432 



Fishery Bulletin 89(3). 1991 



The National Oceanic and Atmospheric Administra- 

 tion (NOAA) satellites record image data in four or five 

 bands: two in the reflected solar region of the spec- 

 trum and two or three in the emitted thermal region. 

 The spatial resolution of the sensors is 1.1km at nadir, 

 increasing to about 5.5km at the edges of the 2580km- 

 wide scan. The AVHRR data were processed into sea- 

 surface temperature (SST) images using a TeraScan* 

 computer system operated by the School of GeoSci- 

 ences at Louisiana State University. The TeraScan 

 system consists of a computer, color image display 

 device, and other custom hardware and software 

 designed to process AVHRR digital imagery. The 

 images were digitally cut to fit a master image that 

 encompassed the entire Gulf of Mexico (latitude 

 17°45.30'N to 30°44.70'N and longitude 81°0.33'W to 

 97°59.67'W). 



Daytime images were calibrated and converted to 

 SST using the multichannel SST algorithm (MCSST) 

 described by McClain et al. (1985). Radiometric noise 

 in channel 3 caused some difficulties in deriving SST 

 from nighttime images. Spatial filtering techniques 

 (Schowengerdt 1983) performed on the channel 3 

 image prior to computing the SST had little or no effect 

 on the noise and often resulted in significant degrada- 

 tion of the information content in the completed image. 

 Although image restoration techniques such as filter- 

 ing in the frequency domain appear to have been suc- 

 cessful in minimizing noise in channel 3 (Warren 1989), 

 they were not a practical consideration for this project 

 because of the large number of images to be processed 

 and computer software limitations. Thus, channel 3 

 data were removed from each nighttime image file 

 prior to processing data into SST. Since the nighttime 

 processing technique was an untested modification of 

 the MCSST algorithm, a linear regression analysis was 

 used to compare the satellite-derived SST data with 

 in situ Gulf of Mexico SST data obtained from NOAA 

 weather buoys. Residuals were plotted by the date and 

 time of image acquisition, satellite number, and buoy 

 locations to look for potential bias in SST values that 

 may have been related to the processing technique. 



Since atmospheric effects can significantly reduce the 

 reliability of satellite-derived SST measurements, par- 

 ticularly as the viewing angle increases from nadir, a 

 threshold was defined to identify image data acquired 

 at a satellite zenith angle of greater than 53 degrees. 

 These pixels were digitally masked and therefore ex- 

 cluded from further processing. 



The SST images were coregistered to an equidistant 

 cylindrical projection (Snyder 1987) using least-squares 



transformation equations and the nearest-neighbor 

 resampling technique (Lillesand and Kiefer 1979). They 

 were then reformatted for additional processing with 

 version 8.0 of the Science and Technology Laboratory 

 Applications Software (ELAS) (Beverly and Penton 

 1989). ELAS was installed on a MicroVAX 3600 com- 

 puter and a MicroVAX 3500 workstation and provided 

 advanced spatial processing utilities required during 

 the second phase of the analysis. A processing protocol 

 was developed using sequential ELAS commands and 

 VAX software utilities to analyze the SST images and 

 transfer selected data and tabulations to the Statistical 

 Analysis System** for statistical analysis and plotting. 

 A binary mask constructed from the World Data Bank 

 II digital coastline file (Gorney 1977a, b) was used in 

 each processing stream to exclude land pixels from the 

 analyses. This protocol was used for all subsequent pro- 

 cessing during the study. The initial analysis of each 

 SST image consisted of Gulf-wide tabulations of tem- 

 perature frequencies and cloud pixels. 



The magnitude of surface temperature gradients 

 (MSTG) was derived from each SST image using Sobel 

 operators and simple image arithmetic (Gonzales and 

 Wintz 1977). Sobel operators use the 3x3 moving win- 

 dow technique to extract vertical (north-south) and 

 horizontal (east-west) temperature gradients from 

 digital images. The MSTG was computed by summing 

 the absolute value of the horizontal and vertical gra- 

 dient information (Gonzales and Wintz 1977). The 

 result of this operation is that each pixel location is 

 assigned a numerical value that indicates how greatly 

 SST at that location differs from that of surrounding 

 pixels. This gradient value is independent of the direc- 

 tion of temperature change. An additional masking 

 operation was performed on each MSTG image to 

 exclude contaminated pixels adjacent to the land and 

 cloud masks that were created as an artifact of the 

 moving window technique. 



The possible relationships among thermal features 

 and yellowfin tuna CPUE was examined by computing 

 summary statistics (e.g., mean, median, coefficient of 

 variation) of the SST and MSTG data for circular 

 regions of sea surface encompassing each longline set. 

 Since the orientation and initial and final geographic 

 coordinates of the sets were not available, we defined 

 three concentric circular areas encompassing each 

 reported location of a longline set. The recorded loca- 

 tion of the set was the center of these circular areas, 

 while the radii were specified as the length of the 

 longline set, one-half the length of the set, and one- 



* TeraScan is a proprietary computer system marketed by 

 SeaSpace, 3655 Nobel Drive, Suite 16o! San Diego, CA 92122. 



* * The Statistical Analysis System is a proprietary computer soft- 

 ware package marketed by the SAS Institute, Box 8000, Cary 

 North Carolina 27511-8000. 



