This page provides information about artifacts discovered in Landsat Collection 2 products. Note: These artifacts will also be visible in Landsat Collection 2 U.S. Analysis Ready Data (ARD) products.
This page provides a detailed description of the known issues about the following Landsat Collection 2 (C2) products. The known issues will be updated as more details are documented.
NOTE: Landsat sensor and satellite-level known issues and anomalies are listed on the Landsat Known Issues webpage.
Landsat Collection 2 Surface Temperature data gaps due to missing ASTER Global Emissivity Dataset (GED)
Landsat Collection 2 (C2) Surface Temperature (ST) products are generated from several input data sources and atmospheric profiles, this includes NASA’s Advanced Spaceborne Thermal Emission and Reflection Radiometer Global Emissivity Dataset (ASTER GED). ASTER GED contains areas of missing mean emissivity data required for successful ST product generation. If there is missing ASTER GED information, there will be missing ST data in those areas. Visit Landsat Collection 2 Surface Temperature data gaps due to missing ASTER GED page to learn more.
Blockiness Artifact in Landsat Collection 2 Surface Temperature Products
The Landsat C2 ST product may appear ‘blocky’ over small surface targets with contrasted thermal infrared signature. This artifact traces back to the Landsat emissivity (EMIS) band, which is derived from the Advanced Spaceborne Thermal Emission and Reflection Radiometer Global Emissivity Dataset (ASTER GED) Version 3 (GEDv3) emissivity data as well as ASTER Normalized Difference Vegetation Index (NDVI). The blockiness results, in part, from nearest neighbor interpolation of 100-m ASTER GED products (emissivity and NDVI) to the finer Landsat 30-meter (m) grid. For additional information about the Landsat emissivity calculation procedure, see Section 18.104.22.168.4. of the Landsat 8-9 Calibration Validation Algorithm Description Document.
The visual blocky artifact in Landsat C2 ST is more noticeable in locations where a geometric mis-registration exists between the ASTER NDVI and Landsat data (as seen in the image below). The NDVI data from ASTER visible and near infrared (VNIR) bands are registered to the thermal infrared data which has limited geolocation accuracy. Additional information about the geometric accuracy of ASTER can be found in the ASTER User Handbook.
Vegetation Adjustment Anomaly in Surface Emissivity input for Landsat Collection 2 Surface Temperature
The ASTER GEDv3 is a static dataset representing the average emissivity from 2000-2008, and in order to use it in the Landsat ST algorithm it needs to be adjusted for land surface conditions at the time of Landsat overpass. The emissivity adjustment procedure is based on Malakar et al. (2018) which takes into account the impact of time-varying vegetation and snow cover on effective surface emissivity using an estimate of vegetation density and snow/ice cover change from ASTER era to Landsat overpass conditions.
An anomaly was uncovered in the C2 implementation of the vegetation cover adjustment. The Landsat emissivity of a mixed pixel is computed using the linear relationship between the bare soil and vegetation fraction components of the pixel where the bare soil component is estimated using ASTER GEDv3 emissivity and mean NDVI products and then adjusted based on the vegetation condition at the time of Landsat observation. The locations of ASTER bare pixels (ASTER NDVI < 0.5) are then overwritten and set to bare soil emissivity. This leads to erroneous Landsat emissivity and subsequently surface temperature estimates wherever vegetation has changed significantly between the ASTER and Landsat eras and will be most notable where there are large spatial contrasts in surface temperature; for example, where new irrigated fields have been established in semi-arid landscapes. The image below displays this anomaly.
The vegetation adjustment procedure will be refined for reprocessing of the Landsat ST product in Collection 3.
Overcorrection of aerosol path radiance in Landsat 8-9 Collection 2 Surface Reflectance resulting in nonphysical values over bright surfaces and over water
The C2 Landsat 8-9 data are atmospherically corrected using the Land Surface Reflectance Code (LaSRC) (Vermote et al. 2016) version 1.5.0. The USGS LaSRC code version 1.5.0 is a C translation of the NASA Fortran-based LaSRC version 3.5.5 and is modified to make the code run more efficiently. The aerosol optical thickness (AOT) in USGS LaSRC code is retrieved with respect to 3x3 30-m pixel windows instead of the original per-pixel retrieval implementation. In addition, a semi-analytical approach is used to estimate the atmospheric variables (transmission, intrinsic reflectance, and spherical albedo) more efficiently based on a cubic polynomial fit of the AOT Lookup Table (LUT) values instead of iterative LUT searches.
However, the LaSRC aerosol inversion is fine-tuned for the dark targets. In Collection 1 LaSRC (C code version 1.3.0), the cloud, cloud shadow, cirrus, and water pixels were excluded from aerosol retrieval routines. Instead, the median AOT value of the clear pixels within the scene was assigned to these pixels for atmospheric compensation. A climatological AOT (0.05) was used if there were insufficient clear sky AOT retrievals. For the cloud and cirrus pixels, the climatological AOT-corrected surface reflectance values were then passed through to the Level-2 product. These changes were a deviation from the original NASA Fortran algorithm and established a dependency on the Level-1 cloud mask algorithm.
The AOT retrieval in C2 LaSRC was modified to be more consistent with the original Fortran LaSRC code and to eliminate the uncertainty associated with the Level-1 cloud masking algorithm. In Collection 2 LaSRC (C code version 1.5.0), any non-fill pixels are now utilized for aerosol retrieval regardless of the cloud, cirrus, shadow, or water condition. The algorithm searches for non-fill pixels within 3x3 30-m pixel windows, starting at the center and then moving to non-center pixels, inverts the AOT for the first non-fill pixel encountered and assigns it to the center of the 3x3 pixel window. The AOT is inverted using spectral ratios of the OLI coastal aerosol, blue, and red spectral bands and is optimized by iteratively calling the atmospheric correction function until the AOT inversion converges. The spectral ratio method is designed for non-water pixels so for the pixels that AOT inversion failed to converge, a subsequent water-based approach is attempted which uses a red spectral band AOT LUT. Window locations where the AOT retrieval failed are flagged as failed and their AOT values are gap filled from surrounding 15x15 pixel windows. After this step if there were any remaining windows without AOT, an AOT=0.05 (climatology) is assigned to the center of 3x3 pixel window. Finally, an inverse distance weighted interpolation is then applied to obtain AOT estimates for every non-center 30-m pixel.
The Aerosol QA (SR_QA_AEROSOL) band, delivered with Landsat 8-9 Level-2 Surface Reflectance provides per-pixel information about the validity of the aerosol retrieval and a qualitative measure of aerosol level. It also shows if a water-based algorithm was used for aerosol determination or if the aerosol was interpolated from the center of 3x3 windows. The Landsat Quality Assessment ArcGIS toolbox can be used to unpack the Aerosol QA band.
In summary, the C2 Landsat 8-9 Level 2 Surface Reflectance is produced globally and in general performs as expected particularly over dark land targets. However, the aerosol retrievals often fail over and near bright snow/ice and cloud pixels, due to excess path radiance. Thus high AOT values can be inappropriately retrieved near the edges of the cloud and over bright targets. In this scenario, the gap filling procedure propagates these inappropriately high AOT values over to any failed retrievals within the surrounding 15x15 pixel windows. This can lead to an undesired overcorrection for aerosol scattering and absorption over the larger area. The yellowish or dark color at the cloud edges in Figure 1 indicates low surface reflectance due to aerosol path radiance overcorrection. The Aerosol QA layer can be used to somewhat exclude invalid aerosol retrievals resulting in high aerosol levels from any data analysis.
A subset of the atmospherically corrected pixels may have values outside their theoretical limits, i.e., reflectance >1.0 or <0.0 (unitless reflectance scale) due to overcorrection associated primarily with incorrect atmospheric AOT characterization over surfaces that reflect light in the satellite observation direction more strongly than a Lambertian surface. Therefore, a known C2 issue for the Landsat 8 and 9 Level 2 surface reflectance products occurs over snow and ice where the AOT inversion can fail and the AOT interpolated from neighboring values is insufficiently representative, causing excessive reflectances > 1.0, particularly in the shorter wavelength spectral bands. Users are encouraged to examine the per-pixel Aerosol QA (SR_QA_AEROSOL) information, especially the validity of aerosol retrieval (bit 1) and the qualitative aerosol level (bits 6 and 7). Pixels with invalid AOT retrieval and high aerosol levels indicate greater uncertainty in surface reflectance and can be used to mask most reflectance values >1. Figure 2 shows the extent and magnitude of the Landsat 8 and 9 surface reflectance >1.0 pixels in March 2022 over the North America. Nearly all these pixels are flagged as cloud or snow/ice in pixel QA (QA_PIXEL) layer.
Correspondingly another known C2 issue is that the aerosol retrievals frequently fail over water pixels. If the aerosol retrieval fails over water pixels, and if the water pixels are adjacent to cloud or land pixels with valid high AOT, the high AOT from the land or cloud pixels can get propagated into the water by the local averaging fill method. The local averaging fills the invalid aerosol in a top to bottom and left to right image operation. As a result, the high aerosol values on the land or cloud will be propagated into the water pixels on the right and bottom. Thus, the high aerosol values over the water will leave a dark trail (low surface reflectance) towards the bottom right of the Landsat 8 image. The aerosol overcorrection usually results unphysical surface reflectance values (surface reflectance <0), especially over shadowed water. Figure 3 shows examples of a trail artifact over the Gulf of California and the Atlantic Ocean.
'NoData' pixels in Landsat 8-9 Collection 2 Surface Reflectance
Landsat 8 and Landsat 9 C2 Surface Reflectance products may contain 'NoData' pixels on the edges of clouds even though the QA band does not indicate a NoData pixel. This artifact is more frequent at the shorter wavelengths and usually occurs over dark water pixels and shadowed land pixels under low solar illumination conditions.
This is partially due to the use of '0' as fill value in the product. The surface reflectance values for shorter wavelength bands over dark targets may fall below the minimum valid range of -0.2 in the L8-9 Land Surface Reflectance Code (LaSRC). Such pixels are then mapped to -0.2 to be within the valid range. After applying the C2 offset and scale factor, the surface reflectance =-0.2 will be scaled to zero, which is the 'NoData' value in C2. Therefore, the pixels that have a calculated surface reflectance value lower than the valid minimum, will be set to 'NoData'.
The highlighted pixels in the images to the right indicate 'NoData' pixels in shadowed areas over land and water. (In the Collection 1 surface reflectance product, the valid minimum was scaled to -2000 for such pixels, which is distinct from the NoData value of -9999.)
Masked Atmospheric Opacity in Landsat 4-7 Surface Reflectance
NOTE: This artifact has also been noticed in Landsat 4-7 Collection 1 Surface Reflectance data, and propagated to Landsat Collection 1 and Collection 2 U.S. Analysis Ready Data.
The atmospheric opacity band in the Landsat 4-7 C2 Surface Reflectance product may be completely masked. The masked atmospheric opacity is due to aerosol retrieval failure. The Landsat Ecosystem Disturbance Adaptive Processing System (LEDAPS) algorithm identifies dark dense vegetation (DDV) pixels (clear land pixels where TOA reflectance at Band 7 <0.15) that fall within 1.2 km grids (40 x 30-m pixels). Aerosol optical thickness is inverted by using 6S radiative transfer and iteratively varying AOT to correct TOA reflectance until atmospherically compensated surface reflectance in the Blue spectral band is equal to 0.66 times the surface reflectance in the Red spectral band. If LEDAPS fails to find adequate “dark targets” required for aerosol retrieval, the atmospheric opacity will be entirely NoData. For these scene instances, a default AOT = 0.01 value is used for atmospheric compensation of aerosols. Users can refer to the DDV bit (bit 0) in LEDAPS internal QA band (SR_CLOUD_QA) to identify valid DDV pixels. The image below shows the natural color composite for a Landsat 5 Thematic Mapper image in north Africa where the atmospheric opacity band is completely masked.
The atmospheric opacity band may also be completely masked in wintertime Landsat scenes at high latitudes where the solar elevation angle is low. In these images, adequate “dark targets” are present, but the algorithm fails to find clear pixels that meet Short Wave Infrared (SWIR) 2 and Near Infrared (NIR) thresholds. The AOT of the 1.2 km grids are set to fill if the pixels in that grid are fill, water, cloud, adjacent to the cloud, cloud shadow, or snow, or if the clear pixels have low TOA reflectance at SWIR2 (2.2 µm) or NIR (0.83 µm) spectral bands (reflectance in the SWIR2 >0.015 and reflectance in the NIR >0.10). Due to the lack of clear pixels above the SWIR2 and NIR thresholds, the aerosols are set to fill value, and that triggers the atmospheric opacity to be set to fill. The image to the right shows an SLC-off Landsat 7 Enhanced Thematic Mapper Plus (ETM+) image from December 2012 on the US/Canada border. It has a mixture of clouds, snow, and water. In this image, there are not enough pixels that meet the criteria to allow aerosols to be computed for the 1.2 km grids, and therefore the final atmospheric opacity band is completely masked.
Degraded AURA OMI Ozone Files with Anomalous Values
The total column ozone data products provided by AURA Ozone Monitoring Instrument (OMI) have been shown to exhibit erroneously high values between May 18, 2021 through August 10, 2021 spanning from 30 degrees South to 50 degrees South. (See the red pixels in the map below, and this NASA Forum page)
A total of 64 Landsat 7 ETM+ scenes processed to Surface Reflectance using the Landsat Ecosystem Distribution Adaptive Processing System (LEDAPS) algorithm were found to have been impacted by the coinciding degraded ozone data. The scenes listed in this file have since been reprocessed using corrected OMI measurements. Users of the affected scenes are encouraged to re-download the products to preserve continuity with previous and future Collection 2 Surface Reflectance data.
Landsat 4-7 QA Pixel Clear Flag Discrepancy
In February 2021, the USGS identified an issue with the Landsat 4-7 Collection 2 Quality Assessment (QA) Pixel "Clear" bit (bit 6) under certain conditions over water. This issue does not impact over land pixels nor does it impact the Landsat 8 QA Pixel. The issue relates to the Landsat 4-7 Clear bit being incorrectly set to OFF when it should be ON when the Cloud Shadow (bit 4), Snow/ice (bit 5), and Water (bit 7) are all ON. Additionally, the Clear bit is ON when it should be OFF when the Dilated Cloud (bit 1) is ON.
The USGS recommends users cease reliance on the QA Clear bit (bit 6) alone in all Landsat 4-7 Collection 2 data products. Instead, users should engage the QA “Dilated Cloud” (bit 1) AND “Cloud” (bit 3) OFF condition to correctly identify clear pixels over water. Users who have accessed or downloaded Landsat 4-7 Collection 2 data products since December 1, 2020 should consider accounting for this discrepancy in their analysis. Failure to do so may reduce the number of valid clear pixels while increasing the inclusion of dilated cloud pixels.