Improving Landsat land surface temperature estimation in Google Earth Engine using NDVI-based emissivity

Land surface temperature (LST) data derived from satellite images are important for various applications, including mapping urban heat islands, analysing temporal and spatial temperature patterns, assessing the cooling effect of urban greenery, and developing population vulnerability indices for heat waves. Thermal sensors aboard Landsat satellites provide the most spatially detailed data with the longest temporal continuity. Although Landsat Surface Temperature (ST) is already available as a standard product, and a code for estimating the Landsat LST using the statistical mono-window method has been implemented in the Google Earth Engine, these approaches rely on the ASTER Global Emissivity Dataset, which has certain limitations, including missing values. In Google Earth Engine, we developed an approach to calculate land surface emissivity using various NDVI-based methods, combined with the statistical mono-window and radiative transfer equation methods for LST calculation. Validation against in situ measurements from the SURFRAD network revealed that the statistical mono-window method proved to be more accurate than the Landsat ST product and radiative transfer equation methods, regardless of the emissivity data source. The NDVI-based emissivity combined with the statistical mono-window method yielded higher LST precision than the approach using ASTER GED emissivity. These results were consistent across all Landsat missions. Furthermore, we demonstrate that the lowest accuracy is achieved in calculating LST on mixed surfaces and the highest on bare soil. The overestimation of satellite LST measurements at high temperatures was only apparent on mixed and vegetated surfaces, while it was more pronounced in the Landsat ST product and other radiative transfer equation methods. These findings and the publicly available Google Earth Engine code can lead to more accurate LST mapping and analysis results.