On the land emissivity assumption and Landsat-derived surface urban heat islands: A global analysis

The prescription of surface emissivity (epsilon) strongly controls satellite-derived estimates of land surface temperature (LST). This is particularly important for studying surface urban heat islands (SUHI) since built-up and natural landscapes are known to have distinct epsilon values. Given the small signal associated with the SUHI compared to LST, accurately prescribing urban and rural epsilon would improve our satellite-derived SUHI estimates. Here we test the sensitivity of SUHI to the epsilon assumption made while deriving LST from Landsat measurements for almost 10,000 global urban clusters for summer and winter days. We find that adjusting the epsilon values from the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) dataset based on pixel-level normalized difference vegetation index (NDVI) increases the summer to winter contrast in daytime SUHI, a constrast that has been noted in previous studies. Overall, the difference between the two methods of prescribing epsilon, one from ASTER and one after NDVI-adjustment, is moderate; around 10% during summer and around 20% during winter, though this difference varies by climate zone, showing higher deviations in polar and temperate climate. We also combine five different methods of prescribing emissivity to provide the first global estimates of SUHI derived from Landsat. The global ensemble mean SUHI varies between 2.42 degrees C during summer to 0.46 degrees C in winter. Regardless of the surface emissivity model used, compared to Moderate Resolution Imaging Spectroradiometer (MODIS) Terra observations, Landsat data show higher SUHI daytime intensities during summer (by more than 1.5 degrees C), partly due to its ability to better resolve urban pixels. We also find that the epsilon values prescribed for urban land cover in global and regional weather models are lower than the satellite-derived broadband epsilon values. Computing sensitivities of urban and rural LST to epsilon, we demonstrate that this would lead to overestimation of SUHI by these models (by around 4 degrees C for both summer and winter), all else remaining constant. Our analysis provides a global perspective on the importance of better constraining urban epsilon for comparing satellite-derived and model-simulated SUHI intensities. Since both the structural and geometric heterogeneity of the surface controls the bulk epsilon, future studies should try to benchmark the suitability of existing LST-epsilon separation methods over urban areas.