Online Global Land Surface Temperature Estimation from Landsat

This study explores the estimation of land surface temperature (LST) for the globe from Landsat 5, 7 and 8 thermal infrared sensors, using different surface emissivity sources. A single channel algorithm is used for consistency among the estimated LST products, whereas the option of using emissivity from different sources provides flexibility for the algorithm’s implementation to any area of interest. The Google Earth Engine (GEE), an advanced earth science data and analysis platform, allows the estimation of LST products for the globe, covering the time period from 1984 to present. To evaluate the method, the estimated LST products were compared against two reference datasets: (a) LST products derived from ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer), as higher-level products based on the temperature-emissivity separation approach; (b) Landsat LST data that have been independently produced, using different approaches. An overall RMSE (root mean square error) of 1.52 °C was observed and it was confirmed that the accuracy of the LST product is dependent on the emissivity; different emissivity sources provided different LST accuracies, depending on the surface cover. The LST products, for the full Landsat 5, 7 and 8 archives, are estimated “on-the-fly” and are available on-line via a web application.