Satellite Underflight Utility for Thermal Sensor Harmonization

TL;DR

This paper demonstrates that Landsat 8 and Landsat 9 underflight data enable pixel-level radiometric cross-validation of Thermal Infrared Sensor (TIRS) bands. By constructing 1k×1k cutouts, applying QA masks, and restricting to near-simultaneous acquisitions, the study identifies a practical RMSD threshold of $<5%$ to select scene pairs suitable for calibration validation. The results show that selected scene pairs achieve TOA radiance agreement with average brightness-temperature differences of approximately $0.123$ K (10.9 μm) and $0.066$ K (12.0 μm), indicating high fidelity in cross-calibration for near-identical sensors. This approach supports using full-orbit data to harmonize calibrations across sensors, informs Cal/Val planning for current and future missions, and suggests automation to scale the method to more diverse sensor pairs.

Abstract

Underflight maneuvers provide a unique opportunity to harmonize calibration of on-orbit sensors. Due to their similar sensor technologies, their near-identical transmission profiles, orbital properties and platform operations, the underflight data of Landsat 8 and 9 instruments stand out as a qualifier to test proposed metrics, methods, and the extent over which to compare two independently calibrated sensors across their similar operating bandpasses. This study performed a pixel-to-pixel comparison of thermal imagery of TIRS and TIRS-2 (aboard Landsat 8 and 9, respectively) during their five-day underflight maneuver in November 2021, with the ultimate goal of identifying the key site/scene-selection criteria for a subset of images that are suitable for radiative calibration validation purposes. If a group of near-coincidentally observed images by two identical underflying sensors fail to show consistent Top-of-Atmosphere (TOA) Brightness Temperatures for the same exact geographical locations, then the scenes with those shared properties and/or observing conditions will prove unreliable for cross calibration validation of less-similar underflying sensor pairs. This study demonstrates that near-coincidental images with Root Mean Square Deviations (RMSDs) of less than 5\% between their TOA radiances are optimum candidates for cross-validation of radiative calibration between two independently calibrated sensors. This criterion is shown to be reliable for coincidental acquisitions with a wide range of overlapping area, terrain type, land-to-water fraction and cloud coverage. Important considerations include any time gap between near-coincident acquisitions as well as the application of pixel quality masks. The analysis of the selected underflight scenes demonstrated an agreement between the TIRS on Landsat 8 and 9 to within 0.123 K and 0.066 K for the 10.9um and 12.0um bands respectively.

Figures (12)

• Figure 1: Schematic illustration of the orbital tracks of Landsat 8 and Landsat 9 during the November 11-16, 2021, underflight operational period. As Landsat 9 was maneuvered to its final orbit, the ground track drifted east-to-west across the Landsat 8 ground track with the nearest simultaneity on November 14.
• Figure 2: Global coverage of the 408 scene pairs in the original sample used for this study (open blue squares), and the selected subset of those scene pairs that exhibited observing condition similarity for unbiased pixel to pixel comparison.
• Figure 3: Matched cutouts of TIRS and TIRS2 images that are precisely cropped to the same latitudes and longitudes in the coincidentally observed scenes with Path/Row of 30/40 and 119/38. The color-mapped cutouts show the TOA radiances at B10 (top) and B11(bottom). The right-most panels on the top and bottom rows are their pixel-to-pixel brightness temperature differences. Pixel quality maps have been applied to the images and all the pixels that are not categorized as clear are removed from both cutouts.
• Figure 4: Selected scene pairs (green) cover the full range of overlaps as the pairs in the parent sample of all scene pairs. This is one illustration of the notion that the selection criterion is not biased towards highly overlapping scene pairs. Eligible scene pairs are a subset of the parent sample for which a fit to their radiance and temperature distributions did not fail but implied dissimilarities between the two distributions.
• Figure 5: Comparisons for the two example cutout image pairs shown in Figure \ref{['fig3']}. The observed good agreement with the equality line is not unexpected.