Surface Signature of the Chesapeake Bay Outflow Variability Observed with Coastal Radar
Abstract
The present work is an observational analysis of the small-scale variations induced by Chesapeake Bay outflow on coastal surface circulation off Virginia. The variation in freshwater discharge into Chesapeake Bay is investigated as a triggering mechanism for local coastal jets. While it is known that an extra amount of freshwater received from the tributaries will lead to an increased total outflow, the effect on the outflow surface signature (the upper meter) has not been extensively explored.
Our analysis is based on surface currents, wind data, and freshwater discharge records. The surface currents were derived from two high-frequency Doppler radars deployed on the Virginia coast during September and October 1996 and 1997. The wind data were obtained from a nearby weather Coastal-Marine Automated Network station, and the freshwater discharge data were from the U.S. Geological Survey stations located at the mouth of each tributary. The freshwater discharge into the bay observed during 1996 decreased from 4200 to 1000 m3/s in 10 days, while for the 1997 period it varied between 500 and 5500 m3/s in 1 week.
The present analysis shows that sudden variations of freshwater flux will drive the plume further offshore before its normal southward turning, in the same way upwelling-favorable winds do as reported in all the available literature of the area. Strong freshwater variations on weekly scales are then shown to be an important factor in determining the extent of the surface plume displacement and the fate of the shelf currents or jets.
INTRODUCTION
The interaction between the continental freshwater discharge (FWD) and the ocean waters influences the shelf circulation and its variability. In an estuarine area, the outflow is subject to several mechanisms, giving such outflow an interior dynamics that can be in some circumstances sensed through its surface signature. The surface boundaries of plumes can be identified directly by satellite observation and surface radars and indirectly by measuring different parameters in situ, such as salinity, temperature, and sediment concentration, (Fong and Geyer, 2001; Marmorino et al., 2000; Wang, 1979; Wang and Elliot, 1978). These plumes play a crucial role in the coastal dynamics because they allow the exchange of properties as momentum, salt, and temperature between the fresh continental waters and the oceanic shelf waters. Moreover, they allow the transport of nutrients and sediments toward the shelf. The plume exiting Chesapeake Bay is of particular importance for the east coast of the United Estates due to its large size and to the interactions with shelf waters in determining the existence of coastal jets off Virginia. These southward-flowing coastal jets are confined to the first 50 km from the coast and have typical speeds of around 75 cm/s (Shay et al., 2001).
The Chesapeake Bay system is a partially mixed estuary. In this type of estuary, the tidal movements are appreciable, and considerable friction occurs between the bed of the estuary and the tidal currents (Chao, 1990; Garvine, 1987; Valle-Levinson and Atkinson, 1999; Wang, 1979; Wheless and Valle-Levinson, 1996). The main tributaries to Chesapeake Bay are the Susquehanna, the Potomac, and the James rivers, which contribute 60, 17, and 12%, respectively, of the total discharge (U.S. Geological Survey, USGS, 2003). Each of the other tributaries brings just 2 or 3% of the total. The mean annual discharge to the bay was calculated as 2500 m3/s (Marmorino et al., 1999; Valle-Levinson et al., 1998). The Chesapeake Bay is subject to strong variations in river discharge throughout the year. Accurate knowledge of the fluctuations from the mean discharge is of fundamental interest, since they induce significant variability to the surface water dynamics at the Chesapeake Bay entrance (Valle-Levinson et al., 1998). Typically, the largest amount of freshwater influx occurs during springtime, facilitating the stratification of the water column and lowering the salinity values at the mouth. In autumn, the freshwater influx diminishes, the water column becomes more homogeneous, and the salinities at the plume exit increase from September through November. This last period, however, coincides with periods of episodic heavy rains associated with tropical storms (Valle-Levinson et al., 1998), which vastly increase the otherwise low discharge values. Nevertheless, the effect of the long-term variability (in periods longer than a day) over the surface outflow is not yet completely understood.
The local winds are primarily from the NE and SW, and they vary seasonally (Valle-Levinson and Lwiza, 1998; Valle-Levinson et al., 2001). The most energetic winds are usually from the NE or NW during late fall and winter and contribute to the water column destratification; weaker SW winds prevail during the summer, with typical magnitudes less than 4 m/s (Valle-Levinson et al., 1998).
Inside Chesapeake Bay, tidal forcing is predominantly semidiurnal, with the lunar constituent (M2) being the most energetic. The interaction among the three semidiurnal tidal constituents (M2, N2, and S2) produces fortnightly and monthly variability in the tidal currents (Valle-Levinson et al., 1998). Due to the dominance of the N2 constituent over the S2, there is a primary and a secondary spring (or neap) tide during each month. During spring tides, the stratification is reduced and the subtidal flows are weaker than during neap tides.
The behavior of the exiting plume has been studied by Marmorino et al. (1999), 2000, Shay et al. (2001), Valle-Levinson and Lwiza (1998), Valle-Levinson et al. (1998), 2001, and Haus et al. (2006). These studies investigated the wind and tidal influence on the plume characteristics, showed that tidal currents are responsible for most of the dynamic variations occurring at the mouth exit or in the first few kilometers offshore.
Winds affect the exiting plume in different ways and over different frequencies. Northwesterly winds (downwelling favorable winds) can produce a strong outflow if in resonance with tides inside the bay (Boicourt et al., 1987) and confine the plume to a narrow band along the coast (Haus et al., 2003). Conversely, southwesterly winds can repress the out-flow and reduce the size of the plume: they oppose the expected anticyclonic turn and force the plume to spread seaward, allowing upwelling between the plume and the coast (Valle-Levinson et al., 1998, 2007).
Few studies have investigated the role of the freshwater fluctuations on the plume dynamics. Ruzecki (1981) investigated the temporal and spatial variations of the plume with periods smaller than 1 day, i.e., in relation to ebb and flood. Valle-Levinson et al. (1996) studied the influence of freshwater discharge (FWD) pulses through numerical experiments. Li et al. (2005) included FWD variations in their prediction model of vertical stratification and salinity distribution in the Chesapeake Bay. Their results were compared to salinity data obtained at depths greater than 5 m; therefore, no surface characteristics could be identified. Valle-Levinson et al. (2007) analyzed different parameters of the plume at depths greater than 2 m. Therefore, to the best of our knowledge, this work is the first attempt to quantify the influence of the FWD variability over the plume surface signature (in the upper meter) in a weekly (or greater) timescale with two-dimensional data.
The Chesapeake outflow plume experiments (COPE1–3) performed between 1995 and 1997 sought to understand the shelf dynamics. Three types of high-frequency (HF) coastal radar and acoustic Doppler current profilers (ADCPs) were deployed near the estuary mouth. The HF coastal radar (ocean surface current radar, or OSCR) from University of Miami was operated during autumn 1996 and 1997. These deployments were designed to capture the main coastal circulation under the low-outflow conditions typical of autumn. Tropical storms during September 1996 dramatically increased the mean discharge volume for most of the observed period (Marmorino et al., 1999; Shay et al., 1995). The mean annual discharge was normal for 1997, but unexpectedly high values of discharge occurred for 1 week during the fall, at the time of the experiment. This provided an excellent opportunity to study the influence of a sudden pulse of FWD over the plume displacement over the shelf. Such “extra” buoyancy can affect the outflow not only in the total volume of flow but also in the turning point, that is, the point where surface current vectors change direction toward a coastal current (for clear sketches of such a turning point, see Valle-Levinson et al., 2001).
The present work aims at understanding the role of the estuarine discharge variability over the coastal dynamics by performing a high-resolution statistical analysis of the data. Surface currents obtained with the HF radar for 1996 and 1997 were compared to wind speed data from the Coastal-Marine Automated Network (C-MAN) weather station located offshore within the zone (Figure 1) and to the inflow data derived from the USGS stations located at the exit of each tributary. Note that the present is a two-dimensional study of the plume surface signature, which is useful for comparison with satellite observations.
Citation: Journal of Coastal Research 25, 6; 10.2112/08-1027.1
A brief overview of the data sets can be found in next section. Then the statistical tools utilized and the results obtained for each year are presented. In the discussion, a comparative analysis between both years' results is presented, including a comparison with previous works.
DESCRIPTION OF THE DATA
Surface velocities were obtained with OSCR from September 13 through October 6, 1996, and from October 14 through November 29, 1997. Each of two stations consisted of 4 transmitter antennas in a Yagi directive configuration and 16 receiver antennas in a linear phased array. The measurement domain consisted of 700 grid points 1.5 km apart (depicted as dots in Figure 1). Around 20% of the time series were ignored to avoid interpolation over large gaps; the rest were averaged over 20 minutes.
The wind speed and direction data were obtained with a local weather station, the National Data Buoy Center's C-MAN station CHLV2, situated at 36°54′ N, 75° 42′ W, off Chesapeake Bay (Figure 1, denoted by T). The C-MAN wind data set was adjusted to a 10 m height with a modified version of the log equation, as in Marmorino et al. (1999). Sampling rate corrections and linear interpolation for gaps of less than 5 hours were applied where needed (Figures 2a and b).
Citation: Journal of Coastal Research 25, 6; 10.2112/08-1027.1
Daily mean discharges into the Chesapeake Bay were obtained from the USGS and are available at va.water.usgs. gov/chesbay/RIMP/index.html. For the calculation of total discharge, the Susquehanna, James, and Potomac stations were used. They supply 60, 17, and 12%, respectively, of the total discharge into the bay (USGS), while the rest of the tributaries provide less than 2% each. By continuity of volume principles, we use this total discharge into the bay as a proxy for total outflow at the exit of Chesapeake Bay (Figure 2c).
Four bottom-mounted upward-looking ADCPs (A1 to A4 in Figure 1) obtained vertical current profiles every 20 minutes. The bins started at 2 m under the ocean surface, with 1 m vertical spacing. Hallock and Marmorino (2002) and Marmorino et al. (1999), among others, investigated the characteristics of the water column observed through the ADCP data. The excellent agreement between the OSCR series and the velocities derived from the first bin of the ADCPs validates the OSCR measurements for the present experiment (Gremes Cordero, 2004; Shay et al., 2001).
STATISTICAL ANALYSIS
1997 Data Set: COPE3
The 1997 HF radar data set was previously analyzed only in relation to wave shoaling (Haus et al., 2006; Ramos, 2006). Here we focus on the relationship between FWD variations and surface currents. Moreover, the COPE3 data covers a longer period than the COPE1 data (for a total of 37 days), allowing for a more comprehensive analysis of the results. The first week was eliminated due to data intermittency.
Of the 700 OSCR grid points, 4 were selected that coincided with the ADCP moorings (Figure 1). The power spectra of the surface velocity components (U and V) for each point showed that the most energetic signal was the one associated with the M2 component of the tides, as previously demonstrated for this region by Shay et al. (2001). There were two other peaks whose frequencies correspond to 1-week and 2.5-day periods, respectively (Gremes Cordero, 2004). The diurnal signal of the tides (period = 1 day) had less than 30% of the energy corresponding to the other peaks. The same characteristics appeared in both spectra, for zonal and meridional currents. On the other hand, spectral analysis of the wind data revealed that one of the significant frequencies in both components was around 6 days, coinciding with the meteorological synoptic scales (Pierson, 1983). For the north–south component, the energy corresponding to a 3-day period was also significant. For the east–west component, there was significant energy at 3 and 4 days, but it contained about 25% of the energy relative to the 6-day period. There was a noticeable similarity between the power spectral density (PSD) of OSCR currents and winds for the U component.
Note that the characteristics of the wind-driven currents are generally less evident when tidal components are still present in the data. To extract the HF motions, including diurnal and semidiurnal tidal effects, from the original surface currents, a low-pass Hanning-Finite Impulse Response filter with a 0.025 frequency cutoff (40 hours) was applied to the OSCR data set. Then, the empirical orthogonal function (EOF) analysis, sometimes referred to as eigenvector or principal component analysis, was carried out. The temporal variance of the data was split into orthogonal modes, which represented the axes of greatest variance (Venegas, 2001; Von Storch and Frankignoul, 1998).
Note that Marmorino et al. (1999) applied the “real vector” EOF analysis to the original OSCR data set obtained for COPE1 during 1996, with tidal flow included. This method was preferred because of its proven ability to separate more cleanly unidirectional flow from eddy-like flows (Kaihatu et al., 1998). By retaining tidal effects during the calculations, they assured an unbiased analysis of spatial and temporal structure in the data set (Bendant and Piersol, 1986; Helstrom, 1984; Marmorino et al., 1999). The same type of EOF analysis was applied here, but to the detided data, to isolate the lower frequency modes.
The first two modes accounted for 82% of the total variability of the field (the first 46% and the second 36%), while the third mode contributed only 5% of the total. The circulation patterns resulting from our EOF analysis are illustrated in Figure 3.
Citation: Journal of Coastal Research 25, 6; 10.2112/08-1027.1
Interestingly, two zones were present in the first mode (Figure 3a), suggesting two different regimes. One was situated near the Chesapeake Bay mouth, where the plume geometry was well defined, while the other was outside this region. This suggests the action of more than one forcing over the area: one near the coast and the other further offshore. This idea was also supported by the EOF error calculation described later. Another possibility is that a wind direction shift was responsible for the delineation. However, no geological features over the land, such as mountains or hills, support such veering. This is explained in more detail in the next section.
The errors in the EOFs were calculated following the method proposed by North et al. (1982), which takes into account the effect of estimating the EOF when too small a sample is available. In practice, when the error bars of two eigenvalues overlap, they influence each other. In our case, there was a significant overlap between modes 1 and 2 (Figure 4), suggesting that the forcing affecting mode 1 was also affecting mode 2, as intuitively observed in Figure 3.
Citation: Journal of Coastal Research 25, 6; 10.2112/08-1027.1
To identify the dominant forcing, a principal components (PCs) analysis was performed. The PCs are obtained by projecting the original data series onto each eigenvector of the EOF analysis (Ng, 1993; Venegas, 2001; Xue and Brooks, 2000). A PC does not necessarily have a dominant timescale since the EOF decomposition is designed to optimally separate spatial patterns, not frequencies or timescales (Venegas, 2001). The first principal component (PC1) was highly negatively correlated with the meridional component of wind (V), with an absolute value of the correlation coefficient (CC) of 0.74, indicating that the first mode was wind driven. Note that the minimum significant value of correlation was calculated as 0.40 for a 95% of confidence level. The highest spectral energy for PC1 was concentrated around 3 days, coinciding with the second (in magnitude) spectral energy peak of meridional winds, possibly related to a local variation in winds. For the second mode, the highest CC value was with the U component of the wind (0.57). Note that even if the correlation was not especially high, the minimum significant value was 0.40. The dominant period in the wind U spectrum is around 6 days, and secondarily dominant are the time variations of 3-day periods coinciding with the PC cycles.
This same comparative analysis was performed between the PCs and the streamflow conveying into Chesapeake Bay from the USGS stations. The daily mean of the FWD was correlated with PC1, with a CC of approximately 0.60. Actually, an outflow increase in each tributary is not necessarily associated with an immediate increase in surface velocities over the shelf: the changes can happen over a long period and could be altered by some other forcing before arriving at the area of interest. Note, however, that the time required for the flow to travel from the northern part of the bay to the exit, with typical current velocities, is only about 1 day (Boicourt et al., 1987), minimizing the possibilities of such modifications.
The power spectra revealed that the second energy peak of the first two PCs was in agreement with the second peak of energy of the FWD, which occurred in cycles of 2.5 days. Note that 2.5 days coincides with the second peak of energy for the winds, which was more important for the V component. This is an alternative mechanism for the overlapping effect mentioned earlier.
To confirm the validity of the correlation values, the confidence intervals of the correlations were calculated through a resampling (or Monte-Carlo) method called “bootstrapping” (Wilks, 1995). In each case, the confidence interval was less than 5% of the correlation value; thus, the coefficients could be considered valid.
To identify the subzones primarily influenced by the discharge, the cross-correlation between the surface velocities and the discharge was computed for each of the 700 points forming the OSCR grid. The results are presented in Figures 5 (U) and 6 (V) for different lags.
Citation: Journal of Coastal Research 25, 6; 10.2112/08-1027.1
The maximum correlations for each lag were also determined. Note that the highest positive correlations were those for the U component. In Figure 5, we can observe what appear to be the plume contours delineated by the values of cross-correlation, following the bathymetry. The maximum values of CC corresponded to lags of 1 and 2 days (Figures 5b and c). After 4 days, the plume was swept out by other forcings. These timescales coincide with those found by Marmorino et al. (2000) through airborne radar imagery analysis.
Note that a minimum in CC appeared in the zone where a small eddy is usually present (lower left of the plot). This low correlation with FWD suggests that this feature was more related to winds than to FWD or tidal effects (since they were removed), as hypothesized by Boicourt et al. (1987). However, one must take into account that tidal residual motions with periods greater than 3 days were still present in the series and hence can be responsible for this structure as well. Indeed, Marmorino et al. (1999) and Valle-Levinson et al. (2007) predicted the appearance of such a tidal residual eddy resulting from nonlinear effects. It is also important to note that a small topographic depression was present (of about 40% difference), so the eddy could be topographically driven. Not enough data is available in this study to determine the origin of the eddy.
The V component of the radar velocities was negatively correlated to the FWD (Figure 6) for most of the grid. This result supports the idea that an enhancement of FWD will result in smaller meridional current velocities; it is discussed in the next section. Maximum absolute values of CC are around day 1, and the contour of the maximum was in agreement with previous observations of the outflow obtained with airborne radar systems (Marmorino et al., 2000) and with the bottom topography. From day 2, the outline of this maximum splits into two parallel shapes, a characteristic not documented or found in any previous literature. Differently from Figure 5, the structure became completely unattached from the original point after the day 3, instead of forming one continuous shape. This interesting feature may involve recirculation processes and will be the subject of future work.
Citation: Journal of Coastal Research 25, 6; 10.2112/08-1027.1
1996 Data Set: COPE1
As in the COPE3 data set of 1997, a week of data was eliminated from the analysis due to sparse measurements. The final COPE1 data set used here is 17 days long. For easy comparison with the previous results, the same points of the radar grid chosen for 1997 were analyzed.
The PSD for the U component revealed that the variability in the currents appears at periods between 2 and 3 days. For the surface current V component, the energy was concentrated around 3 days but decreased offshore and shifted to shorter periods of around 2 days.
The EOF analysis was applied to the low-pass filtered data set using the same Hanning-Finite Impulse Response filter as for the 1997 radar velocities as was explained in the previous section. The first mode accounted for 64% of the total variability of the field, the second for 21%, and the third for 7%. Therefore, the first two modes explained 85% of the total variability. The errors calculated by means of North's method show an important difference with the 1997 results: none of the first three modes overlap. This suggests that the mechanisms that influence the modal variability should be separable, in contrast to the 1997 results. However, the vector field for the first mode in 1996 (Figure 7) resembles the maps corresponding to the 1997 EOF analysis (Figure 3).
Citation: Journal of Coastal Research 25, 6; 10.2112/08-1027.1
To understand the mechanisms driving the variability described by the EOFs, the PCs of the surface currents for the first two modes were calculated and correlated to the other data available (winds and discharge) as in the 1997 case. The CC between the zonal component of winds (U) and the PCs of currents had a maximum for the second mode of 67%. For the V component of the winds, the maximum value of correlation was obtained for the first mode, with CC equal to 0.92. The PSD of the PCs showed the highest energy around 3 days for the first mode and at 2 and 3 days for the second mode. Note that for this year the minimum valid correlation was 0.57. The CC between the PC of surface currents and the river discharge rate was maximum for the first mode (CC = −0.42). The PSD of the discharge identifies the major variations with periods of 1.5 and 4 days for the first mode and 1.5 days for the second mode. A summary of the correlation values is presented in Table 1.
Finally, the cross-correlation between the original surface velocities' PCs and the discharge was performed over the 700 points forming the OSCR grid (Figures 8 and 9). The highest correlation can be found after 3 days, as in the case of COPE3. The main difference with the former case, however, was the shape of the plume. Indeed, fore COPE1 the plume was not as well delineated as in 1997 but had a meandering shape. Moreover, the direction was not toward the south but mostly toward the east; i.e., it tended to flow further offshore, as can be observed in Figure 8 for the U component. Figure 9 shows the correlation of surface currents and FWD for the V component. Considering that the significant value of correlation for this experiment was 0.57, a significant correlation existed only in some grid points near the mouth, and only in day 3. All these results are compared in the next section.
Citation: Journal of Coastal Research 25, 6; 10.2112/08-1027.1
Citation: Journal of Coastal Research 25, 6; 10.2112/08-1027.1
COMPARISON OF DATA SETS AND RESULTS
The EOF analysis revealed some unsuspected characteristics of the surface signature of the plume. It was expected that the plume would turn to the right as soon as it exited the Chesapeake Bay mouth due to Coriolis effects. However, the analysis showed that the outflow extended further off-shore than anticipated, and additional calculations were performed to understand its variable horizontal shape.
Two zones appeared in the first mode map for 1997 (Figure 3a), suggesting two types of regimes. From the analysis presented in the previous section, we know that the highest correlation between the U component (winds) and the PCs corresponded to the second mode for 1997 (Table 1). There was also a high negative correlation between the first mode and the V component. This implied that an enhancement of the wind speed correlates with a reduction of the PC value, which grows positively toward the south, in opposition to V. In addition, the FWD was well correlated with the second mode for 1997, in the same percentage as winds (56 and 57%, for valid correlations of 40%). This illustrates the combined role of both forcings in driving the local coastal circulation. In addition, these values of CC emphasized the role of FWD in the local circulation. For 1996, the first mode was related to the V component of winds and the second mode was related to the U component of winds. The second mode was also correlated with the FWD, but since the significant CC for 1996 is 0.55, the correlation between PC and FWD cannot be considered meaningful. Note that in 1996 the first and second modes did not overlap, as in 1997.
The idea that both wind and discharge acted in synchronism during the 1997 period was also supported by the energy concentrations of the PC spectra (wind and FWD had significant frequencies in common) and by North's method of error calculation, as explained in the previous section (Figure 4).
An alternative explanation, however, is that the circulation resulting from the first mode was solely related to a shift in wind direction as it approached the coast; i.e., there was no FWD influence, just a veering of the wind. To test this hypothesis, we performed a cross-correlation between the wind series from the C-MAN tower analyzed in the previous section and the wind series from a station located inside the bay, near Cape Henry (the Chesapeake Bay Bridge Tunnel Station of the National Oceanic and Atmospheric Administration). Those calculations result in values of 72 and 82% for the cross-correlations, for both U and V, for both years. Note that Paraso and Valle-Levinson (1996) found similar results. Such values suggest a very small wind variation throughout the 40 km of the OSCR coverage and do not support the hypothesis of the wind steering as the driving mechanisms of the outflow. Moreover, no geological features over the land, such as mountains or hills, support the veering.
It can also been hypothesized that a rotation of the winds to a new spatial system coinciding with the coastal geometry will produce higher correlations of the currents with the wind, as proved in many coastal studies. A singular value decomposition analysis was then applied to the data set as in Bjornsson and Venegas (1997) to determine whether the axis of maximum wind variability coincides with the coastline geometry. To apply the singular value decomposition method, we constructed a temporal cross-variance matrix between the two space- and time-dependent data fields: the U and V components of the surface current vector. Therefore, only one pair of singular vectors was obtained; it forms angles of 150° and 210° in normal Cartesian coordinates (counted counterclockwise from the positive x-axis in a Cartesian coordinate plane, or a west–east direction). These directions of maximum wind variability did not coincide with the coastline angle (285°, for the same frame) visible in Figure 1; therefore, the rotation of the winds for our particular case will not produce higher values of correlation.
In an attempt to isolate the effect of winds from the FWD influence, the winds during fall 1997 were compared to those of fall 1996 in both speed and direction (Figure 2). The wind speed variations were smaller than 5 m/s throughout both periods of observation. In 1997, the predominant wind direction was NW and SW, with nearly the same percentage (35%) for both directions. For 1996, the prevalent winds were divided into three quadrants: NE, SE, and NW (around 30% each). Note that although SW winds were not important for the weeks of 1996 analyzed here, the plume still behaves as if under the influence of upwelling-favorable (SW) winds, as explained in Valle-Levinson et al. (1998), 2001. There was a slight rotation on the mean direction of winds, from 1996 to 1997. However, the averaged monthly wind speed over the plume did not show significant differences between the two years.
By assuming the wind differences for both years are negligible, we can focus on the FWD variability between the two years as the cause of the discrepancy found in the surface signature of the plume (Figures 5 and 6 compared to Figures 8 and 9). The discharge rate for both periods of observation revealed two main differences (Figure 2). First, the peak of discharge (the maximum value) for 1997 occurred in the middle of the period observed, while in 1996 this peak occurred in the beginning of the data set and persisted for a shorter period. Second, the maximum discharge was 25% greater in fall 1997 than in fall 1996. Such variation is larger than the variation in winds between the two years.
The area where the FWD was more effective in driving the surface circulation can be also observed in the figures of cross-correlation between surface velocities and discharge (Figures 5 and 6 compared to Figures 8 and 9). The plume exhibited a meandering character during 1996 in comparison with a well-defined shape in 1997. The behavior of the plume during 1996 corresponded to the expected circulation patterns under SW winds, as presented previously by Valle-Levinson et al. (1998) and Boicourt et al. (1987). But as mentioned before, SW winds are almost not present during this period (1996). It is likely then, that FWD has more than just a complementary roll in driving the plume offshore.
The numerical experiments of Valle-Levinson et al. (1996) demonstrated how a sudden pulse of FWD onto the bay could produce the same effects over the plume that SW winds do. Of particular interest is the experiment that Valle-Levinson et al. (1996) performed by prescribing a pulse of high FWD during the first 10 days of simulation and then setting it to zero. This produced a similar surface outflow pattern as the one we found here for the 1996 case. Indeed, Valle-Levinson found that at the entrance the volume exchange between estuary and shelf shifts from outflow dominated, during the first 10 days, to a balance between inflow and outflow, after 20 days. The response time for the reversal of volumetric in-flow was approximately 1 day after cessation of the increased discharge. They established that time variations of barotropic discharge at the estuary upstream boundary influence the volumetric exchange at the entrance to the estuary. They also determined that a similar effect should be expected after the cessation of a continuous outflow generated by strong SW winds (Valle-Levinson et al., 1996).
By comparing the CCs for 1997 and 1996, the effect that we should observe after removing the tidal effects, and after assuming the similarity in winds for both periods, is the pulse of FWD. The reason is that the 1996 data started in the moment that a peak on the FWD occurred, followed by a decreasing discharging rate of about 400% (which can be compared to an ebb flow). Alternatively, the 1997 data set contained average FWD levels, followed by a slow increase and then by a slow decrease in FWD. This means that, when the spatial variability was calculated for the EOF maps (Figures 3 and 7), the opposing effects of the diminishing flow balanced the effects of the enhancing flow. This is not the case for 1996, where the pulse (high volume) of FWD was not compensated by low flow (Figure 2). That is why the difference between the results for the two years seems to be only the effect of the high FWD pulse in the surface currents.
Note that while we have shown that pulses of high FWD drive the plume in the same way that SW winds do, the quantity of the discharge also affects the plume geometry. Quantifying such effect is of interest in future research.
CONCLUSIONS
The present study consists of an analysis of the coastal circulation off Chesapeake Bay, focusing on the effect of the river discharge over the shelf. Our ability to observe the surface characteristics of the shelf circulation demonstrates the ability of HF radar to describe coastal features. Determining the reliability of such data is important (Graber et al., 1997), and for the data sets used in this study such information appears in detail in Gremes Cordero (2004).
The present analysis explored the relative importance of FWD variations and wind forcing in the surface signature of the Chesapeake Bay outflow. It established that the FWD variations are as important as the wind forcing in defining the plume dynamics, as demonstrated by the EOF analysis. We also determined the areas over the shelf where the different forcing is more effective in driving the local circulation (Figures 5 and 6 compared to Figures 8 and 9).
We corroborated our hypothesis that there was a relationship between river discharge and plume surface geometry over the shelf. Note that Garvine (1987), Chao (1988a, 1988b), and other collaborators performed a similar effort in theoretical grounds or performed data analysis only near the mouth, as did Valle-Levinson et al. (2007). However, their calculations require an accurate knowledge of salinity, water density, or both types of values. Instead, the present study brings the possibility of establishing the plume geometry only through two-dimensional information, essentially HF radar velocities, wind data sets, and streamflow measurements. We showed how the shape of the plume was influenced by an extreme increase of FWD, and the time response of shelf waters to such variation was established to be around 2–3 days. In addition, our analyses of the surface characteristics of the plume resemble the radar observations of Marmorino et al. (2000). We also validated the results of the numerical experiments performed by Valle-Levinson et al. (1996) in which the characteristics of the plume they described were in agreement with the shape of the plume observed in 1996.
Map of the zone studied, with the ocean surface current radar stations (blue dots), the resulting grid (black dots), Chesapeake outflow plume experiment 3 moorings (A1, A2, A3, and A4), and weather station localization (T). Bathymetry is in meters.
Wind speed and direction from Coastal-Marine Automated Network station for (a) Chesapeake outflow plume experiment 1 (COPE1) and (b) COPE3. (c) Total freshwater discharge for both periods investigated, given in cubic meters per second.
Empirical orthogonal function calculations for 1997: (a) mode 1 and (b) mode 2.
Amount of variance explained by each mode in the empirical orthogonal function (EOF) analysis for 1997 and EOF error calculations as in North et al. (1982).
Maps of correlation coefficients between the U component of surface current PCs and the discharge for 1997: (a) at zero lag, (b) at lag 1 = 1 day, (c) at 2 days, (d) at 3 days, and (e) at 4 days.
Maps of correlation coefficients between the V component of surface current PCs and the discharge for 1997: (a) at zero lag, (b) at lag 1 = 1 day, (c) at 2 days, (d) at 3 days, and (e) at 4 days.
Empirical orthogonal function results for 1996: (a) mode 1 and (b) mode 2.
Maps of correlation coefficients between the U component of surface current PCs and the discharge for 1996: (a) at zero lag, (b) at lag 1 = 1 day, (c) at 2 days, (d) at 3 days, and (e) at 4 days.
Maps of correlation coefficients between the V component of surface current PCs and the discharge: (a) at zero lag, (b) at lag 1 = 1 day, (c) at 2 days, (d) at 3 days, and (e) at 4 days.
