HORUS data description & availability

The image repository contains a total of 10,060 non map-projected HORUS (Hyper-effective nOise Removal Unet Software) post-processed images (2,515 individual images). The repository holds 4 variants of each individual HORUS image, representing different clipping values. ‘Clipping’ is a HORUS pre-processing step that omits all pixels with DN (digital number) counts over a given, fixed threshold. The suppression of very bright pixels helps to reduce the contrast between sunlit and shadowed regions, ultimately allowing HORUS to discern more detail in the sun-shadow transition zone (terminator) while preventing synthetic ‘pixel bleed’ and other artifacts that might suppress or cover physical features – which is particularly important for very small PSRs (<~500 m, PSR = permanently shadowed region). Clipping is performed at 4 pre-defined thresholds: 5, 10, 20, and 50 DN. Particularly poorly secondary-lit PSRs benefit from lower thresholds, well secondary-lit PSRs benefit from higher thresholds.

The different variants have the file names:

<M> <NAC image unique number > <NAC L or R> <C> <HORUS> <clipping value (DN)> <clip> <.cub>

Each HORUS-processed image preserves its unique NAC (Narrow Angle Camera) image ID number. Note that clipping can occasionally remove high-reflectance features within PSRs – we recommend to always consider all image variants before using any of the images for science- or exploration-related purposes.

The number of available images per site depend on the local secondary illumination conditions (for exact numbers see, e.g., the PSR Atlas that is associated with Bickel et al., 2022). Note that PSRs X01a and X01b are jointly listed as ‘X01a’. The total dataset size is about ~5 TB. We additionally provide shape files (.shp, Moon2000 polar stereographic projection) that document the location and extent of all 58 studied PSRs (Bickel et al., 2021: 14 PSRs; Bickel et al., 2022: 44 PSRs).

All images are default USGS Isis3 (Integrated Software for Imagers and Spectrometers, https://isis.astrogeology.usgs.gov/) cube files with the file ending .cub. HORUS-processed images do not need to be imported with, e.g., lronac2isis and do not require any further calibration (lronaccal) or echo correction (lronacecho). HORUS images can be map-projected to any user-defined system using, e.g., the default cam2map command. We recommend initializing the SPICE kernels prior to map-projection (spiceinit). We recommend to view the map-projected images and manually optimize their brightness and contrast in, e.g., a Geographic Information System to maximize the number of visible features and facilitate image analysis & cross-comparison.

The image repository further holds browse versions of all regular HORUS images, for quick review and selection. Note that the browse product visualizations do not always feature the optimal histogram stretch (brightness & contrast) for technical reasons (especially for very small PSRs); all available HORUS images provide at least some coverage of their target PSR, even if their browse products indicate otherwise. This dataset is a supplement of the two papers “Peering into lunar permanently shadowed regions with deep learning” (Bickel et al., 2021) and "Cryogeomorphic Characterization of Shadowed Regions in the Artemis Exploration Zone" (Bickel et al., 2022) published in Nature Communications and Geophysical Research Letters, respectively. Please cite these references for the dataset:

Bickel, V.T., Moseley, B., Lopez-Francos, I., Shirley, M. (2021). Peering into lunar permanently shadowed regions with deep learning. Nature Communications, 12, https://doi.org/10.1038/s41467-021-25882-z 

Bickel, V. T., Moseley, B., Hauber, E., Shirley, M., Williams, J.-P., & Kring, D. A. (2022). Cryogeomorphic characterization of shadowed regions in the Artemis exploration zone. Geophysical Research Letters, 49, e2022GL099530. https://doi.org/10.1029/2022GL099530

Supporting Information for Cryogeomorphic Characterization of Shadowed Regions in the Artemis Exploration Zone

For specific details about HORUS and the processing pipeline refer to Moseley et al. (2021):

Moseley, B., Bickel, V.T., Lopez-Francos, I., Rana, L. Extreme Low-Light Environment-Driven Image Denoising Over Permanently Shadowed Lunar Regions With a Physical Noise Model. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)(2021). URL: https://openaccess.thecvf.com/content/CVPR2021/html/Moseley_Extreme_Low-Light_Environment-Driven_Image_Denoising_Over_Permanently_Shadowed_Lunar_Regions_CVPR_2021_paper.html