Evaluation of split-window land surface temperature algorithms for generating climate data records

Land surface temperature (LST) is a key indicator of the Earth's surface energy and is used in a range of hydrological, meteorological, and climatological applications. As needed for most modeling and climate analysis applications, LST products that are generated from polar-orbiting meteorological satellite sensors have spatial resolutions from several hundred meters to several kilometers and have (quasi) daily temporal resolution. These sensors include the National Oceanic and Atmospheric Administration Advanced Very High Resolution Radiometer (AVHRR), the Earth Observing System Moderate Resolution Imaging Spectroradiometer (MODIS), and the forthcoming Visible/Infrared Imager Radiometer Suite (VIIRS) series, to be flown onboard the National Polar-Orbiting Operational Environmental Satellite System (VIIRS flights begin approximately 2009). Generally, split-window algorithms are used with these sensors to produce LST products. In this paper, we evaluated nine published LST algorithms (or, in some cases, their slight variants) to determine those that are most suitable for generating a consistent LST climate data record across these satellite sensors and platforms. A consistent set of moderate-resolution atmospheric transmission simulations were used in determining the appropriate coefficients for each algorithm and sensor (AVHRR, MODIS, and VIIRS) combination. Algorithm accuracy was evaluated over different view zenith angles, surface-atmosphere temperature combinations, and emissivity errors. Both simulated and actual remote sensing data were used in the evaluation. We found that the nine heritage algorithms can effectively be collapsed into three groups of highly similar performance. We also demonstrated the efficacy of an atmospheric path-length correction term that is added to the heritage algorithms. We conclude that the algorithms depending on both the mean and difference of band emissivities (Group 1 in our nomenclature) are most accurate and stable over a wide range of conditions, provided that the emissivity can be well estimated a priori. Where the emissivity cannot be well estimated, the Group 3 algorithms (which do not depend on the emissivity difference) modified with the path-length correction term perform better.