Optimization of DirecTES thermal infrared land surface temperature and emissivity separation algorithm for the upcoming TRISHNA mission

The inversion of the radiative transfer equation to retrieve both the surface temperature (LST) and emissivity (LSE) values from top-of-atmosphere (TOA) radiances in the thermal infrared (TIR) domain (8-14 µm) is a not straightforward problem. Marcq et al. (2023) proposed the algorithm DirecTES to invert LST using a spectral library of emissivity of various materials, to be applied on several TIR channels.  The algorithm consists in inverting the radiative transfer equation for the LST, for each material of the library. A threshold criterion selects materials of the library for which the standard deviation of retrieved LST across different TIR channels is below 3K. The final LST and LSE are the median of the values retrieved for the selected materials. However, a constant threshold is problematic because sometimes no material in the library may match the criterion and thus the LST may not be retrieved on some pixels of a satellite image. Moreover, DirecTES’s original spectral library (SAIL179) is only composed of vegetation and arid surface materials and performs poorly on desertic surface pixels.

This study focuses on optimizing DirecTES in the TIR channels of the upcoming TRISHNA instrument conjointly developed by CNES (France) and ISRO (India). A new universal spectral library of emissivity that could be applied to any type of observed land surface of the globe is built with 150 emissivity spectra from the CAMEL database, categorized into four main classes (arid, desert, snow-covered or vegetated). In most cases, the category of the observed surface in a satellite image pixel is not known. We propose an optimization of DirecTES’s criterion that consists in selecting from the spectral library only the 10 materials with the lowest LST standard deviation between TIR channels. This new approach efficiently selects materials in the appropriate emissivity category on any surface, thus reducing the bias and RMS error on the retrieved LST and LSE. In addition, this new approach corrects the limitation of the original DirecTES criterion and can retrieve the LST and LSE on every pixel of the processed image.

The performances of the new DirecTES criterion and spectral library are evaluated, using TOA radiances simulated from the CAMEL emissivity database and the TIGR atmospheric database. LST is retrieved with a biais < 0.1K and a RMSE < 0.6K on vegetated surfaces and < 0.8K on arid and desert surfaces. LSE is retrieved with a RMSE < 0.02 for all TRISHNA TIR wavelengths. For desertic areas, performances are further improved when adding a few more emissivities from these specific regions to the spectral library used by DirecTES, while not affecting the performances on the other regions.

Finally, DirecTES is validated with match-up data of TOA radiances measured by ECOSTRESS and LST ground measurement at La Crau, France. For 53 match-ups dates of 2023, the LST is retrieved with a bias < 0.15K and RMSE < 0.9K.