Calculation of land surface emissivity and retrieval of land surface temperature based on a spectral mixing model

Abstract

Land surface emissivity (LSE) is a key parameter for the retrieval of land surface temperature (LST). Using the linear spectral mixing model (LSMM) to calculate the LSE can estimate the pixel composition at the sub-pixel level, which effectively solves the problem of mixed pixels when using the Landsat thermal infrared band to calculate the surface specific emissivity. In this paper, Landsat OLI/TIRS was used as the data source, surface emissivity was calculated using the LSMM of mixed pixels, and the LST was then obtained using the radiative transfer equation algorithm. At the same time, three threshold methods of the normalized difference vegetation index (NDVI) were used to calculate surface emissivity and then retrieve the LST. By comparing and analysing the inversion results and verifying the accuracy of the measured ground data, the surface emissivity and LST results obtained by the four methods were similar overall. Specifically, the maximum and average values of LST obtained by the LSMM were the highest, while those obtained by the Sobrino threshold method were the lowest, with an average difference of 0.63 °C. The difference between the LST inversion results of the LSMM and the other three methods were calculated separately, and the maximum change in LST reached 2.82 °C. All three ΔLST results showed high values in densely urbanised areas and low values in vegetation-covered areas. For urban areas with complex structures, the LSMM can be used to estimate the abundance of each component in a pixel at the sub-pixel scale, which can significantly improve the calculation accuracy of surface emissivity and lead to better LST inversion results.