1. Introduction
The Gulf Stream turns eastward into deep water at Cape Hatteras and flows toward the tail of the Grand Banks of Newfoundland as a meandering warm jet between the Sargasso Sea to the south and the cooler slope water to the north. The meanders grow with time and propagate eastward. Large looping meanders may remain stable for weeks or they may pinch off at the base in a matter of days, forming warm core rings to the north and cold core rings to the south. The rings generally propagate back to the west, interacting with other rings and meanders in the process until they are reabsorbed. These energetic mesoscale events, including meander growth and propagation, ring formations, ring–stream or ring–ring interactions, and ring absorptions, make the prediction of Gulf Stream and ring evolution on timescales of a few days to weeks an interesting and challenging task for both observationalists and modelers.
The Data Assimilation and Model Evaluation Experiment (DAMEE) was designed to evaluate the nowcast and forecast capabilities of dynamical ocean models in the Gulf Stream region (GSR) described above (Willems et al. 1994). To focus the evaluation on model physics under optimal conditions, a 6-week duration test case was formulated during a data-rich time period from the late 1980s. Satellite infrared imagery, Geosat altimetry, and numerous temperature–salinity profiles obtained from several dedicated observation programs were used to construct a series of snapshots of the Gulf Stream north wall and ring locations at approximately weekly intervals. These surface snapshots, called feature analyses, were used to generate a series of three-dimensional initialization and verification fields for the numerical models using the Optimum Thermal Interpolation System (OTIS) (Clancy et al. 1990). The forecast model performance was then evaluated based on the model’s ability to reproduce observed events (e.g., ring formations and absorptions) and to predict the location of the evolving Gulf Stream north wall.
A series of accurate Gulf Stream and ring feature analyses clearly are vital to the success of DAMEE GSR. Development of the 6-week duration case study presented here was aided by a new image compositing technique called “patching,” more accurate altimetric geoids (Glenn et al. 1991), recently unrestricted U.S. Navy temperature data, and a Geographical Information System (GIS) to generate data overlays. The new case study was generated for 25 May through 4 July 1988 to take advantage of (a) the most cloud-free satellite infrared coverage of the Gulf Stream during the entire Geosat Exact Repeat Mission, (b) the overlap of several in situ observation programs that provided a greater number of temperature–salinity casts than normally available, and (c) the availability of an existing case study covering 4–18 May 1988 (Glenn and Robinson 1995) that made it possible to extend the long-term evaluation period to 8 consecutive weeks.
The purpose of this manuscript is to (a) introduce the new patching technique for image composites, (b) present the results of the DAMEE GSR feature analyses, and (c) compare those results to other analyses available for the same time period. The individual datasets used here and the choice of the case study time period are briefly described in section 2. Section 3 discusses the patching technique, demonstrates its utility through a comparison with warmest pixel composites, and uses the technique to compare the two different sources of satellite imagery available for the feature analysis. Section 4 discusses construction of the GIS data overlays and the resulting evolution observed during the 6-week duration case study. Section 5 compares the DAMEE GSR results to two other Gulf Stream analysis products available for the same time period. Section 6 summarizes and concludes.
2. Datasets
The fortuitous overlap of several oceanographic observation efforts in the late 1980s resulted in the collection of one of the most extensive Gulf Stream datasets to date. Between November 1986 and January 1990, the Gulf Stream was monitored from space by both the Advanced Very High Resolution Radiometers (AVHRR) on the NOAA polar-orbiting satellites and the radar altimeter on Geosat. Numerous in situ observation programs were active, including the Synoptic Ocean Prediction (SYNOP) experiment (Watts 1991), the Regional Energetics Experiment (REX) (Mitchell et al. 1990), the Naval Oceanographic Office underflights of Geosat (Horton et al. 1992), and the Gulf Stream Forecasting Project (Glenn and Robinson 1995). The different remote sensing and in situ datasets are complementary. The AVHRR imagery provides broad spatial coverage in cloud-free regions but is patchy because of cloud cover. The along-track Geosat data lacks the synoptic spatial coverage of the AVHRR imagery, but once an accurate geoid is established, the altimetrically derived dynamic topography can be used to locate circulation features obscured by clouds or surface thermal processes. In situ data also are limited in spatial extent, yet they provide often critical subsurface information. Previous Gulf Stream forecast model validation studies (e.g., Glenn and Robinson 1995) used this multiplatform dataset to develop several 1–2-week duration case studies scattered throughout the Geosat mission based solely on the availability of air-deployed expendable bathythermograph (AXBT) or conductivity–temperature–depth (CTD) data. These case studies consist of a best estimate of the Gulf Stream and ring locations on both the model initialization day and the verification day exactly 1 week later. Although Geosat altimetry or in situ observations can greatly reduce Gulf Stream location errors locally, Glenn and Robinson (1995) found that the broad spatial coverage provided by cloud-free AVHRR imagery was the most effective means for reducing the regionally averaged error. Since the purpose of this study is to evaluate forecast model physics with the best available datasets rather than forecast system performance under typical conditions, time periods with the most extensive AVHRR coverage were sought.
With the aim of maximizing AVHRR coverage over an extended time period during the Geosat mission, the midweek NOAA–National Weather Service (NWS) oceanographic analysis charts (Auer 1980) were examined for completeness of Gulf Stream AVHRR coverage. The midweek charts were chosen since the AXBT deployments discussed below are clustered with midweek maximums. Gulf Stream coverage statistics (observed length divided by total estimated length) between Cape Hatteras and the Grand Banks were calculated for 1987 through 1989 (Fig. 1). Average coverage for the entire 3-yr time period was 47.7%. The coverage plots indicate that day 140 through day 190 (mid-May through early July 1988) was the only time period during which the coverage remained consistently above 60% for greater than 2 weeks. This AVHRR data-rich period therefore was chosen to develop the long duration case study for DAMEE–GSR.
a. NORDA AVHRR
The primary AVHRR dataset used here was collected by the Naval Oceanographic Research and Development Activity (NORDA) for the Geosat Ocean Applications Program (GOAP) as described by Lybanon et al. (1990). The AVHRR imagery was earth located and converted to sea surface temperatures by NORDA and transferred in real time to the Gulf Stream Forecasting Project (Glenn and Robinson 1995) at Harvard University where the data was archived. The images cover the region between 75° and 49.425°W with a resolution of about 2.25 km. Because the real-time activities of both the GOAP and the Harvard groups were relying on this AVHRR dataset for Gulf Stream locations, NORDA was careful to process at least the best image of the day if any portion of the Gulf Stream was visible at all. If different segments of the Gulf Stream were visible throughout the day, multiple images were occasionally processed and transferred. The time of each processed image within the above-defined model evaluation period is shown in Fig. 2. A total of 42 images are available over 50 days.
b. Pathfinder AVHRR
The Pathfinder sea surface temperature (SST) project was designed to produce a long time series of global AVHRR-derived SSTs for climate change studies. Toward that end, NASA’s Jet Propulsion Laboratory began a reanalysis of the historical AVHRR data (PO-DAAC 1994). Improvements include a detailed reanalysis of the calibration data (Brown et al. 1993), resulting in a time-dependent term in the nonlinear SST algorithm to compensate for instrument drift, and the specification of three atmospheric water vapor regimes with different corrections in each. SST values are grouped into daily ascending and descending passes, and mapped onto a 9-km equal angle grid. Two data quality tests (PO-DAAC 1994) are then performed to help flag cloud-contaminated pixels. The first, described as a satellite test, compares channels 4 and 5 differences with nearby pixels. The second, called a reference field test, is a spatial and temporal homogeneity test that checks if returns are within ±2°C of the expected value. Passing both tests returns a data quality value of 3, failing either one results in a data quality of 2, and failing both returns a data quality of 1.
c. Geosat
Geosat was in a 17-day repeat orbit during the Exact Repeat Mission (ERM) from November 1986 through January 1990. At the latitude of the Gulf Stream, each successive overpass of the altimeter was displaced 1500 km to the east. Since the Gulf Stream region is only about 1700 km wide, perfect data recovery would typically result in about one ascending and one descending pass per day over the Gulf Stream. During the 17-day repeat period, the grid would be filled in, resulting in a zonal track separation of about 120 km with adjacent tracks about 3 days apart. During this 50-day period, however, only 42 passes (38 ascending and 4 descending) are available at the times shown in Fig. 2. Data dropouts were often caused by the altimeter failing to maintain a lock on vertical.
d. (A)XBT/CTD
(A)XBT/CTD data were acquired from the in situ observation programs discussed in the introduction and from the data archives of the National Oceanographic Data Center (NODC) and the Fleet Numerical Oceanographic Center (FNOC). As expected, many of these datasets overlapped. Figure 2 contains a histogram of the number of independent (A)XBT/CTD observations available each day during the DAMEE GSR study period (548 total). Days with over 30 observations correspond to NORDA or Naval Oceanographic Office AXBT surveys along Geosat groundtracks. Although numerous AXBTs were available on these days, an along-track flight typically results in only one Gulf Stream crossing, and the AXBTs are partially redundant if data were also received from the Geosat altimeter on the same day. The most useful AXBTs for defining Gulf Stream features were those deployed by the U.S. Navy as part of the Gulf Stream Forecasting Project (Glenn and Robinson 1995). These weekly flights of 30 AXBTs were designed specifically to define Gulf Stream and ring features, typically resulting in six Gulf Stream crossings and three to four ring crossings in the vicinity of critical interactions or in areas obscured by clouds.
3. AVHRR data comparisons
a. Patching versus warmest pixel composites
Single AVHRR images of the entire Gulf Stream are quite rare due to the variable cloud cover that obscures portions of the ocean surface. Cloud tops viewed by the AVHRR sensor are much colder than the ocean below and, assuming the ocean temperature is approximately uniform, partially cloud-covered pixels are colder than cloud-free pixels. Because the clouds often move faster than the oceanographic features, more continuous views of the ocean surface are often constructed by compositing several images collected over a few days. One standard method is to construct warmest pixel composites from a time series of several images simply by retaining for each pixel the warmest temperature observed in the series. The assumption here is that the warmest pixel is the most cloud free and that the oceanographic features remain stationary.
The Gulf Stream north wall, however, can move quite rapidly. Lee (1994) documents meander propagation speeds over 15 km day−1. Frontal eddies and shingles can move even faster. Although there is little documentation on frontal eddy propagation speeds downstream of Cape Hatteras, propagation speeds over 50 km day−1 are well documented in the South Atlantic Bight [see Glenn and Ebbesmeyer (1994) for a recent review]. The movement of the north wall or shingles along the north wall can result in a spatial smearing of the Gulf Stream features over the time period of the composite. For example, a warmest pixel composite of a propagating meander of constant form will result in a composite meander with a wider crest and a narrower trough than the true meander. Similarly, a warmest pixel composite of a growing but stationary meander will not look like a snapshot of the true meander at any given time. Instead, the composite meander will be constructed from the largest crest observed at the end of the time period and the smallest trough from the beginning.
Since the numerical models being evaluated require the best estimate of the true shape of a meander system at an arbitrary time to predict its ensuing evolution, a new image compositing technique called patching was developed. The patching technique first requires elimination of all the cloud-contaminated pixels from a series of images. In our case, a simple temperature threshold of 10°C was used. Typical summer slope water temperatures range from 12° to 15°C. The most cloud-free image was chosen for the central time. The patched composite is assembled beginning with the cloud-free pixels in the image furthest from the central time. Cloud-free pixels from the remaining images are then overlaid in order of decreasing time difference. Since the cloud-free pixels from the central image are overlaid last, most of the Gulf Stream is defined by this single image with only the cloud gaps filled with data from surrounding days.
While both patched and warmest pixel composites smear features in time—an unavoidable problem when compositing is necessary—the patched composites do eliminate most of the spatial smearing associated with warmest pixels. Figure 3 is one example of different composites for 4 July 1988 that was constructed from images collected on 2–6 July 1988. The actual image from 4 July (Fig. 3a) and the digitized north wall defined by the strongest temperature gradient (white line) indicate there are four meander crests, and the meanders have similar wavelengths. Cloud gaps are indicated in gray. The cloud gaps are filled in the patched composite for 2–6 July (Fig. 3b), but the digitized north wall remains the same as it appears in the single 4 July image. Figure 3c is the warmest pixel composite constructed from the images available on 3–5 July. The actual time difference between the first and last image is approximately 48 h. The digitized north wall from the warmest pixel composite (black) can be compared to the actual north wall from 4 July (white). The warmest pixel composite has five meander crests, with three medium length waves and two short waves. In contrast, the 2–6 July (96 h) warmest pixel composite (Fig. 3d) has three meander crests with two medium length waves and one long wave. Depending on the image compositing technique and the size of the data window, between three and five meanders are observed in this limited region.
After examining the AVHRR dataset for the most complete coverage of the Gulf Stream at approximately weekly intervals, similar 96-h patched and warmest pixel composites were constructed for each of the central days marked in Fig. 2. The time between central days ranges from 5 to 9 days and averages 6.7 days. The average and maximum offsets between the patched and warmest pixel north walls for each central day are given in Table 1. The average offset is defined as the area between the two north walls divided by the average length of the north walls. Although the average offset values are small (0.4–4.7 km), the maximum values are much larger (24.0–40.0 km). These large maximum offsets can significantly change the shape or number of features locally, which in turn could have a major impact on predicted evolution. Figure 3 is a striking example that illustrates the importance of careful image compositing for the development of model evaluation test cases. Given an initial small amplitude meander, the tendency for most numerical models is to let the meander grow and propagate downstream (see, e.g., Glenn and Robinson 1995). Even though the average offset between the warmest pixel and the patched north walls is only a few kilometers, initialization with data from Figs. 3c or 3d would cause the wrong number of meanders to grow in the model results.
b. NORDA versus Pathfinder AVHRR
Because the NORDA AVHRR imagery has a higher resolution than the Pathfinder imagery (2.25 km vs 9 km), the NORDA dataset is preferred for the construction of precision estimates of the Gulf Stream north wall location. Figure 2, however, indicates that the NORDA dataset only contains approximately one image per day during the DAMEE GSR time period. The Pathfinder dataset contains data from both the ascending and descending passes of the satellite, that is, two images per day. Patched composites therefore were constructed from both the NORDA and the Pathfinder datasets over identical time periods surrounding each of the central days identified in Fig. 2. The actual feature analysis, that is, digitization of the Gulf Stream north wall and ring edges, was conducted using the higher-resolution NORDA dataset. The digitized locations were then overlaid on the Pathfinder composites for a consistency check. The Gulf Stream north walls determined from the NORDA data were found to be in excellent agreement with the Pathfinder composites, indicating that the once per day imagery was sufficient during this time period to locate the Gulf Stream north wall with similar accuracy. However, one major addition to the feature analyses did result from the Pathfinder comparisons. A warm core ring obscured by clouds in the available NORDA imagery is clearly visible in the more complete Pathfinder data for 13 June.
The process leading to the discovery of the additional warm ring is illustrated in Fig. 4. The Gulf Stream north wall and ring edges determined for 13 June from the NORDA, Geosat, and (A)XBT/CTD data are plotted over the NORDA composite image in Fig. 4a. The NORDA image is nearly complete west of 58°W and is partially cloudy to the west. The region of concern is the cloudy area between 40° and 42°N, 55° and 57°W. As previously discussed, each pixel in the Pathfinder dataset is given a data quality flag between 1 and 3, with 3 being the highest quality cloud-free pixels and 1 being the lowest quality, probably cloud-contaminated pixels. Figure 4b illustrates the patched composite constructed using the level 3 pixels only. Although the two cold rings visible in the NORDA imagery are observed here, the Pathfinder cloud detection algorithm (most likely the reference field test) clearly is mistaking for clouds some of the sharp fronts across the Gulf Stream north wall and the warm rings. Figure 4c was constructed using all the levels 1–3 pixels with the “cloudy” pixels removed using the same 10°C temperature threshold as in Fig. 4a before compositing. The Gulf Stream north wall, four warm rings, and two cold rings digitized from the NORDA composite are in excellent agreement with the Pathfinder data throughout, but the region east of 58°W remains partially cloudy and somewhat obscured. Finally, a third Pathfinder composite was constructed using the level 2 and 3 pixels as shown in Fig. 4d. All the features observed in the NORDA composite are visible, but the level 2 and 3 Pathfinder data clearly define a newly formed warm core ring centered near 41°N, 56°W. Failing both the Pathfinder cloud detection algorithms appears to be a better cloud removal technique than simple thresholding. The pixels responsible for identifying the additional warm core ring can be traced back to a single Pathfinder image not contained in the NORDA dataset. Comparisons of the Pathfinder level 2 and 3 composites with the other NORDA composites indicate that this is the only ring missed by the less frequent NORDA imagery. However, the discovery of even one additional ring highlights the importance of full data coverage in the more cloudy regions in the eastern third of the DAMEE GSR region.
4. Feature analyses
The Gulf Stream and ring feature analyses were conducted by overlaying the Geosat-derived dynamic topography and the (A)XBT/CTD data on the NORDA patched composites in a Geographic Information System as illustrated in Fig. 5. The Gulf Stream surface north wall observed in the satellite imagery is defined as the maximum temperature gradient (Cornillon and Watts 1987) when shingles are not present. Gulf Stream shingles often cause multiple large gradients in the vicinity of the north wall. When shingles are present, the most southern of the multiple large gradients is usually associated with the Gulf Stream north wall (Porter et al. 1996).
The edges of rings similarly are defined by a closed contour that closely follows the maximum thermal gradient and appears to best define the ring circulation (Glenn et al. 1990). Although multiple contours around rings are readily defined, Glenn et al. (1990) have shown this single contour to be sufficient to define the ring center and orientation. Ring size, of course, does depend on which contour is chosen. Previous experience indicates that the Geosat altimetric signature of rings often extends beyond the observed AVHRR signature (Lybanon et al. 1990). For the purpose of this study, the closed contour that best defines the eddy circulation is assumed to correspond to the radius of maximum velocity (Glenn and Robinson 1995). The outer edge of the ring is assumed to be a constant multiple of this distance based on the observed velocity structure of typical rings (Joyce 1984; Joyce and Kennelly 1985).
The locations of the Geosat ground tracks within ±3 days of the central day are given by the nearly straight white lines in Fig. 5. Along-track dynamic topography was calculated using the Glenn et al. (1991) geoid, which has been validated to better than 10 cm rms in the Gulf Stream region (Glenn et al. 1991; Porter et al. 1996). For analysis purposes, the along-track dynamic topographies were plotted perpendicular to each groundtrack with positive heights to the right and negative to the left. The arbitrary constant in each file was fixed by assigning the Gulf Stream maximum velocity axis a height of zero. Gulf Stream axes locations are thus readily identified in Fig. 5 as the location where the topography line crosses the groundtrack. The maximum surface velocity axis is located on average about 17 km south of the surface north wall (Porter et al. 1996), but this offset is modulated by Gulf Stream curvature. An approximately 1-m-height difference is observed across the Gulf Stream, the warm core rings appear as highs, and cold core rings appear as lows. The Geosat topographies are especially useful for locating cold core rings that have migrated away from the Gulf Stream and are no longer visible in the AVHRR imagery.
A temperature of 15°C at 200 m is one commonly used historical definition of the subsurface Gulf Stream north wall (Cornillon and Watts 1987). The subsurface north wall is typically located an average of about 9 km south of the subjectively determined, AVHRR surface north wall (Cornillon and Watts 1987), but this offset also is modulated by Gulf Stream curvature (Horton 1987; Gangopadhyay 1990). All (A)XBT/CTD profiles acquired for DAMEE GSR were plotted for removal of obvious spikes, and the temperature at 200 m was determined by linear interpolation. For clarity, the (A)XBT/CTD locations within ±3 days are displayed in Fig. 5 simply as yellow crosses for 200-m temperatures less than 15°C or as squares for 200-m temperatures greater than 15°C. In actual applications, four different temperature-keyed icons were used, and the (A)XBT/CTD profiles and cross sections were plotted and available.
A single color table was used for all the composite images in Fig. 5 to facilitate week-to-week comparisons. When the actual analyses were performed, however, the color table was varied to enhance specific features in small subsections of the data. Some features in Fig. 5 therefore were more clearly defined when they were digitized. The digitized Gulf Stream north wall and ring edges are shown as the heavy black line. In several cases, the outer edge of the rings was not visible in the AVHRR imagery, so the ring centers (black X) and radii were either estimated from the Geosat and (A)XBT/CTD data or were interpolated/extrapolated from adjacent composites.
In subsequent discussions, the DAMEE GSR domain is divided into western, central, and eastern subregions. The western subregion extends from Cape Hatteras to about the New England Seamounts (about the 67.5°W longitude line in Fig. 5). The Gulf Stream in this region is relatively benign, often characterized by small, propagating, and growing meanders. Most of the dynamical events such as large meander and ring formations occur in the central region between the seamounts and roughly the 57.5°W longitude line in Fig. 5. Both of these regions have excellent data coverage during the DAMEE GSR time period. The remaining eastern region also experiences large meanders and numerous ring formations–absorptions, but the AVHRR and (A)XBT/CTD data coverage drops off quickly.
a. 25 May (day 146)
The test period begins with three meander crests in the western region, a central region dominated by a large meander near 62°W, and several small meanders in the eastern region. Numerous (A)XBT/CTDs and Geosat tracks aid AVHRR data interpretations in the often cloudy eastern region and in the center of the domain where complicated interactions between the large meander and the rings are anticipated. While portions of all four warm rings are visible in the imagery, only one of the six cold rings (C3) is observed. The other cold ring locations are extrapolated or estimated. Because the cold rings are often obscured from the AVHRR by surface warming and the (A)XBT/CTDs are sporadic, cold ring locations are highly dependent on the Geosat data that sweeps through the region once every 17 days.
b. 30 May (day 151)
The three western meanders have propagated downstream, with the middle meander propagating the farthest as its amplitude is reduced. In the central region, the large meander crest has formed a new elliptical warm ring (W5). The small-amplitude meanders remain along the eastern portion, with a northeastward shift of the north wall east of 53°W. The westernmost cold ring (C1) is now visible in the imagery. The two Geosat tracks near 65°W are beginning a sweep across the Sargasso, their flat topographies south of the Gulf Stream confirming that no hidden cold rings are present. Geosat also helps fill the AVHRR data gap in the Gulf Stream near 59°W and aids in the identification of another cold ring (C4) near 57°W.
c. 5 June (day 157)
In the western region, the rapidly moving middle meander crest catches up to the stalled crest near 68°W, so only two crests remain in the area. The meander crest near 65°W has grown and interacts with the nearby warm ring (W2). The newly formed warm ring (W5) has become more symmetrical. The short-wavelength meanders between 55° and 57°W are growing and possibly interacting with the warm ring near 58°W; an interpretation of the patchy imagery supported by two Geosat tracks. The small meander trough near 53°W begins to deepen. AXBTs aid in the location of the Gulf Stream north wall as it exits the domain. Geosat continues its sweep across the Sargasso, indicating there are no invisible cold rings near 63° or 55°W.
d. 13 June (day 165)
A new small meander trough develops just south of the warm ring (W1) near 70°W. The large meander crest that was interacting with the warm ring (W2) near 66°W has propagated rapidly downstream and is now approaching the recently formed warm ring (W5) near 63°W. The meander crest just downstream has stalled, and the trough between these two crests is deepening rapidly. Between 55° and 57°W, the rapidly growing short-wavelength meanders that were located just north of 40°W have formed the new warm ring (W6) located in the Pathfinder dataset, leaving behind a flat stream just south of 40°W. Farther to the east, the small meander trough near 53°W continues to deepen. Meanwhile, Geosat completes its sweep across the Sargasso between 61° and 70°W, confirming that no cold rings exist within this region.
e. 22 June (day 174)
Meander amplitudes in the western region remain small, with a fourth crest propagating into the region from Cape Hatteras. The crest–trough–crest system in the central region continues to evolve as a classical cold ring formation event (Robinson et al. 1988). The downstream crest remains stalled, the upstream crest continues to propagate and close the gap, while the trough continues to deepen. In this case, however, the trough interacts and absorbs the nearby cold ring, resulting in the rapid southerly extension of the meander. Just downstream, the Geosat altimetry contains a surface elevation peak halfway between the old warm ring (W3) near 58°W and the new warm ring (W6) near 57°W, signaling a strong interaction. The numerous small amplitude meanders in the eastern region have been replaced by a crest–trough–crest meander system between 50° and 56°W that has grown significantly. The easternmost crest is interacting with the warm ring (W4) near 53°W, which is visible again for the first time since 30 May. The Geosat tracks clearly identify the cold ring (C5) near 53°W and a sixth cold ring that has propagated into the domain from the southeast near 62°W.
f. 29 June (day 181)
(A)XBT/CTD data availability declines rapidly beginning 29 June, forcing a much greater reliance on Geosat to fill gaps and resolve ambiguities in the AVHRR data. The first three small meanders in the western region propagate slowly downstream, while the fourth interacts with and sprints past the warm ring (W2) near 67°W. The gap between the two crests in the central region continues to narrow due to the propagation of the upstream crest. One of the Geosat tracks crosses directly through the center of this deep meander trough, which is the flat negative topography along that section confirming the AVHRR-derived shape. In the vicinity of the two strongly interacting warm rings observed in the previous analysis, only one ring is present with its center located between the original two, suggesting a ring merger event. The two Geosat tracks confirm that the new warm ring (W6) did not propagate eastward. Farther east, three Geosat tracks confirm the interpretation of the AVHRR imagery. The continuing interaction of the easternmost crest with the warm ring (W4) is observed.
g. 4 July (day 186)
Small propagating meanders continue to dominate the western region. The meander crest that was interacting with the warm ring near 67°W has flattened out and has been replaced by two shorter wavelength, smaller amplitude meanders. In the central region, an elongated cold ring has formed from the deep meander trough. Once again, only one warm ring is visible near 57°W, which is consistent with the ring merger hypothesis. The warm ring in the far eastern portion of the region has been absorbed.
h. Six-week composite
Figure 5h illustrates the observed evolution of the Guff Stream and its rings during the 6-week DAMEE GSR test period. From Cape Hatteras to about 64°W, the Gulf Stream formed 3–4 small propagating meanders with a narrow envelope. The meander envelope was quite wide in the central region due to the warm and cold ring formation events. Beyond this region to the east, the meander envelope again narrowed, but the meanders were no longer the simple propagating wave patterns observed in the western region. They instead took on a wider variety of shapes and underwent stronger interactions with existing rings. To quantify the approximately weekly changes in the Gulf Stream location, the average and maximum offsets of the north wall from one analysis to the next were calculated (Table 2). The average offset is defined as the area between the two north walls divided by their average length. Table 2 indicates that the largest offsets occur after ring formation–absorption events.
Six warm rings were identified in the DAMEE GSR analyses. The two western warm rings were observed to simply propagate westward over the duration of the experiment. The warm eddy that initially formed near 63°W on 30 May was forced northeastward during an interaction with a propagating meander. The warm ring near 58°W was propagating westward until an interaction with a newly formed ring just to the east resulted in a ring merger. The warm ring on the far eastern side of the domain moved eastward as it was absorbed by an upstream meander.
Six cold rings were observed. The cold ring near 73°W was observed to remain relatively stationary, as cold rings often do near Cape Hatteras. The nearby cold ring by 71°W propagated westward. A wide region between 70° and 61°W was found with no cold rings within 250 km of the Gulf Stream. The central cold ring near 60°W was absorbed and reborn during the study period. The remaining two eastern rings and ring C6 were located well outside the meander envelope and simply propagated westward with little apparent effect on the evolution of the Gulf Stream.
5. Comparisons to other analyses
The DAMEE GSR feature analyses were then compared to similar analyses of the Gulf Stream north wall performed by scientists at NOAA and the University of Rhode Island/University of Miami (URI/Miami). Both of these analyses were based on the AVHRR data only. The NOAA Oceanographic Analysis charts (Auer 1980) were constructed in real time three times per week using data available up until about noon on the day of issue. The NOAA analysts used the most recent data available to define the Gulf Stream over its entire length, with regions in which the data were more than a few days old identified. In 1988, the images were earth located by hand using a clear plastic overlay keyed to identifiable land features. The potential for large navigational errors existed, possibly resulting in Gulf Stream meanders with the proper shapes but shifted in location. For comparison to the DAMEE GSR analyses, the Gulf Stream north walls were digitized from the NOAA charts closest in time to the DAMEE GSR central days. If the closest NOAA charts were available both the day before and after the central day, the day after was chosen since its data is more representative of the data used in the DAMEE GSR analysis.
The URI/Miami analyses were constructed from 2-day warmest pixel composites of the AVHRR imagery acquired by the University of Miami. Before compositing, the 1-km data were decimated to 4-km resolution by retaining the warmest pixel in each 4-km square (Cornillon et al. 1987). Both of these processes have the potential for introducing a northward shift in the Gulf Stream north wall. The composite images were displayed on a computer and the north wall was digitized where it was visible, leaving gaps where the north wall was obscured by clouds. The cloud gaps were filled using a space–time interpolation scheme (Chin and Mariano 1997) based on contour analysis (Mariano 1990).
The Gulf Stream north walls derived from the three different datasets are compared in Fig. 6. The DAMEE GSR north walls often are located between the more northerly URI/Miami and the more southerly NOAA analyses. In the western half of the DAMEE GSR region, all three analyses usually agree on the type and number of meanders, differing mostly in subtleties of shape. In the eastern half, significant feature differences are observed. Data quality for the DAMEE GSR analyses can be judged directly from Fig. 5. As in Fig. 5, the thin lines in Fig. 6 indicate that most of the cloud gaps occur in the eastern half of the region. Larger discrepancies therefore are expected in the eastern half, simply due to the manner in which the cloud gaps are filled. However, even in regions of good data coverage, occasional discrepancies can be quite large. Examples include the northward shift of the URI/Miami analysis in the eastern region on 25 May, the apparent navigational errors in the 13 June NOAA north wall, and the attached warm rings in the URI/Miami north walls for 30 May, 22 June, and 29 June. The attached warm rings may be the direct result of the decimation and warmest pixel compositing applied to the original data. Both processes can potentially obscure a small separation zone between a warm core ring and a nearby meander.
To quantify the differences from the DAMEE GSR analyses, the average and maximum offsets of the URI/Miami and NOAA north walls were calculated (Table 3). As before, the average offset is defined as the area between the two north walls divided by their average length. The maximum offset was subjectively determined by measuring the approximately perpendicular distance between the two north walls. The average offsets between the DAMEE GSR and the URI/Miami analyses are remarkably similar to the evolutionary offsets presented in Table 2, with nearly identical means. The means of the NOAA average and maximum offsets are slightly smaller.
6. Summary and conclusions
The purpose of this study was to develop a series of high quality test cases for the evaluation of Gulf Stream forecast models. The model initialization and verification procedure relies heavily on accurate snapshots of the Gulf Stream north wall and ring locations throughout the evaluation region on an approximately weekly basis. To provide the best possible data coverage, the evaluation time period was chosen from the data-rich late 1980s due to the availability of AVHRR imagery, Geosat altimetry, and numerous in situ observation programs. The actual 6-week time period chosen was 25 May–4 July 1998 because it contained the most complete and prolonged AVHRR coverage during the entire Geosat Exact Repeat Mission.
Good AVHRR coverage is essential for defining synoptic snapshots of the Gulf Stream over a large area such as the DAMEE GSR region. Similar coverage cannot be realistically acquired by altimeters or (A)XBT/CTDs. Because of the importance of the AVHRR imagery, a new “patching” method for compositing images was developed to fill cloud gaps. Patching was demonstrated to be superior to standard warmest pixel composites because it reduces the spatial smearing of Gulf Stream features. The NORDA AVHRR imagery, Geosat altimetry, and the (A)XBTS/CTDs were then combined in a Geographic Information System that enabled easy construction of data overlays for precise determination of the Gulf Stream north wall and ring locations. Digitized Gulf Stream north wall and ring edge locations were in excellent agreement with the newly available Pathfinder dataset, but more importantly, the twice per day Pathfinder imagery providing vital data on a newly formed but short-lived warm core ring that was not observed in the once per day NORDA imagery.
The approximately weekly evolution of the Gulf Stream and rings was documented with seven snapshots during the DAMEE GSR evaluation period. The western third of the DAMEE GSR domain between Cape Hatteras and the New England Seamounts was dominated by small propagating meanders and only weak interactions with nearby warm rings. In the central region between the seamounts and about 57.5°W, the largest amplitude meanders were observed. Major events included the formation of a new warm ring, followed by the rapid deepening of a meander trough, its interaction with a nearby cold ring, and the eventual breaking off of a new, larger cold ring. The eastern third of the domain initially consisted of numerous small amplitude meanders, but after a warm ring formation event in the middle of the study period, the Gulf Stream configuration was dominated by the growth of a single large crest–trough–crest meander system. The mean weekly change in the streamwise average location of the Gulf Stream north wall was about 27 km. A total of three ring formation events (two warm, one cold), two ring absorption events (one each), and one ring merger were observed. Rings not interacting with the Gulf Stream or other rings generally drifted to the west over the study period.
Comparisons with two other Gulf Stream north wall analyses prepared solely from AVHRR data from the same time period indicate that there were significant differences in the perceived location of the north wall. The mean of the weekly average offsets between the DAMEE GSR results and the two AVHRR analyses are about 28 and 23 km. These values are quite similar in magnitude to the observed weekly evolution. Some of the differences between the north wall analyses may be attributable to the different procedures applied to cloudy regions. There is generally good agreement in the western half of the DAMEE GSR region, but the differences are much greater in the more data-poor eastern half. The large differences that occur in data-rich areas, however, may be associated more with the subjective interpretations by experienced analysts of the sometimes ambiguous AVHRR data. The best solution to resolve these ambiguities in the critical AVHRR data appears to be the careful construction of the image composites and the use of auxiliary datasets to aid in their interpretation as was done for DAMEE GSR.
Acknowledgments
This work was supported by the Institute for Naval Oceanography and the Office of Naval Research. Datasets used in the analysis were provided by the Remote Sensing and the Ocean Modeling Branches of NORDA, The Johns Hopkins University/Applied Physics Laboratory, Harvard University, the University of Rhode Island, Woods Hole Oceanographic Institution, the Naval Oceanographic Office, and NASA’s Jet Propulsion Laboratory. The URI/Miami Gulf Stream north walls were provided by Peter Cornillon of URI (observed) and Arthur Mariano (interpolated) of the University of Miami.
REFERENCES
Auer, S. J., 1980: New daily oceanographic analyses. Gulfstream,6, 3–7.
Brown, J. W., O. B. Brown, and R. H. Evans, 1993: Calibration of AVHRR infrared channels: A new approach to non-linear correction. J. Geophys. Res.,98, 18 257–18 268.
Chin, T. M., and A. J. Mariano, 1997: Space–time interpolation of oceanic fronts. IEEE Trans. Geosci. Remote Sens.,35, 734–746.
Clancy, M. P., P. A. Phoebus, and K. D. Pollack, 1990: Operational global scale ocean thermal analysis system. J. Atmos. Oceanic Technol.,7, 233–254.
Cornillon, P., and D. R. Watts, 1987: Satellite thermal infrared and inverted echo sounder determinations of the Gulf Stream northern edge. J. Atmos. Oceanic Technol.,4, 712–723.
——, C. Gillman, L. Stramma, O. Brown, R. Evans, and J. Brown, 1987: Processing and analysis of large volumes of satellite-derived thermal infrared data. J. Geophys. Res.,92, 12 993–13 002.
Gangopadhyay, A., 1990: Wind forcing of the Gulf Stream: A space-time analysis. Ph.D. dissertation, University of Rhode Island, 175 pp. [Available from Graduate School of Oceanography, University of Rhode Island, Narrangansett, RI 02882.].
Glenn, S. M., and C. C. Ebbesmeyer, 1994: The structure and propagation of a frontal eddy cold dome along the North Carolina shelf. J. Geophys. Res.,99, 5029–5046.
——, and A. R. Robinson, 1995: Validation of a Gulf Stream forecasting model. Qualitative Skill Assessment for Coastal Ocean Models, D. Lynch and A. Davies, Eds., American Geophysical Union, 469–499.
——, G. Z. Forristall, P. Cornillon, and G. Milkowski, 1990: Observations of Gulf Stream ring 83-E and their interpretation using feature models. J. Geophys. Res.,95, 13 043–13 063.
——, D. L. Porter, and A. R. Robinson, 1991: Synthetic geoid validation of GEOSAT mesoscale dynamic topography in the Gulf Stream region. J. Geophys. Res.,96, 7145–7166.
Horton, C. W., 1987: Modulation of Gulf Stream surface-subsurface frontal separation by path curvature. J. Phys. Oceanogr.,17, 596–603.
——, D. L. Porter, R. W. deWitt, and W. E. Rankin, 1992: Airborne expendable bathythermograph survey of the Kuroshio Extension and comparison with simultaneous altimeter measurements during the GEOSAT exact repeat mission. J. Geophys. Res.,97, 7447–7463.
Joyce, T. M., 1984: Velocity and hydrographic structure of a Gulf Stream warm core ring. J. Phys. Oceanogr.,14, 936–947.
——, and M. A. Kennelly, 1985: Upper-ocean velocity structure of Gulf Stream warm core ring82B. J. Geophys. Res.,90, 8839–8841.
Lee, T., 1994: Variability of the Gulf Stream path observed from satellite infrared images. Ph.D. dissertation, University of Rhode Island, 188 pp. [Available from Graduate School of Oceanography, University of Rhode Island, Narragansett, RI 02882.].
Lybanon, M., R. L. Crout, C. H. Johnson, and P. Pistek, 1990: Operational altimeter-derived oceanographic information: The NORDA GEOSAT ocean applications program. J. Atmos. Oceanic Technol.,7, 357–376.
Mariano, A. J., 1990: Contour analysis: A new approach for melding geophysical fields. J. Atmos. Oceanic Technol.,7, 285–295.
Mitchell, J. L., J. M. Dastugue, W. J. Teague, and Z. R. Hallack, 1990: The estimation of geoid profiles in the northwest Atlantic from simultaneous satellite imagery and airborne expendable bathythermograph sections. J. Geophys. Res.,95, 17 965–17 977.
PO-DAAC, 1994: NOAA/NASA Pathfinder AVHRR Oceans-Sea Surface Temperature Data Set users guide. Jet Propulsion Laboratory, 23 pp. [Available from Physical Oceanography, Distributed Active Archive Center, Jet Propulsion Laboratory, 4800 Oak Grove Drive, Pasadena, CA 91109.].
Porter, D. L., S. M. Glenn, E. B. Dobson, and M. F. Crowley, 1996: Extension and validation of a Gulf Stream GEOSAT synthetic geoid. J. Atmos. Oceanic Technol.,13, 514–531.
Robinson, A. R., M. A. Spall, and N. Pinardi, 1988: Gulf Stream simulations and the dynamics of ring and meander processes. J. Phys. Oceanogr.,18, 1811–1853.
Watts, R., 1991: A synoptic view of the Gulf Stream. Maritimes,35, 3–6.
Willems, R. C., and Coauthors, 1994: Experiment evaluates ocean models and data assimilation in the Gulf Stream. Eos, Trans. Amer. Geophys. Union,75, 385–394.
Percent AVHRR coverage of the Gulf Stream derived from the NOAA oceanographic analysis charts during the Geosat Exact Repeat Mission. The short dashed lines indicate the time period chosen for DAMEE GSR.
Citation: Journal of Atmospheric and Oceanic Technology 14, 6; 10.1175/1520-0426(1997)014<1366:GSARFA>2.0.CO;2
Percent AVHRR coverage of the Gulf Stream derived from the NOAA oceanographic analysis charts during the Geosat Exact Repeat Mission. The short dashed lines indicate the time period chosen for DAMEE GSR.
Citation: Journal of Atmospheric and Oceanic Technology 14, 6; 10.1175/1520-0426(1997)014<1366:GSARFA>2.0.CO;2
Percent AVHRR coverage of the Gulf Stream derived from the NOAA oceanographic analysis charts during the Geosat Exact Repeat Mission. The short dashed lines indicate the time period chosen for DAMEE GSR.
Citation: Journal of Atmospheric and Oceanic Technology 14, 6; 10.1175/1520-0426(1997)014<1366:GSARFA>2.0.CO;2
AVHRR, Geosat, and (A)XBT/CTD data availability for the DAMEE GSR time period. Crosses indicate times of AVHRR images or ascending Geosat tracks; X’s indicate times of descending Geosat tracks. Short dashed lines indicate the central days chosen for the DAMEE GSR analyses.
Citation: Journal of Atmospheric and Oceanic Technology 14, 6; 10.1175/1520-0426(1997)014<1366:GSARFA>2.0.CO;2
AVHRR, Geosat, and (A)XBT/CTD data availability for the DAMEE GSR time period. Crosses indicate times of AVHRR images or ascending Geosat tracks; X’s indicate times of descending Geosat tracks. Short dashed lines indicate the central days chosen for the DAMEE GSR analyses.
Citation: Journal of Atmospheric and Oceanic Technology 14, 6; 10.1175/1520-0426(1997)014<1366:GSARFA>2.0.CO;2
AVHRR, Geosat, and (A)XBT/CTD data availability for the DAMEE GSR time period. Crosses indicate times of AVHRR images or ascending Geosat tracks; X’s indicate times of descending Geosat tracks. Short dashed lines indicate the central days chosen for the DAMEE GSR analyses.
Citation: Journal of Atmospheric and Oceanic Technology 14, 6; 10.1175/1520-0426(1997)014<1366:GSARFA>2.0.CO;2
Comparison of patching vs warmest pixel composites for 4 July 1988: (a) Single NORDA AVHRR image for 4 July and the digitized north wall (white); (b) patched composite for 2–6 July and the north wall from (a) (white); (c) warmest pixel composite for 3–5 July the resulting north wall (black) and the north wall from (a) (white); (d) warmest pixel composite for 2–6 July, the resulting north wall (black), and the north wall from part a (white).
Citation: Journal of Atmospheric and Oceanic Technology 14, 6; 10.1175/1520-0426(1997)014<1366:GSARFA>2.0.CO;2
Comparison of patching vs warmest pixel composites for 4 July 1988: (a) Single NORDA AVHRR image for 4 July and the digitized north wall (white); (b) patched composite for 2–6 July and the north wall from (a) (white); (c) warmest pixel composite for 3–5 July the resulting north wall (black) and the north wall from (a) (white); (d) warmest pixel composite for 2–6 July, the resulting north wall (black), and the north wall from part a (white).
Citation: Journal of Atmospheric and Oceanic Technology 14, 6; 10.1175/1520-0426(1997)014<1366:GSARFA>2.0.CO;2
Comparison of patching vs warmest pixel composites for 4 July 1988: (a) Single NORDA AVHRR image for 4 July and the digitized north wall (white); (b) patched composite for 2–6 July and the north wall from (a) (white); (c) warmest pixel composite for 3–5 July the resulting north wall (black) and the north wall from (a) (white); (d) warmest pixel composite for 2–6 July, the resulting north wall (black), and the north wall from part a (white).
Citation: Journal of Atmospheric and Oceanic Technology 14, 6; 10.1175/1520-0426(1997)014<1366:GSARFA>2.0.CO;2
Comparison of patched composites for 13 June (±2 days) obtained from (a) NORDA imagery; (b) Pathfinder imagery with data quality flag 3 (best data only); (c) Pathfinder imagery with data quality flags 1, 2, and 3 (all data); (d) Pathfinder imagery with data quality flags 2 and 3 only. The Gulf Stream north wall and ring edges obtained from the NORDA imagery are drawn as solid black. The additional warm ring located only in the Pathfinder imagery is highlighted in white (d).
Citation: Journal of Atmospheric and Oceanic Technology 14, 6; 10.1175/1520-0426(1997)014<1366:GSARFA>2.0.CO;2
Comparison of patched composites for 13 June (±2 days) obtained from (a) NORDA imagery; (b) Pathfinder imagery with data quality flag 3 (best data only); (c) Pathfinder imagery with data quality flags 1, 2, and 3 (all data); (d) Pathfinder imagery with data quality flags 2 and 3 only. The Gulf Stream north wall and ring edges obtained from the NORDA imagery are drawn as solid black. The additional warm ring located only in the Pathfinder imagery is highlighted in white (d).
Citation: Journal of Atmospheric and Oceanic Technology 14, 6; 10.1175/1520-0426(1997)014<1366:GSARFA>2.0.CO;2
Comparison of patched composites for 13 June (±2 days) obtained from (a) NORDA imagery; (b) Pathfinder imagery with data quality flag 3 (best data only); (c) Pathfinder imagery with data quality flags 1, 2, and 3 (all data); (d) Pathfinder imagery with data quality flags 2 and 3 only. The Gulf Stream north wall and ring edges obtained from the NORDA imagery are drawn as solid black. The additional warm ring located only in the Pathfinder imagery is highlighted in white (d).
Citation: Journal of Atmospheric and Oceanic Technology 14, 6; 10.1175/1520-0426(1997)014<1366:GSARFA>2.0.CO;2
Composite patched AVHRR image, Geosat-derived dynamic topographies, and (A)XBT/CTD locations for the DAMEE GSR analysis days. The Gulf Stream north wall and observed ring edges are shown in black. Black X’s mark the observed or interpolated ring centers. Geosat groundtracks and the dynamic topography (plotted perpendicular to the straight groundtrack with positive to the right) are shown in white highlighted by black. (A) XBT/CTD locations are shown in yellow, with temperatures greater than 15°C at 200 m indicated by a square, and temperatures less than 15°C at 200 m indicated by a cross. The 200-m isobath is shown as the solid white line. The additional warm ring identified in the Pathfinder dataset is highlighted in white.
Citation: Journal of Atmospheric and Oceanic Technology 14, 6; 10.1175/1520-0426(1997)014<1366:GSARFA>2.0.CO;2
Composite patched AVHRR image, Geosat-derived dynamic topographies, and (A)XBT/CTD locations for the DAMEE GSR analysis days. The Gulf Stream north wall and observed ring edges are shown in black. Black X’s mark the observed or interpolated ring centers. Geosat groundtracks and the dynamic topography (plotted perpendicular to the straight groundtrack with positive to the right) are shown in white highlighted by black. (A) XBT/CTD locations are shown in yellow, with temperatures greater than 15°C at 200 m indicated by a square, and temperatures less than 15°C at 200 m indicated by a cross. The 200-m isobath is shown as the solid white line. The additional warm ring identified in the Pathfinder dataset is highlighted in white.
Citation: Journal of Atmospheric and Oceanic Technology 14, 6; 10.1175/1520-0426(1997)014<1366:GSARFA>2.0.CO;2
Composite patched AVHRR image, Geosat-derived dynamic topographies, and (A)XBT/CTD locations for the DAMEE GSR analysis days. The Gulf Stream north wall and observed ring edges are shown in black. Black X’s mark the observed or interpolated ring centers. Geosat groundtracks and the dynamic topography (plotted perpendicular to the straight groundtrack with positive to the right) are shown in white highlighted by black. (A) XBT/CTD locations are shown in yellow, with temperatures greater than 15°C at 200 m indicated by a square, and temperatures less than 15°C at 200 m indicated by a cross. The 200-m isobath is shown as the solid white line. The additional warm ring identified in the Pathfinder dataset is highlighted in white.
Citation: Journal of Atmospheric and Oceanic Technology 14, 6; 10.1175/1520-0426(1997)014<1366:GSARFA>2.0.CO;2
(Continued)
Citation: Journal of Atmospheric and Oceanic Technology 14, 6; 10.1175/1520-0426(1997)014<1366:GSARFA>2.0.CO;2
(Continued)
Citation: Journal of Atmospheric and Oceanic Technology 14, 6; 10.1175/1520-0426(1997)014<1366:GSARFA>2.0.CO;2
(Continued)
Citation: Journal of Atmospheric and Oceanic Technology 14, 6; 10.1175/1520-0426(1997)014<1366:GSARFA>2.0.CO;2
Comparison of Gulf Stream north wall locations derived by DAMEE GSR (solid line), URI/Miami (long dashed line), and NOAA (short dashed line). For URI/Miami, thick lines indicate observed fronts, and thin lines indicate space–time interpolated fronts. For NOAA, thick lines indicate fronts observed within the last 2–3 days, and thin lines represent fronts observed more than 2–3 days ago.
Citation: Journal of Atmospheric and Oceanic Technology 14, 6; 10.1175/1520-0426(1997)014<1366:GSARFA>2.0.CO;2
Comparison of Gulf Stream north wall locations derived by DAMEE GSR (solid line), URI/Miami (long dashed line), and NOAA (short dashed line). For URI/Miami, thick lines indicate observed fronts, and thin lines indicate space–time interpolated fronts. For NOAA, thick lines indicate fronts observed within the last 2–3 days, and thin lines represent fronts observed more than 2–3 days ago.
Citation: Journal of Atmospheric and Oceanic Technology 14, 6; 10.1175/1520-0426(1997)014<1366:GSARFA>2.0.CO;2
Comparison of Gulf Stream north wall locations derived by DAMEE GSR (solid line), URI/Miami (long dashed line), and NOAA (short dashed line). For URI/Miami, thick lines indicate observed fronts, and thin lines indicate space–time interpolated fronts. For NOAA, thick lines indicate fronts observed within the last 2–3 days, and thin lines represent fronts observed more than 2–3 days ago.
Citation: Journal of Atmospheric and Oceanic Technology 14, 6; 10.1175/1520-0426(1997)014<1366:GSARFA>2.0.CO;2
Average and maximum offsets of the Gulf Stream north wall determined from patched and warmest pixel composites.
Average offsets between subsequent Gulf Stream north walls determined from the DAMEE GSR analyses.
Average and maximum offsets of different Gulf Stream north wall analyses.
