An Efficient Framework for Producing Landsat-Based Land Surface Temperature Data Using Google Earth Engine

Abstract

A long time-series land surface temperature (LST) product is useful for ecological and environmental studies. However, current LST products cannot provide a global coverage at a fine spatial resolution (~100 m) over a long period (>30 years). Landsat series satellites that have been launched since 1972 provide a unique opportunity to fill the gap. Here, we proposed a single-channel framework for producing global long time-series Landsat LST retrievals on a Google earth engine (GEE) cloud computing platform. This framework unifies the LST, land surface emissivity (LSE) and atmospheric water vapor (AWV) estimation algorithms, as well as the emissivity and atmospheric input data for the Landsat LST retrievals from the entire Landsat thermal infrared image archive. In situ LST measurements and the MODIS LST products were employed to evaluate Landsat LST retrievals using the proposed framework over land and water surfaces, respectively. In total, 1317 clear-sky LST samples were collected from the Landsat 5-8 series after spatiotemporal registration with seven sites, and the average bias and root-mean-square error (RMSE) were 0.33 and 2.01 K, respectively. Intercomparison between Landsat and MODIS LST retrievals based on 100 clear-sky scenes over 12 inland lakes showed an average bias of 0.17 K and RMSE of 1.11 K. We conclude that the proposed single-channel framework can produce Landsat LST with high accuracy following a simple yet robust way. Implementation of the single-channel method on GEE shows promise in providing the community with freely accessible and global long time-series (>30 years) LST data.