This site has moved to www.usgs.gov/landsat
Please visit the new site and update any bookmarks you have. 
 
Top Page    Section 1    Section 2    Section 3    Section 4     Section 5    Section 6    Appendix A     Appendix B     Appendix C    Appendix D    Appendix E    References

Section 4 - Long-Term Acquisition Plan (LTAP)

4.1 Overview

Note: The following description was applicable to Landsat 7 operations from launch through December 2012, and is included for historical interest. In 2012, to make best use of an aging instrument, a new approach was implemented, dubbed “continental landmass scheduling”. The new approach is described in Section 4.2.

The Landsat 7 LTAP, in conjunction with the Mission Operations Center (MOC) Scheduling System software, automates the acquisition of Landsat scenes to periodically refresh the global archive of sunlit, substantially cloud-free land images. (From this point on, the use of “scheduler” includes both the scheduling software and the LTAP data files.) By applying a set of algorithms on a daily basis, the scheduler is designed to ensure optimal collection of Landsat 7 ETM+ imagery for scientific applications, while minimizing the effects of cloud-cover and system constraints. This section provides a brief overview of the LTAP strategy and details the specific algorithms and input files (see Section 4.1.8) used for calculating scene acquisition priorities. For a more complete review of the LTAP, including the science justification for the approach and performance results, see Goward et al. (1999), Gasch and Campana (2000), and Arvidson et al. (2001).

The ETM+ sensor does not continually acquire imagery data as it orbits the Earth. Instead, acquisitions are scheduled in advance using the LTAP data files in conjunction with a software scheduler. The WRS-2 system divides the Earth into a grid of 57,784 scenes—233 paths by 248 rows. However, due to the satellite’s slight inclination away from North-South, the Polar Regions are not completely covered. The Landsat 7 satellite is operated such that it follows the WRS-2 grid within tight tolerances, overflying the entire scene grid every 16 days. A pre-launch database of just over 14,000 scenes containing "land" was compiled for the LTAP, including continental areas, shallow coastal waters, Antarctic sea-ice, and all known islands and reefs. Within any given 24-hour period, approximately 850 of these land scenes (located along descending, sunlit paths only) are in view of the ETM+ and are candidates for acquisition. At launch, mission resource limitations (see Factor 7) restricted the daily acquisition volume for the U.S. archive to 250 scenes; improved resources now allow Landsat 7 to collect ~440 scenes per day (ca. 2017). Given the resources available initially, the mission scheduler selected the "best" 250 scenes for acquisition each day within these constraints. The scheduler automatically selected the best scenes in accordance with the LTAP, basing these decisions on cloud-cover forecasts, urgency of acquisition, and availability of resources to optimize fulfillment of the overall Landsat 7 mission goals.

From the outset of the mission, the basic scheduling approach has been to identify the candidate scenes to be considered, to raise or lower their base priority (as assigned in the LTAP data files) through consideration of several factors, to use the resulting “dynamic” priority to form a priority-ordered list, and to schedule the top ~250 (now ~440 (ca. 2017)) scenes to be collected. The factors considered during scheduling are:

  1. Seasonality of vegetated regions, as well as niche-science communities with specific acquisition needs
  2. Predicted versus nominal cloud-cover (forecast versus statistical assessments)
  3. Solar elevation angle (especially important at high latitudes)
  4. Missed opportunities for previous acquisitions
  5. Quality (based on cloud-cover assessments) of previous acquisitions
  6. Scene clustering (for continuity of swath observations)
  7. System constraints (e.g., instrument duty cycle, ground station locations and functionality, recorder capacity)

Factors 1-6 are used directly for calculating the priority of each potential acquisition, while the actual acquisition list is obtained by incorporating system constraints (Factor 7). Factor 4 requires feedback from previous scheduling runs, while Factor 5 requires feedback from cloud cover assessment of archived scenes. Each of these factors are discussed in the following sections.

4.1.1 National Centers for Environmental Prediction (NCEP)

The NCEP provides timely and continually improving worldwide forecast guidance products. NCEP, a critical part of NOAA’s National Weather Service, is the starting point for many of the weather forecasts in the U.S. NCEP generates weather related products including the cloud cover predicts used by the MOC for ETM+ image scheduling.

4.1.2 Seasonality

The LTAP was designed to help the scheduling software select acquisitions more frequently during periods of land cover change, such as growth and senescence of vegetation, and less frequently during relatively stable periods, such as when full growth canopy exists or during winter quiescence. An eight-year Advanced Very High Resolution Radiometer (AVHRR) Normalized Difference Vegetation Index (NDVI) data set was used to determine when and where change was occurring around the globe (Goward, et al., 1994, 1999). To flag seasonal change within a WRS-2 scene, a statistical test was applied to each NDVI 45-km sample mapped to that scene. Comparing each AVHRR-derived imagery sample or pixel to the same pixel two months later revealed periods of significant change in the NDVI's mean and standard deviation. This approach identified when multiple ETM+ acquisitions may be needed, versus periods of minimal change that may require only a single acquisition.

Based on this analysis, the year is broken into a set of temporal "windows" for each path-row location. During each window, a location may be labeled as "acquire once", "acquire always", or "never acquire", and this information is stored in the LTAP seasonality file. It should be noted that "acquire always" does not imply that an actual acquisition will occur for every overpass during that time window. Rather, that scene will always be a candidate for acquisition, and other factors within the scheduling process will govern whether or not an acquisition actually takes place. Conversely, "acquire once” means that once a quality, cloud-free scene is acquired, that path-row location is no longer eligible for acquisition until the next time period begins.

Niche science communities are defined as users with specific acquisition requirements that do not necessarily correlate with those derived using the NDVI approach. Basically, defining acquisitions solely through semi-monthly NDVI change does not fully capture the science and user interest in the Landsat mission. To supplement the NDVI-based seasonality metric, niche science communities have defined time windows more appropriate to their land cover and application requirements. These niche communities initially sponsored the following locations and associated time windows:

  • 282 agriculture areas (acquire every season if cloud cover predict < 60%)
  • 35 calibration sites (acquire "always")
  • 896 reefs (acquire from 2x to 6x each year)
  • 30 fire-impacted areas (acquire "always")
  • 1392 land ice scenes other than Antarctica (acquire once during certain months)
  • 3601 Antarctica scenes (acquire once during January-February)
  • 60 oceanic islands (acquire twice each year)
  • 1175 rainforest areas (acquire "always" all year)
  • 352 sea ice scenes (acquire from 1x to 3x each year)
  • 11 Siberia scenes (acquire "always" over 9 months)
  • 72 volcanoes (acquire from 2x to 12x during year, including night)

In general, approximately 27 percent  of the 250 scenes/day acquired by Landsat 7 are devoted to satisfying "niche" requests, approximately 70 percent are devoted to satisfying routine acquisitions from the seasonality file, and 3 percent are devoted to night images or special high-priority acquisitions (natural disasters, national needs, etc). Requirements for night imaging and high-priority acquisitions, such as for disaster response, are outside the scope of the LTAP and are handled by the USGS Data Acquisition Manager (DAM) at the Goddard Space Flight Center (GSFC) who may submit special requests to the MOC for inclusion in the schedule.

4.1.3 Predicted Cloud Cover

A basic goal of Landsat 7 scheduling is cloud avoidance. The scheduling software compares near-term predictions from NCEP to the historical average cloud cover for that scene for that month, to determine acquisition dynamic priorities each day. Scenes with better-than-average cloud-cover predictions are given a priority increase, while those with poorer cloud cover predictions are given a lower priority.

Global historical cloud data (climatology) are obtained from the International Satellite Cloud Climatology Project (ISCCP) records (Rossow et al., 1996), which give monthly estimates of cloud-cover percentage for 2.5° grid cells from 1989 to 1993. Monthly averages over the initial five-year dataset (1989-1993) were computed and mapped to the Landsat WRS-2 grid to produce a cloud climatology. This data set has been extended over the years. Studies have been made to evaluate replacement of the ISCCP data with a Terra Moderate Resolution Imaging Spectroradiometer (MODIS) data set of assessed cloud cover data. The advantages of using MODIS data are the higher resolution and that MODIS values are computed at the same time of day as Landsat data are acquired (see Section 2.1.1) whereas the ISCCP data was based on diurnal (24-hour) composites. The MODIS data set under consideration is currently 13 years long; 2001–2014.

Once received from NCEP, these cloud predictions are translated to the Landsat WRS-2 coordinate system. The forecast nearest in time to a candidate acquisition is compared to the climatology value for that day to determine whether the predicted observing conditions are better or worse than typical for that location. Acquisition dynamic priority is then adjusted upward to favor the better conditions or downward to avoid the worst conditions. Cloud avoidance has so far been successful.

4.1.4 Solar Elevation Angle

Requests for high-latitude scenes are typically rejected during local winter when the solar elevation is below a certain threshold. This angle, between the scene center, the horizon, and the center of the Sun's disc, varies with time and with latitude and local elevation. This threshold was set at 5° during the initial years of the Landsat 7 mission in LTAP. On July 24, 2002, the cutoff angle was changed to 15° for the northern hemisphere because of duty cycle concerns and the fact that snow dominates scenes acquired at lower (i.e., winter) solar elevation angles. The cutoff angle remains at 5° for the southern hemisphere to avoid coverage impacts over the bottom third of Argentina and bottom half of New Zealand’s South Island.

4.1.5 Missed Opportunities

An acquisition request is granted a priority increase as a function of the number of consecutive past cycles in which the opportunity to acquire the requested scene was not fulfilled. This occurs when the scene was not scheduled for acquisition due to duty cycle or other impacts, or when the acquired image fails to meet minimum quality standards, typically due to cloud cover. For example, if the last successfully acquired image of a scene was 48 days ago, then a request for this scene is granted a priority increase based on two missed opportunities from 32 and 16 days ago. In addition, all new requests submitted when an acquisition window opens are also given a priority increase to help them compete with established requests.

4.1.6 Image Quality

Image quality is determined during image processing at EROS and expressed as the amount of cloud cover assessed within the image, as well as image integrity in terms of missing data and other image artifacts (see Appendix A)—is fed back to the MOC daily. The quality of the most recently acquired image of a path/row from a previous cycle also factors into the dynamic priority for another acquisition. Taking advantage of scene overlap at high latitudes, the scheduler also considers the quality of the best image that could be spliced together as a mosaic of imagery of adjacent scenes acquired in the recent past.

A previous acquisition with a cloud cover score of 10 percent or less is considered a successful acquisition and resets the priority for another acquisition attempt back to the base value. Conversely, a previous acquisition with a cloud cover score of greater than 60 percent is considered a missed opportunity and the dynamic priority for another acquisition attempt is raised. A sliding scale is applied for values between 10 and 60 percent.

After the failure of the Scan Line Corrector (SLC) mechanism (see Section 2.2.3.3), another quality parameter was added in 2004; a gap phase statistic. It characterizes the location of the scan gap pattern within each image relative to the scene center. This enables users to identify image pairs that when used together would fill in the gaps left by the mechanism failure. The scheduler uses the gap phase statistic to identify potential gap-closing image pairs in previous acquisitions. When a new acquisition candidate is being considered for scheduling, the existence of a good gap-closing image pair lowers the desirability of acquiring a new candidate, while lack of a good image pair raises the priority for acquiring a new candidate that might better pair with an archived acquisition.

4.1.7 Clustering

Within the Landsat 7 scheduling process, a higher priority is given to scenes that form "clusters", or contiguous groups of along-path acquisitions. Most importantly, this reduces the on-off cycling of the ETM+ power supply and prolongs instrument life. It also promotes archiving of continuous swath data.

4.1.8 Long-Term Acquisition Plan Files

The contents of the following files represent the underpinnings of the LTAP:

  1. Seasonality file
  2. WRS-2 land data base
  3. Nominal cloud cover file
  4. Nominal cloud cover daylight additions file
  5. Nominal gain settings file
  6. Maximum solar zenith angle

1. The seasonality file specifies which WRS-2 scenes are to be acquired during which periods of time (request period), and the frequency of acquisition during those periods. Frequency of acquisition is defined as either "once" during the request period, or "all" opportunities during the request period.

The following is the warning label associated with the seasonality file:

The scheduler has many inputs and priorities that it must process during consideration of requests for scheduling, including:

  • predicted cloud cover as compared to the nominal
  • number of missed opportunities for this request since it was opened
  • how good the cloud cover assessment score was for the last acquisition
  • nearness to end of the request period
  • availability of resources including duty cycle, onboard recorder space, and station contact time

The result of this decision process is that scenes marked with an "all" opportunities frequency are usually acquired every 4-5 cycles (64-80 days) instead of every cycle (16 days). Another outcome is that scenes marked with a "once" frequency may be acquired multiple times within the request period in an attempt to acquire an image with 10 percent or less cloud cover—the definition of a successful acquisition. As a consequence, science data users should treat the frequency assignment in the seasonality file as a guide, not a rule.

2. The WRS-2 land database identifies those WRS scenes containing land or shallow water, which is imaged at least once every year. At the end of the file are separate tables for the niche communities and special interests. These include: Earth Observing System (EOS) validation sites, glaciers, MODIS fire validation sites, political (disputed sites), volcanoes, and oceanic islands. A comprehensive list of coral reefs was added after the launch of Landsat 7.

3. and 4. The nominal cloud cover file reports the average cloud cover for each WRS-2 scene for each month of the year. The file spans one year and addresses the descending rows only. The nominal cloud cover daylight additions file adds those ascending rows that will be in daylight during some part of the year and therefore may be imaged. The initial set of average cloud cover for years 1989-1993 was derived from the ISCCP-D2 data set, described in Section 4.1.3. This file is updated as subsequent years are processed; currently, it covers the period between July 1983 and June 2006. The columns in the nominal cloud cover files are: day of the year for the first day of the month | day of the year for the last day of the month | path | row | cloud cover value. All the cloud cover values are given for Row 1 across all paths, followed by Row 2, etc.

5. The nominal gain settings file identifies the gain settings used as defaults for each WRS-2 scene for each day of the year. This file spans one year. Although the scheduler uses the Day Of Year (DOY) value, the DD-MMM values are also included for those who do not easily equate a DOY value of 177 with the date 25-JUN. The file contains settings of H for high gain and L for low gain for both descending and ascending rows. The order in which the settings are specified for each scene refers to bands 123456678, where Band 6 is repeated twice, once for Format 1 and once for Format 2 (see Section 2.2.2), and Band 8 is the panchromatic band. Band 6-1 and Band 6-2 are constant at LH and Band 8 is constant at L. The other bands are changed in groupings, with Bands 1, 2, and 3 changed together, Band 4 changed on its own merits, and Bands 5 and 7 changed together. Ascending rows that are imaged at night use the following default settings: HHHHLLHLL. The file only contains entries showing when the default gain setting has changed from its previous setting. The default gain settings are generated using rules that take land cover type and sun angle into account. Other files influenced by these rules are:

  • desert mask
  • arctic mask
  • monthly gain maps

6. Initially, the maximum solar zenith angle was a single value for all scenes at all times of the year. Daylight imaging was not scheduled if the angle was 85 degrees or more. (This is the same as 5 degrees or less in solar elevation angle from the Earth's surface.) In 2002, this was split into two values: 75 degrees for the northern hemisphere and 85 degrees for the southern hemisphere (15 and 5 degrees of solar elevation angle, respectively).

4.2 Continental Landmass Scheduling

In February 2014, ETM+ continental landmass scheduling was implemented to help extend the ETM+ lifetime and make best use of aging resources. These changes are partly due to the availability of Landsat 8 imagery in 2013. The primary characteristics of this approach are:

  1. Only continental land (not including Antarctica) from Row 10 southward and a few major islands are imaged (not Greenland). No small islands or reefs are acquired. No night imaging occurs except by special request. This results in longer intervals and fewer ETM+ on/off cycles.
  2. The daily quota of scenes acquired is essentially disabled due to these changes. Onboard resources (ETM+ duty cycle, Solid State Recorder (SSR) capacity)) and communication link availability are the determining factors for how many scenes are acquired daily. Additional downlink opportunities were added to support increased acquisitions.
  3. All daylight scenes are candidates at every opportunity; there are no longer any seasonality windows being applied.
  4. Cloud avoidance is active.
  5. Use of a gap phase statistic to acquire gap-filling image pairs has been disabled.

In March of 2016, the ETM+ duty cycles were relaxed by 5 percent to allow occasional excursions as needed to acquire the maximum scenes possible. By 2017, acquisitions under this approach were at an average of ~440 images a day, with peaks just over 500 images per day during the Northern Hemisphere summer. See Section 5.7.3.2 for an illustration of continental coverage into mid-2017.


Top Page    Section 1    Section 2    Section 3    Section 4     Section 5    Section 6    Appendix A     Appendix B     Appendix C    Appendix D    Appendix E    References