Temporal and Spatial Variations of the Hydrogen Content in the Martian Subsurface as Measured by the Neutron Spectrometer onboard Mars Odyssey: The Hellas Planitia Case

Abstract:

By analyzing Mars Odyssey Neutron Spectrometer (MO/NS) measurements, we have unveiled a distinct seasonal variation of epithermal count rates (ECR) in the shallow subsurface (~1 m) of Hellas Planitia (centered at 42°S, 70°E). This variation is caused by seasonal changes in the water content, possibly in the form of water ice. In this project, we loaded MO/NS data from the original source (NASA’s Planetary Data System, where each archived file includes observations for a 24h-time span stored in binary format) to a non-tabular and non-transaction orientated (no-SQL) database, which significantly eases and quickens the analysis of this dataset. Based on the footprint of the NS instrument (~550 km at the surface), and to ensure a statistically significant approach, we grouped ECR measurements in spatial-temporal bins of various widths, from Martian Year (MY) 26 to MY 34, encompassing more than 16 terrestrial years. Then, we calculated statistically-significant differences between the seasonal and climatological average in each bin to unveil seasonal variations in the water content.

From mid to equatorial latitudes, Hellas Planitia was the only region in which statistical tests based on the nearest pixels deviation revealed statistically-significant seasonal variations in ERCs. Specifically, this region presented higher hydrogen content in southern Autumn and lower in southern Summer, suggesting the formation and sublimation of water ice. To understand this unique behavior, we plan to fit a dynamic mathematical model to a phase space with the relative variation of ECRs as the dependent variable, and the Mars's surface temperature, elevation, and pressure as independent variables, among others. If this model returns a relationship between the independent variables and the relative variation of ERCs, then we will search for new regions showing the same pattern by using deep learning modeling based on a Convolutional Neuronal Network linked to a Recurrent Neuronal Network (CRNN). These models have shown a good capability to extract information from climatic and meteorological systems. Else, we will apply CRNN to analyze the pattern found in Hellas Planitia. In both cases, an explainable artificial intelligence model will be developed for extracting climatic and meteorological information from CRNN.