In this project, we developed an inversion model for extracting the physical properties of the surface using deep learning benchmarked against lunar sample drive core measurements.
-
First, we simulated the data for training of the DL model from the parameterization of Integral Equation Model (IEM) for rough surfaces.
-
Second, testing was done on the S-band Mini-RF data to retrieve the dielectric constant of the lunar surface at global and polar scales.
Note: The forward model incorporates the radar backscatter contribution from the subsurface and buried heterogeneities (like volatiles and rocks). We also accounted for the differences in the concentration of volatiles by modelling the dielectric interface of silicate-ice-helium mixtures.
We explicitly tested our model for lunar science application however it is possible to transfer and reproduce this algorithm for a diverse range of earth and planetary science base applications such as:
- Dielectric Properties
- Surface roughness
- Soil moisture
The global dielectric maps of the lunar surface have been generated using this model as shown below:
Additionally dielectric maps of the lunar poles have been generated to avoid extreme projective distortions in these these regions:
License: CC-BY