A General Framework for Retrieving Land Surface Emissivity and Temperature Using Sensors With Split-Window Thermal Infrared Channels: A Case Study With Landsat 9

Abstract

Land surface temperature (LST) and emissivity (LSE) are the crucial parameters for thermal infrared (TIR) remote sensing. However, the coupling of the two parameters presents a challenge to achieving high-accuracy retrieval, particularly for sensors with only one or two TIR channels. Following the launch of Landsat 9, there has been a rapid increase in demand for methods to accurately estimate LSE and LST for sensors with high spatial resolution but limited TIR channels. Therefore, this article proposes a two-step framework to retrieve LSE and LST for Landsat 9 only using data of its own. First, the data in visible-to-near-infrared (VNIR) to short-wave infrared (SWIR) channels of Landsat 9 were used to retrieve LSEs based on a machine learning method. Subsequently, the split-window (SW) method was employed to retrieve LST based on the estimated LSEs. As a result, the retrieved LSE exhibits high accuracy across the cross and direct validation, with RMSEs all below 0.01 for the two TIR channels. For LST, the retrieved result was validated by the existing products and in situ LSTs from surface radiation budget (SURFRAD), demonstrating excellent accuracies, with RMSE of 1.86 K, which is superior to the LST product of Landsat 9, with RMSE of 2.14 K. Therefore, the proposed framework is feasible for LSE and LST retrieval without support of auxiliary data from other origins, which is of great significance for the sensors with limited TIR channels to produce accurate LSE and LST products.