A mixture emissivity analysis method for urban land surface temperature retrieval from Landsat 8 data

Land surface temperature (LST) retrieval from satellite imagery is one of the most practical ways to consistently monitor urban thermal environment. Given the heterogeneous nature of urban landscape, an implicit assumption should be considered in remotely sensed LST determinations that a mixed urban land cover aggregation is the combination of its constituent components. Currently, the common LST retrieval method which utilize emissivity measures estimated by NDVI threshold method (NDVITHM), including mono window (MW), single channel (SC), and split window algorithms (SW), does not take into account heterogeneity of pixels. While in this study, a new approach, the mixture analysis of emissivity (MAoE), is proposed to calculate temperature by estimating pixel emissivity from mixed land cover classes. We conduct a comparison of six approaches by the combinations of three LST retrieval algorithms with NDVITHM and MAoE respectively. The differences among strategies are characterized and analyzed by comparing LST estimates from Landsat 8 thermal images. The LST gradients derived from transect analysis are found consistently similar for combinations of two LST algorithms (MW and SC) and the two emissivity estimation algorithms (MAoE and NDVITHM). LSTs derived from SW algorithms using band 10 have the highest mean values, while the SC algorithms have moderate mean values and the MW algorithms have the lowest values. Standard deviations of estimated LST from MAoE are smaller compared with NDVITHM methods, SC retrieval algorithm with MAoE has the smallest standard deviation, and NDVITHM temperature estimation could be more impacted by different land use land cover types.