Low Lunar Surface Temperature Retrieval From LRO Diviner Radiometer Observation Data

The daytime and nighttime lunar surface temperatures (LSTs) are crucial for investigating lunar surface environment and lunar mineral composition. The Diviner sensor provides global lunar surface observation in seven thermal infrared (TIR) channels from 8 to
$400~\mu $
m, but the existing LST retrieval methods are more suitable for daytime pixels with high temperature rather than the nighttime or shadowed pixels with low temperature. This letter develops a new method, called as TES-GBR, by combining the conventional temperature-emissivity separation (TES) and gradient boosting regression (GBR) method, to retrieve low LST (e.g., nighttime or shadowed regions) from Diviner’s four longwave infrared channel data. The new method used three emissivity curve shape parameters, maximum-minimum apparent emissivity difference (MMD), maximum-minimum ratio (MMR), and emissivity variance (VAR), to establish their relationship with the minimum emissivity (
$\varepsilon _{\mathbf {min}}$
). Results indicate that the TES-GBR method can reduce the emissivity error to 0.005 from 0.029 obtained by the conventional TES method and get a general retrieval accuracy of 1.0 K for the low LST. Finally, the TES-GBR method was applied to retrieve the nighttime LST of the year 2015, and it found that there was a period variation in the nighttime temperature.