1 Introduction

The Chesapeake Bay (CB) is the largest estuary in the U.S., and the large population living on its shores are vulnerable to increased flooding due to sea level rise, SLR (Boon et al. 2010; Ezer and Corlett 2012; Ezer and Atkinson 2014; Sweet and Park 2014; Valle-Levinson et al. 2017; Domingues et al. 2018; Ezer 2022, 2023) and storm surges during hurricanes (Ezer et al. 2017; Ezer 2020b). In addition to global SLR (Kopp et al. 2014; Dangendorf et al. 2019) and land subsidence (Boon et al. 2010; Eggleston and Pope 2013; Karegar et al. 2016; Bekaert et al. 2017; Buzzanga et al. 2020), local sea level variability in CB may also be affected by other factors that are not clearly understood. For example, potential weakening of the Atlantic Meridional Overturning Circulation, AMOC (Smeed et al. 2014; Robson et al. 2014; Ezer 2015) and the Gulf Stream (Ezer et al. 2013) may increase coastal SLR. There are periods when weak AMOC transport is correlated with higher sea level along the U.S. East coast (Ezer 2015; Goddard et al. 2015), but finding direct links between AMOC and coastal sea level may be elusive, since the pattern of sea level variations associated with AMOC varies by location, forcing, and timescale (Ezer 2013; Little et al. 2019; Wang et al. 2024). Piecuch et al. (2019) for example, found that anticorrelation between AMOC and sea level on the New England coast is linked to zonal wind and pressure, but not much to Gulf Stream variability. Internal variability in the Atlantic Ocean may amplify SLR along the East and Gulf Coasts (Dangendorf et al. 2023) through connections between the open ocean variability and the coast (Dangendorf et al. 2021). On interannual to decadal time scales, teleconnection between coastal sea level, AMOC, and Atlantic Ocean dynamics may involve large-scale heat divergence over the subtropical gyre (Volkov et al. 2019) which could influence frequency of floods along the U.S. East Coast (Volkov et al. 2023).

Unlike coasts that are directly exposed to the Atlantic Ocean, sea level variations in a semi-enclosed bay like CB is more complicated due to local estuarine dynamics, local tides, river discharge, and exchange of heat and salt across the mouth of the bay (Valle-Levinson 1998, 2003), which motivated the current study. Recent analysis of sea level in CB using NOAA tide gauges (Fig. 1) found large spatial variations within the bay and an unusual pattern of long-term sea level variability in which stations in the upper bay and stations in the lower bay are anticorrelated (Ezer 2023). On seasonal time scales, differences in sea level between the upper and lower bays were explained by the pattern of annual and semi-annual tides (Ezer 2020a, 2023), but the spatial variations within CB on interannual to decadal time scales could not be explained. Therefore, the goal of this study is to assess these long-term variabilities in the bay in the context of large-scale Atlantic variability as captured by the latest AMOC observations (Moat et al. 2023). Two stations, one in the upper bay (Baltimore) and one in the lower bay (Norfolk) were compared to examine potential mechanisms and forcing that may explain the different response to Atlantic Ocean variability at the two ends of CB. Surface currents near the mouth of the CB were also analyzed to see if there are long-term variations or trends in the exchange processes connecting the Atlantic Ocean and the CB.

Fig. 1
figure 1

A topographic map (depth in meters) of the Chesapeake Bay and locations of tide gauge stations. Stations 1 and 8 at the two ends of the bay were used in this study

Full size image

The study is organized as follows. First, the data sources and analysis methods are described in Section 2, then results are presented in Section 3, focusing on variations in sea level, water temperature, and surface currents, finally a summery and conclusions are offered in Section 4.

2 Data sources and analysis methods

Water level and water temperature records from CB tide gauge stations (Fig. 1) are available from NOAA (https://tidesandcurrents.noaa.gov/). Data at 6-min intervals for 2005–2021 were obtained for Norfolk and Baltimore from the NOAA server (https://opendap.co-ops.nos.noaa.gov/axis/webservices/). The data were first detrended (removing linear trend for the period of our data) and then daily and yearly mean anomalies were calculated. A 10-day low-pass filter was applied to the daily data (Fig. 2a and b) to be consistent with the AMOC data that were filtered at the source (the filter is similar, but not necessarily identical to that used by the RAPID/AMOC group). The climate related linear trends were neglected in the analysis of the anomaly data shown in Fig. 2. It is of interest to acknowledge that the removed trend from 2005 to 2021 was downward -0.1 Sv/y for AMOC, and upward SLR rate of 6.33 mm/y for Baltimore and 6.83 mm/y for Norfolk. These SLR rates are about twice the global SLR rates due to land subsidence in the region (Boon et al. 2010; Eggleston and Pope 2013; Karegar et al. 2016; Bekaert et al. 2017; Buzzanga et al. 2020). Moreover, SLR is accelerating in CB- for example, SLR rates since 1975 were lower, at 4.5 mm/y and 6.1 mm/y for Baltimore and Norfolk, respectively (Ezer 2023). These two stations were chosen following the analysis of Ezer (2023) which shows the different patterns of sea level at the north and south edges of the CB.

Fig. 2
figure 2

Daily sea level anomaly (detrended) for a Baltimore and b Norfolk; green lines are the raw data and black lines are 10-day low-pass filtered data. c AMOC daily total transport anomaly (also after 10-day low-pass filtered at the source)

Full size image

The AMOC observations for 2005–2021 at 26°N are available at twice daily intervals from the AMOC-RAPID site (https://rapid.ac.uk/; see Moat et al. 2023, for the latest data release). Daily and yearly mean anomalies were calculated for each of the AMOC components: the Gulf Stream (GS) transport (measured at the Florida Strait), the wind-driven Ekman transport (EKM) the density driven upper mid-ocean transport (UMO), and the total transport of the meridional overturning circulation (MOC) (Fig. 3).

Fig. 3
figure 3

The components of the AMOC transport: the Gulf Stream transport at the Florida Strait (GS), the Ekman wind-driven transport (EKM), the density-driven upper mid-ocean transport (UMO) and the total transport (MOC)

Full size image

To obtain information on the inflow/outflow at the mouth of the CB, hourly surface currents in this region were obtained from the Coastal Ocean Dynamics Application Radar (CODAR; http://www.ccpo.odu.edu/currentmapping/); 197 spatial data points were averaged to create time series of surface currents (see Atkinson et al. 2009 and Ezer et al. 2022, for details). The current vectors were transformed by rotating the axis 45° southeastward so that negative values represent currents out of the bay (like most estuaries, due to river discharge into the bay there is a mean outflow, i.e., mean current is negative). Daily and yearly detrended mean anomalies were also calculated to be compared with the other data. Monthly river streamflow into the CB (2007–2021) was obtained from USGS (https://www.usgs.gov/media/images/estimated-monthly-mean-streamflow-entering-chesapeake-bay).

Empirical Mode Decomposition (EMD; Huang et al. 1998; Wu et al. 2007; Wu and Huang 2009) was used to analyze modes of variability on different time scales. EMD is a nonstationary nonlinear method that breaks time series records into Intrinsic Mode Functions (IMFs) representing oscillations with time-dependent amplitudes and frequencies, Ci(t), and a long-term trend, r(t). Therefore, the time series is represented by.

$$\eta \left(t\right)=\sum_{i=1}^{N}{C}_{i}\left(t\right)+r(t)$$
(1)

where N is the total number of oscillating modes. Note that since here linear trends have been removed from all data, the trend r(t) represents the remaining nonlinear trend that points for example of sea level acceleration (Ezer and Corlett 2012; Ezer 2023). The EMD method has been used in many studies of sea level variability (Ezer and Corlett 2012; Ezer et al. 2013; Ezer 2013, 2015). Here EMD is used to explore the relation and correlation between different data sets and to see at which time scales different data are linked or not. Note that each EMD mode may not represent a specific process, so often several modes within a window of frequencies are added together, for example, high-frequency, middle-frequency, and low-frequency modes. Ensemble EMD was used to calculate the statistical significance of EMD modes following Wu and Huang (2009), and the statistical significance of correlations between EMD modes was estimated based on the degrees of freedom dependency on autocorrelation scales following the method of Thiebaux and Zwiers (1984). However, one should keep in mind that estimated confidence levels on low-frequency modes may not be as accurate as for high-frequency modes. Note that unlike standard spectral analysis that can only find cyclic variability of constant frequency at each frequency band and is limited to periods much shorter than the record length, in the EMD analysis frequency can change within each mode and non-linear trends with incomplete cycles can also be detected.

3 Results

3.1 Modes of variability in daily sea level and AMOC data

Qualitatively, the daily sea level anomaly (Fig. 2a, b) and AMOC (Figs. 2c and 3) data show considerable variations on a wide range of scales including apparent seasonal variations, but also high-frequency variations and interannual variability; the sea level variations represent combined impacts from many potential factors such as tides, winds, storm surges, river discharge, and thermal changes. A particular interesting period with significant anomalies is 2009–2010 when AMOC was especially weak and sea level in Norfolk especially high. This period, during a low phase of the North Atlantic Oscillation (NAO), has been mentioned in past studies (Ezer 2015; Goddard et al. 2015). To further investigate the nonlinear nonstationary variability at different time scales, the ensemble EMD analysis (Wu and Huang 2009) was employed, producing 10 modes (Eq. 1) for each daily time series. Figure 4 shows the normalized spectral energy of the EMD modes of sea level (Fig. 4a) and the AMOC components (Fig. 4b). High frequency modes with periods less than ~ 1 month are not significant due to the 10 days low pass filter (high-frequency modes of sea level before the filtering are significant, but not discussed here). Sea level variability in Baltimore has higher relative energy only around the annual cycle, while in the lower bay, in Norfolk, relative energy is higher in all other time scales (especially large difference between the stations is seen for the longest time scales). As discussed later, this may indicate source of variability in the lower CB that originated from the Atlantic Ocean, while the upper bay is more influenced by local dynamics. The energy of the AMOC components shows that the wind-driven Ekman transport is relatively more energetic at short-term monthly time scales (weather systems), while the Gulf Stream and mid-ocean transports are more energetic at longer time scales. The relatively high energy of the total AMOC at decadal time scales seems to be a combination of low-frequency modes of several of its components, especially the mid-ocean transport associated with long-term density variations. For periods longer than 5 years, sea level is several orders of magnitude more energetic at Norfolk than at Baltimore (Fig. 4a), which suggests that the long-term variability seen in Norfolk may have originated outside the CB and potentially linked with AMOC (Fig. 4b).

Fig. 4
figure 4

Spectral energy (normalized) of EMD modes calculated from ensemble simulations of a daily sea level for Norfolk (blue) and Baltimore (red), and b AMOC components. The estimated mean periods are indicated, as well as the significance level

Full size image

Figure 5 shows the correlation between the EMD modes of sea level in Norfolk and Baltimore. The overall correlation of daily detrended anomalies between the two locations is R = 0.8; while this correlation is significant at over 95% confidence, it also means that about 33% of the variability cannot be explained by a common forcing (i.e., involved local forcing or local estuarine dynamics). The highest positive correlation is near the annual cycle, which is expected, though it is interesting to note that the seasonal sea level cycle in CB is driven mostly by the annual and semiannual tides and not by the seasonal temperature cycle (Ezer 2020a, 2023). Except the annual cycle, positive, but not very high correlations (~ 0.2–0.4), are seen in modes representing variability with periods of weeks to interannual, demonstrating that local forcing within the CB has significant spatial variability (local rivers, winds, etc.). The very high negative correlation of the lowest frequency mode was indicated before (Ezer 2023), but not fully explained yet. Since AMOC has significant energy at low-frequency modes (dash line in Fig. 4b), the link of AMOC with CB variability is further explored below.

Fig. 5
figure 5

Correlation between EMD modes of daily sea level of the two locations. Estimated mean periods are indicated

Full size image

3.2 Comparison of sea level and AMOC

To look at the relationship between AMOC and sea level at different time scales, the EMD modes were composed into three bands, high-frequency modes (mean periods of weeks to ~ 6 months), mid-frequency modes (mean periods of ~ 2–5 years), and low-frequency modes (mean periods of ~ 10 year and longer nonlinear trends). The mode representing the annual cycle (mode 6 in Fig. 5) was not used. The comparisons of the EMD modes of sea level anomaly and AMOC are shown for Baltimore (Fig. 6) and Norfolk (Fig. 7); cross correlations were also shown to indicate potential lags (in the bottom two panels vertical stem-lines are dense, looking like a solid plot). Negative correlations between sea level and AMOC are expected due to the relation with the Gulf Stream (Ezer et al. 2013, 2017; Ezer and Atkinson 2014; Ezer 2020b) and is seen in the high-frequency modes (upper panels of Figs. 6 and 7). The mid- and low-frequency modes show very different AMOC-sea level relations between Baltimore and Norfolk with positive lag of 3–5 years for Baltimore and similar negative lag for Norfolk. In fact, the nonlinear trends in the low-frequency modes of sea level (blue lines; bottom panels of Figs. 6 and 7) are almost in opposite phase (minimum sea level in Baltimore around 2018, but maximum sea level in Norfolk around 2016, about 5 years after maximum in AMOC).

Fig. 6
figure 6

Comparison between EMD modes of AMOC and sea level in Baltimore; left panels are time series and right panels are cross correlations. Modes were grouped according to mean frequency, top to bottom: high-frequency modes (periods ~ 20 days to ~ 6 months), mid-frequency modes (typical periods ~ 2–5 years), and low-frequency modes (typical periods of 10 years and longer nonlinear trends). The mode near the annual cycle was skipped (mode 6 in Fig. 5)

Full size image
Fig. 7
figure 7

Same as Fig. 6, but for sea level in Norfolk

Full size image

Figure 8 examines how the 3 components of AMOC (Fig. 3) linked with sea level at the two locations (Norfolk/Baltimore in left/right panels; frequency bands from high to low in top to bottom panels). Almost all correlations are higher in Norfolk than Baltimore, suggesting that the AMOC influence affects the CB through exchange in the mouth of the bay in the south (Fig. 1). High-frequency Gulf Stream variability (Fig. 8a) and mid-frequency wind-driven Ekman transport variability (Fig. 8c) have especially significant correlation with sea level in Norfolk. The pattern of AMOC-sea level correlations in the low-frequency band is exactly opposite in sign between Norfolk and Baltimore for all three components (Fig. 8e and f), owing to the phase difference in sea level at the two ends of the CB as seen before (Figs. 6 and 7). It should be noted however that the estimated confidence level for the low-frequency modes is probably not very accurate, but the sign of the correlation is of interest. For this low-frequency band, higher mid ocean transport is linked with lower sea level in Norfolk and higher sea level in Baltimore, but higher Gulf Stream and Ekman transports are linked with higher sea level in Norfolk and lower sea level in Baltimore. The impact of the Gulf Stream on low-frequency sea level in Norfolk (green bar in Fig. 8e) is interestingly opposite to its impact on high-frequency variability (green bar in Fig. 8a), pointing to a different mechanism than the simple geostrophic argument of sea level slope across the GS, as found in past studies (Ezer et al. 2013; Ezer 2015). It is important to acknowledge that correlation does not necessarily means cause and effect, and the mechanism involved could be a complex combination of several factors.

Fig. 8
figure 8

Linear correlations between sea level and AMOC components for the 3 frequency bands as in Figs. 6–7 (high to low frequencies from top to bottom); left/right panels are for Norfolk/Baltimore. The AMOC components are mid-ocean transport (red bars), Gulf Stream transport (green) and Ekman transport (blue). Dash lines represent estimated correlation values for 95% confidence level when considering the loss of degrees of freedom for lower frequency modes

Full size image

3.3 Comparison of temperature variability and AMOC

Climatic change and variability over the North Atlantic can affect the CB through local atmospheric heat fluxes and heat transport exchange through the mouth of the bay—temperature changes can also affect steric sea level in the bay. Studies also show that remote influence from wind pattern and heat exchange over the subtropical gyre can impact the coast (Piecuch et al. 2019; Volkov et al. 2019, 2023; Wang et al. 2024). Therefore, daily water temperature records for Norfolk and Baltimore were analyzed. While dominated by the seasonal cycle, temperature variations from year to year are significant (Fig. 9). The seasonal range of temperature is larger in Baltimore than Norfolk, which is consistent with the larger energy at seasonal time scales in Baltimore (Fig. 4a). To remove the dominant seasonal cycle, yearly mean values of AMOC components are compared with sea level (Fig. 10a) and temperature (Fig. 10b); shown are only the AMOC components with significant correlations with either sea level or temperature. On annual basis, sea level correlation between Norfolk and Baltimore is larger (R = 0.84) than temperature correlation (R = 0.6). Both correlations are significant at the 95% confidence level, but temperatures seem to be affected more by local conditions than sea level. While sea level variability is dominated by an 8-year cycle, temperature variability is dominated by a 5-year cycle. Correlations of sea level and temperature with AMOC components (R ~ 0.3–0.45) are only significant at the 90%-95% level. A much longer record than the 17-year record here is needed for obtaining higher significance; nevertheless, the pattern of correlation sign and amplitude can help us understand the potential processes involved. First, AMOC is negatively correlated with sea level (especially with the Gulf Stream and Ekman transports, as mentioned before) while positively correlated with temperatures (in particular, with the mid-ocean transport). This result implies a transport-driven sea level versus a heat flux driven temperature. Second, sea level correlations with AMOC are higher in Norfolk, but temperature correlations are higher in Baltimore, suggesting that sea level in the bay is largely driven by transports from the Atlantic Ocean while temperature in the upper bay is driven more by local atmospheric conditions. A particular period of interest is 2009–2010 when AMOC and NAO were extremely low and sea level anomalously high along the U.S. East Coast (Ezer 2015; Goddard et al. 2015). At the same time, water temperatures were colder than normal across the bay when sea level was raising in Norfolk, so thermosteric effect is not a factor. However, while sea level in Norfolk reached a peak in 2009–2010, sea level in Baltimore peaked only 1–2 years later in 2011, creating phase difference as noted before. In 2019, when AMOC was in another low phase, temperature in Baltimore was low again and sea level was high across the CB. The latter case is somewhat different than the more dramatic AMOC low in 2009–2010, because in 2019 only the Gulf Stream transport was low (Fig. 10a), with no significant change in the wind-driven Ekman component and the density-driven mid-ocean transport.

Fig. 9
figure 9

Daily water temperature in Norfolk (red) and Baltimore (blue) calculated from NOAA 6-min data

Full size image
Fig. 10
figure 10

Yearly mean sea level (a) and temperature (b) anomalies in Baltimore (blue) and Norfolk (red). AMOC components with significant correlations are Gulf Stream transport (dash black line in a), Ekman transport (dotted black line in a), total AMOC transport (dash black line in b), and mid-ocean transports (dotted black line in b). Correlations are also shown

Full size image

3.4 Surface currents at the mouth of the Chesapeake Bay

The data analyzed so far show that the lower CB is more closely linked with variability over the Atlantic Ocean than the upper bay, suggesting that the exchange of water and heat through the mouth of the CB may play a role in the dynamics of the bay. Observations of surface currents by CODAR stations have been carried out systematically since ~ 2007 (for details see Atkinson et al. 2009 and Ezer et al. 2022). These currents are dominated by the semi-diurnal and spring/neap tidal cycles, as well as large variability during storms. Therefore, a 30-day low-pass filter was applied to remove high-frequency variability (Fig. 11a), and yearly mean anomalies were calculated and compared with the Gulf Stream transport and with observations in Norfolk (Fig. 11b). There is a persistent increased outflow trend of 0.22 cm/s per year (negative/positive values indicate outflow/inflow direction) that will be discussed later. The interannual variations in currents include a cycle with period of ~ 3–5 years and a strong outflow in 2010, around the time of colder temperatures, weaker Gulf Stream, and higher sea level, that were discussed before. Because the record is relatively short (only 15 years) the correlations between the surface currents and the other observations are only significant at 80%-90% confidence level, though the pattern of variations is interesting. During years of anomalous surface currents, sea level and the Gulf Stream show the following patterns: in 2010 when the outflow was maximum (largest negative anomaly), the Gulf Stream was weak and sea level peaked, but in 2013 and 2017 when the outflow was especially weak (positive anomaly) the Gulf Stream was stronger and sea level lower. This pattern is also consistent with the trend of increased outflow (Fig. 11a) when AMOC and the Gulf Stream are weakening over time due to climate change (Ezer et al. 2013; Ezer 2015; Piecuch and Beal 2023). The results here show that surface outflow change of 1 cm/s is roughly equivalent to ~ 2 Sv change in AMOC transport. Reconstruction of AMOC using sea level data estimated recent AMOC decline of ~ 0.44 Sv per year (Fig. 9 in Ezer 2015), which based on the results here is equivalent to 0.22 cm/s per year change in surface currents, exactly the trend shown in Fig. 11a. The proximity between this estimate and the observational trend in outflow could be a coincidence, and does not imply cause and effect, i.e., it does not necessarily imply that AMOC directly drives surface currents in the CB, but rather that maybe both are affected by climate change.

Fig. 11
figure 11

a daily (green) and annual (blue) mean surface velocity measured by CODAR stations near the mouth of the CB; downward linear trend (dash line) indicates increased outflow from the CB toward the Atlantic Ocean. b Yearly mean data for CODAR surface velocity (dash black line), Norfolk sea-level (blue) and water temperature (red), and the Gulf Stream transport (green). Y-axis for sea-level in on the right and for all other variables on the left

Full size image

A more direct connection to increase in net outflow is the increase in precipitation in the northeastern U.S. region and increased river discharge into the CB (Rice et al. 2017) – if inflow from rivers increases, outflow from the mouth of the bay should increase as well. The analysis of Rice et al. (2017) indicated about 30–40% increase in discharge from 1927 to 2014, or about 0.4% increase per year, compared with ~ 2.5% increase in surface outflow per year from 2007 to 2021. Figure 12 shows a comparison between the monthly river inflow into the CB and the surface currents out of the CB—both inflow and outflow show positive trends of increased flow. The 1.4% per year increased river flow into the CB during 2007–2021 is closer to the increased surface outflow during this period (2.5%/y), than the previous estimates of Rice et al. (2017) for 1927–2014. Therefore, it seems that the trend of increased flow in/out of the CB have accelerated in recent years. While correlations between surface currents and AMOC or between currents and river discharge are statistically significant (Fig. 12), the correlation is not very high, indicating potential combination of several factors. Additional factors that can impact surface velocities and should be further explored in future studies include for example, the decrease in tidal range in the CB (Cheng et al. 2017), impact of sea level rise and inundation on the estuarine dynamics (Ezer 2023), coastal erosion, change in stratification, etc.; all these changes can affect the complex dynamics of the CB as seen in past observations of currents near the mouth of the CB (Valle-Levinson et al. 2003).

Fig. 12
figure 12

Monthly streamflow into the CB (blue) and surface outflow out of the CB (red). The monthly outflow is calculated from the daily data (Fig. 11a with reverse sign). Dash heavy lines are the linear trends which are indicated (both positive, 1.4% increase inflow per year and 2.5% increase outflow per year). The correlation between the data and the P-value are also indicated

Full size image

4 Summary and discussion

Understanding the impact of potential climate change on the CB’s ecosystem and population is important for planning mitigation and adaptation options. The acceleration in flooding due to fast sea level rise has been documented in many studies (Ezer and Corlett 2012; Ezer and Atkinson 2014; Boesch et al. 2018; Sweet and Park 2014; Park and Sweet 2015; Ezer 2022, 2023). While global sea level continues to rise, local variations in sea level are more difficult to predict – Ezer (2023), for example, shows large differences withing the CB between SLR prediction based on climate models and SLR based on local statistics of past data. A particular difficulty in SLR prediction is the impact of remote, large-scale oceanic variations such as changes in ocean currents, impact of Rossby Waves on coastal sea level (Dangendorf et al. 2021, 2023) and changes of heat flux divergence over the subtropical gyre (Volkov et al. 2019). Potential slowdown of AMOC and the Gulf Stream (Ezer et al. 2013; Ezer 2015; Piecuch and Beal 2023) already showed their impact on increased SLR and coastal flooding, but the impact may be very different along different sections of the U.S. East Coast (Ezer 2013; Little et al. 2019; Volkov et al. 2023). Moreover, most previous studies of the links between remote Atlantic variability and the coast include only coasts that are directly in contact with the open ocean, while the remote impact inside estuaries like the CB is more complicated due to local dynamics (e.g., tides, rivers, and estuarine circulation). The exchange of water, heat, and salt through the mouth of a bay is of particular interest as it provides a link between a bay and the open ocean.

In the CB, for example, a recent study (Ezer 2023) found some unexplained long-term sea level variability that has opposite phases in stations in the upper bay compared with stations in the lower bay. Such long-term variability is likely linked to large-scale climate variability such as variations in AMOC (Ezer 2015) and NAO (Ezer and Atkinson 2014), though observations of AMOC are quite short (about two decades) so studying decadal variations is challenging. To examine these variations, daily, monthly and annual data of various observations were analyzed – they include sea level, surface temperature, river discharge, surface currents near the mouth of CB bay, and the RAPID/AMOC transport (which also includes the Gulf Stream transport). The results suggest that part of the observed long-term variabilities in the CB originated in the Atlantic Ocean and impacted the CB by creating variations in the exchange of flow at the mouth of the bay. Several results support this hypothesis:

  1. 1.

    EMD analysis shows more energetic low-frequency variability in the lower CB (at similar periods as seen in the AMOC variability) than in the upper CB. The upper bay seems to be more affected by local dynamics and the seasonal cycle than by flow exchange with the Atlantic Ocean.

  2. 2.

    When comparing sea level in the upper and lower CB, the lowest frequency modes and trends are out of phase, indicating different forcing, while variabilities in all other modes from periods of days to few years are positively correlated.

  3. 3.

    Correlation between sea level and the 3 components of AMOC are more significant in Norfolk than Baltimore and the sign of correlations of the low frequency modes are exactly opposite between the two locations. The mechanism is still not completely understood. For example, negative correlation between sea level variability in Norfolk and the Gulf Stream transport has been found before for high-frequency variability such as following hurricanes (Ezer and Atkinson 2014; Ezer 2020b; Park et al. 2022, 2024), due to weakening of the Gulf Stream post hurricanes and reduced sea level gradients across the Gulf Stream. However, why is there a positive correlation between the Gulf Stream transport and Norfolk’s sea-level (and negative correlation in Baltimore) for the low-frequency modes?

  4. 4.

    Analysis of yearly mean temperatures and their comparison with AMOC shows positive correlations that are larger in Baltimore than Norfolk – This contrasts with sea level-AMOC correlations that are negative, with higher values in Norfolk. This may suggest that sea level variations are driven by transports coming from the south, when temperature variations are driven by local air-sea interactions, which are stronger in the more isolated northern bay.

  5. 5.

    Finally, the most convincing argument for remote influence on the CB comes from the surface currents near the mouth of CB, which show low-frequency variability resembling the AMOC variability. Periods of increased outflow from the CB coincide with periods of colder water temperatures in the CB, higher sea level in Norfolk, and a weaker Gulf Stream transport. The relation between the observations is complicated and does not necessarily indicate cause and effect. For example, the interannual variability is dominated by a 5-year cycle in temperature, surface currents and the Gulf Stream, and an 8-year cycle in sea level. On the other hand, the wind-driven Ekman transport component of AMOC shows more energy at higher frequencies. Past studies identified 6–8 years cycle in the Gulf Stream transport (Ezer et al. 2013) that influence sea level in the Mid-Atlantic coast, but inside the CB it appears that sometimes there is a delayed response in the upper bay, which can explain the observed anticorrelation between sea level in the upper and lower bay.

Another interesting finding is the continuous increase in surface outflow from the CB since observations started in 2007 (Fig. 11a). It can be assumed that this trend represents increased net outflow transport, though there are no continuous observations of the entire water column and the flow there is quite complicated (Valle-Levinson et al. 2003). This trend is consistent with the increase in precipitation and river discharge into the CB as previously reported by Rice et al. (2017) and seen in recent streamflow data (Fig. 12). The net outflow from any bay is generally ΔV = R + P-E (R, P and E are river discharge, precipitation, and evaporation rates, respectively); assuming no significant change in evaporation, outflow should increase with increased R + P. However, one cannot eliminate the possibility that other climate variability factors may also contribute to this trend, such as the decline in tidal amplitude (Cheng et al. 2017) and impact of sea level rise on the estuarine dynamics. A link was found here between variations in the Gulf Stream transport and variations in surface currents near the mouth of the bay. There are now evidences that there has been a climate-related decline in the observed Florida Straits transport over the last four decades (Piecuch and Beal 2023) and a weakening of the Gulf Stream in the Mid-Atlantic Bight since the 1990s (Ezer and Dangendorf 2020). While the surface current record of 15 years was too short to establish undisputed significant statistical confidence about its relation to AMOC, the trends and variability do point to climate related influence on the environment in CB from remote variations in the Atlantic Ocean, and these variations seem to enter the bay at its mouth, affecting the lower bay. The forcing of the CB is complicated though with different forcing acting on the northern and southern CB at different time scales which result in spatial variations within the bay. The study provides further support to the notion that the lower bay is affected by large-scale, long-term Atlantic variability. The study also found another potential detector for climate change: changes in the outflow from estuaries and bays, and suggests that future projections of climate change in bays and estuaries must consider local dynamics.