Comparing Methods for Land Surface Temperature Retrieval over Heterogeneous Land Cover Using Landsat-5 TM Thermal Infrared Data

Abstract:

Among other applications, remotely sensed land surface temperature (LST) has become critical for monitoring the surface urban heat island (SUHI) effect in cities across the world. While daily MODIS thermal infrared data is invaluable for examining changes in LST over time, the large 1 km spatial resolution makes studying the spatial patterns of LST in a heterogeneous urban environment difficult. The 120 m spatial resolution of Landsat 4-5 TM, as well the archive of data stretching back to 1982, make Landsat 4-5 TM sensors valuable resources for thermal data, especially in urban areas. However, the difficulty accurately correcting for atmospheric effects with only one thermal band, as well as the necessity for a priori knowledge of land surface emissivity (LSE), mean it is underutilized. Research to determine best practices for deriving LST from Landsat TM data given homogenous, usually vegetated land cover is relatively extensive; however, the accuracy of these methods given heterogeneous land cover is less well known, especially given Land Surface Emissivity (LSE) calculations that often rely heavily on NDVI. In order to determine the best methodology for measuring LST across heterogeneous land cover in the central United States, this study derives LST from Landsat 5 TM band 6 for Oklahoma City and the surrounding countryside on a fall and a spring date using three different methods: no atmospheric correction, the radiative transfer equation, and the mono-window algorithm. With all three methods, the common NDVI-based approach for estimating LSE is used; a fourth LST calculation with no atmospheric correction and an assumed emissivity of one is therefore included as contrast. Using regression analysis, these four LST measurements are compared to air temperatures recorded concurrently by approximately 40 Oklahoma Mesonet stations across the study area, and results are broken down by land cover type to explore potential biases or variations in accuracy.