Cross-Frontal Velocity and Temperature Structure of the New Jersey Midshelf Frontal Zone

The cross-shelf structure of fronts, which occur during the winter months along the 30 m and 50 m isobath, are examined using remote sensing observations of the southern New Jersey shelf region. Shipboard observations show that cooler, fresher, less dense water is located inshore of the front ; however, little is known of the velocity structure in the frontal zone. Surface current fields from high-frequency radar are analyzed along with surface thermal front observations to describe the cross-shelf spatial variability of surface flow with regard to the fronts. Cloud-cleared, Level 2 MODIS Thermal IR sea-surface temperature (SST) data from AQUA and TERRA from winters 2003 2007 are processed using an edge-detection algorithm to determine the frequency, location, strength and orientation of the fronts. The record and seasonal progressions of the temperature and velocity fields are analyzed. No evidence is found to support the Ou Tidal Diffusivity theory that predicts that front location is a function of the spring-neap cycle. Cross-shelf locations of convergence due to the surface velocity appear to increase front occurrence, which is in agreement with Hoskins' frontogenesis theory. Shear/voriticity also seems to play a role in front occurrence and front orientation. Al so of note is that the location of minimum cross-shelf velocity variance is coincident with 30 m isobath fronts , while the minimum in along-shelf velocity variance is at the 50 m isobath. Furthermore, the 30 m (50 m) is the location at which the cross-shelf (along-shelf) wind-driven surface velocity component becomes greater (less) than the residual surface velocity component. Additionally, the front strength, measured by SST gradient magnitude, is inversely related to heat flux in the 50 m isobath region and nearly unrelated to the heat flux in the 30 m isobath region.

along the continental shelf of southern New Jersey. The focu s of the grant work is to characterize the vertical structure of these fronts. The focus of this work is the analysis of the surface structure/ frontal signature using remote sensing.

Background Definitions and Equations
Fronts ex ist throughout the ocean and range in length from I to 1000 km (Federov, 1986).
They can mark the boundary between two water masses and are identifiabl e by large horizontal gradients in physical, chemical, and/or biological properties. There are a variety of frontal types which have di stinct locations, vertical signatures, duration s, intensities, and formation mechanisms.
Prominent frontogenesis mechani sms are tidal friction, river di scharge, convergence, and/or wind. The most common types are tidal mixing fronts (TMF), buoyant outflow fronts (BOF), shelf break fronts (SBF), and MSFs. A schematic of where these fronts manifest on the New Jersey continental shelf is di spl ayed in figure I . Tidal mixing fronts form between stratified and unstratified regions. They develop as a result of solar heating dominating the stratified region and vertical mixing dominating in the unstratified region (Hill and Simpson, 1989). Buoyant outflow fronts arise as a result of significant freshwater discharge from coastal estuaries (Yanko vsky and Chapman , 1997). In the case of the New Jersey MSF, the Hudson River outflow is considered to be a surface-advected BOF (Ou et al., 2003). However, Narayanan and Garvine (2002) found that a surface-advected BOF can become a bottom-advected BOF at large di stances from the source. The isopycnals of a BOF are nearly vertical, sloping from the bottom to the surface (figure 2). Shelf break fronts have a similar vertical structure and mark the barrier between cool, fresh inshore waters and warm, salty slope waters. They occur year-round offshore of the 200 m isobath and have an associated geostrophic jet. This jet is approximately 10 to 25 km wide and flows southwestward at 20 to 50 emfs according to Fratantoni et al. (200 I) and Oey ( 1986). Furthermore, the cross-front velocities are on the order of I emfs while along-front velocities are one order of magnitude larger ( 0 ( 10) emfs) ( Oey, 1986).
Midshelf fronts have been observed in the shelf seas around the Briti sh Isles (Robinson, 1985), the Gulf of Mex ico (Huh et al., 1978), the East China Sea (Hickox et al., 2000) (Chang et al. , 2006), the South Atlantic Bight (Oe), 1986), and the Mid-Atlantic Bight (MAB ) (Ullman and Cornillon , 2001). The MSF on the New Jersey coastline is a wintertime (January -March) feature that separates cooler, fresher, less den se in shore water from warmer, saltier outer shelf water and tends to be aligned with the local bathymetry (Ullman and Cornillon , 2001). Also, since it appears only in the winter, the significant decrease of solar heating and weak stratification of the offshore waters suggests that thi s feature is not a tidal mixing front. Cross-shelf hydrographic sections (figure 2) show how the cross-shelf structure is similar to a type l front as defined by (Hill and Simpson, 1989), in which isopycnals extend from the bottom to the surface in the frontal zone.
Like the SBF, the MSF is believed to have an associated geostrophic velocity jet. This jet would be smaller in magnitude, due to weaker density gradients in the midshelf region. The magnitude of such a jet is e ·ti mated using the thermal wind equation (equation I). So as density (p) varies in the cross-shelf (x ) direction, the along-shelf geostrophic velocity (v 9 ) varies in the vertical (z). The other variables are the coriolis parameter (f), average density (p 0 ), and gravity (g). Using the winter 2007 research cruise data and assuming vg = 0 at the bottom, I calculated an approximate value of -5.71 cm/s for the v 9 (g = 9.8 mis, f = 0.92* 10-4 s-1 , p 0 = 1026.48 kg/m 3 , ap = 0. 11 kg/m 3 , ax = 10 km, az = 50 m).
A major hypothesis of MSF forcin g mechani sm is the Tidal Diffusivity (TD) theory described by Ou et al. (2003). While this theory attempts to explain why fronts form , there is another theory to note by Hoskins ( 1982), w hich di scusses how an ex isting front can be affected.
Of particular interest to thi s work is how Hoskins ( 1982) describes the temporal variation of a horizontal gradi ent as a functi on of the velocity gradient fi eld. It has been shown that fronts occur along isobaths, therefore my concern is primarily with the cross-shelf temperature gradient, 8T /ox . (2) In this equation , u and v are horizontal ve locities in the x (cross-shelf) and y (along-she lf) directions. F is the turbulent heat flux at the sea surface and S is the source term of radiative heating.
For the purposes of thi s anal ys is, I will concentrate on the first two term s on the right hand side (RHS). The first term on the RHS is a measure of convergence by the horizontal velocity field upon a temperature gradient. If convergence (divergence) occurs, thi s term tends to strengthen (weaken) the total cross-shelf temperature gradient (on the LHS ). And the second term on the RHS quantifies the effects of shear flow in the horizontal velocity upon a temperature gradient. Assuming 8T / 8y < 0, when positive (negative) shear occurs, thi s term tends to strengthen (weaken) the total cross-shelf temperature gradient (on the LHS ).
Ou's analytical model assumes a dominance of cross-shelf tidal motion . He assumes that buoyancy flux, Fx, is constant at across the shelf and equal to its value at the coast. He argues that diffusivity, and Px is the cross-shelf density gradient. Furthermore, Ou et al. (2003) assert that there is a total diffusivity minimum and consequently a density gradient maximum at midshelf.
Most importantly, frontal depth is predicted to have a square-root dependence on tidal amplitude (Ou et al., 2003). They assert that a front should migrate in shore/offshore with the spring-neap tidal cycles (of 14 and 28 days). Al so, In thi s case, the existence of a front has no effect on the rate of cross-shelf exchange of properties.

t.J Importance of Fronts
If the MSF is a smaller version of the SBF, than there may be enriched biological activity in the vicinity of MSFs due to vertical advection of nutrients. For instance, on the Argentinean coast, a band of high phytoplankton concentration, due to nutrient upwelling, appears along the SBF (Romero et al., 2006). Also, large quantities of ti sh, scallops, and zooplankton are found along this SBF region (Brunetti et al., 1998). Another example of the influence of fronts on primary production is off the coast of California. Albacore tuna are located on the warm, offshore of an upwelling front, while salmon are on the colder, inshore side (Breaker et al., 2005).
Additionally, fronts can act as a barrier to cross-shelf exchange of chemical and geo logical constituents, and thi s cross-shelf variation can be traced via water mass characteristics (Blanton , 1986).

Fronts
The along-shelf coastal current in the MAB is influenced by the bathymetry, since the inner shelf is quite shallow and gently sloping (Whitney and Garvine, 2005). Ullman and Cornillon (2001) found that winter (January-March) MSFs along the northeastern U.S. coast predominantly occur in the vicinity of the 50 m isobath (figure 3), where bottom depth on the shelf changes most rapidly. A second region further in shore in the vicinity of the 30 m isobath, and with slightly lower frontal probabilities, is also noted by Ullman and Cornillon (200 1). The 50 m isobath MSF appears to be approximately 5 km wide (Ullman and Cornillon,200 I). This MSF was first noted in Ullman and Cornillon ( 1999), in which they used an edge-detection alogorithm to perform a 12 year ( 1985 -1996) time series analysis of sea surface temperature (SST) images to obtain frontal length, width, depth, and other coordinating information. The surface signature of these midshelf SST fronts appear and di sappear on a regular basis throughout the winter and take approximately 3 to 5 weeks to form or decay (Ullman and Cornillon, 2001).
Not only have 50 m isobath fronts been found along the eastern U.S. coast, but they have 4 also been observed along the eastern coast of China. Hickox et al. (2000) found seasonally persistent cold surface thermal fronts that align along the 50 m isobath in the East China Sea (ECS). Just south of the ECS, in the Taiwan Strait, wintertime (December -February) 50 m fronts were detected (Chang et al., 2006). They noted a widening and shoreward movement of the frontal band over the winter period.
The relationship between wind and SST front dynamics was another aspect of the work done by Ullman and Cornillon (2001). They concluded that offshore winds result in stronger fronts, and vice versa with regard to onshore winds. Also, offshore winds during the months of January to March are typically caused by Cold Air Outbreaks (CAO) and tend to induce rapid surface cooling, especially over the shallower, well-mixed waters (Ullman and Cornillon ,200 I). Csanady ( 1978) found that a coastal front is strengthened by down-shelf winds and is weakened by up-shelf winds.

Surface Circulation and Wind Influence
The high-frequency (HF), Coastal Ocean Dynamics Application Radar (CODAR) manufactored by CODAR Ocean Sensors has been used extensively along the MAB , particularly in the Block lsland and the New Jersey shelf region. Thi s equipment enables constant observation of coastal flow with high temporal resolution (hourly) and wide spatial coverage. The velocity vectors have a 6 km resolution out to approximately 200 km from the coast.
According to Beardsley et al. ( 1976), the MAB sub-tidal shelf currents are predominately wind-driven. Furthermore, winds in thi s region are most frequently from the W and NW (Saunders, 1977) and vary on time scales of 0(2 -I 0) days (Mooers et al., 1976). But currents not only are influenced by winds, but are also large scale free-waves from the far field that migrate along the coast and affect the current in this region (Ou et al., 1981 ).
This area off the coast of New Jersey is very well-sampled. Kohut et al. (2004) noted two dominant regimes of hydrographic variability, which are summer stratified and winter mixed.
They found that in the winter, the current response is less correlated with the wind and there is an influence of bottom topography on current variability. LaTTE (Lagrangian Transport and 5 Tranformation Experiment) was conducted during the spring of 2005, from which Gong et al.
(Z006) di scovered five major current pathways visible in the NJ shelf, which tended to flow along the SBF, the MSF, and inner shelf. Additionally, they concluded that the pathways were strongly wind driven and influenced by bottom topography. Surface currents, sea level and winds from August 2002 -January 2004 in thi s region were analyzed by Dzwonkowski et al. (2008a).
They suggested that shelf wide cross-shelf surface flow is driven by cross-shelf winds. Further analysis of thi s data via an EOF analysis of the sub-inertial surface velocity structure identified flow patterns (Dzwonkowski et al., 2008b). They found shelf wide and point flow impact offshore transport. Also, Castelao et al. (2008) analyzed data from satellites, gliders, buoys, and the HF radar during the summer/spring of 2006 and found an offshore jet south of the Hudson shelf valley that was correlated to upwelling winds. These are just a g limpse of the abundant research conducted in thi s region.
6 2 Data and Methodology

.1 Study Area and Analysis Period
The analysis region is the shelf off southern New Jersey in the MAB (figure 4). These waters were selected due to the continuous monitoring program carried out by Rutgers University' s Coastal Ocean Observation Lab (COOL) . This includes a five year record of surface currents measured via HF radar. Another reason for selection is that thi s region was found to exhibit a strong MSF signal in winter (Ullman and Cornillon , 1999). The general orientation of the coastline in this area is 38°T, defined as Bcoast · Preliminary analysis of SST imagery found that fronts were present along the 50 m isobath and, almost as frequently, along the 30 m isobath.
Jn order to separately describe the fronts in the vicinity of each of these isobaths, two 30 km  ( 1999) found that fronts along the shelf were common during January -March. Preliminary analysis of SST imagery showed that front occurrence increased in December and decreased in March. In order to analyze the frontogenesis (front generation) and frontolysis (front dissipation), the winter period was defined to be from 0000 0 I December Further mention of a record mean will be with regard to these years. The winter mean consists of d t · h. a a wit in the four months of that winter. The monthly mean consists of the average 7 of values found during the days within that month . And the daily mean will be defined as the ge O f the given property from 0000 to 2359 of that day. avera 2 • 2 Sea Surface Temperli)ture Fronts Sea surface temperature (SST) imagery was obtained from NASA's Ocean Color Web The swaths obtained had latitudinal bounds of 38°N to 42°N, a longitudinal bounds of 74°W and 68°W and a resolution of I km.

Ullman and Cornillon
The processing of SST data began with converting the files from hdf to netCDF. This is the format required by the hi stogra m based edge-detection algorithm, which is expl ained in the next paragraph. [mages were cloud cleared using the inherent MODIS quality assurance variable known as quality flag. For each SST pixel within each image, there is a coordinating quality fl ag value ranging from 0 to 3. The most liberal quality fl ag value of 3 was used to mask the SST values at which there were clouds and swath edge errors. Examples of raw and cloud masked images are shown in figure 5.
After cloud-masking, eac h SST image was processed by the edge-detection algorithm (E DA) developed by Cayuta and Cornillon ( 1995) to objectively detect front s. Thi s EDA was tirst developed by Cayula and Cornillon (1992) to detect surface thermal fronts in single AVHRR images and was later improved upon to include multi-image processing using a five day sliding temporal window and a 32 by 32 pixel sliding window ( Cayula and Cornillon, 1995).
The EDA outputs two ascii files of front pixel locations, one with latitude and longitude, the other with pixel coordinates, for every input SST image. Figure Sb is an example cloud-masked SST image with fronts overlaid. Front locations, along with bathymetry data and the original SST images were used to make a front database. The database includes a range of information about each front pixel: satellite passage time, latitude, longitude, water depth, depth gradient magnitude, depth gradient direction, SST gradient magnitude, and SST gradient direction.
Given that thi s shelf region is cloud-free only 15 % to 30% of the time over the months of my analysis, it is necessary to take into account the fact that a pixel cannot be detected as a front if the pixel is cloud covered. This is done using a quantity called percent front probability.
Percent front probability over a given time period is calculated as the number of times a given pixel is identified as a front divided by the number of times that same pixel is identified as clear, in other words has a valid SST value.
The cross-shelf temperature structure in this region is such that warm waters are generally found offshore of cold waters. Consequently, the analysis of front location and SST gradient magnitude will be limited to front pixels that meet thi s criterion, which have been referred to as cold fronts (Ullman and Cornillon, 1999).
From the database information, a variety of variables were calculated and analyzed to describe the temporal variation in location and strength of the cold fronts. Using the front pixel latitudes and longitudes, the intersection location of line B and front segments within Box 30 (figure 4) are averaged to make daily, monthly, yearly and record mean front cross-shelf locations. While for SST gradient magnitude and SST gradient direction all front pixels within Box 30 were averaged into daily, monthly, yearly and record mean values. A si milar methodology was used for Box 50.

Surface Velocities
Rutgers University maintains a HF radar array off the coast of New Jersey. The coverage array extends from Long lsland to a few nautical miles past Delaware Bay, and from a few kilometers offshore out to the shelf break. Current vectors were estimated along each of the three cross-shelf lines every 5 km using the least squares method of Lipa and Barrick ( 1986), using an averaging radius of I 0 km. Points with > 10% temporal coverage and ::; 1.25 map error are used. Miss ing data were linearly interpolated. Each component, Ucodar (East) and Vcodar (North) was low-pass filtered using a 4 th order, 36-hour cutoff period, Butterworth filter. The velocity vector (resultant of Ucodar and ) was resolved into cross-shelf (u), positive offshore, and along-shelf (v ), positive up-shelf Vcodar components towards () coas t (equations 3 and 4).
The cross-shelf divergence and shear, defined as ou/ ox and ov /ox, were calculated from u, v and the distance between points.
The spatial mean current is defined as the average current vector calculated from all 75 points (on all three lines) . Due to reliability and consistency of surface velocity vectors, a portion of the analysi s will be limited to line B. There are several lengthy temporal gaps in these data, which are due to one or more of the CODAR sites having no radials. Due to the geometric positioning of the sites, it is frequently the case that the in shore 4 to 5 km portion of line C does not have data. It is assumed that winds are spatially uniform over the region of interest and are calculated d · inaly First winds were converted from meteorological convention, where wind direction accor o · is defined as the direction from which the wind is blowing, to the oceanographic convention, where wind direction is defi ned as towards which the wind is blowing. The wind speed and direction were converted into U wi ncl and Vwin cl components, with 90°T as the positive x-direction.

Buoy
Each of the components from each buoy were low-pass filtered using the same filter as the surface velocities. The Uwincl from buoy 44009 and U wincl from buoy 44025 were averaged. Thi s was also done with Vwincl to obtain a single hourly wind vector for the entire region . The resultant hourly wind vector (made by 'l.Lwi n cl and VwincL) was resolved into cross-shelf and along-shelf components just as surface velocity had been in equations (3) and ( 4 ).

2.S Tidal Amplitude
Sea level data were obtained from Atlantic City, NJ (figure 4), Station ID 8534720, from NOAA's website (http://www.co-ops.nos.noaa. gov/) . Hi storic, verified, hourly water leve ls in meters with a MLLW datum were downloaded. They were then complex demodulated at the M 2 tidal frequency to get the tidal amplitude (m). Complex demodulation uses a least squares algorithm to determine the amplitude change at a specific frequency over the course of a particular time series.
In this section, I will describe the temporal variability of fronts using averaged statistics.
The four variables, which are derived solely from the SST images, are percent front probability, front location, SST gradient magnitude, and SST gradient direction. They will be described in terms of their record, annual , and monthly variability.

Front Probability
Examination of front probability maps (figure 6) shows that there is significant variability in position, width, and occurrence of frontal zones over the five winters. The record mean percent probability (figure 6f) shows that along the 30 m isobath, the highest percent probabilities occur directly on that isobath, whereas, for the 50 m isobath, the highest probabilities occur 5 km to I 0 km offshore of that isobath. It is important to note that high front probabilities are not observed in the vicinity of the SBF (offshore of the 200 m isobath) due to the method of cloud-clearing used. By using a mask value of 3, pixels with strong temperature gradients, on the order of 1°C/km, are masked. I verified that this was not occuring in the middle and inner shelf, since SST gradients are smaller in magnitude than the SBF. Just south of 39°N there is a bend in the 50 m isobath. South of this location, the magnitude of front percent probability of fronts along the 50 m isobath decreases while the width of the frontal band increases. Similarly, south of 39.3°N there is a bend in the 30 m isobath, south of which the frontal band widens and front probability decreases. Also note that the 30 m isobath appears to have an additional, less frequent band about 20 km inshore of it. This feature is more apparent in front probability for each winter (figure 6a-e), and is the next focus of my analysis.
In the record mean front probability map, fronts are on both the 30 m and 50 m isobath, however, the front probability images for each of the five winters show that this is not the case during every year. In 2003 , fronts are well defined on the 30 m and 50 m isobath south of the bend in each · · respective 1sobath. North of these points, there appears to be an increased number of fronts in between these isobaths. In 2004 the frontal bands are most di stingui shable along the 2 001, the 30 m isobath front is more frequent south and inshore of its bend. The 50 m isobath during thi s year is most coherent from the Hudson Canyon to its bend, south of which frontal bands are more patchy and di splaced offshore.
A goal is to find out when these fronts develop or di ssipate over the winter months. A five year record monthly average of percent front probability was cal cul ated for each of the four calendar months (fi gure 7). In December, there is a more narrow region of hi gh probability, which for the 50 m isobath is positioned north of the bend. ln January, the high probability frontal band widens, and alon g the 50 m isobath it di sconnects just north of the bend. ln February, the 30 m isobath frontal band has its hi ghest percent probability, particularly on and in shore of the isobath. Also, the 50 m frontal band is almost twice as wide as it was during January. This is due to high variability over that month of each gi ven year, rather than the variability between years. This was confirmed by examin ing the indi vidual monthly percent front probabilities (not shown). In March, the front probabilities decrease and the frontal bands narrow. Note that between 39.2°N and 39.3°N there is a high probability of fronts in between the 30 m and 50 m isobaths.

Front Location
In attempts to more accurately quantify the cross-shelf movement of each frontal band, the mean front location was calculated for each day, month, and winter. From thi s I am a ble to see that, in Box 30, fronts have a seasonal perturbation of less than +5 km offshore of the isobath and have a record mean front locati on of 47 km offshore (figure 8) . On the other hand, fronts in Box SO have a seasonal perturbation of less than +IO km offshore of the isobath and a record mean 13 of 98 km offshore. lt is important to note that in 2005, the 30 isobath front was farthest offshore during that year, while the 50 m isobath front is farthest in shore during that year. However, thi s I ·s biased given that there are no fronts observed in Box 50 along line B during the month va ue 1 ofFebruary, which is the month with the greatest offshore di.stance over the record (figure 9b).
In the front probability plots, there appears to be no consistent pattern of front location for a specific calendar month ( figure 7). Also, the record mean change in distance offshore over the winter season is less than 5 km (figure 9). Note that between year, the monthly mean decreases by 5 km.
In the study area, the water depth generally increases with distance offshore (figure I 0).
The Ou theory argues that front location (depth) will be a function of tidal amplitude and this should vary over the spring-neap cycle ( 14 and 28 day periods). Time series of daily mean front location (figure 11 ) do not exhibit visual periodicity, but the gapiness of the data does not allow definitive conclusions to be drawn. For this reason, a spectral analysis was performed. The spectral analysis of the daily mean front location over the entire record was performed usi ng the Lomb method (Press et al. , 1992). The tomb normalized periodogram is ideal for time series with long gaps. The benefit of using this over a fast fourier transform (FFT), is that the FFT spectrum of an interpolated time series can result in false low-frequency bands of high power.
Given the uneven temporal spacing and long gaps between seasons, the Lomb method was used.
In Box 30 the only spectral estimate greater than the significance level is at 90 days, which is the seasonal/winter cycle ( figure I 2a). Although there are bands of increased significance near the !4 days and 28 days, there is not conclusive evidence of spring-neap frontal movement that would support Ou's theory. Con-elation coefficients were calculated for the 30 m front location and tidal amplitude, the 50 m front location and tidal amplitude, and the 30 m front location and the SO m front location. They are 0.06, -0.07, and -0.08 respectively, which are not significantly « t from zero Thus, the front location time series do not indicate significant movement di 11 eren · over the spring-neap cycle.

Sea Surface Temperature Gradient Magnitude
The SST gradient magnitude is used as a measure of front strength. The annual mean SST gradient magnitude of front pixels in Box 30 seems to vary inversely with mean gradient which is believed to be due to warmer than average December and January during that winter (visible in the monthly mean SST, not shown).

Sea Surface Temperature Gradient Direction
The gradient direction is defined as the direction towards warmer water. The winter mean SST gradient direction ranges between 99°T and 110°T (figure 15), which are in agreement with the values found by Ullman and Cornillon (200 I). Given that the cross-shelf direction is I 28°T (8coast+90°), this means the gradient direction is slightly upshelf, such that ~;; > 0. Therefore in order for ~~ to increase (decrease), the shear at that cross-shelf location must be negative (positive). Also, SST gradient direction in Box 30 is typically more in the along-shelf direction than the SST gradient direction in Box 50, every year except for 2005.
Over the winter months, there is not much greater than a I 0° change in orientation (

• 2 Surface Velocities
This section describes how the current components, cross-shelf and along-shelf velocity, vary in magnitude, direction, and their gradient magnitudes vary with distance offshore (along the 3 cross-shelf lines) over the winter months of the record. The majority of the analysis focuses on the cross-shelf and along-shelf surface velocities individually, and towards the latter part of this section just line B is discussed. The relationship between wind and surface current, particularly the variation of wind influence with distance offshore is also described.

Annual Variations Surface Velocity
The winter record mean surface flow is predominately downshelf and offshore, and there is

Surface Velocity and Wind
Since surface currents in thi s region are predominately wind-driven (Ou et al., 1981) , the next portion of my analysis focused on the cross-shelf variability of wind influence upon surface currents.
A cross-correlation analysis was performed on the spatial mean current and the wind to find that the current lags the wind by one hour. This is in agreement with Dzwonkowski et al. However, the correlation coefficients change by less than 0.0 1 for a 0 -3 hour lag range. With these results , in addition to the fact that the velocities are low-pass filtered which makes a one hour lag rather irrelevant, instantaneous wind and current vectors are compared.
First, for each winter, for each point along line B, a complex correlation was performed between the demeaned wind and demeaned surface velocity. From this, a magnitude and phase of correlation, for each point, for each winter was obtained.
The degree to which the cross-shelf variability of the surface velocity is due to wind is visible in figure 28. The wind/current correlation magnitude has a maximum just offshore of 80 km at 0.69 and decreases to approximately 0.49 at 150 km offshore. This is a proxy for the portion of the low-pass filtered current that can be attributed to winds. In these calculations, a positive phase angle is defined as the current being to the left of the wind. Also, the wind/current correlation phase ranges from -60°T to I 0°T. It decreases (becomes more to the right of the wind) as depth increases. In the record mean phase there is a plateau around midshelf. Furthermore, the greatest variation in phase between the years is offshore of 90 km. It is important to acknowledge that the flow in thi s region is not in Ekman dynamical balance, such that the surface velocity would be 45° to the right of the wind. However thi s is due to the fact that shallower depths are more affected by bottom friction, hence the larger (more negative) phase angle with di stance offshore and the current is more Ekman-like.
Then the instantaneous wind-driven velocities at each point for each surface velocity component, u and v were computed. At each point, a linear regress ion was performed between u and the cross-shelf and along-shelf components of the wind. The same was done for v . From this, coefficients a and b, for u and v respectively, describe the winter mean influence of the wind components at that point. With these coefficients equations (5) and (6) were then used to calculate the instantaneous wind-driven velocity at each point.
Uwfad-driven = a1 * Uw ind + a2 * Vwind + a3 Vwind -driven = b1 * Uw ind + b2 * Vwi nd + b3 The residual velocity was calculated by taking the original , low pass filtered u and v and subtracting out the wind-driven velocity, for each point, at each time. The residual is assumed to be influenced by non-tidal and non-wind forcings, of which include the far-field dynamics. Figure 29 shows velocity variance, separated into wind-driven and residual components, as function of cross-shelf distance. In subfigures 29a,c,e, g,i and k, it is evident that the wind-driven portion of cross-shelf velocity generally increases with di stance offshore. On the other hand, in subfigures 29b,d,f,j and I, it is evident that the wind-driven portion of along-shelf velocity generally decreases with di stance offshore. Note the cross-shelf location at which the winddriven velocity variance is equal to the residual velocity variance. For the cross-s helf velocity variance, the wind-driven portion is at a minimum at 30 km and overtakes the residual portion at 45 km . From that point on, the wind-driven portion tracks with the observed current. For the along-s helf velocity variance, the wind-driven portion has peaks in the in shore regions and linearly decreases as a function of distance offshore. It is around 90 km that the wind-driven portion is less than the residual portion. Note that in 2003 a nd 2004 there is not a minimum in along-shelf velocity variance (noted by the blue line). This can be attributed to the lower residual (green line) offshore of 120 km.
Regardless of year, winds are most frequently in the positive cross-shelf direction, particularly betwee n 90°T and 140°T (fi gure 30). Not only are winds from this direction more frequent , but they are 3 mis stronger wind speed on average. A second band for more frequent winds is towards the NNE, between 10°T and 40°T. Note that di stribution of wind direction in 2006 and 2007 is most similar and in agreement with Saunders ( 1977)

Comparisons
This section analyzes the temporal and spatial relationships between front location, surface velocities, wind and heat flux. It will discuss the features found using the aforementioned data, in the vicinity of the 30 m and 50 m isobaths, and how they seem to relate to one another.

Cross-shelf Convergence
Along line B, there appears to be a positive correlation between surface velocity gradients and front probability. The seasonal mean divergence was calculated at each point along line B.
The offshore distance at which the winter mean velocity is convergent (negatively divergent) then the rate of change in cross-shelf temperature gradient due to convergence is 2.3 * Io-7 °Clkrn!s. And in one day, the convergence could change the temperature gradient by +0.20°C/km.

Assuming
.:!__ (aT) ~ _ avaT dt ax ax ay (8) then the rate of change in cross-shelf temperature gradient due to shear is 9.8* I o-8 °C/krn/s. And in one day, the negative shear/vorticity could change the temperature gradient by +8 .5 * I 0-3 ocfkm. Given these values for ft ( ~) , convergence is more influential than shear. Futhermore, a sustained convergence can double the existing gradient (0.25°C/km) in approximately 1.25 days. However, it takes about 30 days of shear/vorticity to cause the existing gradient to double.

Composite Analysis
The In winter 2003, it is evident that winds to the S to SW results in strong convergence in the cross-shelf velocity anomaly, at midshelf (figure 32b). This is most distinguishable from January 29th to February 2nd, from February 15th to February 19th, and from February 26th to March 2nd. Note that it is frequently cloudy during these periods. But, once the percent clear increases after these incidents, fronts become apparent. There are times when the cross-shelf location of maximum convergence (blue offshore and red inshore) is coincident with the cross-shelf location of maximum shear.
In December 2005, periods of sustained cross-shelf convergence are less frequent and more inshore (figure 33) than in the previous example. But there is a consistent shear in the along-shelf velocity, particularly during the latter one third of the month that is coincident with fronts along the 50 m isobath.
The last example (figure 34) includes the "Upshelf Incident," which takes place between

Heat flux, Wind, and Fronts
Another factor affecting SST gradient magnitude, which we will discuss briefly, is heat flux

Discussion and Conclusions
It is difficult to distinguish the exact time, movement, and consequently the cause of the fronts due to the short periods with which they form or decay largely due to cloudiness. There is significant variability from year to year. Though varying in time, space, orientation, and strength over a season , as well as over the five year record, winter fronts are located just offshore of the 30 m and 50 m isobaths. lt is still unclear as to the entire dynamical picture, but l believe the surface currents are a major forcing function in front positioning and strength. lt was theorized that a geostrophic surface jet is co-located with fronts, but there is no evidence of that. lf a jet does occur, either it is narrower than the 6 km CODAR resolution can distinguish or it is subsurface.
There does seem to be a wider region of strong down-shelf flow, 0(20 km) wide offshore of the 50 m isobath, just at the conclusion of the " Upshelf incident." Al so, the record mean change in front location offshore over the four winter months is less than 5 km. Though it was found by Ullman and Cornillon (200 I ) that midshelf fronts progress offshore over the winter months, the data here are incongruent with that finding. This discrepancy may be due to the difference in the geographical extent of my analysis compared to Ullman and Cornillon (200 I). Their region covered the entire shelf width and a further extent in the along-shelf direction, rather than 25 km by 200 km boxes around specific isobaths.
Furthermore, the bathymetry in thi s region is gently sloping and upon cursory analysis it does not appear that fronts are coincident with high bathymetry gradients. There is a great deal of noise in the daily mean front location data (figure 11) displaying the fronts have significant variability in position. l am not convinced that relatively high bathymetry gradient regions of the shelf are where fronts occur, as claimed by Ullman and Cornillon (200 I) and Gong et al. (2006).
Similarly, there is no evidence of fortnightly or monthly movement of front location, as is evident by the spectral analysis results, that would support Ou 's theory. Furthermore, Ou argued that a minimium in tidal diffusivity times depth would result in a maximum in density gradient at midshelf. There are minima found in the plots of velocity variance as a function of cross-shelf distance. However, since the surface velocities were low-pass filtered it is unlikely that this is related to tidal diffusivity. Despite this fact, it should not be neglected that this could be another 24 form of diffusivity.
It is also important to note that the range of SST gradient magnitudes (0.23 °C/km to 0.32 oc/km) found in thi s work are comparable to those found for winter fronts in Taiwan Strait (Chang et al. , 2006).
In terms of forcing in the cross-shelf, offshore (positive cross-shelf) winds see m to be correlated with stronger fronts, expressed as higher SST gradient mag nitudes. In the case of 2007, when the fronts are stronger in the 50 m isobath region , thi s is likely due to the relatively longer period of sustained winds to the SE that cooled the shelf more rapidly and farther out than in the previous four years. Also Oey ( 1986)  If these two fronts are type I as described by Ou et al. ( 1981 ), then there is not simply a single causation that determines their formation. There is much more in terms of coastal dynamics and mesoscale processes that come into play, particularly the vertical structure and the influence of the far field, that make it difficult to state with certainty the relationship between surface fronts and surface currents.

Ongoing Research
Researchers at Rutgers University and the University of Rhode Island have a wealth of information in thi s region at their disposal. They are able to capture a more complete spatial and temporal resolution of these fronts from moored and shipboard Acoustic Doppler Current Profi 1ers (ADCPs), Autonomous Underwater Vehicles (AUYs), thermi stors, as well as from satellites and CODAR. Further studies should incorporate as many of these data sets, and particularly those that are able to characterize front location on a timescale on the order of several days.