Abstract
Two Moderate Resolution Imaging Spectroradiometer (MODIS) sensors have been in operation for over 19 and 17 years respectively, providing 36 combined years of data as part of NASA's Earth Observing System (EOS). Accurate geolocation is essential for retrieving global biogeophysical parameters. This paper details the evolution of MODIS geolocation accuracy through Collections 5 (C5), 6 (C6), and 6.1 (C6.1). We discuss improvements in re-processed data, corrections for operational biases—such as those induced by the AMSR-E instrument on Aqua—and the discovery of scan-to-scan underlaps caused by Earth's rotation and focal length variations. Finally, we outline potential enhancements for the upcoming Collection 7.
The Geolocation Challenge
For Earth-observing satellites, the utility of an observation is fundamentally tied to its spatial accuracy. Geolocation serves as the foundation for monitoring critical global changes, including deforestation, ice melt, and urbanization. However, maintaining sub-pixel accuracy over decades is a complex engineering feat. Satellites are dynamic structures that experience thermal flexing as they transition between sunlight and shadow, and onboard instruments can exert mechanical influences on one another.
The MODIS team specifically addressed two major hurdles:
- Long-term Drift: Correcting pointing errors resulting from the spacecraft's thermal environment.
- Geometric Gaps: Identifying the cause of "underlaps"—small physical gaps between consecutive scans—which can result in missing data at the pixel level.
Methodology: Ground Control Point Matching
To quantify and correct these errors, researchers employed a Ground Control Point Matching (CPM) program. This method correlates high-resolution (30m) "chips" from Landsat imagery with MODIS Band 1 data. By identifying the point of maximum correlation, the team can precisely measure the displacement (error) in the MODIS observations.
Key technical refinements included:
- Bias Stratification: Categorizing errors by time, scan angle, and sun angle to build robust correction models.
- On-Orbit Focal Length Measurement: Developing a technique to measure the Effective Focal Length (EFL) while in orbit. Even a 0.5% deviation in EFL can translate to a 50-meter shift on the ground.
- Inter-instrument Modeling: Accounting for the mechanical torque produced by the Advanced Microwave Scanning Radiometer for EOS (AMSR-E) on the Aqua platform, which caused "pointing jumps" when its antenna changed rotation status.
Mission History and Data Collections
The analysis spans the entire mission history of both sensors, comparing performance across iterative data "Collections."
- Terra MODIS: Launched in December 1999, with data records beginning in February 2000.
- Aqua MODIS: Launched in May 2002, with data records beginning in July 2002.
- Validation: A global library of over 1,000 Landsat ground control points was used for continuous validation.
Key Insights: Mechanical and Geometric Anomalies
The long-term trending revealed several subtle but significant phenomena:
- The AMSR-E Effect: The rotation of the AMSR-E instrument on Aqua physically nudged the satellite. When AMSR-E's status changed, it induced a torque that shifted the MODIS line-of-sight, necessitating specific corrections in the C6.1 data stream.
- Earth Rotation Oversight: "Scan-to-scan underlaps" were found to be a design oversight regarding Earth's rotation. Near the equator, the Earth rotates beneath the sensor faster than the scan captures it, creating gaps of up to 170 meters.
- Focal Length Sensitivity: Measurements showed that Near-Infrared (NIR) focal planes have slightly shorter focal lengths than nominal, contributing to these underlaps and affecting Band-to-Band Registration (BBR).
Results: Achieving Sub-Pixel Accuracy
Through successive collections, the MODIS team has driven geolocation errors down to impressive sub-pixel levels.
- Residual Reduction: Root-Mean-Square Error (RMSE) has been stabilized at approximately 43–46 meters in the track direction and 45–54 meters in the scan direction. For a 250m resolution pixel, this represents an error of only ~20%.
- Bias Elimination: Mean geolocation errors have been reduced to near-zero (0–1 meter) in the latest C6.1 re-processing.
- Underlap Quantification: Maximum gaps at nadir were quantified at ~100m for Terra and ~170m for Aqua, primarily occurring near the equator.
Future Outlook: Collection 7
While current accuracy is high, the upcoming Collection 7 (C7) aims for further refinement. Planned improvements include:
- Format Transition: Moving from HDF4 to HDF5 formats for better data interoperability.
- EFL Integration: Formally incorporating measured on-orbit focal lengths into Look-Up Tables (LUTs) to further mitigate scan underlaps.
- Enhanced Resolution: Introducing geolocation products at 250m resolution to match the highest-resolution MODIS bands.
- Robust Maneuver Handling: Expanding the CPM search area to maintain accuracy during satellite maneuvers.
Summary updated