Artifacts discovered in Landsat Collection 1 and/or Landsat Collection 2 U.S. ARD are listed below. Note: Some of the artifacts described on this page trace back to and are also visible within the source Landsat Level-2 Surface Reflectance or Surface Temperature data and/or the supporting files that are delivered with those scene-based products.
Artifacts affecting both Landsat Collection 2 and Collection 1 U.S. ARD
- Quality Band Discontinuity (all sensors)
- Surface Reflectance Cloud Quality Assessment Anomaly (Landsat 4-5 TM, Landsat 7 ETM+)
- Cloud Shadow Placement Shift Between Projections
- Incorrect Classification of Dark Lava
- Masked Atmospheric Opacity in Landsat 4-7 Surface Reflectance
- Missing Surface Temperature data due to missing ASTER GED
Artifacts affecting Landsat Collection 2 U.S. ARD
- Inconsistent downsampling level of Cloud-optimized Geotiff (COG) Overviews
- Inaccurate Corner Longitudes in Antemeridian-crossing Tiles
Please visit Landsat Collection 2 Known Issues for details on these Collection 2 artifacts:
- 'NoData' pixels in Landsat 8 Collection 2 Surface Reflectance
- Overcorrection of aerosol path radiance in Landsat 8 Collection 2 Surface Reflectance
- Landsat 4-7 QA Pixel Clear Flag Discrepancy
Artifacts affecting Landsat Collection 1 U.S. ARD
- Incorrect Scaling of Downwelled Radiance used in Single Channel Surface Temperature
- Underestimation of Surface Temperature Uncertainty
- Incorrect Precision of WGS84 Axis Values
- Incomplete File Names in the U.S. Landsat ARD Tile XML Metadata
- Incorrect Unit of Atmospheric Transmittance in XML Metadata
- Unused Landsat Scene Listed in Collection 1 Landsat ARD Tile XML Metadata File
Discontinuities were discovered in U.S. Landsat ARD PIXELQA and SRAEROSOLQA optical thickness bands. The image to the right displays a diagonal line indicating the discontinuity.
Cloud detection and aerosol retrieval are performed at the Landsat scene level, before ARD tile processing begins. The discontinuities in PIXELQA are likely due to temperature differences between the northern and southern Landsat scenes, resulting in the cloud confidence to be identified differently.
This discontinuity is visible in both Landsat Collection 1 and Landsat Collection 2 U.S. ARD products.
The false-positive identification of cloudy pixels was discovered in the Landsat Ecosystem Disturbance Adaptive Processing System (LEDAPS) algorithm’s internal detection of clouds, which is output as the SRCLOUDQA band in Landsat 4-5 TM and Landsat 7 ETM+ Surface Reflectance data products. An algorithmic calculation is incorrectly identifying clear irrigated field and mountain shadows as clouds in high desert regions. Analysis has determined that compared to the surrounding ambient temperature, these areas meet the difference criteria used by LEDAPS in its cloud cover assessment, and are therefore incorrectly marked as clouds.
The image below shows an example of an SRCLOUDQA band (left window) compared to a surface reflectance natural color composite image (bands 3,2,1) (right window). The misidentified clouds in high desert regions in LEDAPS-based U.S. Landsat ARD products does not significantly impact their scientific integrity.
This anomaly is visible in both Landsat Collection 1 and Landsat Collection 2 U.S. ARD products.
Cloud shadows are identified in Landsat Collection 1 and Collection 2 Level-1 and Level-2 products by the C Function of Mask (CFMask) algorithm, and then projecting cloudy pixels onto where their shadows should fall on the earth's surface. In different geometric projections (e.g., Albers vs. UTM), cloud shadows will be projected into slightly different positions on the earth's surface. This has the effect of changing the estimate of cloud height, which can then cause a large shift in the position of the final cloud shadow identification. This happens most often over high-altitude, partially transparent clouds such as cirrus. Because of this behavior, the position of cloud shadows in UTM and Albers products may differ (see the images below). The identification of cloudy pixels themselves is not affected.
Both Landsat Collection 1 and Landsat Collection 2 U.S. ARD contain this placement shift.
The land areas around dormant volcanoes that are covered by dark lava may be incorrectly classified as water in the U.S. Landsat ARD Pixel Quality Assessment (PIXELQA) band. This is a known issue of the C Function of Mask (CFMask) algorithm. The low reflectance at the near infrared and red wavelengths is the reason for this commission error in the CFMask water detection function.
The images below shows the PIXELQA over an area in Hawaii where the brown/dark lava pixels are incorrectly classified as water.
This classification error can be found in both Landsat Collection 1 and Landsat Collection 2 U.S. ARD products.
The Advanced Spaceborne Thermal Emission and Reflection Radiometer Global Emissivity Dataset (ASTER GED) provided by the NASA Land Processes Distributed Active Archive Center (LP DAAC) is used in the generation of Landsat Surface Temperature (ST) products. Where ASTER GED data is missing, there will be data gaps in the Landsat U.S. ARD ST product.
Note : To investigate the availability of any given pixel, the “ASTER Global Emissivity Dataset 100 meter HDF5 V003” data can be attained from https://www.doi.org/10.5067/Community/ASTER_GED/AG100.003.
Due to the size of Landsat Collection 2 U.S. ARD tiles (5000 x 5000 pixels) and rounding, the coarse-resolution overview of the tile-based COGs contains a resampling level of 63, which differs from the scene-based COG resampling level of 64. Details about the COG creation specifications and downsampling levels can be found in Landsat COG Data Format Control Book.
The metadata values for the longitudes of the upper and lower right corners "CORNER_UL_LON_PRODUCT", "CORNER_LR_LON_PRODUCT" for antemeridian-crossing Landsat Collection 2 U.S. ARD tiles over the Alaska region are inaccurate. The correct longitudes can be obtained by projecting the X-values of the Albers Equal Area (AEA) into Geographic Coordinate System (GCS).
The downwelled radiance used in the Single Channel Surface Temperature algorithm is erroneously not normalized to provide hemispherical radiance, resulting in ~0.5 Kelvin underestimation on average in the Landsat Provisional Surface Temperature product. This underestimation is almost negligible over water surfaces due to high emissivity, while it may be more pronounced over desert and arid regions where the atmospheric relative humidity is high. Landsat Collection 2 U.S. ARD is not affected by this issue.
Due to a miscalculation in the propagation of uncertainty in the Landsat Collection 1 Provisional Surface Temperature product, the uncertainty value (STQA) is underestimated by approximately 0.5 Kelvin on average. The magnitude of the uncertainty underestimation for any given pixel is dependent on the emissivity standard deviation obtained from the ASTER GED, land cover, and distance to the cloud at that pixel. Additional information about ST uncertainty and the standard error propagation can be found at https://doi.org/10.1016/j.rse.2018.06.026. Landsat Collection 2 U.S. ARD is not affected by this uncertainty.
An error that defines map projection geotag values in Landsat Collection 1 Level-2 Albers scenes resulted in assigning inaccurate values to the semi-major and semi-minor axes, lengthening them by ~3-meters (m). This error is carried through only in the Landsat Collection 1 science product Albers production stream, but subsequently does appear in all Landsat Collection 1 ARD tiles.
This geotagging issue does not result in any resampling or impact the validity of the ARD pixel values. The error manifests itself in a sub-2-m offset in each ARD tile. The offset values range geographically across the United States from about 0.75-m in Chesapeake Bay to about 1.5-m in Portland, Oregon.
Minimal impacts on data analysis have been observed, but there is a potential for issues that involve high-resolution sensors (i.e., LIDAR) or vector-based analyses. Any attempts to reproject other data to match an ARD tile using the definitions for WGS84 CONUS Albers will not be successful due to this issue. Any tool, application or software package that utilizes projection information and conducts cross-projection operations and analyses may also be affected by this issue. Landsat Collection 2 U.S. ARD is not affected by this error.
All Landsat Collection 1 U.S. ARD tiles contain incomplete file names in the XML metadata file. The <file_name> parameter in the tile_metadata section of the XML are missing the Tile ID and file format of the band file name.
Below is a correct and incorrect example of Collection 1 Landsat 4 ARD tile h019v011 for surface reflectance band 1.
This artifact does not affect the U.S. Landsat ARD LINEAGEQA band. This was corrected in Landsat Collection 2 U.S. ARD processing.
The data unit of the Atmospheric Transmittance (ATRAN) band in the Landsat Collection 1 U.S. ARD XML metadata is incorrect. Atmospheric Transmittance is an intermediate band of the Provisional Landsat Surface Temperature Science Product and is unitless.
Incorrect: <data_units>radiance (W m^(-2) sr^(-1) mu^(-1))</data_units>
This error was corrected in Landsat Collection 2 U.S. ARD processing.
The Lineage index band (LINEAGEQA) identifies which Landsat Collection 1 Level-2 Albers scene was the source for each pixel in a U.S. Landsat ARD tile by referencing scene identifiers listed in an associated XML metadata file. However, a small number of Landsat 8 and Landsat 7 ARD XML metadata files processed prior to April 16, 2019 have been found to list the incorrect Level-2 Albers input scenes.
This image illustrates the unused Level-2 Albers input scene listed in ARD tile (LC08_CU_027010_20170923_20181130_C01_V01; outlined in red) XML metadata file. Scene 1 and Scene 2 are incorrectly listed in the XML as contributing to ARD tile h027v010, whereas the correct input scenes listed should be Scene 2 and Scene 3.
Landsat Collection 1 U.S. ARD tiles created after April 16, 2019, and Collection 2 U.S. ARD tiles are not affected by this error.