Retrieval and validation of the land surface temperature derived from Landsat 8 data: A case study of the Heihe River Basin

Abstract

Land Surface Temperature (LST) is an important parameter in land surface physical processes in regional and global scales. LST plays an crucial role in the interaction and energy exchanges between the atmosphere and land. LST has been widely used in weather forecasting, ocean circulation, drought monitoring, and energy balance. At present, the spatial resolution of LST products is relatively low, which can no longer meet the demand for monitoring urban heat island and estimating regional evapotranspiration. Landsat satellites offer numerous high-spatial-resolution data of land surface. However, an operational Landsat8 LST product is unavailable, thereby limiting the use of the data. In this study, we developed a physical single-channel algorithm to retrieve LST from Landsat 8 TIRS database, which can be used for retrieving Landsat LST with long time series. A physical single channel algorithm was developed to retrieve the LST from Landsat 8 TIRS data. First, ASTER Global Emissivity Database and vegetation cover method were used to calculate the land surface emissivity. Then, MERRA reanalysis data and fast radiative transfer model RTTOV 11.3 were utilized for atmospheric correction of Landsat8 thermal infrared images. The validation results were divided into two parts: (1) validation using the simulated data and (2) validation based on the ground measured data. First, simulation data calculated by TIGR atmospheric profile and MODTRAN were used to validate the accuracy of the algorithm, and the in situ LSTs between 2013 and 2015 acquired from the HiWATER experiment were used to evaluate Landsat LST. The transmittance and atmospheric upward radiance simulated by RTTOV are close to those of MODTRAN. The mean bias of transmittance between RTTOV and MODTRAN is approximately 0.01, and that for atmospheric upward radiance is approximately 0.04 W/(m2•sr•μm). Compared with the NDVI threshold method, the proposed method can both reflect the dynamic change trend of land surface emissivity over vegetations and reflect the variation among different soil types. The validation results show that the overall deviation for both PSC algorithm and JMS method is within ±0.2 K. Moreover, the RMSE of the PSC algorithm is around 2.2 K, whereas that for JMS algorithm is 2.4 K. LST with high spatial resolution and precision could be obtained using the proposed method for monitoring urban heat island and estimating regional evapotranspiration.