A Forward Modeling Approach for the Vicarious Validation of the Thermal Infrared Sensor On-Board Landsat 8 and 9

The Landsat program has maintained a continuous, global thermal record for over forty years, supporting various long-term earth observation scientific studies. We present an in-flight validation of Landsat 8 and Landsat 9 Thermal Infrared Sensor (TIRS) data using a physics-based forward-modeling approach that converts in situ measurements from open-water buoys distributed across the Continental United States (CONUS) into predicted at-sensor radiance. Previous vicarious validation approaches relied on radiosonde atmospheric profiles that are sparsely sampled in time and space, often mismatched to Landsat overpass conditions, and typically lacked a comprehensive, end-to-end uncertainty characterization which limited quantitative assessment of radiometric accuracy. Unlike these approaches, the framework presented here uses high spatiotemporal resolution GEOS-5 Forward Processing (FP) reanalysis data to accurately represent the atmospheric column at the time of satellite overpass. We detail this improved forward-modeling framework and develop a comprehensive uncertainty budget for the process in accordance with the Guide to the Expression of Uncertainty in Measurement (GUM) guidelines. Our results confirm that Landsat 8/9 performance from the vicarious validation remain within each sensor's absolute radiometric uncertainty, with Landsat 8 mean biases of -0.020 and 0.013 [$W/m^{2}/sr/\mu m$] for Bands 10 and 11, and Landsat 9 mean biases of -0.002 and -0.009 [$W/m^{2}/sr/\mu m$], demonstrating continued radiometric stability across successive missions. Although our open-access workflow is demonstrated here using the Landsat thermal bands, the underlying process and uncertainty methodology provide a reproducible vicarious validation/calibration framework applicable to current and future thermal sensors, including NASA missions (e.g., Landsat Next) and emerging commercial platforms that may rely heavily on external validation of their data products.