Download Landsat 8 (L8) Data User Handbook (.pdf 4.39 MB)
The geometric algorithms used by LPGS at EROS were originally developed for the L8 IAS. The overall purpose of the IAS geometric algorithms is to use Earth ellipsoid and terrain surface information in conjunction with spacecraft ephemeris and attitude data, and knowledge of the OLI and TIRS instruments and L8 satellite geometry, to relate locations in image space (band, detector, sample) to geodetic object space (latitude, longitude, and elevation).
These algorithms are used to create accurate L1 output products, characterize the OLI and TIRS absolute and relative geometric accuracy, and derive improved estimates of geometric calibration parameters such as the sensor to spacecraft alignment.
The Level-1 processing algorithms include the following:
Precision, terrain-corrected image resampling
Figure 4-1 shows LPGS standard product data flow, which includes radiometric and geometric processing.
Figure 4-1. LPGS Standard Product Data Flow
The following paragraphs describe the purpose and function of each of the major LPGS Subsystems. Complete detailed designs for each Subsystem are presented in subsequent sections.
The L8 OLI and TIRS geometric correction algorithms are applied to the wideband (data contained in Level-0R (raw) or 1R (radiometrically corrected) products. Some of these algorithms also require additional ancillary input data sets. These include the following:
One of the goals of L8 is the provision of high-quality, standard data products. About 400 scenes per day are imaged globally and returned to the United States archive. All of these scenes are processed to a Level-1 standard product and made available for downloading over the Internet at no cost to users.
The L1T available to users is a radiometrically and geometrically corrected image. Inputs from both the sensors and the spacecraft are used, as well as GCPs and DEMs. The result is a geometrically rectified product free from distortions related to the sensor (e.g., view angle effects), satellite (e.g., attitude deviations from nominal), and Earth (e.g. rotation, curvature, relief). The image is also radiometrically corrected to remove relative detector differences, dark current bias, and some artifacts. The Level-1 image is presented in units of DNs, which can be easily rescaled to spectral radiance or TOA reflectance.
Figure 4-2. Level-1 Product Ground Swath and Scene Size
A complete L1 product consists of 13 files, including the 11 band images, a product-specific metadata file, and a Quality Assessment (QA) image. The image files are all 16-bit GeoTIFF images. The OLI bands are Bands 1-9. The TIRS bands are designated as Bands 10 and 11.
The QA image is a 16-bit mask, which marks clouds, fill data, and some land cover types. Subsection 5.4 gives a full description of the L8 QA mask.
The metadata (MTL) file contains identifying parameters for the scene, along with the spatial extent of the scene and the processing parameters used to generate the Level-1 product. This file is a human-readable text file in ODL format.
The product delivered to L8 data users is packaged as Geographic tagged image file format (GeoTIFF) (a standard, public-domain image format based on Adobe's TIFF)and is a self-describing format developed to exchange raster images. The GeoTIFF format includes geographic or cartographic information embedded within the imagery that can be used to position the image in a geographic information display. Each L8 band is presented as a 16-bit grayscale image. Specifically, GeoTIFF defines a set of TIFF tags, which describes cartographic and geodetic information associated with geographic TIFF imagery. GeoTIFF is a means for tying a raster image to a known model space or map projection and for describing those projections. A metadata format provides geographic information to associate with the image data. However, the TIFF file structure allows both the metadata and the image data to be encoded into the same file.
The L8 CCA system uses multiple algorithms to detect clouds in scene data. Each CCA algorithm creates its own pixel mask that labels clouds, cirrus, and other classification types. The separate pixel masks are then merged together into the final L1 quality band.
The separate masks are merged together via a weighted voting mechanism. Each algorithm is assigned weights for every class (cloud, cirrus, water, and snow / ice), which indicates how accurate that algorithm is expected to be when classifying that type of target. These weights are defined in the CPF. Then, for each pixel, the confidence value in each mask is used to sum the algorithm weights together:
This final score is assigned to the pixel in the quality band. Note that the mid-confidence score – which is equivalent to 'ambiguous' – wins all ties among the CCA algorithms.
The L8 CCA system was designed to be modular, providing the ability to quickly add or remove CCA algorithms as desired. It is expected that more CCA algorithms will be added to the L8 CCA system in the near future. At launch, there were four CCA algorithms, including: ACCA, See-5, Cirrus, and AT-ACCA.
ACCA is an algorithm used to generate scene-wide cloud scores for previous Landsat instruments. The L8 implementation of ACCA is similar, but it uses only Pass 1, the spectral cloud identification component of the algorithm, which generates a per-pixel cloud mask. Later ACCA passes are used to bias the pass 1 results to create a single cloud score for the entire scene, so they are not useful for creating per-pixel masks.
ACCA works Bands 3-6 of Level-1 imagery that is converted to TOA reflectance using a scene-center solar elevation angle. It also uses TIRS Band 10 imagery that have been converted to at-satellite brightness temperature. The formula for these conversions is in Subsection 5.3 of this document.
Once the TOA reflectance for Bands 3-6 and the brightness temperature of Band 10 is available, ACCA uses eight different filters to classify the pixels in the scene:
Figure 4-3. A Diagram of the First Pass ACCA Algorithm
The ACCA algorithm labels each pixel in the Level-1 scene as non-cloud, ambiguous, or cloudy. These labels are analogous to the low, medium, and high confidence cloud designations in the L1 quality band. The CCA control system takes the output from the ACCA algorithm and other CCA algorithms to generate the final L1 quality band.
See-5 CCA is a cloud algorithm developed using the concept of information entropy to split training data into thresholds that provide the maximum information. Its name comes from the software used to train the algorithm, the C5.0 software package from Rulequest Research.
The See-5 algorithm is a very large decision tree of 244 threshold tests, whose input is OLI Bands 2-7. For every pixel in an L1 scene, See-5 CCA generates a confidence score that is used to label the pixel as low, medium, or high confidence cloud. Masks created by See-5 are very accurate – this algorithm outperforms ACCA in most situations – but the algorithm itself is too convoluted to present in detail. For more information about the See-5 CCA algorithm, see Scaramuzza et al., 2012.
High, thin clouds are detected using OLI Band 9, which is centered on shortwave infrared light at 1.38 µm wavelength. Light at this wavelength is strongly absorbed by water vapor in the Earth's atmosphere, so that sunlight that travels too deep into the atmosphere cannot reflect back to the satellite. Cirrus clouds are high in the atmosphere and thus above most of the water vapor, so they are strong reflectors of this 1.38 µm radiation.
The L8 Cirrus CCA algorithm is a simple threshold test for reflectance in OLI Band 9. The threshold is set in the CPF; the at-launch threshold is 2 percent reflectance. Any object with greater reflectance in Band 9 is marked as possible cirrus.
Cirrus if: ρ9 > Tcirrus
ρ9 = TOA reflectance in OLI Band 9.
(Note that this is calculated from the Level-1G product using a scene-center solar elevation angle.)
Tcirrus = Cirrus threshold from the CPF. (0.02 at launch)
This test is very permissive, and it detects false positives over high-altitude arid regions and over the poles, where the air is very cold and dry. The cirrus test is very accurate over low to moderate elevation areas in non-polar regions.
Figure 4-4. Temperate Region Affected by Cirrus Clouds. Top image is RGB composite of OLI Bands 4,3,2; bottom image is OLI Band 9, the cirrus detection band
The quality band in Level-1 products contains a cirrus flag that reports high, medium, or low probability of cirrus for each pixel in the scene. This is intended as an aid to users, to indicate how each pixel may be affected by thin, high-altitude clouds that may or may not be visible in other bands of the imagery.
The 'Artificial Thermal' (AT) variant of the ACCA algorithm is intended for use only when thermal data from the TIRS instrument are not available. It uses the reflectance of Bands 2-7, calculated from L1 data using a scene-center solar elevation angle, to approximate the brightness temperature of the missing thermal band. This AT band is intended only for use in the AT-ACCA algorithm, and is not available to the public. Once the AT band is created, the AT-ACCA algorithm uses it as a replacement for BT in the same threshold filters as the normal ACCA algorithm. Note that ACCA and AT-ACCA are never run together. If a thermal band exists, ACCA is run; if no thermal band exists, then AT-ACCA is used.
A large number of pixels are labeled ambiguous by the filters in the AT-ACCA algorithm. To reduce the number of ambiguous pixels, a second pass is made using an additional suite of disambiguation threshold tests. These disambiguation tests are only applied to pixels labeled ambiguous by the AT-ACCA. The tests are simple thresholds that are statistically unlikely to indicate clouds among data normally marked as ambiguous by ACCA. If a pixel passes any two of these tests, it is labeled as non-cloud. Pixels that pass only one or zero tests remain ambiguous. Similar to the ACCA algorithm, the AT-ACCA's labels of non-cloud, ambiguous, and cloud pixels are analogous to the low, medium, and high confidence cloud designations in the L1 quality band.
For details on the AT band, the AT-ACCA algorithm, and the AT-ACCA disambiguation tests, see Scaramuzza et al., 2012.
The quality algorithm is calculated by the following two formulas:
SIQS = 9 - [ SNF/ANF * ( NDF / DFBP + NCF / CFBP)]
IIQS = 9 - [(NDF/DFBP + NCF/CFBP)
SIQS = Scene Image Quality Score
IIQS = Interval Image Quality Score
SNF = Standard number of video frames in a scene [7001 for OLI, 2621 for TIRS]
ANF = Actual number of frames in the scene / interval
NDF = Number of dropped frames in the scene / interval
DFBP = Dropped Frame Break Point: dropped frame count at which the quality score drops by one point. 
NCF = Number of video frame CRC failures in the scene / interval
CFBP = CRC Failures Break Point: Video frame CRC failure count at which the quality score drops by one point. 
Numbers in brackets are configurable. If changed, the quality algorithm version is updated and the new values documented
Values: 0–9, where
9 = Best
0 = Worst
-1 = quality not calculated or assessed
Values: 0–9, where
9 = Best
0 = Worst
-1 = quality not calculated or assessed
The L8 instruments, OLI and TIRS, represent an evolutionary advance in technology. OLI builds upon Landsat heritage and technologies demonstrated by the ALI. As such, OLI is a push-broom sensor with a four-mirror telescope and uses 12-bit quantization. The OLI collects 30-meter data for visible, near infrared, and short wave infrared spectral bands as well as provides for a 15-meter panchromatic band. New with OLI is the addition of a 30-meter deep blue Coastal Aerosol band (Band 1) for coastal water and aerosol studies and a 30-meter Cirrus band (Band 9) for cirrus cloud detection. Additionally, the bandwidth has been refined (narrowed) for six of the heritage bands.
The TIRS instrument collects data for two narrow spectral bands in the thermal region, formerly covered on previous Landsat instruments by one wide spectral band. Although TIRS is a separate instrument, the 100-meter TIRS data are registered to the OLI data in order to create radiometrically, geometrically, and terrain-corrected 12-bit data products.
These sensors both provide improved SNR radiometric performance quantized over a 12-bit dynamic range. This translates into 4096 potential grey levels in an image compared with only 256 grey levels in previous 8-bit instruments. Additionally, improved signal-to-noise performance enables better characterization of land cover state and condition.
In addition to Table 2‑1 in Section 2, Figure 4-5 compares L8 spectral bands and wavelength to that of L7 ETM+.
Figure 4-5. Landsat 8 Spectral Bands and Wavelengths compared to Landsat 7 ETM+
L8 acquires high-quality, well-calibrated multispectral data over the Earth’s land surfaces. On average, over 500 unique scenes are acquired per day across the globe and sent to the USGS EROS Center for storage, archive, and processing. All of these scenes are either processed to an L1Gt product or to a standard L1T product. The highest available product derivative is made available for download over the Internet at no cost to users. A complete standard Level-1 product consists of 13 files, including OLI Bands 1-9 (one file per band), TIR Bands 10 & 11 (one file per band), a product-specific metadata file, and a QA file.
LSDS-809 Landsat 8 (8) Level-1 (L1) Data Format Control Book (DFCB) is an excellent reference for L8 product format and information. The following paragraphs summarize the Level-1 data format.
In addition to GeoTIFF, the data incorporate cubic convolution resampling, North Up (Map) image orientation, and Universal Transverse Mercator (UTM) map projection (Polar Stereographic projection for scenes with a center latitude greater than or equal to -63.0 degrees) using the WGS84 datum.
The format of the final output product is a tar.gz file. Specifically, the files are written to a tar file format and then compressed with the gzip application. Of note, the tar file does not contain any subdirectory information. Therefore, uncompressing the file places all of the files directly into the current directory location.
Table 4-1 and Table 4-2 address the standard and compressed file-naming conventions for L8 L1 products.
The MTL file is created during product generation and contains information specific to the L1 product ordered. The MTL file contains identifying parameters for the scene, along with the spatial extent of the scene and the processing parameters used to generate the Level-1 product. This file is a human-readable text file in ODL format. LSDS-809 Landsat 8 (L8) Level-1 (L1) Data Format Control Book (DFCB) provides a complete description of the metadata file. In general, the MTL file includes the following parameters:
The QA Band contains quality statistics gathered from the image data and cloud mask information for the scene. The QA file is a 16-bit image with the same dimensions as the standard L1T scene. Bits are allocated for some artifacts that are distinguishable after the systematic correction (Level-1G) stage of processing. The first bit (bit 0) is the least significant. Subsection 5.4 provides a full description of the L8 QA band.
Used effectively, QA bits improve the integrity of science investigations by indicating which pixels might be affected by instrument artifacts or be subject to cloud contamination. For example, Normalized Difference Vegetation Index (NDVI) calculated over pixels containing clouds will show anomalous values. If such pixels were included in a phenology study, the results might not show the true characteristics of seasonal vegetation growth. Cloud-contaminated pixels will lower NDVI values, and measures such as the timing of ‘green up’ or peak maturity would appear later than they actually occurred. A worse consequence would be that the reported reduction of vegetation growth would be taken as an indicator of environmental change, potentially prompting unnecessary land management policies or practices.
Rigorous science applications seeking to optimize the value of pixels used in a study will find QA bits useful as a first-level indicator of certain conditions. Otherwise, users are advised that this file contains information that can be easily misinterpreted and it is not recommended for general use. Robust image processing software capable of handling 16-bit data is necessary to compute statistics of the number of pixels containing each of the designated bits.
Figure 4-6 . Quality Band (BQA.TIF) displayed for Landsat 8 Sample Data (Path 45 Row 30) Acquired April 23, 2013
The QA image can be stretched to emphasize the light ("1"s) and dark ("0") pixels for a quick view of general quality conditions. In the Crater Lake, Oregon, image above, the lighter pixels are likely to be affected by snow or clouds.