucket presented in Wang et al. (2022). We then conclude the paper in Section 9.

2 Model and Methodology

Following Wang et al. (2022), we use the adjoint-based method to reconstruct and decompose interannual sea-level variations by convolving forcing anomaly and adjoint sensitivity. This method identifies contributions to sea-level variations as a function of forcing type, space, and time lag. Unless otherwise specified, sea level in this study, as in Wang et al. (2022), is ocean dynamic sea level, inverse barometer corrected sea level.

Here we briefly describe the adjoint-based reconstruction method. Readers are referred to Fukumori et al. (200720152021) and Wang et al. (2022) for more detailed descriptions about the methodology and its applications to study casual mechanisms of various ocean quantities.

Assuming sea-level anomaly (SLA), J, has stationary, linear response to forcing anomalies, we can expand J in a Taylor series as follows:
J ( t ) i s t J F i ( s , t ) δ F i ( s , t t ) , $J(t)\approx \sum\limits _{i}\sum\limits _{s}\sum\limits _{{\increment}t}\frac{\partial J}{\partial {F}_{i}(s,{\increment}t)}\delta {F}_{i}(s,t-{\increment}t),$ (1)
where J F i ( s , t ) $\frac{\partial J}{\partial {F}_{i}(s,{\increment}t)}$ is the adjoint sensitivity of SLA to forcing Fi at lead time ∆t and location s; and δFi(s,t − ∆t) is the anomaly of forcing Fi at some prior time t − ∆t (i.e., target time, t, minus lead time, ∆t).

To verify the validity of the stationary and linear response assumptions, we compare the SLA time series reconstructed by the convolution in Equation 1 against SLA from observations and a nonlinear primitive equation numerical ocean circulation model. Because Equation 1 is linear, one can quantify the relative contributions of forcing anomalies as a function of forcing type, space, or time lag.

In Wang et al. (2022), we chose the objective function J as interannual sea-level variations at Nantucket, MA. Here, we conduct the same adjoint-based analysis except for defining the objective function J as interannual sea-level variations at Charleston, SC. We then compare the results from the two adjoint-based analyses to shed light on forcing mechanisms affecting the interannual sea-level co-variability between the U.S. Northeast and Southeast Coasts. As in Wang et al. (2022), sea-level anomalies (SLAs) in this study are referenced to its global mean and 1992–2015 time mean. A linear trend and the mean seasonal cycle have also been removed. The time series are 13-month low-pass filtered to focus on the interannual to decadal variability. The model is the flux-forced version of the global, data-constrained ocean and sea-ice state estimate of the project of Estimating the Circulation and Climate of the Ocean (ECCO) Version 4 Release 3 (hereafter ECCO or ECCO V4r3) (Forget et al., 2015; Fukumori et al., 2017). It has a nominal 1-degree grid size. Wind stress, heat flux and freshwater flux are the forcings used to reconstruct J according to Equation 1. Here, buoyancy forcing is defined as the combined contribution of heat flux and freshwater flux. Weekly-mean forcings, with lead times ranging from 0 to 1253 weeks, are used to reconstruct sea-level variations from 1992 through 2015. See Wang et al. (2022) for more details.

3 Results

3.1 Reduced Correlation for Sea-Level Anomalies Across Cape Hatteras

As shown in many past studies, interannual to decadal sea-level variations on the U.S. Northeast and Southeast Coasts are less correlated than those within each of these sectors. Figures 1a and 1b show the correlation coefficients between interannual SLA at each grid point and that at Nantucket (Figure 1a) and Charleston (Figure 1b) based on estimate from European Union's Copernicus Marine Environment Monitoring Service (CMEMS) merged-altimetry gridded product. Similarly, ECCO shows interannual sea-level variations within each sector is more correlated than those across the two sectors (Figures 1c and 1d).

Details are in the caption following the image

Correlation coefficients of CMEMS SLAs at various locations with that at Charleston (a) and Nantucket (b). Panels (c) and (d) are the same presentations as (a) and (b), but for ECCO. A 13-month low-pass filter has been applied to the monthly mean SLAs after removing the global mean, the mean seasonal cycle, and a linear trend. The original 0.25° × 0.25° gridded CMEMS data has been bin-averaged to 1° × 1° to roughly match the size of the ECCO V4 grid.

We also compare interannual sea-level variations (Figures 2a and 2b) among tide gauge (gray) (Holgate et al., 2013), CMEMS altimetry (black), and ECCO (blue) for Charleston and Nantucket. While the three estimates are overall similar between each other, ECCO is more different than the other two. This is likely due to the limitations of coarse-resolution ECCO V4r3 model to resolve some coastal processes. Certain aspects of sea-level variations may need higher-resolution models for accurate representation (see Section 9). ECCO sea level is slightly more similar to observations in Nantucket (correlation coefficient r = 0.79 with tide gauge and 0.80 with CMEMS) than Charleston (r = 0.77 and 0.79). The correlation between tide gauge and CMEMS is 0.93 for Nantucket and 0.95 for Charleston. The variations between Charleston and Nantucket are noticeably disassociated in some years, for instance, during 2011–2015 when sea level sharply rises in Charleston but falls in Nantucket.

Details are in the caption following the image

Sea-level anomaly (cm) from tide gauge (gray), CMEMS altimetry (black), ECCO (blue), and total reconstruction (orange) from all contributions for (a) Charleston and (b) Nantucket. Panels (c) and (d) show decomposition of the total reconstruction into various contributions for (c) Charleston and (d) Nantucket. The two numbers in the legend are standard deviation (cm) and explained variance of the total reconstruction by each contribution. Panels (e) and (f) show separation of total reconstruction (blue) into 13-month to 10-year band-pass filtered (orange) and decadal (green) variations. The numbers in the legend of (e) and (f) are standard deviation (cm) over 1997–2010. Panels (b) and (d) are adapted from Figure 2 of Wang et al. (2022).

3.2 Overall Reconstruction Skill and Contribution by Forcing Type

Having established that ECCO SLA is less correlated between the two sectors across Cape Hatteras as suggested by observations, we apply the adjoint-based reconstruction method to reproduce interannual variations of ECCO SLA in Nantucket and Charleston (blue in Figures 2a and 2b). Total reconstructed SLA (orange in Figures 2a and 2b), using all forcings from all regions and time lags by including all summations in Equation 1, reproduces the ECCO-simulated SLA. This verifies the reasonableness of our linearity and stationarity assumptions. Reconstructed SLA explains similar variance of ECCO-simulated SLA (∼90%) for both Nantucket and Charleston.

We further decompose the total reconstructed SLAs into contributions from each forcing by excluding the partial summation over forcing type i in Equation 1. Figures 2c and 2d show total reconstructed SLAs (orange; same as in Figures 2a and 2b) as well as reconstructed SLAs by wind stress (dashed), buoyancy forcing (dotted), and the contribution of SLA adjustment to the initial state (purple). The two numbers in the legend are standard deviation (σ; cm) of each time series and explained variance of the total reconstruction by each contribution.

The decomposition of total reconstructed SLAs indicates that wind stress is the dominant contributor while overall the contribution from buoyancy forcing is small. Wind stress explains 77% of the total reconstruction variance in Charleston and 67.5% in Nantucket, respectively, much larger than buoyancy forcing (5.3% and −1.0%). The standard deviation of wind stress contribution (1.55 for Charleston and 1.88 cm for Nantucket) is similar to that of the total reconstruction (1.69 and 1.94 cm). Buoyancy forcing contribution has smaller, but sizable, standard deviation (0.83 and 1.19 cm) than wind stress contribution does. Wind stress and buoyancy forcing contributions are weakly anticorrelated (r = −0.18 for Charleston and −0.26 for Nantucket, which are insignificant since the 95% confidence level is 0.41), as indicated by the fact that the variance of the total reconstruction is less than the sum of variances of wind and buoyancy contributions. Buoyancy forcing contribution thus compensates wind stress contribution. Such compensating effects have been noticed by other studies (e.g., Piecuch & Ponte, 2012). As expected, the magnitude of the contribution of adjustment to the initial state is small and limited to the first few years (see Wang et al. (2022) for how the contribution to initial conditions was quantified).

Valle-Levinson et al. (2017) investigated the large increase in sea level at Charleston between 2011 and 2015 and posited that it was due to the cumulative effect of wind anomalies related to NAO and ENSO. Our adjoint-gradient based decomposition of Charleston SLA shows that wind stress was, in fact, the dominant driver of SLR over 2011–2015, as the magnitudes of SLR in this period are comparable between the total and wind stress reconstructions (Figure 2c).

Note that the 13-month low-pass filtered monthly time series shown in Figures 2a–2d are 24-year long and thus include some decadal variability. Figures 2e and 2f show the separation of the total reconstruction into interannual variability (13-month to 10-year band-pass filtered) and decadal variabilities. While the interannual variability is dominant (σ = 1.34 and 1.76 cm over 1997–2010 for Charleston and Nantucket), the decadal variability is small but not negligible (σ = 0.39 and 0.45 cm).

3.3 Comparing Reconstructed SLAs Between Charleston and Nantucket

To give a better understanding of how different atmosphere forcing anomalies contribute to SLAs in Charleston and Nantucket, we compare the reconstructed SLA time-series between the two locations for wind stress, buoyancy, and total forcing (Figure 3). Although the total reconstructed SLAs appear anticorrelated between Charleston and Nantucket (Figure 3a), the correlation coefficient (r = −0.22) is statistically insignificant at the 95% confidence level. The confidence level is relatively high because the length of the monthly SLA time series over 1992–2015 is short and the 13-month low-pass filtering further reduces the number of degrees of freedom. The reconstructed SLAs associated with wind stress forcing are also statistically uncorrelated (r = 0.03; Figure 3b). In contrast, buoyancy-reconstructed SLAs are significantly correlated between Charleston and Nantucket (r = 0.57) at the 95% confidence level (Figure 3c). Buoyancy forcing is more important at lower frequencies than wind stress (cf. Figures 3c and 3b).

Details are in the caption following the image

Comparison of SLAs (cm) between Charleston and Nantucket reconstructed using (a) all forcings, (b) wind stress, and (c) buoyancy forcing. The numbers are the correlation coefficients for each pair. Asterisk indicates that the correlation coefficient is insignificant at the 95% confidence level.

We then compare spatial patterns of forcing contribution, the so-called forcing influence maps (FIMs), for Charleston and Nantucket SLAs (Figure 4 where panels b and d are modified from Figure 3 of Wang et al., 2022). The values are fractional variance per unit area (km−2) of total reconstructed SLA at Charleston or Nantucket explained by forcing at a particular grid point. The explained variance of time-series X by Y represents how well X is predicted by Y in terms of both phase (e.g., correlation) and magnitude (e.g., variance or standard deviation). The time-series X is total reconstructed interannual SLA in Charleston or Nantucket, while time-series Y is reconstructed SLA at a specific location. A positive value means the variance of the residual (X-Y) is smaller than the variance of X, indicating a good fit. Conversely, a negative value would mean the variance of the residual (X-Y) is larger than the variance of X, indicating a poor fit. For further details, see Fukumori et al. (2015; their Appendix A) and Wang et al. (2022). Our discussion primarily focuses on the regions with large explained variance values.

Details are in the caption following the image

Forcing influence maps for Charleston SLAs due to (a) wind stress and (b) buoyancy forcing. Panels (c) and (d) are the same as (a) and (b) but for Nantucket. The values represent fractions per unit area (km−2) of variance of total reconstructed interannual SLA at Charleston or Nantucket explained by reconstructed SLA using forcing at each location.

Consistent with the reconstructed SLA variations versus time in Figures 2c and 2d, wind stress (Figures 4a and 4c) overall has larger contributions than buoyancy forcing (Figures 4b and 4d) for both Charleston and Nantucket. Nearshore wind stress north of their respective locations both has relatively large contributions to SLAs, attributable to coastal trapped waves that propagate counterclockwise along the coast in the Northern Hemisphere and reach the two locations from north. Adjoint sensitivities to wind stress show that within a lead time of a few weeks there are large values along the shelf to the north of both locations, respectively (see Figures S1a and S1e in Supporting Information S1 for Charleston and Figures 1i and 1m of Wang et al. (2022) for Nantucket). We have analyzed the temporal evolution of adjoint sensitivities to wind stress at weekly intervals within a lead time shorter than a few weeks (not shown). The analysis suggests that the time scale and the regions of these large sensitivities are consistent with the well-known theory of coastal trapped waves (Hughes et al., 2019; Pujiana et al., 2023). The signs of these adjoint sensitivities indicate that both eastward and northward wind stresses cause negative sea level variations. As the coast near Charleston and Nantucket is aligned in the southwest-northeast direction, both eastward and northward wind stresses have a northeastward along-shelf component. Northeastward along-shelf wind stress anomalies trigger seaward Ekman transport that creates negative nearshore sea level anomalies. These anomalies then propagate southward, leading to the negative sensitivities of sea level to wind stress in both study areas.

Buoyancy forcing contribution overall has similar patterns for both Charleston (Figure 4b) and Nantucket (Figure 4d). For both locations, there are positive explained variance in the Labrador and Irminger Seas and negative values in the nearshore region between Charleston and Nantucket, in the Flemish Cap, on the outer continental shelf of Grand Banks, and in the Iceland Basin. Charleston and Nantucket SLAs reconstructed using buoyancy forcing are also highly correlated (r = 0.57; Figure 3c), even including buoyancy forcing contributions from regions where the two buoyancy FIMs differ most—the Florida shelf south of Charleston, the Gulf of Mexico and the Caribbean Sea (Figures 4b and 4d). Thompson (1986) identified coherent signals at periods longer than 6 years between Charleston and Boston. Wang et al. (2022) used a forward perturbation experiment to investigate how a sea-level signal, related to the perturbation of heat flux in the subpolar North Atlantic, propagates to Nantucket. The signal from the subpolar North Atlantic takes about 2–3 years to reach Nantucket, likely via the Labrador Current, given that the propagation timescale aligns with the Labrador Current's mean speed and the distance from the subpolar North Atlantic to Nantucket. From Nantucket, the signal continues to propagate southward to Charleston and beyond (their Figures 6d and 6e). It is thus suggested that buoyancy forcing, particularly from the subpolar North Atlantic, results in correlated, relatively lower-frequency interannual variations across Cape Hatteras. Coherent low-frequency (>6-year) signals, as identified by Thompson (1986), are likely caused by buoyancy forcing in the subpolar North Atlantic. Further study is needed to investigate the detailed oceanic processes related to the steric anomaly's propagation from Nantucket across Cape Hatteras: for example, advective-diffusive processes associated with the steric anomaly versus the role of higher-mode coastal trapped waves induced by the steric anomaly. Direct equatorward nearshore advection of the steric anomaly cannot be the cause because the background nearshore current between Charleston and Nantucket is northward in the model (Figure S2 in Supporting Information S1). Whether southward transport of buoyancy anomalies from the subpolar gyre in the deep western boundary current plays a role (Andres et al., 2018; Le Bras et al., 2023) also needs to be investigated in future work.

Despite some similarity between Charleston and Nantucket FIMs as described above, there are some pronounced differences. In particular, wind stress offshore of Charleston has a larger contribution to Charleston SLA than offshore wind stress to Nantucket SLA. In fact, the red zonal band off Charleston is the most pronounced pattern in the wind stress FIM for Charleston (Figure 4a). Rossby waves induced by wind stress curl likely cause the westward prorogation of this offshore contribution (Calafat et al., 2018; Dangendorf et al., 2023; Domingues et al., 2016). The wave-like patterns shown in maps of sea level sensitivity to wind stress as a function of forcing lead time are consistent with this mechanism (Figure S1 in Supporting Information S1; also see Figure 4 of Frederikse et al. (2022)). We have analyzed the propagating wave patterns shown in these sensitivity maps. The length scale (larger than a few hundred kilometers) and westward-propagating speed (a few centimeters per second) of these wave patterns are consistent with first baroclinic mode Rossby waves at this latitude (Chelton & Schlax, 1996). The contribution of offshore wind stress for Charleston is mainly due to zonal wind stress, as opposed to meridional wind stress (Figure S3 in Supporting Information S1). Contributions of zonal (σ = 0.82 cm) and meridional wind stress (σ = 0.22 cm) explain 94% and 18%, respectively, of the variance of offshore wind stress contribution in Charleston (σ = 0.9 cm). The patterns for near shore wind stress contributions are also noticeably different: widespread over the shelf for Nantucket and more along a stripe over the continental slope for Charleston.

Buoyancy FIMs also have some noticeable difference between Charleston and Nantucket. South of Charleston, there are large positive buoyancy forcing contributions to Charleston SLA (Figure 4b). In contrast, buoyancy forcing in that same region contributes minimally to Nantucket SLA (Figure 4d). Such buoyancy forcing contribution to Charleston SLA, however, is smaller than the offshore wind stress contribution (see later in Section 7 and Figure S5 in Supporting Information S1). The contributions from south of Charleston suggest an advective nature, probably by mean flows of the Florida Current and Loop Current, as coastal trapped waves would propagate southward (counterclockwise along the coast) away from Charleston. The time evolution of adjoint sensitivities of Charleston sea level to heat flux (Figures S1i–S1l in Supporting Information S1) supports the hypothesis of an advective process. As the time lag increases, large sensitivities appear along the Florida Current, the Loop Current, and the Caribbean Sea within less than a year. The propagation speed of these significant sensitivities, inferred from the time lag and the distance between Charleston and the locations where these large sensitivities occur, aligns with the average current speed of the Florida Current and its upstream currents, which is a few tens of centimeters per second (Garcia & Meinen, 2014; Volkov et al., 2020). This advective nature of heat flux contributions south of Charleston contrasts with the westward propagation of Rossby waves suggested by offshore winds discussed above.

In summary, nearshore wind stress anomalies north of each respective study area contribute positively to interannual sea-level variations for both Charleston and Nantucket. Overall, buoyancy forcing causes correlated SLA variations at both locations. Offshore wind stress has a greater contribution to interannual sea-level variations for Charleston than Nantucket. Buoyancy forcing along the Florida shelf south of Charleston, the Gulf of Mexico, and the Caribbean Sea also has a larger contribution for Charleston than Nantucket. For Charleston, offshore wind stress has a greater contribution to sea-level variations than buoyancy forcing from the Florida shelf south of Charleston, the Gulf of Mexico, and the Caribbean Sea.

As pointed out in Section 4, there are some small decadal variability signals in the time series. Figure S4 in Supporting Information S1 shows the same FIMs as Figure 4, but for the time series excluding the decadal variability. Excluding the decadal variability, the FIMs overall show the same patterns as in Figure 4. The mixture of decadal variability in the time series thus does not substantially affect the origins of the salient contributions.

3.4 How Wind Stress Reduces Charleston and Nantucket SLA Correlation

We further analyze and quantify the differential contributions of offshore and nearshore wind stress between Charleston and Nantucket. Figures 5a and 5b show the offshore and nearshore contributions to Charleston SLA, while Figures 5c and 5d are for contributions to Nantucket SLA. These figures are obtained by taking Figures 4a (Charleston) and 4c (Nantucket) and masking out the values outside their respective regions. Nearshore versus offshore is defined by seafloor depths shallower than or deeper than 2,000 m, respectively, within some latitudinal and longitudinal limits. The latitudinal and longitudinal limits were chosen based on Figure 4 so that the pattern inside each region is coherent. The latitudinal limits for the nearshore and offshore regions for Charleston are the same as the South Atlantic Bight. The eastern longitude limit for the offshore region is where the values of the FIMs become close to zero. The nearshore region for Nantucket includes the shallow region between the Mid-Atlantic Bight to the Scotian Shelf (same as the regional forcing box in Wang et al., 2022). The offshore region is from Nantucket to the northern limit of the Flemish Cap. The exact regions in Figure 5 are defined as follows: 82°–45°W, 28°–36°N, ≥2,000 m for Figure 5a; 82°–72°W, 28°–36°N, 0–2,000 m for Figure 5b; 82°–30°W, 40°–49°N, ≥2,000 m for Figure 5c; 80°–60°W, 35°–45°N, 0–2,000 m for Figure 5d.

Details are in the caption following the image

Forcing influence maps for Charleston due to (a) offshore (a) and (b) nearshore wind stress. As (a) and (b), (c) and (d) are for Nantucket.

For Charleston, offshore wind stress contribution is larger than or at least comparable to nearshore wind stress contribution. The offshore wind stress explains 49% of total reconstructed Charleston SLA variance, while nearshore local wind stress in the South Atlantic Bight explains 42% of variance. For Nantucket, offshore wind stress contribution is much smaller than nearshore wind stress contributions, explaining 36% and 66% of the variance of total reconstructed Nantucket SLA, respectively. Note that these numbers are obtained by first calculating a regional-forcing reconstructed SLA using wind stress from a specific region and then computing how much variance of Charleston or Nantucket SLA is explained by the regional-forcing reconstructed SLAs. They are not obtained by calculating area-weighted sum of values in Figure 5, as explained variance is not additive. In summary, offshore wind stress contributes much more to interannual sea-level variation at Charleston than at Nantucket.

Does offshore wind stress off Charleston tend to reduce the correlation between Nantucket and Charleston SLAs? To answer this question, we compare time series of Nantucket SLA reconstructed using wind stress from all regions (blue curve in Figure 6 that is the same as the dashed curve in Figure 2d) and Charleston SLA reconstructed using wind stress from all regions (orange), the offshore region east of Charleston enclosed by the polygon shown in the inset (green), and regions outside the polygon (red). The two curves reconstructed using wind stress from all regions (blue and orange for Nantucket and Charleston that are the same as those in Figure 3b) have a correlation coefficient r = 0.03. The correlation coefficients between Nantucket SLA reconstructed using wind stress from all regions and regional wind-stress reconstructed Charleston SLA are −0.1 for the offshore region and 0.14 for all areas outside that region, respectively. Among the three Charleston SLA time series, the one reconstructed using offshore wind stress has the smallest correlation coefficient with the Nantucket SLA time series. Note, however, that the three aforementioned correlation coefficients are not significant even at the 68% confidence level. Later in this subsection, we further identify additional regions where wind stress might cause wind-reconstructed Charleston SLA to correlate or be uncorrelated/anticorrelated with Nantucket SLA. We then investigate, in all regions where wind stress might reduce the correlation between Charleston and Nantucket SLAs, the extent to which the offshore region would contribute.

Details are in the caption following the image

Nantucket SLA reconstructed using wind stress from all regions (blue) in comparison with Charleston SLA reconstructed using wind stress from all regions (orange), the offshore region east of Charleston enclosed by the polygon shown in the inset (green), and regions outside the polygon (red). The numbers in the legend are correlation coefficients between blue and other curves with asterisk indicating insignificant correlation. The inset is a zoom-in of Figure 4a.

Although Nantucket SLA reconstructed using wind stress from all regions and Charleston SLA reconstructed from all areas outside the offshore region is not significantly correlated, the correlation is expected to be higher if only nearshore wind stress is used to reconstruct both Nantucket and Charleston SLAs. Wind-induced coastal trapped waves can cause correlated sea-level variations over a large swath of coastal regions. Indeed, reconstructed SLAs in Nantucket and Charleston using only nearshore wind stress are higher correlated (r = 0.35; significant at the 90% confidence level) than the reconstructions using the full wind-stress forcing (r = 0.03; cf. Figures 7 and 3b). The nearshore regions are shown in the polygon of the inset diagrams (left for Nantucket and right for Charleston). Note that the nearshore regions are chosen to be only north of Nantucket and Charleston, respectively, to align with the propagation direction of coastal trapped waves. The two nearshore regions are defined as 75°–40°W, 41.5°–60°N, 0–2,000 m for Nantucket and 75°–40°W, 32.5°–60°N, 0–2,000 m for Charleston (excluding a small region south of Greenland).

Details are in the caption following the image

Wind-reconstructed SLAs (m) for Nantucket (blue) and Charleston (orange) using nearshore (<2,000 m) wind stress north of their respective locations to 60°N. The two insets are zoom-in of Figures 4a and 4c, with the polygon showing where the near shore wind stress is. The number (0.35) is the correlation coefficient between the two curves which is significant at the 90% confidence level, but not at 95%.

We further decompose wind-stress contributions to Charleston SLA by using wind stress from additional subregions and then compute the correlation coefficient between each of the regional-forcing reconstructed Charleston SLAs and Nantucket SLA reconstructed using wind stress from all regions. The purpose is to identify how wind stress from each subregion tends to improve or degrade the correlation of regionally forced reconstructions of Charleston SLA with Nantucket SLA.

We define six subregions, including four nearshore regions along the U.S. East Coast, the offshore region off Charleston (same as that in Figure 6), the Gulf of Mexico and the Caribbean Sea. The four nearshore regions are the regions having a depth of 0–2,000 m between 60°N and Nantucket, Nantucket and Cape Hatteras, Cape Hatteras and Charleston, Charleston and Key Largo, Florida (the northernmost of the Florida Keys).

Figure 8 shows the map of correlation coefficients between Nantucket wind-reconstructed SLA from all regions and Charleston reconstructed SLA using wind stress from one of the six subregions. The nearshore regions have decreasing correlation from north to south. Charleston SLA reconstructed with wind stress from north of Cape Hatteras is correlated with Nantucket wind-reconstructed SLA (using wind stress from all regions), while those from south of Cape Hatteras are uncorrelated or anticorrelated with Nantucket wind-reconstructed SLA.

Details are in the caption following the image

Correlation coefficient between wind-reconstructed Nantucket SLA (global) and regional wind-reconstructed Charleston SLA from a specific region.

Among the six subregions, only wind stress from the two nearshore regions located between Nantucket and 60°N (region 1) and Cape Hatteras and Nantucket (region 2) can cause Charleston SLA to significantly correlate with Nantucket SLA reconstructed with wind stress from all regions. In contrast, Charleston SLA reconstructed using wind stress from all regions excluding the two nearshore regions (hereafter “the other regions”) is uncorrelated with Nantucket wind-reconstructed SLA. Using the latter as the reference, Table 1 compares Charleston SLAs, which are reconstructed using wind stress from the three regions. Charleston SLAs reconstructed using wind stress from the two nearshore regions between Cape Hatteras and 60°N are not only significantly correlated with Nantucket wind-reconstructed SLA, but also explaining some of its variance. In contrast, Charleston SLA reconstructed using wind stress from “the other regions” is not correlated with Nantucket wind-reconstructed SLA, showing negative explained variance.

Table 1. Correlation Coefficient, r, Between Nantucket Wind-Reconstructed SLA (Global) and Charleston SLAs Reconstructed Using Wind Stress From Various Regions (Each Represented by a Row in the Table), Along with Additional Statistics
Region Description r Influence on SLA co-variability between sectors Std. Dev. (cm) Explained variance
1 Nearshore wind stress north of Nantucket to 60°N 0.77 Improve 0.19 14%
2 Nearshore wind stress between Nantucket and Cape Hatteras 0.59 Improve 0.15 9%
Other Wind stress from all regions other than the nearshore region between Cape Hatteras and 60°N −0.13* Degrade 1.49 −82%
  • Note. Standard deviation (cm) for Charleston SLAs reconstructed using wind stress from various regions are also provided. The last column shows the percentage of variance of Nantucket wind-reconstructed SLA (global) explained by Charleston SLAs reconstructed using wind stress from various regions. Nearshore versus offshore is defined by seafloor depths shallower than or deeper than 2,000 m, respectively. Asterisk indicates insignificant correlation at the 95% confidence level.

Having established that wind stress from “the other regions” collectively results in Charleston SLA becoming uncorrelated with Nantucket SLA, we now use Charleston SLA, reconstructed using wind stress from “the other regions,” as the reference to further investigate its contributions (Table S1 in Supporting Information S1). The offshore wind stress off Charleston contributes the most to the reference. It is highly correlated with the reference (84%), has a relatively large standard deviation (0.9 vs. 1.49 cm for the reference as shown in Table 1), and explains 65% of the variance. The nearshore region between Charleston and Cape Hatteras contributes far less than the offshore wind stress in terms of standard deviation and explained variance. Other regions have even smaller contributions to the reference. This suggests that offshore wind stress is the major factor causing the less correlated interannual sea-level variations between Nantucket and Charleston.

Our finding that offshore wind stress reduces the correlation between Nantucket and Charleston SLAs confirms previous studies using a simple barotropic model (Piecuch et al., 2016) and correlation analysis (Woodworth et al., 2017) that suggested sea level south of Cape Hatteras is more likely influenced by offshore processes. Hong et al. (2000) found the decadal sea-level variations along the U.S. Southeast Coast is highly correlated with open-ocean wind stress forcing. Calafat et al. (2018) showed Rossby waves can modulate the decadal change of the sea-level annual cycle along the U.S. Southeast and Gulf Coasts. Our study suggests offshore wind stress, given their strong contribution to the variance of Charleston sea level, plays an important role and partly reduces the correlation of sea-level variations between the Northeast and Southeast Coasts.

The impact of wind stress at different lead times on the correlation between Nantucket and Charleston SLAs is shown in Figure S6a in Supporting Information S1. The largest correlation associated with wind stress occurs at very short lead times (<1 month), consistent with the mechanism of coastal trapped waves discussed earlier.

3.5 Buoyancy Forcing Contribution

The buoyancy FIM for Charleston (Figure 4b) shows that buoyancy forcing has large positive contributions to Charleston SLA from the Florida nearshore region south of Charleston, the Gulf of Mexico, and the Caribbean Sea (hereafter called “upstream FC and LC”). Its pattern suggests that the contributions are from the Florida Current, the Loop Current, and the Caribbean Current. In contrast, buoyancy forcing from these regions has barely any contribution to Nantucket SLA (Figure 4d), likely because the signals from these regions are advected offshore by the Gulf Stream near Cape Hatteras, instead of northward to reach Nantucket, where the Florida Current separates from the coast to become the Gulf Stream.

The contributions of buoyancy forcing in the upstream FC and LC region to sea-level variations at Charleston are still smaller than those of offshore wind stress in terms of variability, maximum magnitude of the reconstructed sea-level variations, and explained variance (Figure S5 in Supporting Information S1). The former has a standard deviation of 0.52 cm and maximum magnitude of 1.6 cm, explaining 28% of the variance of the total reconstructed Charleston SLA. In contrast, the offshore wind stress causes sea-level variations with a standard deviation of 0.9 cm and a maximum magnitude of 2.1 cm, explaining 49% of the variance.

In the upstream FC and LC region, the buoyancy forcing contribution (σ = 0.52 cm) is dominantly due to heat flux (Figure S7 in Supporting Information S1). Contributions of heat flux (σ = 0.48 cm) and freshwater flux (σ = 0.12 cm) explain 95% and 17%, respectively, of the variance of the buoyancy forcing contribution.

Despite the notably different buoyancy forcing contribution in these regions, the reconstructed SLAs using buoyancy forcing from all areas are significantly correlated (r = 0.57) between Charleston and Nantucket at the 95% confidence level. The main reason for this relatively high correlation is because Charleston SLA reconstructed using buoyancy forcing outside the “upstream FC and LC” region has an even higher correlation coefficient (r = 0.67) with Nantucket reconstructed SLA using buoyancy forcing from all regions. In contrast, the correlation coefficient for Charleston SLA reconstructed using buoyancy forcing from inside the “upstream FC and LC” region is only 0.16, insignificant at the 68% confidence level. The variability of the reconstructed Charleston SLA using buoyancy forcing outside the “upstream FC and LC” (σ = 0.58 cm) is comparable to that using buoyancy forcing in that region (σ = 0.52 cm). Figure S6b in Supporting Information S1 shows that the correlation between Nantucket and Charleston SLAs associated with buoyancy forcing is large at longer lead time (>2 years).

Domingues et al. (2018) attribute the accelerated SLR along the U.S. Southeast Coast during 2010–2015 (Valle-Levinson et al., 2017) to warming of the Florida Current. Our study suggests buoyancy forcing from the upstream FC and LC region caused accelerated SLR only over the period from mid 2010 through mid 2013 with a trough-to-peak magnitude of 2.7 cm (green curve in Figure 9). In contrast, offshore wind stress off Charleston caused a more persistent SLR from late 2010 through end of 2015 when the ECCO V4r3 model integration ends, with a larger trough-to-peak SLR of 3.5 cm (green curve in Figure 6). Total wind stress caused even larger trough-to-peak SLR for the 2011–2015 event (orange curve in Figure 6). Domingues et al. (2018) utilized a simple method developed by Piecuch et al. (2016) to estimate wind-induced coastal sea level variations. This method specifically associates coastal sea level variations with changes in alongshore wind stress across the shelf's width (chosen as 200 km; see their Supporting Information in Section Datasets and Methods). Consequently, in Domingues et al. (2018) the sea level variations due to alongshore wind stress changes account only for wind contributions within 200 km from the coast. Our study suggests that wind stress further offshore is a larger contributor to sea level variations at Charleston than nearshore wind stress (refer to Figure 4a and Table S1 in Supporting Information S1).

Details are in the caption following the image

Nantucket SLA reconstructed using buoyancy forcing from all regions (blue) in comparison with Charleston SLA reconstructed using buoyancy forcing from all regions (orange), the upstream Florida Current and Loop Current region enclosed by the polygon shown in the inset (green), and the regions outside the polygon (red). The numbers in the legend are correlation coefficients between blue and other curves with asterisk indicating insignificant correlation at the 95% confidence level. The inset is a zoom-in of Figure 4b.

4 Concluding Remarks

We investigated the forcing mechanisms affecting the interannual sea-level co-variability between the Northeast and Southeast Coasts of the United States, using Nantucket and Charleston as the representative locations for the two sectors. Interannual sea-level variations between the two sectors are distinctively less correlated than within each sector. Following Wang et al. (2022) that investigates local and remote forcings for interannual sea-level variations at Nantucket, we employ the same rigorous adjoint-based method to first identify causal mechanisms of interannual sea-level variations in Charleston, SC. The drivers of interannual sea-level variations at Nantucket and Charleston are decomposed into separate contributions from wind stress and buoyancy forcing, as a function of space and lag time, and then compared.

We found that nearshore winds north of Cape Hatteras through 60°N cause correlated interannual sea-level variations between Nantucket and Charleston. This is consistent with the theory of coastal trapped waves, which propagate counterclockwise along the coast in the Northern Hemisphere. In general, buoyancy forcing, especially that from the subpolar North Atlantic, also causes correlated interannual sea-level variations between Nantucket and Charleston.

Offshore winds contribute much more to interannual sea-level variations at Charleston than to those at Nantucket. For Charleston sea-level variations, offshore wind stress has larger contributions than nearshore wind stress, explaining 49% and 42% of total reconstructed Charleston SLA variance, respectively. For Nantucket, offshore wind stress has much smaller contribution than nearshore wind stress, explaining 36% and 66% of total reconstructed Nantucket SLA variance, respectively. Our study confirms previous studies that suggest offshore processes have important influence on sea-level variations along the U.S. Southeast Coast in contrast to the U.S. Northeast Coast (Hong et al., 2000; Piecuch et al., 2016; Woodworth et al., 2017). Our quantitative result discussed in relation to Figure 5a pinpoints offshore wind stress as the key offshore process.

Offshore wind stress is not only the largest contributor to Charleston interannual sea-level variations but is also the major factor in lowering the correlation between interannual sea-level variations between Nantucket and Charleston. Charleston SLA, when reconstructed using wind stress from all regions excluding the nearshore region between Cape Hatteras and 60°N, is uncorrelated with Nantucket SLA reconstructed using wind stress from all regions (Table 1). For the former, the offshore wind stress off Charleston is the largest contributor, explaining 65% of its variance and exhibiting the highest correlation (0.84). Furthermore, Charleston SLA reconstructed using offshore wind stress has much larger variability (σ = 0.9 cm) than those reconstructed using nearshore wind stress from some nearshore regions, the largest being σ = 0.41 cm from that between Cape Hatteras and Charleston (Table S1 in Supporting Information S1). The explained variance, correlation, and variability magnitude all support the conclusion that offshore wind stress off Charleston is the major factor causing interannual Charleston sea-level variations less correlated with those at Nantucket.

In contrast to wind forcing, buoyancy forcing overall increases the correlation in interannual sea-level variations between the U.S. Northeast and Southeast Coasts. Buoyancy forcing contributions for the two sectors have similar spatial patterns, especially in the subpolar North Atlantic (Figures 4b and 4d). Buoyancy forcing from the Florida shelf south of Charleston, the Gulf of Mexico, and the Caribbean Sea has notably positive contributions to sea-level variations at Charleston, but less so at Nantucket. Therefore, buoyancy forcing from these regions reduces the correlation for sea-level variations between the U.S. Southeast and Northeast Coasts. However, this forcing mechanism is secondary to that of offshore wind stress off Charleston in reducing the correlation of sea-level variations between the two coastal sectors. Sea level reconstructed using buoyancy forcing from these regions has substantially smaller variability than (σ = 0.52 cm) that reconstructed using offshore wind stress off Charleston (σ = 0.9 cm).

Our study significantly advances the understanding of the similarity and difference in forcing mechanisms for sea-level variations between the U.S. Northeast and Southeast Coasts. The results can benefit the development and improvement of sea-level prediction models, including those using machine-learning methods.

It is worth noting that the ocean model used to create the ECCO V4r3 state estimate in this analysis does not resolve the full range of ocean coastal dynamical processes that contribute to interannual sea level variations. While ECCO, tide gauge data, and CMEMS generally align in their representations of interannual sea-level variations, some discrepancies between the model and observations are evident (Figures 2a and 2b). Future data-constrained ECCO state estimates constructed with higher-resolution ocean models are expected to have a better fit to observed coastal sea level variations.

Acknowledgments

This research was carried out at the Jet Propulsion Laboratory, California Institute of Technology, under a contract with the National Aeronautics and Space Administration (80NM0018D0004). © 2023. All rights reserved.

    Data Availability Statement

    This study uses ECCO V4r3 product (https://ecco.jpl.nasa.gov/drive/files/Version4/Release3) (Fukumori et al., 2017), tide gauge data (https://psmsl.org/data/obtaining/) (Permanent Service for Mean Sea Level, 2023), and CMEMS altimetry data (https://resources.marine.copernicus.eu/product-detail/SEALEVEL_GLO_PHY_L4_MY_008_047/INFORMATION) (E.U. Copernicus Marine Service Information, 2023). The adjoint sensitivity, forcing, and adjoint-convolution script are available on Zenodo at https://zenodo.org/records/10084712 (Wang, 2023).