When a satellite instrument maps the Earth, it rarely looks straight down at every point. To capture global data efficiently, wide-swath sensors sweep back and forth across their flight paths. This scanning motion introduces a fundamental physical challenge: the instrument's viewing geometry changes continuously from the center of the track (nadir) to its outer edges.
As the scan angle increases, the physical path length of the light traveling through the atmosphere to the sensor grows longer. This alteration changes the air mass factor—the relative path length of light through the atmosphere—and introduces directional radiance and reflectance effects. If the mathematical models or radiometric calibrations do not perfectly account for these angular shifts, the resulting data will contain artificial spatial gradients. These spurious trends reflect the movement of the instrument rather than actual geophysical changes on the Earth's surface or in the atmosphere.
The Anatomy of Cross-Track Biases
This geometric distortion is frequently observed as a "cross-track bias," where retrieval accuracy systematically varies across the instrument's field of view. As Antje Inness et al. explain in Monitoring and assimilation tests with TROPOMI data in the CAMS system: near-real-time total column ozone, wide-swath satellite sensors exhibit systematic retrieval dependencies based on the instrument's scan position. Without specialized calibration, these dependencies introduce artifacts into atmospheric analyses, making observations from the edges of the sensor's view less reliable than those from the center.
These distortions manifest in several ways depending on the sensor design:
- Edge-to-Center Gradients: Scanning geometries frequently produce artificial gradients across the viewing swath, creating false spatial variations that must be smoothed out to ensure accurate global mapping, as noted by Nan Hao et al. in GOME-2 total ozone columns from MetOp-A/MetOp-B and assimilation in the MACC system (Atmospheric 2014).
- Asymmetrical Swath Biases: The bias is not always symmetrical around nadir. In Comparison of GOME-2/MetOp total ozone data with Brewer spectroradiometer data over the Iberian Peninsula, M. Antón et al. identified systematic discrepancies between the eastern and western edges of a scanning swath, pointing to underlying limitations in radiometric calibration or the mathematical models used to calculate air mass factors at varying observation angles.
- Surface Radiance Distortions: The problem is equally acute when measuring surface properties. Xiaofeng Zhao et al. demonstrate in Single-channel method based on temporal and spatial information for retrieving land surface temperature from remote sensing data that neglecting the variation in viewing angles across different satellite overpasses introduces significant errors in land surface temperature retrievals due to uncorrected directional reflectance and radiance effects.
Discrepancies Across Orbits and Campaigns
Viewing geometry issues extend beyond the sweep of a single sensor's mirror; they also emerge when comparing different satellite vantage points or distinct observation periods.
Consider the difference between geostationary satellites (which hover over a fixed spot on the equator) and polar-orbiting satellites (which sweep from pole to pole). Because they view the same patch of Earth from entirely different angles, their retrievals can systematically diverge. The authors of Evaluation of CLARA-A2 and ISCCP-H Cloud Cover Climate Data Records over Europe with ECA&D Ground-Based Measurements showed that distinct viewing geometries between geostationary and polar-orbiting satellites lead to systematic variations in observed cloud properties, creating artificial north-south gradients. Similarly, Alexander Ignatov et al. found in Preliminary Inter-Comparison between AHI, VIIRS and MODIS Clear-Sky Ocean Radiances for Accurate SST Retrievals that new geostationary imagers can exhibit a systematic cold bias in infrared brightness temperatures when compared directly to polar-orbiting sensors, necessitating sensor-specific calibration adjustments.
Furthermore, when a single mission operates over many years, or when multiple missions are stitched together into a long-term climate record, "inter-campaign" or "inter-satellite" biases arise. In MASS BALANCE OF ANTARCTIC ICE SHEET FROM 2003 TO 2008: A SYSTEMATICALLY IMPROVED NEW ESTIMATION, Rongxing Li et al. highlight that satellite altimetry measurements suffer from systematic differences between observations taken during different satellite passes or campaigns, which can distort estimations of ice sheet mass balance if left uncorrected.
Strategies for Bias Correction
To resolve these geometric and instrumental artifacts, researchers employ a mix of empirical modeling, inter-sensor calibration, and ground-truth validation.
Empirical Polynomial Adjustments
One effective way to neutralize cross-track gradients is to model the bias mathematically as a function of the scan angle and latitude. Because the geometric distortion varies smoothly across the swath, a polynomial curve can be fitted to the observed discrepancy and subtracted from the raw data. Nan Hao et al. utilize a latitude-dependent empirical polynomial correction to eliminate artificial gradients across the instrument's viewing swath, successfully reducing cross-track biases to less than 0.4%. By adjusting the data based on scan position, this calibration ensures that observations from the edges of the scan are consistent with those from the center, preventing artificial gradients in data assimilation systems, as shown by Antje Inness et al.
Inter-Sensor Calibration
When a sensor exhibits an initial bias, it can be systematically corrected by cross-referencing it with a highly stable reference instrument. For example, Lei Guan demonstrates in First Assessment of HY-1C COCTS Thermal Infrared Calibration Using MetOp-B IASI how initial biases dependent on latitude and brightness temperature can be corrected using inter-calibration techniques against a trusted reference sensor. For multi-satellite time series, Jinyang Wang et al. apply a systematic deviation correction in Study on the Variations in Water Storage in Lake Qinghai Based on Multi-Source Satellite Data by calculating and adjusting for common-period observations between adjacent satellites, reconciling the discrepancies introduced by distinct sensors.
Ground and Airborne Validation
Integrating ground-truth or high-resolution survey data helps anchor satellite observations. In Performance Evaluation of LMI Based on Low-Frequency Three-Dimensional Total Lightning Flash Location Data, Mengjin Zou et al. show that correcting systematic biases using ground-based reference data significantly improves spatial alignment and detection accuracy. In oceanography, Alexander Ignatov et al. utilize Sensor-Specific Error Statistics (SSES) bias corrections in AVHRR GAC SST Reanalysis Version 1 (RAN1) to minimize discrepancies between satellite sea surface temperatures and in situ measurements. Similarly, Hamish D. Pritchard et al. note in Bedmap3 updated ice bed, surface and thickness gridded datasets for Antarctica that satellite altimetry data must be calibrated against available survey data to correct for spatially varying biases in ice shelf thickness.
However, researchers must be cautious when tuning models empirically. Florent Garnier et al. warn in Advances in altimetric snow depth estimates using bi-frequency SARAL and CryoSat-2 Ka–Ku measurements that calibrating satellite products against localized airborne datasets can introduce a systematic high bias if the validation campaigns primarily sample unrepresentative regions (such as thicker snow cover). They argue that independent physical comparisons, rather than localized empirical tuning, are essential for generating globally consistent geophysical products.
Key Foundational Work
Comparison of GOME-2/MetOp total ozone data with Brewer spectroradiometer data over the Iberian Peninsula (2009) — contributed 3 concept elements to this direction.
GOME-2 total ozone columns from MetOp-A/MetOp-B and assimilation in the MACC system (2014) — contributed 3 concept elements to this direction.
Single-channel method based on temporal and spatial information for retrieving land surface temperature from remote sensing data (2014) — contributed 1 concept elements to this direction.