Effective October 1, 2013, the calibration of the Landsat 7 Enhanced Thematic Mapper Plus (ETM+) Band 6 will be updated to correct a bias error that has been present in all thermal data since the last calibration update in 2010. Vicarious calibration teams detected a bias error of 0.036 watts per square meter steradian micron (W/(m2 sr µm)) that causes overestimation of top-of-atmosphere temperatures by approximately 0.26 Kelvin (K) at 300K. Calibration parameters within the Calibration Parameter File (CPF) have been adjusted to correct for this bias error.
All data processed by USGS/EROS from October 1, 2013 forward will be correctly calibrated within 0.4K at 300K using updated CPF coefficients. Users can correct for the bias error themselves by subtracting 0.036 W/(m2 sr µm) from the top-of-atmosphere radiance product for data products processed between January 1, 2010 and September 30, 2013.
The vicarious calibration teams, the Rochester Institute of Technology and the NASA/Jet Propulsion Laboratory, have been collecting field data coincident with the satellite overpass since the ETM+ launched in 1999. Early in the mission, an offset error of 0.31 W/(m2 sr µm) was detected [Barsi, 2003], and subsequently a correction was applied to the processing system in October 2000. Further updates were made in January 2010 [Schott, 2012] to correct a 5.8% gain error.
Continued calibration monitoring has revealed that the 2010 calibration update resulted in a small but statistically significant bias error of 0.036 W/(m2 sr µm). The most recent October 2013 update corrects this error by adjusting the shutter view coefficient in the Calibration Parameter File (CPF).
The CPF biases were also adjusted for non-default processing conditions, but for most users, the view coefficient is the only parameter that will affect the product. Users should note that with the current distribution system, the downloadable product may not have been processed with the latest CPF. Please check the “Product Creation Time” in the metadata file (MTL) that comes with the downloaded product to ensure that the product is processed after October 1, 2013.
Users can correct their own data by subtracting 0.036 W/(m2 sr µm) from the top-of-atmosphere radiance product for data products processed between Jan 1, 2010 and Sept 30, 2013.
J.A. Barsi, J.R. Schott, F.D. Palluconi, D.L. Helder, S.J. Hook, B.L. Markham, G. Chander, E.M. O’Donnell, “Landsat TM and ETM+ thermal band calibration,” Canadian Journal of Remote Sensing, 29(2), 141-153 (2003) John R. Schott, Simon J. Hook, Julia A. Barsi, Brian L. Markham, Jonathan Miller, Francis P. Padula, Nina G. Raqueno, “Thermal Infrared Radiometric Calibration of the Entire Landsat 4, 5, and 7 Archive (1982-2010).” Remote Sensing of Environment, July 2012.
Analysis of large, unchanging desert sites in the Sahara has shown that the gains of the ETM+ instrument have degraded over the lifetime of the sensor. The degradation has been slow, a maximum of -0.21% change per year, but over 13+ years of operation, this amounts to calibration errors of up to 2.8%. The degradation is significant in all reflective bands.
The effect of the degradation is to make targets appear darker over time. For example, in the Blue band, a 30% reflector in 1999 would appear to have a reflectance of 29.5% in 2006 and 29.2% in 2012.
Our current, best estimates of the gain change come from the ETM+ lifetime response to path 181 row 40, encompassing the Libya-4 site (referenced in Cosnefroy et al. 1996). The gains are modeled as linear functions of time. The gain changes below are listed here as percent (of the starting gain) change per year along with a 2-sigma uncertainty and the degradation in percentage over the lifetime (relative to the current CPF):
A more detailed description of the processing and analysis can be found in the proceedings of IGARSS 2012, though the numbers here are slightly different due to number of data points being used in the final assessment.
As of April 1, 2013, Landsat 7 processing will correct data for this degradation.
H. Cosnefroy, M. Leroy, and X. Briottet, "Selection and characterization of Saharan and Arabian desert sites for the calibration of optical satellite sensors," Remote Sensing of Environment, Vol. 58, pp. 101-114, 1996.
Barsi, Markham, Helder. "In Flight Calibration of Optical Satellite Sensors Using Pseudo Invariant Calibration Sites" IGARSS 2012, Munich, Germany, July 22-27, 2012.
Effective January 1, 2010, the calibration of the ETM+ thermal band was modified to correct for a lifetime gain error detected by the vicarious calibration teams. Changes were made to the Calibration Parameter File (CPF) to correct a gain error that has been present since the launch of the instrument. This gain error causes the thermal band to predict too hot at cold temperatures and too cold at hot temperatures. For instance, a target with an at-satellite temperature of 0C will appear to be 0.8C based on the image data and a target with an at-satellite temperature of 27C will appear to be 26.3C. The error at 12C is approximately 0C. The vicarious calibration data were used to generate a correction to the calibration and the corrected coefficients were implemented in the processing system on Jan 1, 2010. All data processed by the USGS/EROS processing system after Jan 1, 2010 are correctly calibrated.
The vicarious calibration teams, the Rochester Institute of Technology and the NASA/Jet Propulsion Laboratory, have been collecting data under the satellite since the instrument launched in 1999. Early in the mission, they detected an offset error of 0.31 W/m2 sr um [Barsi, 2003]. This was corrected in the processing system in October 2000. Since then, the teams have continued to collect field data. The addition of a site on the Salton Sea, which can get as hot as 35C in the summer, expanded the temperature range over which the teams acquired ground truth. With these hot points, a gain error of 5.8% became statistically significant. The effect of this gain error is that the instrument was estimating that hot targets were too cold and cold targets were too hot. See the table below for estimates of the error on specific target temperatures.
The error was corrected by adjusting the internal-to-external gain ratio in the Calibration Parameter File (CPF). Other parameters needed to be adjusted for non-default processing conditions, but for most users, the internal-to-external gain ratio is the only parameter that will affect the product. Users should note that with the current distribution system, the downloadable product may not have been processed with the latest CPF. Please check the "Product Creation Time" in the metadata file (MTL) that comes with the downloaded product to ensure that the product is processed after January 1, 2010.
This update will be modified when the correction is published in an official paper.
Table 1. Error estimates for at-satellite radiances. The error varies with target temperature with the satellite estimating too hot for cold targets and too cold for hot targets.
[W/m2 sr um]
|273K / 0C||6.0||0.8 too hot|
|285K / 12C||7.5||0|
|300K / 27C||9.0||0.7 too cold|
J.A. Barsi, J.R. Schott, F.D. Palluconi, D.L. Helder, S.J. Hook, B.L. Markham, G. Chander, E.M. O'Donnell, "Landsat TM and ETM+ thermal band calibration," Canadian Journal of Remote Sensing, 29(2), 141-153 (2003)
As of April 1, 2007, the Landsat 7 ETM+ Calibration Parameter File (CPF) format was updated to include several new parameters and groups, in support of the L7 ETM+ change to bumper mode operations.
For a full description of the contents and format of the L7 ETM+ CPF file, see the CPF Definition Document - .pdf (1.00 MB).
Previously, two Landsat 7 science team investigator groups reported a bias of 0.31 w/(m^2 sr um) in the calibration of Band 6 of the Enhanced Thematic Mapper Plus (ETM+). To compensate for this bias, the following occurred with the October 1, 2000 release of the CPFs:
Users need to be aware of which changes were made and how these changes might affect the distributed data.
The software change that incorporated the Band 6 bias calibration fix was installed in the IAS system October 30, 2000, and in the LPGS system December 20, 2000. Because the National Landsat Archive Production System (NLAPS) relies on the CPF for Band 6 bias calculation, NLAPS was corrected with the October 1, 2000 release of the CPFs. Data processed after these dates are calibrated correctly. For data processed before these dates, subtract 0.31 W/(m^2 sr um) from the L1 image radiance to get calibrated radiance.
Note that the distinction between calibrated and non-calibrated data is based on processing date. This has no relation to the acquisition date (the date that the ETM+ acquired the image). Data acquired before 01-Oct-00 but processed after the calibration correction are calibrated correctly.
For users of other processing system Band 6 products, the effect depends on the calibration equation used for processing. (1) If the CPF gains and CPF biases are used for normal operational processing, this calibration correction occurs immediately on all images ordered. (2) If the original calibration equation used the shutter view factor (Vsh) (and IC biases are used), the new Vsh values have an effect, although the specific implementation of Vsh varies and has differing effects. Because only the IAS and LPGS were developed under the Landsat Project Science Office, other processing software is not under configuration and verification control. In addition, when changes are made to LPGS software, other processing systems are under no obligation to make similar changes to their software, so other systems may never implement new calibration methods.
Summary of Changes
Update to the CPF (view coefficients and biases): A number of coefficients were determined during the prelaunch calibration of ETM+, including the view factors of the individual instrument components and the ratio between the internal and external gain. Based on the error in calibration, the prelaunch data are now considered suspect. Therefore, it was decided to empirically re-calculate one of the coefficients to correct the bias. Also during this time, the groups observed that the Vsh was not used correctly in the calibration software algorithms. This algorithm was corrected, and Vsh is the coefficient that was empirically adjusted to yield a 0.31 w/(m^2 sr um) lower bias.
Biases: Accompanying this change in Vsh was a change in the CPF biases. This is the simplest way to adjust for the bias error, but the LPGS processing system defaults to using the IC as the bias source. The LPGS only uses these biases for special requests. However, other processing systems may use them, so updating them was necessary. The biases were adjusted by determining the DN equivalent of the 0.31 w/(m^2 sr um) radiance change from the current gain, then adjusting the bias by that amount.
Update of the Calibration Equation
The processing software originally used a calibration equation that relied heavily on information collected through prelaunch tests. Because the prelaunch calibration results are believed to be the cause of this bias error, the calibration equation was simplified to its most basic form.
Lsc = (Qsc - Qsh) / RGi + (Vsh/R)Lsh - sum(ajLj)
In the ideal situation, new on-orbit data will be collected with the instrument sitting at a range of temperatures so the ajs, Vsh, and R can be re-derived. Until that opportunity becomes available, the solution is to re-solve for Vsh based on the ground look calibration that the science team members already perform; keeping the result of the equation the same as the current equation:
Vsh(CPF) = R ( Vsh(emp) + sum(aj) )
Vsh(CPF) is the new Vsh that will be put in the CPF.
Vsh(emp) is the Vsh solved for when the ground truth data is used
This new equation was implemented in IAS 4.3 (Oct 30, 2000), LPGS 4.1 (December 20, 2000), and NLAPS (October 1, 2000). Band 6 data processed on or after those dates are correct and need no adjustment. For data processed before those dates, adjust the calibrated radiances by subtracting 0.31 W/(m^2 sr um).
On November 24, 1999 at 00:37:37Z, an incorrect command to the Landsat 7 spacecraft caused the Enhanced Thematic Mapper Plus (ETM+) to switch gain states for Band 6. As the only band that uses both formats, Band 6 Low Gain data normally processes through Format 1, while Band 6 High Gain data normally processes through Format 2. The incorrect command load reversed this arrangement; Format 1 became High Gain, while Format 2 became Low Gain. The original gain settings were restored on December 8th at 16:01:15Z.
Data during this event was reprocessed with an Event-Specific Calibration Parameter File (CPF), and cannot be not be processed to Level 1 by Image Assessment System (IAS) software, and may have incorrect Automated Cloud Cover Assessment (ACCA) scores.
Switching the gains in Band 6 is not a simple matter of substituting Format 1 for Format 2. Different detectors and circuitry are used in the different formats, so the calibration of a gain state in one format does not relate to its calibration in another format. Essentially, two new bands were created by switching the gain states, both of an intermediate gain between High and Low, and both of which were unstudied and uncalibrated at the time of the event.
However, these data cannot be processed to Level 1 with the existing CPF and the current IAS software. The current version of the IAS software makes assumptions about the gain state of Band 6 data, and processing this data to Level 1 Geometrically Corrected (L1G) leads to corrupted imagery.
The High Gain data has striping in the L1Gs 'corrected' image. The Low Gain data is totally saturated, having been scaled to inappropriate values.
The Level 0 data for both formats should be considered to be in new, intermediate gain states. The gain states for the Band 6 data are correctly marked in the L0R metadata. This data should only be processed using the Event-Specific CPF.
In addition, because the Automated Cloud Cover Assessment system uses Band 6 data, the ACCA score for scenes during this event may be incorrect.