AVHRR Data Processing: All AVHRR data starting January, 1991 were acquired directly by Ocean Imaging Corp. in Solana Beach, California. Acquisition and processing were done using Global Imaging/Telonics hardware and software. The raw data were first calibrated using standard NOAA procedures, then run through a Sea Surface Temperature (SST) algorithm, used to generate a cloud mask, georectified, and resampled to a final 8-bit product, as described below:

The SST algorithms used correspond to those developed for the NASA Pathfinder program whenever available. No Pathfinder algorithm exists for the NOAA-12 satellite. The Coastwatch SST algorithm developed by NOAA was thus substituted for NOAA-12 data.

For some time periods, two sets of satellite-specific coefficients used in the Pathfinder algorithm exist. The first set is derived within a year after a new NOAA satellite becomes operational. The second set is derived retrospectively after a few years of operation in an attempt to correct for any drift in sensor performance over time. There is thus a lag of several years between real-time data collection and the availability of the second coefficient set. For that reason, only historical processing of data collected up to 1996 utilize the second coefficient set. All historical processing after 1996 and all processing done in "real-time" during the GLOBEC Project utilize the initial Pathfinder coefficients.

The Cloud Mask file provided with all AVHRR data sets is generated using the CLAVR cloud mask algorithm as described by Stowe et al. - Global Distribution of Cloud Cover Described from NOAA/AVHRR Operational Satellite Data, Advances in Space Research, Vol. 11, No. 3, 1991. Ocean Imaging made the following modifications to the algorithm to maximize its efficiency for the geographical area covered by the GLOBEC Products:

1. As part of both the daytime and nighttime algorithm, the CLAVR routine incorporates a Thermal Uniformity Test (TUT). The TUT requires that the maximum difference in Ch4 brightness temperature within a 2x2 array not exceed a given threshold for the array to be classified as cloud free. It operates on the theory that the ocean is almost always more spatially uniform in the infrared when it is cloud free than when clouds are present. Ocean Imaging found that the default threshold (0.5°K) determined by Stowe et. al. is too low for the N. American coastal Pacific. It results in falsely masking over areas showing strong thermal gradients such as upwelling plume boundaries. The default TUT threshold has been raised to 0.75°K to reduce false masking over large SST thermal gradients. However, the algorithm sometimes still produces lines of cloud masked pixels along strong gradients.
2. The nighttime algorithm for ocean regions includes a test called the Uniform Low

Stratus Test (ULST). The ULST is used to detect low stratus clouds that frequently

escape the first two tests of the algorithm. The ULST is based on the CH3-Ch5

temperature difference and has a Ch4 dependent threshold to allow for the variable

after vapor attenuation between the two channels. This test has been found to

incorrectly classify areas of relatively cold water as clouds under certain situations. It

also falsely classifies noisy pixels resulting from the use of channel 3 in the test (Most AVHRRs suffer from instrument noise in the thermal Channel 3). In

order to reduce the amount of false classification by the ULST, Ocean Imaging

developed a simple restoral test (ULSR) that runs after the ULST which helps

reclassify pixels that should not be considered cloudy:

All 2x2 tile sets that the ULST determines to be cloudy or mixed are subject to the

ULSR test. The ULSR looks at Ch3-Ch5 for pixels in the same 2x2 tile. If the

Ch3-Ch5 difference for any of the 4 pixels in the array is less than X_L or greater than

X_H, the pixels in the tile continue to be classified as cloudy. If all 4 of the Ch3-Ch5

values fall between X_L and X_H, the pixel will be reclassified as clear and passed on to

the next test.

Through a series of empirical tests for the waters off of Northern California and

Oregon, the X_L and X_H values were determined to be -1.0 and 0.5 respectively.

The only drawback to this restoral test is that it occasionally reclassifies pixels as

clear that undoubtedly should be deemed cloudy. The X_L and X_H values above

represent a decided balance between over-masking cold water and noise to under-

masking obvious cloudy pixels.

The resulting cloud masks represent a robust alternative to some of the other algorithms available. To our knowledge, no "perfect" cloud masking methods exists, so any approach necessitates some trade-offs. One notable limitation of the above approach is considerable overmasking in daytime data containing sunglint. The presence of a sunglint swath and its location within the data is determined by the timing of the orbit relative to sun angle. The orbits of some satellites (such as the best afternoon NOAA-14 orbit over the U.S. west coast) coincide closely with the sungtlint-causing geometry. This sometimes results in large areas of the entire data set being masked, while some useful data could have still been obtained on the outskirts of the sunglint swath. However, altering the algorithm to eliminate this problem would cause it to significantly undermask certain cloud types under other conditions.

Georectification is accomplished using the Global Imaging GAE-Focus 9000 software. The positioning of AVHRR data is linked to an on-board clock stamp. The clock is subject to drift which is corrected by NOAA mission control every few days. In the GLOBEC processing, errors due to this drift are minimized by an operator-assisted step involving the computation of a position offset by matching a "master" coastline file to fit each data set prior to rectification. Using a coastline file rather than computing an offset from single or multiple landmarks was found to be more consistent and more efficient under partly cloudy conditions. Each full data set is then georectified into two slightly overlapping sectors using Stereographic Linear latitude/longitude projection. The areas are defined by:

Northern Sector:

Pixel (1,1) - top left: 56.23N / 138.23W

Pixel (1024,1024) - center: 46 .00N / 128.00W

Pixel (2048, 2048) - bottom right: 35.76N / 117.76W

Southern Sector:

Pixel (1,1) - top left: 39.23N / 132.29W

Pixel (1024, 1280) - center: 29.00N / 119.50W

Pixel (2048, 2560) - bottom right: 18.76N / 106.70W

Resampling is done as the final step to minimize file size for electronic transfers and maintain universal file compatibility. The images are resampled from 16-bit real values to 8-bit values in a temperature range that preserves the original AVHRR thermal resolution of 0.1 deg. C. The two regional sectors are resampled from different ranges to match the temperature range generally existing in each area. The resampling ranges are:

Northern Sector: From 0.0º - 25.5º to byte values 0 -255

Southern Sector: From 10.0º - 35.5º C to byte values 0 - 255

True SST values can be derived by reversing the relationships.

Comparisons of satellite-derived SST and buoy temperatures: OI periodically checks the performance of the SST algorithms by comparing the computed results with measurements recorded by U.S. and Canadian buoys moored off the N. American west coast. Detailed information about the buoys, location maps and maintenance schedules can be obtained at www.nws.fsu.edu/buoy. OI obtains the archived buoy measurements via ftp at www.ndbc.noaa.gov.

The comparisons are made by manually selecting within each satellite data set valid SST/buoy pairs whose location and vicinity (approx. 10x10 pixels) are clear of clouds and have no thermal gradient intersecting the location. A satellite SST value for the buoy location is then computed as the average of a 3x3 pixel matrix centered on the buoy. This is done to minimize the effects of pixel noise present in the original SST files. The buoys record temperatures once each hour and the reading closest to the time of the satellite overpass is used for the comparisons.

The satellite SST and buoy values, the resulting differences and related statistics for each buoy, satellite and algorithm (Pathfinder and Coastwatch) are available as MS Excel files on this web page.