An Operational Land Surface Temperature Product for Landsat Thermal Data: Methodology and Validation

Thermal sensors onboard Landsat satellites have been underutilized due to the lack of consistent and accurate methodologies for retrieving the land surface temperature (LST) at global scales over all land cover types. We present an operational algorithm for generating Landsat LST consistently for all sensors that will be implemented by the United States Geological Survey/The National Aeronautics and Space Administration and made available at the Land Processes Distributed Active Archive Center. The LST algorithm involves three steps. The observed thermal radiance is atmospherically corrected using a radiative transfer model and reanalysis data. The Advanced Spaceborne Thermal Emission and Reflection Radiometer Global Emissivity Data Set version 3 is spectrally adjusted and then modified to account for vegetation phenology and snow cover using Landsat visible-shortwave infrared data. The LST is retrieved by inverting the atmospherically and emissivity corrected Landsat radiances with a lookup-table approach. Landsat-derived emissivities were validated at two pseudoinvariant sand dune sites within an average absolute error of 0.54% when compared with laboratory measurements. The Landsat LST retrievals were validated with <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">in situ</i> observations from four surface radiation budget network (SURFRAD) sites, and two inland water bodies (Salton Sea and Lake Tahoe) in the USA. The LST retrievals for Landsat 5 and 7 had a mean bias (root mean square error) of 0.7 K (2.2 K) and 0.9 K (2.3 K) for the SURFRAD sites, and −0.3 K (0.6 K) and 0.4 K (0.7 K) for the inland water bodies, respectively. The operational algorithm will provide a consistent LST record from four decades of historical Landsat thermal data enabling the long-term monitoring of temperature and trends, land cover and land use changes, and improved utilization in models.