Split-Window Algorithm for Retrieval of Land Surface Temperature Using Landsat 8 Thermal Infrared Data

Land surface temperature (LST) plays a vital role in global climate change, radiation budgets, heat balance, vegetation, snowmelt, glacier hydrology, and geo-biophysical processes. It is, therefore, essential to determine LST precisely over large areas. With advancements in remote sensing, LST can now be estimated. In this study, a critical appraisal of various LST inversion algorithms is presented. These algorithms include mono-window (MW), split-window (SW), dual-angle (DA), single-channel (SC), and Sabmao method. The main objective is to derive an SW algorithm to retrieve LST from Landsat 8 satellite data and demonstrate the application to the Beas River basin in India. Located within the Himalayan range, the study area is characterized by heterogeneity and rugged terrain with areas covered by snow and glacier. The satellite imagery is a product of the Optical Land Imager (OLI) with a spatial resolution of 30 m and a thermal infrared sensor (TIRS) having a spatial resolution of 100 m. The SW algorithm requires spectral radiance and emissivity from two bands of the TIRS as input for the estimation of LST. The spectral radiance has been estimated using the TIRS bands 10 and 11. The normalized difference vegetation index (NDVI) and the threshold technique of OLI bands (2 to 5) have been used to derive the emissivity. The estimates of LST from the TIRS and OLI bands using the SW algorithm are found to be accurate and close to the in situ air temperature measurements and the LST values obtained from the MW algorithm. Results obtained show that the values of LST are high in the barren/rocky areas and low in the snow/glacier areas. The study reveals that the LST estimates from SW and MW algorithms are linearly transferable with negligible loss of accuracy. The LST estimates from the SW algorithm differs at most by up to 5 °C with the measured air temperature.