GOES-18 Band 7 Barcode Artifact Removal

Geostationary Operational Environmental Satellite (GOES) 18 is the newest satellite in the GOES-R series of geostationary satellites operated by the National Oceanic and Atmospheric Administration (NOAA). GOES-18 was launched on March 1, 2022 and began producing images from the Advanced Baseline Imagery (ABI) on May 5, 2022. Between May 16 and June 6, GOES-18 drifted from its original 89.5°W position to 136.8°W. On July 27, 2022, GOES-18 imagery reached provisional maturity and beginning August 1, the ABI data was interleaved with GOES-17 due to GOES-17’s Loop Heat Pipe anomaly. GOES-18 replaced GOES-17 as the operational GOES-West satellite on January 4, 2023.

After GOES-18’s drift to 136.8°W, vertically-oriented bars were discovered in GOES-18 ABI Band 7 (3.89 μm) imagery. These vertically-oriented bars have a barcode-like appearance. This “Barcode Artifact”  impacts most products that use Band 7 including Band 7 L1b, RGBs, Band Differences with Band 7, and other multispectral imagery which uses Band 7. The barcode artifact likely does not affect other ABI bands, products that do not use Band 7, and some L2 data products that use Band 7.

Throughout GOES-18's preliminary and operational phases, the barcode artifact's appearance has changed. Between GOES-18's first ABI imagery and September 22nd, 2022, the barcode artifact was decreasing in appearance due to instrument adjustments. Between September 22nd, 2022 and December 25th, 2022, the barcode artifact was mostly removed. On December 25th, the GOES-18 ABI reset, which caused the barcode artifact to change appearance and become more severe. Currently, the barcode artifact's appearance is as a few random larger stripes with more minor striping in colder scenes. More information about GOES-18 ABI changes and anomalies can be found at  https://www.star.nesdis.noaa.gov/GOESCal/goes_SatelliteAnomalies.php .

The barcode artifact has been observed to be most significant in colder scenes such as in cold cloud tops and at night when solar reflection is at its minimum. While the barcode artifact is less visible in warmer scenes, it can still be observed in multispectral imagery such as the Nighttime Microphysics RGB and band differences with Band 7.

While the barcode artifact does not prevent Band 7’s use in operational settings, it does add noise to products which can make interpretation more difficult. The barcode artifact is also not consistent between scan times, causing animations of Band 7 and some products which use Band 7 to contain stripes which move randomly across cloud tops, possibly harming a forecaster's ability to see key features in imagery.

1. Radiance data is converted to brightness temperature (BT) in Kelvin

2. The BT data is masked where values are greater than 273.15K (see Figure 1)

3. The data is sectioned into areas with widths dependent on the image sector (Full Disk, CONUS, Mesoscale) (see Figure 2 image 1)

4. Step 3 is repeated multiple times, depending on the image sector, with the sections staggered. This process reduces the impact of the edges of the sections on the final data (see Figure 2)

5. A Fast Fourier Transform (FFT) is computed for each section and stagger and the FFT data is shifted so that the zero-frequency component of the FFT is centered in the spectrum (see Figure 3)

6. A horizontal wedge of width 34 grid cells and a 6 grid cell gap centered horizontally and a similar vertical wedge with a width of 2 grid cells and 6 grid cell vertically centered gap of FFT data are replaced with a value of 1 (see Figure 4)

7. The inverse FFT is computed for each section and stagger

8. The resulting BT values is adjusted using the following process:

8a. The difference between the original BT values and resulting BT values in calculated

8b. A Gaussian filter is used with a sigma values of 2 to reduce the striping present in difference data

8c. The filtered difference is added to the resulting brightness temperature data

9. The new BT data is inserted into a copy of the original BT data with a 50 grid cell edge removed

10. Steps 5 - 9 are repeated for each stagger and the modified BT data is added to a list

11. The mean value for each grid cell from the list of staggers is computed to produce the final BT data

12. The final BT data is either converted back to radiance data and exported as a NetCDF4 file or plotted as brightness temperatures

The barcode artifact removal method is effective at removing the barcode artifact, especially in colder scenes. In warmer scenes, the artifact removal can create a wrinkle-like pattern in the resulting data. This is caused by too much data being replaced in the FFT data (step 6 in Methodology).

The wrinkle pattern is consistent in location between imagery within a sector. This means that the wrinkle pattern should align if multiple images from the same sector are aligned, even if the sectors are from different locations. The magnitude of this wrinkle is not consistent between imagery and appears related to the brightness temperature values. Due to these properties of the wrinkle pattern, there is the possibility of removing this artifact in the future.

One possible way to mitigate the wrinkle pattern is by adjusting the the brightness temperature mask (see Step 2 in Methodology) to fit certain applications. For example, in convective applications, a lower brightness temperature threshold could be used so that the method only impacts clouds, leaving the minor striping present in clear sky areas.

The artifact removal method is computationally quick enough for use in an operational setting. On a consumer grade computer, the runtime for the full disk sectors is about 1 minute. For the CONUS sector, the runtime is about 5 seconds. For the mesoscale sectors, the runtimes are about 1.2 seconds. Runtimes are dependent on load times for the original satellite imagery.

The FFT barcode artifact removal method significantly reduced the effects of the barcode artifact in GOES-18 ABI Band 7 satellite imagery. The method has introduced a wrinkle pattern into imagery which is most visible in the Nighttime Microphysics RGB and the Night Fog Difference band difference. Due to the wrinkle pattern's properties, the pattern could be removed in the future.

Data is run live for the CONUS and Mesoscale sectors through the process described above. NetCDF files are also available for closer viewing of the modified satellite imagery.

Funding provided by the Pathways Internship through the National Weather Service Office of Observations. The Xarray, Scipy, Numpy, Boto3, S3FS, Matplotlib, and Cartopy Python libraries were utilized for data processing and visualization. The University of Wisconsin-Madison Space Science and Engineering Center’s Satellite Information Familiarization Tool was used to visualize RGBs and band differences.