GEOMETRIC PERFORMANCE COMPARISON BETWEEN THE OLI AND THE ETM+ INTRODUCTION

Transcription

1 GEOMETRIC PERFORMANCE COMPARISON BETWEEN THE OLI AND THE ETM+ James Storey, Michael Choate Stinger Ghaffarian Technologies, contractor to USGS EROS, Sioux Falls, SD Work performed under USGS Contract Number 08HQCN Kenton Lee Ball Aerospace Technology Corp., Boulder, CO ABSTRACT The Landsat Data Continuity Mission (LDCM) Operational Land Imager (OLI) is the successor to the Landsat 7 Enhanced Thematic Mapper Plus (ETM+) instrument and features a pushbroom architecture that is more geometrically stable than the whiskbroom scanner of the ETM+. As a tradeoff of this architecture selection, imagery must be terrain corrected to ensure accurate band registration. This paper compares the geometric performance requirements for the two instruments and discusses the implications for the Landsat user community. INTRODUCTION The launch of Landsat 7 on April 15, 1999 continued the series of moderate resolution Earth observation data acquisitions begun in 1972 with the launch of Landsat 1. The Enhanced Thematic Mapper Plus (ETM+) sensor is the imaging payload on the Landsat 7 spacecraft. It uses the whiskbroom scanner architecture common to the Thematic Mapper (TM) sensor family also flown on Landsats 4 and 5, with several evolutionary refinements including the addition of a 15-meter resolution panchromatic band and a higher resolution (60-meter) thermal band. In the TM-series architecture the spectral bands are distributed in the along-scan direction as shown in Figure 1. The bands sequentially image the same ground area as the ETM+ scan mirror sweeps the field of view along the crosstrack imaging swath. Figure 1. ETM+ Instrument and Focal Plane Layout (Irish, 1998). The ETM+ suffered a critical system failure in 2003 when the Scan Line Corrector (SLC) mechanism, responsible for forward image motion compensation, stopped working (Storey et al., 2005). The SLC was subsequently deactivated and imaging operations have continued. Image data collected since the SLC failure exhibit scan-to-scan coverage gaps due to uncompensated spacecraft motion during the ETM+ bidirectional scan cycle. Another change to the ETM+ operational configuration occurred in April 2007 when the scan mirror was switched from its primary scan angle monitor (SAM) operating mode to the backup bumper mode. This action was

2 taken to extend mission life as the scan mirror was beginning to lose synchronization with the internal calibrator mechanism when operating in SAM mode. Fortunately, since the Landsat 5 Thematic Mapper was switched into bumper mode in 2002, the transition to and operation in bumper mode was well understood prior to the decision to switch Landsat 7 (Storey and Choate, 2004). The Operational Land Imager (OLI) will be the primary payload on the Landsat Data Continuity Mission (LDCM) when it is launched to continue the Landsat series of observations. The OLI uses a pushbroom imaging architecture, similar to the Advanced Land Imager (ALI) technology demonstration sensor flown on the EO-1 spacecraft. In this architecture the spectral bands are distributed in the in-track direction, as shown in Figure 2. The remainder of this paper will compare the ETM+ and OLI geometric characteristics and show how the OLI geometric performance requirements were derived from a combination of ETM+ specifications, ETM+ on-orbit performance, and ALI measured performance. Figure 2. OLI Instrument and Focal Plane Module Layout Concept. ETM+ AND OLI SENSOR CHARACTERISTICS Two characteristics of the ETM+ sensor design have a substantial influence on geometric performance: 1. The along-scan layout of the spectral bands ensures that all bands cover the same image area over an approximately 2-millisecond time period. Very little spacecraft motion, or even jitter, occurs over this short time interval, leading to very stable band-to-band registration. 2. Scanning pattern variations due to scan mechanism instability and jitter make maintaining accurate internal image geometric accuracy challenging, especially since the switch to bumper mode where less withinscan mirror telemetry is available. These characteristics are consequences of the cross-track whiskbroom ETM+ design. The pushbroom OLI architecture leads to a somewhat different set of geometric challenges. Instead of using a small focal plane and a scanning mirror, fourteen separate focal plane modules (FPMs) are required to cover the full Landsat cross-track field of view. Each FPM contains nine spectral bands (the seven heritage ETM+ reflective bands plus coastal/aerosol and cirrus bands) spread out in the along-track direction. The along-track spectral band separation leads to an approximately 0.96-second time delay between the leading and trailing bands. This time delay creates a small but significant terrain parallax effect between spectral bands, making band registration more challenging. The along-track dimension of the OLI focal plane (see Figure 3) also makes it desirable to yaw steer the spacecraft. This means that the spacecraft flight axis is aligned with the ground (Earth fixed) velocity vector, rather than with the inertial velocity vector, in order to compensate for cross-track image motion due to Earth rotation. Maintaining this alignment requires a small spacecraft yaw maneuver that varies continuously over the orbit, from zero near the poles to approximately 4 degrees at the equator. This small yaw accounts for Earth rotation during the time delay between leading and trailing bands and during the delay between the leading and trailing FPMs. Although the pushbroom architecture requires many more detectors and a correspondingly larger focal plane, it also allows for a much longer detector dwell time (~4 milliseconds vs. 9.6 microseconds for ETM+), leading to much higher signal-to-noise ratios. The lack of moving parts in the pushbroom design also allows for a more stable imaging platform and good internal image geometry.

3 Along-track Field of view FPM-to-FPM overlap Band order is reversed in adjacent FPMs Figure 3. Orientation of Focal Plane Modules on the OLI Focal Plane. LANDSAT 7 ETM+ GEOMETRIC PERFORMANCE On-orbit ETM+ performance is monitored by the Image Assessment System (IAS), a component of the Landsat 7 ground system (Lee et al., 2004). The IAS routinely monitors three geometric performance characteristics: 1. Band registration accuracy 2. Absolute geodetic accuracy (geolocation accuracy) 3. Internal image accuracy/image registration accuracy A brief review of the observed ETM+ performance with regard to these characteristics will help demonstrate the heritage of the corresponding OLI geometric requirements. ETM+ Band Registration Accuracy Performance ETM+ band registration performance is monitored by the IAS using the results of band-to-band correlation measurements of systematically corrected images over desert sites (Lee et al., 2004). Sites with little vegetation provide less spectral variation across the ETM+ reflective bands and thus yield more reliable correlation results. One exception to this general principle is the emissive (thermal) band, which can exhibit contrast reversals relative to the reflective bands. This effect is mitigated by favoring winter scenes, using northern hemisphere sites from the autumnal to the vernal equinoxes and southern hemisphere sites between the vernal and autumnal equinoxes. Even so, the correlation measurements for the thermal band are significantly less accurate than those for the reflective bands. Based on over 9 years of on-orbit measurements, ETM+ band registration performance exceeds the 0.28-pixel (LE90), or 8.4-meter specification (NASA, 1996). The on-orbit measurements show that the reflective bands are all registered to 0.10 pixels (LE90), or 3.0 meters, or better, and the thermal band is registered to 0.21 pixels (LE90), or 6.3 meters. This level of performance has been stable over the mission life with only two significant calibration updates required. The first band alignment recalibration was required immediately following commissioning due to overall instrument warming once the full operational duty cycle was achieved. The second recalibration was required following the SLC failure to account for the lack of forward image motion compensation in the along-track band alignment. ETM+ Absolute Geodetic Accuracy Absolute geodetic accuracy refers to the geolocation accuracy of systematically corrected images (i.e., images processed without ground control). Landsat 7 geodetic accuracy is limited mainly by spacecraft attitude and ephemeris knowledge. Landsat 7 does not carry an on-board navigation system (e.g., Global Positioning System); it relies instead, on predicted ephemeris data uploaded daily from the mission operations center and then downlinked

4 with the ETM+ data stream. For ETM+ product generation, ephemeris knowledge was improved by deriving postpass definitive ephemeris data each day from the same S-band ranging measurements used to produce the daily ephemeris predictions for the coming day. Geodetic accuracy performance has varied over the life of the mission (see Figure 4) but has always been much better than the 250-meter (1σ) specification (Lee et al., 2004). Initial on-orbit geodetic accuracy performance of meters (1σ) gradually worsened to meters (1σ) due to degrading gyro performance. On March 14, 2005, nearly 10 months after the problematic redundant gyro unit was shut down (on May 5, 2004), accuracy abruptly improved to meters (1σ). Though this improvement coincided with more stable gyro drift rates and is attributed to the gyro shut down, no convincing explanation for the time lag has been proposed. Since the switch to bumper mode on April 1, 2007, accuracy has degraded due to the poorer mirror angle knowledge available in that mode. Frequent (semi-annual in scan angle monitor mode, quarterly in bumper mode) calibration updates have been required to track the time varying alignment between the ETM+ scanning reference system and the spacecraft attitude control system. This is a particular problem when operating in bumper mode because the scan start/stop angles are no longer held fixed or measured. These alignment calibration operations involve analyzing ground control point measurements collected over a global set of calibration sites. Data from many separate acquisitions are analyzed to allow scene-specific attitude and ephemeris knowledge errors to decorrelate, exposing the more slowly varying systematic alignment biases. Figure 4. Landsat 7 Lifetime Absolute Geodetic Accuracy by Calendar Quarter. ETM+ Image Internal Accuracy ETM+ image internal accuracy is measured by comparing precision (i.e., using ground control) and terrain corrected images to reference data either Digital Orthophoto Quadrangle (DOQ) mosaics or other Landsat 7 images. In particular, within-scan errors are measured by extracting a set of test points down the center of each ETM+ scan and correlating them with the DOQ reference image. ETM+ internal accuracy is limited by the stability of the scan mirror drive mechanism and by unmeasured jitter. Performance was stable in scan angle monitor mode but degraded with the switch to bumper mode. The bumper mode scan mirror parameters change continuously and must be recalibrated frequently (every 2 weeks or so). These calibration parameters are predicted in advance and refined retrospectively (Storey and Choate, 2004). The original Landsat 7 specification was for multitemporal image registration accuracy of 0.4 pixels (LE90), or 12 meters (NASA, 1996). The measured ETM+ within-scan accuracy was 7.1 meters (LE90) in scan angle monitor mode. In bumper mode, the internal accuracy is 10.5 meters (LE90) using the predicted calibration parameters and 8.4 meters (LE90) using the refined calibration parameters.

5 OLI GEOMETRIC PERFORMANCE The LDCM OLI geometric performance requirements were based on Landsat 7 ETM+ and EO-1 ALI actual performance. Landsat 7 performance provided the primary basis for the OLI requirements since data continuity was of paramount importance. Lessons learned from the ALI, particularly with regard to band registration, led to some relaxation in that area, based on the assumption that the OLI sensor architecture would be similar to ALI (as it is). Though ALI performance was evaluated following the EO-1 mission launch, nothing comparable to the multi-year performance history compiled by the Landsat 7 IAS is available. The OLI geometric performance requirements were thus derived from Landsat 7 ETM+ performance, with adjustments for the OLI sensor architecture based upon the ALI experience. The OLI vendor is responsible for the end-to-end geometric performance requirements. Contributions to geometric performance from the spacecraft platform, such as attitude and ephemeris knowledge and stability, are explicitly allocated to the spacecraft through an Observatory Interface Requirements Document (NASA, 2007). In addition to the OLI instrument, the OLI vendor is providing the ground processing algorithms that will be used to geometrically correct the OLI data (NASA, 2008). The LDCM/OLI system is required to be capable of collecting image data at nadir and up to 15 degrees off-nadir (cross-track). While the data processing algorithms must be capable of creating OLI data products for either nadir or off-nadir imagery, the OLI geometric performance requirements apply only to nadir-viewing imagery because that reflects Landsat data continuity. The next four sections present the key OLI geometric performance requirements and provide some of the rationale used to define those requirements. OLI Band Registration Accuracy The OLI spectral bands are required to be registered to an accuracy of 4.5 meters (LE90) or better in terrain corrected products (NASA, 2008). Terrain correction is required to compensate for the small but significant alongtrack band-to-band parallax effect (see Figure 5). The ALI bands were registered to 4.0 meters (LE90) or better in terrain corrected products following on-orbit band alignment calibration. (Storey et al., 2004). These ALI results were based on a small number of data sets over test sites with very accurate ground control and terrain data. The ALI image data were also acquired with the EO-1 solar array drive mechanism halted to ensure a stable imaging platform. This was possible with the low imaging duty cycle of the experimental EO-1 mission but is not practical for the operational LDCM. While some additional allowance was made for a potentially less stable spacecraft environment, this is expected to be largely compensated for by collecting more accurate, and more frequent, attitude data and by mounting the attitude determination system s sensors on the OLI instrument deck. The OLI band alignment calibration will be evaluated and, if necessary, updated during on-orbit commissioning and will be monitored during routine operations with calibration updates issued as necessary. The required on-orbit geometric calibration techniques were developed and tested using the ALI (Storey et al., 2004). Initial analyses of the OLI optical and focal plane designs indicate that the line-of-sight modeling approach used for the ALI is suitable for the OLI. Since the ALI did not and the OLI does not include thermal bands, the associated mensuration and calibration accuracy problems encountered in that regard with the ETM+ are not an issue. Including a separate thermal instrument on the LDCM platform would raise a new set of instrument-to-instrument band registration issues.

6 Figure 5. OLI Band-to-Band Terrain Parallax Effect. OLI Geodetic Accuracy Geometrically corrected OLI data are required to be accurate to 65 meters (CE90) (excluding terrain effects) relative to the WGS84 geodetic reference system, without the use of ground control (NASA, 2008). This is equivalent to an accuracy of 30.3 meters (1σ). Performance is driven largely by spacecraft attitude and ephemeris knowledge, which consume most of the geodetic accuracy error budget. Though the OLI-to-spacecraft relative alignment will be measured prelaunch, achieving this level of geolocation performance will require on-orbit estimation of the alignment between the OLI and the spacecraft attitude determination system. This sensor alignment calibration will be performed during commissioning and as needed during routine operations using techniques similar to those applied to Landsat 7 (Lee et al., 2004). To ensure that most of the allowable geolocation error is due to low frequency along- and cross-track biases, there is also a specification for relative, or within-scene, geodetic accuracy of 25 meters (CE90). This limits the residual geodetic error remaining in a scene after the mean along- and cross-track offsets are removed. The relative accuracy requirement controls the effects of, for example, yaw errors and attitude rate errors. OLI Image Internal Accuracy The 12-meter (LE90) multitemporal image registration accuracy requirement imposed on Landsat 7 ETM+ data is also levied on the OLI. Due to the inherent internal stability of the OLI sensor architecture, it is not expected to be a stressing requirement. This requirement primarily limits the errors associated with short-term attitude knowledge (i.e., gyro performance) but also requires accurate knowledge of the OLI instrument s internal geometry. The relative alignment of the 14 FPMs on the OLI focal plane will be evaluated during on-orbit commissioning. Any residual misalignment will be corrected through updates to the OLI geometric calibration. This focal plane alignment will be monitored and updated as necessary throughout the mission life using techniques developed for the ALI (Storey et al., 2004). OLI Terrain Corrected Product Geometric Accuracy Though necessary to support the image registration requirement, internal image geometry is more tightly constrained by the terrain corrected product absolute geometric accuracy specification. This requirement applies to OLI data that have been geometrically corrected using ground control and digital elevation data. It was written to explicitly assume GPS-quality ground control and SRTM-quality elevation data are used in the correction process. Specifying the expected ground control and elevation data accuracy bounds the error contributions from these supporting data sources. Precision (ground control) and terrain corrected OLI data must be accurate to 12 meters (CE90) or better relative to the WGS84 geodetic reference system (NASA, 2008). There is no corresponding requirement on ETM+ data, but based on the ETM+ internal geometric accuracy results measured using DOQ reference imagery, ETM+ terrain corrected product accuracy is believed to be within 15 meters (CE90) in all operating modes with scan angle

7 monitor mode data being somewhat more accurate than bumper mode data. Achieving this specification requires spacecraft stability and accurate short-term attitude knowledge, OLI internal stability, geometric calibration accuracy, and highly precise geometric correction algorithms. Along with band-to-band registration accuracy, this requirement is believed to be the most challenging of the OLI geometric performance requirements. As a practical matter, GPS-quality ground control is not available globally, so this specification may more realistically be interpreted as registration accuracy to the best available ground control source. This control source will most likely be the Global Land Survey of 2000 (GLS2000) data (Gutman et al., 2008). Given the GLS2000 data accuracy limitations (20 meters RMSENet or 30.3 meters (CE90)), using it as control for OLI data products cannot be expected to yield 12-meter (CE90) absolute (WGS84) accuracy globally, but it will ensure that the OLI data are all registered to a common, internally consistent reference system. It is expected that, over time, the inherent geodetic accuracy of the OLI data will help identify and correct substandard regions in the GLS data set. COMPARISON OF ETM+ AND OLI The preceding sections briefly reviewed the key geometric performance characteristics of the ETM+ and the OLI. Table 1 presents a side-by-side comparison of the ETM+ specifications and actual performance and the OLI specifications. Table 1. ETM+ and OLI Geometric Performance Requirement ETM+ Specification ETM+ Performance OLI Specification Band Registration 8.4 m LE m LE m LE90 Accuracy (see note 1) Absolute Geodetic m CE m CE90 65 m CE90 Accuracy (see note 2) (see note 3) Relative Geodetic N/A 17 m CE90 25 m CE90 Accuracy Image Registration 12 m LE m LE90 12 m LE90 Accuracy (see note 4) Geometric (Terrain N/A 15 m CE90 12 m CE90 Corrected) Accuracy (see note 5) Notes: (1) Reflective band registration. (2) Specified as 250 meters 1σ. (3) Varied with gyro state of health. (4) Bumper mode performance. (5) Estimated. In most cases, the OLI specifications are tighter than their ETM+ counterparts, reflecting actual ETM+ performance and the expected benefits of improved geometric stability offered by a pushbroom sensor architecture (Table 1). The lack of a moving scan mirror and the associated jitter should lead to improved internal image accuracy. A spacecraft with Global Positioning System navigation and modern star trackers should provide geolocation accuracy as good as or better than Landsat 7. The OLI requirements for relative (within-scene) geodetic accuracy and for terrain corrected product absolute accuracy have no ETM+ equivalents. Nevertheless, the OLI specifications were based upon ETM+ performance data collected by the Landsat 7 IAS. The notable exception to this general improvement is band registration accuracy, which is expected to be more challenging for OLI because more time is required for ground targets to be viewed by all of the OLI spectral bands. This time delay leads to band-to-band terrain parallax effects and increases the sensitivity to short-term attitude stability. In the case of band registration, the OLI requirements leaned heavily on the performance of the ALI. Unlike ETM+, all OLI products will be terrain corrected to ensure band registration accuracy. This imposes additional computational burdens on the ground system and requires the assembly and management of a global digital elevation model. Fortunately, the availability of the SRTM data and the inclusion of this new terrain data source in the elevation model used for the GLS 2000 processing has provided just such a data set. Of more direct concern to data users, the band-to-band parallax effect also means that the OLI spectral band data are not, in general, inherently registered and that nearest-neighbor resampling cannot provide band-registered products. The standard

8 OLI product will therefore be a terrain corrected image, generated using GLS2000 control, created using cubic convolution resampling. The availability of these LDCM data products will not present as great a change to the user community as it would have in past years because the Landsat program is already implementing a standard terrain corrected cubic convolution product based on the GLS2000 reference as part of its Web-enabled data initiative (USGS, 2008). SUMMARY The Landsat 7 ETM+ met or exceeded, often by comfortable margins, its geometric performance requirements. The OLI geometric performance requirements are derived from a combination of the ETM+ requirements, ETM+ on-orbit performance measured by the Landsat 7 IAS, and ALI on-orbit performance. In most cases, the OLI specifications are tighter than their ETM+ counterparts. This reflects actual ETM+ performance and the expected benefits of improved geometric stability offered by a pushbroom sensor architecture. The notable exception to this is band registration accuracy, which is expected to be more challenging for OLI. All OLI products will be terrain corrected to compensate for band-to-band parallax effects. This processing will maintain acceptable band registration accuracy in regions of high terrain relief. The OLI specifications include a requirement for terrain corrected product absolute accuracy that has no ETM+ counterpart. REFERENCES Irish, R., Landsat 7 Science Data Users Handbook. Updated January 7, Storey, J., P. Scaramuzza, G. Schmidt, and J. Barsi, Landsat 7 scan line corrector-off gap-filled product development. Proceedings of Pecora 16: "Global Priorities in Land Remote Sensing," October 23 27, 2005, Sioux Falls, South Dakota. Storey, J.C. and M.J. Choate, Landsat-5 bumper-mode geometric correction. IEEE Transactions on Geoscience and Remote Sensing, 42(12): Lee, D.S., J.C. Storey, M.J. Choate, and R.W. Hayes, Four years of Landsat-7 on-orbit geometric calibration and performance. IEEE Transactions on Geoscience and Remote Sensing, 42(12): NASA GSFC, Landsat 7 System Specification, Version H, NASA Goddard Space Flight Center, Greenbelt, MD. NASA GSFC, LDCM Observatory Interface Requirements Document, Revision B, Document # , NASA Goddard Space Flight Center, Greenbelt, MD. NASA GSFC, LDCM Operational Land Imager Requirements Document, Revision C2, Document # , NASA Goddard Space Flight Center, Greenbelt, MD. Storey, J.C., M.J. Choate, and D.J. Meyer, A geometric performance assessment of the EO-1 advanced land imager. IEEE Transactions on Geoscience and Remote Sensing, 42(3): Gutman, G., R. Byrnes, J. Masek, S. Covington, C. Justice, S. Franks, and R. Headley, Towards Monitoring Land-Cover and Land-Use Changes at a Global Scale: The Global Land Survey Photogrammetric Engineering and Remote Sensing, 74(1): U.S. Geological Survey, Landsat Archive to Be Accessible at No Charge. Updated July 31,