Sharon M. Mesick, Martin H. Booda, and Barbara A. Gibson

AUTOMATED DETECTION OF OCEANIC FRONTS AND EDDIES FROM REMOTELY SENSED SATELLITE DATA USING ArcInfo GRID PROCESSING



Abstract

Oceanic fronts can be defined as ocean areas where horizontal gradients of various measurable parameters exceed particular thresholds. Ocean frontal boundaries are located throughout the world's oceans, and represent areas of high interest for Naval Operations and for commercial application. The Naval Oceanographic Office obtained seasonally distributed, remotely sensed Multi Channel Sea-Surface Temperature data for a large study area. The data was processed through the ArcInfo Grid software using basic slope functions, and then favorably compared with manually digitized frontal boundary data. This paper presents the details of the study methodology and the results of the proof of concept.

Background

Ocean Fronts can be defined as areas where horizontal gradients of various measurable parameters exceed particular thresholds. Fronts are located throughout the world's oceans, and define boundaries between different ocean masses which exhibit notable differences in temperature and other properties. Ocean Fronts represent areas of high interest to scientists of many disciplines and to commercial interests because they represent regions of: strong anomalies in the ocean; high biological activity; dynamic chemical processes; change in acoustic propagation. Figure 1 shows the mean locations of the various frontal boundaries as compiled by Department of Defense Major Shared Resource Center (MSRC) personnel.

Figure 1. Global Major Ocean Currents
Figure 1. Global Major Ocean Currents


NAVOCEANO generates products relating to oceanographic fronts which are provided to the Naval Fleet and to commercial enterprise via hardcopy and the World Wide Web. This data also plays an important role in numerical ocean modeling, where the goal is to predict the future locations of ocean fronts and associated features. Data preparation and creation of these products involves expert techniques which are largely manual, labor-intensive, subjective and skill-dependent; therefore automation was seen as beneficial.

This study was initiated to determine the feasibility of employing Commercial-Off-The-Shelf software (COTS) to automate some of the data preparation steps, while continuing to meet established data quality control criteria. Established data processing protocols using ArcInfo Grid functions were shown to work well for subsurface fronts in northern latitudes where data are known to be fairly constant and straightforward, such as the Greenland-Norwegian Sea.

This study sought to prove the concept that this same semi-automated processing protocol could be successfully applied to surface fronts in more complex ocean regions. The increased demand for frontal products in the western North Atlantic Ocean mandated the choice of this ocean region for the study area.

Study Area

The general circulation of the ocean is well known to all mariners. Primarily influenced by the constant equatorial trade winds and the mid-latitude westerlies, ocean currents flow in predictable paths. Seasonal variations in speed, water temperature, and other parameters are exhibited and measured.

The Gulf Stream is perhaps the most renowned ocean current, and the most studied throughout history. Its location and speed were used to advantage by whaling captains in the 18th and 19th centuries. This practice so impressed Benjamin Franklin, then Postmaster of America, that he published a chart for captains sailing aboard mail packets between the New World and England. As study methodologies advanced throughout the 20th century the Gulf Stream has continued to be a study site because of both its proximity to the United States and because of the extensive historical data which exists for this region.

The Gulf Stream is a major component of ocean circulation, involving large-scale transport of tropical water and heat over great distances. Areas of closed circulation near the Gulf Stream, called rings or eddies, are a vital part of the Gulf Stream system. The Gulf Stream is 100 km wide, with surface current velocity as high as 2 m/s. Rings are up to 300 km in diameter, with swirl velocities comparable to Stream velocities. The daily position of the Gulf Stream may change as much as 30 km, while rings move 4-5 km/day on the average. The Gulf Stream and its associated eddies are examples of mesoscale (50-300 km in size) ocean features.

The two most dominant thermal features within this ocean region are the Slope Front, bounding the waters of the Continental Shelf from Nantucket to Norfolk, and the Gulf Stream North wall (GSNW), which separates the Slope waters from the Gulf Stream system. Another front occurs at the southern edge of the Gulf Stream, which separates the Gulf Stream from the Sargasso Sea. This front was not considered in this study because the thermal are weak and surface location is usually undetectable.

In general, thermal gradients across oceanic fronts are determined by the temperatures of the water masses which the fronts separate. The GSNW separates the Gulf Stream from Slope waters with a surface temperature difference ranging between 3-7'C between the two water masses. Based on many historical observations, the sea surface temperature difference is known to be greater in winter than in summer.

Oceanic seasons in this region lag behind atmospheric seasons, and transitional periods (spring and autumn) are shorter in the ocean. The oceanic seasons used in this study are defined as follows: Winter: (January - April); Spring: (May - June); Summer: (July - September); Autumn: (October - December). Within the constraints of available data sets archived at NAVOCEANO, four dates were selected for independent study of processing performance in accordance with these seasons as follows: January 9 (JD97009); April 9 (JD97099); August 22 (JD96234); November 20 (JD96324). Figure 2 graphically identifies the study area.

Figure2. Study Area
Figure2. Study Area

Methodology

Manual methodology: At NAVOCEANO, Human Analysts use Global Imaging software to view AVHRR data at 1.1 km resolution. The data may be either Local Area Coverage (LAC) or High Resolution Picture (HRP) transmission data from AVHRR channels 4 and 5. If a region is poorly covered, data may be supplemented with GAC AVHRR (4.4 km). AVHRR channel 1 data is used to visually explore the amount and extent of cloud cover. If cloud cover obscures an area of interest, the most recent "old" data is used to supplement the new data to maintain product continuity. Finally, a confidence value is assigned to the completed data set, which informs Users about the age of any supplemented data.

By adjusting the contrast and brightness of the imagery, analysts visually detect and manually digitize ocean fronts. Every ocean area is digitized in this manner every day. For this study, digitized data files for the study area were extracted from archived data sets in compliance with the seasonal variations previously noted. The data sets were reformatted into Arc Generate format, and ArcInfo line coverages were then created for each of the four days. These line coverages provided the baseline data for the proof of concept.

Semi-automated Methodology: Semi-automated data processing protocols developed at NAVOCEANO to process data from stable ocean regions resulted in the generation of digital data files representing the fronts and eddies for these regions. Quality control by NAVOCEANO analysts indicated the success of this methodology. In general terms, these protocols involve the automated extraction of subsurface temperature data from the DOD MSRC Supercomputer for the defined region, processing of this data through AMLs employing ArcInfo Grid functions, and the generation of digital products from the resultant data.

Study methodology: Comparative data for this study was created by applying the subsurface data processing protocols to Multi Channel Sea Surface Temperature (MCSST) data. MCSST is created from Global Area Coverage(GAC) AVHRR data obtained from the NOAH-14 satellite. The data has a resolution of 4.4 km; during processing a 2 x 2 array is used to generate an MCSST grid at 8.8 km resolution. The process uses a non linear sea surface temperature algorithm[1]. There is a daytime algorithm which uses AVHRR channels 4 and 5, and there is a nighttime algorithm that uses channels 3, 4, and 5. Cloud screening of the data is a major step in the processing. This is done with several tests using the infrared AVHRR channels and the High Resolution Infrared Radiation Sounder (HIRS) data. HIRS data is collected at the same time as the AVHRR data on the NOAA satellites. Quality control of the MCSST data is done by matching SSTs with fixed and drifting buoy data. A global 10 km gridded database is updated daily with new SST's. In certain ocean regions, LAC AVHRR at 1.1 km resolution, resampled to 2 km resolution, is used to create the MCSST data.

In discussing the transition of these protocols to more active ocean regions, it was understood from the outset that regionally and seasonally dependent parameters would be employed in the existing algorithms. For the proof of concept, a template was prepared for the North Atlantic Ocean Region which included the following data sets:


Each temperature grid was then processed according to the following logic:

A temporary grid was created based on the inverse-distance weighting of the temperatures at depth (zero for surface depth) contained in the projected point cover of all temperatures and bathymetry values, over a 44448-meter radius and gridded to 22224-meter cells; A temporary grid of gradient values was calculated from the above idw grid; A line cover was created (gridline) based on the linear features found in the analysis of the parts of the gradient grid which exceed .0025 'C per meter.

%gridname%_sg = slope((project ( %gridname%, geotoazi.prj, NEAREST, 22224)))
%gridname%_gg = %gridname%_sg > .0025 & tempbathy > %depth%
%gridname%_laz = gridline(%gridname%_gg, POSITIVE , THIN, NOFILTER, ROUND, front)

After ArcInfo Grid data processing protocols were executed on the MCSST data, the resulting fronts and eddies were compared to the manually digitized data sets for the same region and seasonally-dependent time periods(baseline data). Figure 3 presents a flow diagram summarizing this methodology.

Figure 3. Processing Flow Diagram
Figure 3. Processing Flow Diagram

Preliminary Results

The GSNW is defined in several different ways, depending upon the data type and the application. When derived from Advanced Very High Resolution Radiometer (AVHRR) imagery data, the front is defined as the maximum temperature gradient on the north side of the Gulf Stream. The semi-automated processing protocols applied to this problem yielded promising results which were seasonally dependent. Figures 4-7 display the results of the study.

P2524.JPG
Figure 4.

The fronts generated in Figure 4 using the ARC algorithms produced minimal correlation with the digitized fronts. The generated fronts did not appear to follow the pattern produced by the gradient grid, but the digitized fronts did not follow the temperature gradients as described by the MCSST data either. The highest area of correlation between the generated and digitized fronts was in the northeastern, cooler waters of the study area.

P2525.JPG
Figure 5.

The fronts generated in Figure 5 using the ARC algorithms produced a better correlation with the digitized fronts. The generated fronts appeared to more effectively follow the gradient grid. There were several areas of high correlation between the generated and digitized fronts, found in the northeastern, cooler waters and the western coastal region of the study area.

P2526.JPG
Figure 6.

The fronts generated in Figure 6 using the ARC algorithms produced a very good correlation with the digitized fronts. The generated fronts appeared to more effectively follow the gradient grid for the length of the frontal boundary. It is hypothesized that this is due to the dominance of cooler water further south, creating a sharper gradiant between the Gulf Stream and the warmer water throughout the study area.

P2527.JPG
Figure 7.

The fronts generated in Figure 7 using the ARC algorithms produced the best correlation with the digitized fronts. The generated fronts appeared to more effectively follow the gradient grid for the length of the frontal boundary, which now extends throughout the entire study area. As cooler waters continue to expand further south, a pronounced gradient was still present. This again produced a sharp contrast in water temperature in the Gulf Stream which the ARC algorthim was able to distinguish.

Conclusion

In historical studies it has been shown that a strong horizontal temperature gradient is formed by the North Wall during the winter months at approximately 36'55'N. Because of this strong gradient, both the GSNW and the width of the Gulf Stream can be calculated in winter months from temperature data. However, the same studies have shown that during the summer months the horizontal temperature gradients are weak for all water mass boundaries, although the dominant feature is the strong subsurface horizontal temperature gradient at 37'40'N.

Because of the strong temperature gradient which occurs in this region at a depth of 200m, it has been common practice to identify the North Wall position by the 15'C isotherm at 200m. However, sea surface temperature data are more readily available than subsurface data. In the absence of subsurface temperature data, [6] has shown that the position of the North Wall can be estimated from either the surface front or the maximum surface temperature. Maximum surface temperature is a better indicator of the North Wall position than is the surface front (area of maximum horizontal thermal gradient).

When NAVOCEANO analysts compared the semi-automated processing using ArcInfo GRID functions to the manually digitized baseline data sets, the results of historical studies were confirmed. That is, it was found that during the winter months the compliance between the data sets was very high, while during the summer months the software did not distinguish the frontal characteristics as effectively as was desired.

Analysts were encouraged by the confirmation of historical data findings, and resolved to resume prototyping the system with higher resolution MCSST data. It was suggested that an expert system might be developed to compute the required regional and seasonal parameter modifications, resulting in the generation of a frontal boundary coverage for each ocean region. Theoretically, this system would be designed to self-initiate operation for a given region, dependent only upon the arrival of the appropriate satellite data. Such a system would enable the analysts to apply human, subjective skills to data quality control issues, relieving them of the tedious practice of daily visual analysis and digitizing. As higher resolution MCSST data becomes available, the proof of concept will again be tested. Moreover, other parameters of the comparison will be considered in the evaluation process. This will include removing portions of the digitized fronts which have low confidence values from the analysis as they are areas where the fronts are supplimented with older digitized data and thus may not reflect the current frontal boundary.

Acknowledgements

The authors wish to thank the following individuals:

Ludwig A. Goon, Chad Dyle and Michael Adams of the Major Shared Resource Center (MSRC), for use of the graphic in Figure 1.
Bruce Mckenzie, Karrie Shants and Michael Brooking of the NAVOCEANO Warfighting Support Center, for their technical support.
The Staff of the Maury Oceanographic Library, Stennis Space Center, MS.

    References

  1. Algorithm Research Panel for Sea Surface Temperature. Naval Research Lab, Remote Sensing Applications Branch, 1998. Point of Contact: Douglas A. May (228-688-4845). World Wide Web page location: http://web7240.nrlssc.navy.mil/arpsst.htm
  2. Cummings, James A. 1994. "Global and Regional Ocean Thermal Analysis Systems at Fleet Numerical Meteorology and Oceanography Center." IEEE Transactions on Geoscience and Remote Sensing, 32(6).
  3. Hawkins, Jeffrey D. 1985. Multichannel Sea Surface Temperature Retrieval for Navy Utilization. Ocean Sensing and Prediction Division, Naval Ocean Research and Development Activity, Technical Note 312.
  4. Holyer, Ronald and Jeffrey D. Hawkins. 1983. Comparison of Multichannel and Two-Satellite Methods for Remote Measurement of Sea Surface Temperature. Oceanography Division, Naval Ocean Research and Development Activity, Technical Note 162.
  5. Khedouri, E., W. Gemmill and M.Shank. 1976. Statistical Summary of Ocean Fronts and Water Masses in the Western North Atlantic. Naval Oceanographic Office, #NOO RP-9.
  6. Suzanne M. and Matthew Lybanon. 1983. "Automated Boundary Delineation in Infrared Ocean Images." IEEE Transaction on Geoscience and Remote Sensing, 31(6).
  7. Lybanon, Matthew. 1996. "Maltese Front variability from Satellite Observations Based on Automated Detection." IEEE Transactions on Geoscience and Remote Sensing, 34(5).
  8. May, Douglas A. 1993. Global and Regional Comparative Performance of Linear and Nonlinear Satellite Multichannel Sea Surface Temperature Algorithms. Remote Sensing Division, Naval Research Laboratory, NRL/MR/7240--93-7049.
  9. Passi, Ranjit M. and Harsh Anand. 1994. Objective Feature Identification and Tracking: A Review. Mississippi State Center for Air and Sea Technology, Technical Report # 94-4.

Sharon M. Mesick
Computer Scientist
Planning Systems Incorporated
MSAAP, Bldg 9121
Stennis Space Center, Mississippi 39529
Telephone: (228) 689-8753
Email: smesick@nrlssc.navy.mil

Martin H. Booda
Oceanographer
Naval Oceanographic Office
1002 Balch Blvd., Code N212
Stennis Space Center, Mississippi 39522-5001
Telephone: (228) 688-5309
Email: booda@navo.hpc.mil

Barbara A. Gibson
Graduate Research Assistant
Environmental Analysis and Verification Center (EVAC)
University of Oklahoma
710 Asp Avenue, Suite 8
Norman, OK 73069
Telephone: (405) 447-8412
Email: bgibson@ou.edu