Gaussian-process-regression-based periodical variation analysis of the lunar surface temperature with the ESA-Dresden radio telescope

Abstract

Preparation for the extreme range of lunar surface temperature is essential for the success of robot and human exploration on the Moon. Much research on this topic has been carried out over recent years. Generally, these approaches can be divided into three types based on: the observation data, the theoretical model, and the in situ measurements. This paper presents a new Bayesian algorithm to measure the lunar surface temperature and analyze its periodical variation with respect to the Moon phases based on the observation data, which have been taken from the ESA-Dresden 10 GHz radio telescope. As the signal-to-noise ratio of the observation data is very low, and can easily be affected by ambient factors, continuous observations are made when the Moon moves across the main beam of the antenna. These continuous observations are then modeled with Gaussian distribution to improve observation accuracy. Due to the lunar surface temperature varying during the lunations, the Gaussian Process regression, which is coupled with the periodical covariance function, is proposed to analyze the periodical variation of the lunar surface temperature. Both the mean and variance of the temperature can be calculated at any Moon phase. Experiments are conducted on the observation data collected from November 2018 to March 2019 in Strasbourg, France with the ESA-Dresden 10 GHz radio telescope. The results show that the lunar surface temperature is an approximate sinusoidal function of the Moon phase, with the average temperature of about 196 K. The lunar surface temperature reaches the peak 4.80–6.25 days after the full Moon and falls to the bottom 3.67–6.50 days after the new moon.