A Statistical Split Window Method Without Atmospheric Profile Input for Temperature and Emissivity Retrieval from Airborne Long-Wavelenght Infrared Hyperspectral Imagery

The retrieval of Land Surface Temperature (LST) and Emissivity (LSE) from long-wavelength thermal infrared (LWIR) hyperspectral data can be challenging due to uncertainties in atmospheric compensation (AC). While AC is typically performed using atmospheric radiance models, errors in the input atmospheric profiles can lead to significant inaccuracies. In this study, we propose an end-to-end method for retrieving LST and LSE without the need for local atmospheric profile inputs. The method consists of two main steps: first, a statistical split-window (SSW) method is used to estimate the ground leaving radiance, with optimal band configurations and coefficients determined through a trial-and-error approach and statistical regression based on simulation datasets. Second, by integrating the ASTER Temperature And Emissivity Separation (ASTER-TES) method with an atmospheric downwelling lookup table (LUT), the optimal LST and LSE are derived based on the principle of emissivity smoothness. The proposed method is applied to airborne HypercamLW LWIR hyperspectral data. When compared to the MODTRAN-based AC with ASTER-TES (MODTRAN-TES) and the built-in method of FLAASH-IR, the proposed SSW-TES method yields better accuracy, with a LST root mean square error (RMSE) of 1.24 K and an LSE RMSE of 0.016.