SELIS V2 / research / summary

SELIS data sets: summary

18 November 2026. Condensed from research/topics/B-data-sets.md, research/findings/B-data-sets.md (sources read 13 November, corrected 15 November 2026) and research/findings/B2-site-data-coverage.md (sources read 15 November 2026). The "Ref" in the first column gives the finding or table in those files; links, dates and full limits are there. Image counts marked "own count" were computed by the research agent from archive metadata and are not published figures. "Judgement" marks an assessment. Compiled with AI (Claude); not yet re-checked by a person.

Overview

About 20 data sets covering the lunar South Pole were reviewed for archive, licence, resolution, coverage and known errors. Almost all usable South Pole data come from one mission: NASA's Lunar Reconnaissance Orbiter (LRO), and above all its laser altimeter (LOLA) and narrow-angle camera (NAC). LOLA gives the terrain reference at 5 to 20 m/pixel. NASA publishes it with per-cell uncertainty and 100 statistical terrain variants ("clones") for 27 sites. About 90% of the cells in a 5 m LOLA model are interpolated, not measured. Finer terrain (1 to 5 m) exists only as shape-from-shading (SfS) products from NASA and JAXA, plus 11 local NAC stereo models; NAC stereo cannot supply the about 0.5 m site terrain named in the baseline. Thermal data stop at 240 m/pixel, and no rock-abundance map reaches the pole. Kaguya and LRO wide-area products do not reach the pole, or reach it only by filling gaps with LOLA data. NASA data are public domain or CC BY 4.0 and allow commercial use. ISRO (Chandrayaan-2) and ESA archive data do not allow it without permission, and JAXA's Kaguya terms conflict. The best-served site is Connecting Ridge: 403 dated NAC images over its reference point (own count), six terrain products and published illumination results. Malapert Massif is the fallback. The site is still an open decision (O9). The data support a one-site proof of concept of terrain, illumination and Earth visibility with confidence values checked against dated images. They do not support most of the higher-resolution, thermal and geotechnical layers in the RFI draft.

Data sets and their use in SELIS

"Core", "Supporting" and "Not usable" give the role each data set could play in a Type A activity (judgement). The baseline steps refer to CLAUDE.md section 2.

Source What the data is Use within SELIS
LOLA polar DEMs (NASA, PDS and PGDA). B finding 3, table Laser-altimeter elevation grids: 20 m/pixel from 80°S, 10 m from 83°S, a 5 m mosaic for 87 to 90°S. Effective resolution about 20 m poleward of 84°S. Public domain. Core. Regional terrain context and geodetic reference for everything else. Replaces Kaguya and WAC in baseline step 1. Measured where a laser return exists; interpolated elsewhere.
PGDA site LOLA DEMs with clones (NASA). B finding 1; B2 finding 8 5 m DEMs of 27 named sites, with laser-return count maps, height and slope uncertainty maps (typical RMS 0.3 to 0.5 m and 1.5 to 2.5°) and 100 clones per site. About 10% of cells hold a laser return. No licence text on the page; a citation is requested. Core. Terrain layer with per-cell provenance and confidence. The count map separates measured from interpolated cells. The clones allow uncertainty to be carried into slope, illumination and visibility. NASA published clone-based illumination percentiles for four sites (Connecting Ridge in the main text), which a SELIS result can be compared with.
NASA SfS DEMs and image archive (PGDA, Zenodo). B finding 2, table; B2 findings 2, 10 5 m DEMs of 13 Artemis regions, refined from NAC images and controlled to LOLA, with elevation uncertainty maps. The Zenodo archive also holds the input NAC images orthorectified to the 5 m grid, with dates and adjusted cameras: 322 images at Connecting Ridge. CC BY 4.0. Core or supporting. Sharper terrain in sunlit areas; under the glossary definitions these cells are "inferred". The orthorectified images remove most registration work for a lit/shadow check. Images used to build an SfS DEM are not independent checks of that DEM; 1,241 aligned images at Connecting Ridge were not used and can be.
JAXA 1 m SfS DTMs (JLPEDA, Zenodo). B table; B2 finding 6 1 m DTMs and orthomosaics of six ESA Lunar Lander regions (CR1, GR1, GR2, LP1, MP1, MP2), each about 1 to 2 km across, from six NAC image pairs. Uncertainty maps, which the authors call "a sensitivity diagnostic, not a posterior standard deviation". Shadowed pixels (10 to 31%) are empty. CC BY 4.0. Supporting. 1 m terrain at a small area of one site, reused rather than rebuilt; class "inferred". Its uncertainty maps are not directly comparable with the LOLA clones. Useful as a check on SELIS terrain or as a high-resolution inset.
NAC stereo DTMs (LROC PDS). B finding 2; B2 finding 7 11 DTMs poleward of 83.4°S: DLR at 2 m for the ESA Lunar Lander study (Connecting Ridge, Shackleton rim, Malapert), LROC at 3.5 m at Connecting Ridge (2024) and 3 to 4 m at Malapert and Nobile. Tied to LOLA; orthophoto offsets up to 15 m at CR1. Public domain. Supporting. Cross-check for terrain at the sites they cover. Not a main terrain method: stereo pairs are rare at the pole, so baseline step 2 (NAC stereo at about 0.5 m) cannot be met from the archive.
NAC images (LROC PDS). B finding 8; B2 findings 3 to 5, table 3.1 Single images at 1.2 to 1.6 m/pixel near the pole, low Sun (incidence 84 to 90°). 403 images cover the Connecting Ridge reference point, 377 of them from 2009 to 2011, with the Sun direction sampled all round (own count). Few new images near the pole because of LRO's orbit. Public domain from the PDS. Core for validation. Each image is a dated, measured record of lit and shadowed terrain. It can test predicted illumination and the stated confidence. Also input for SfS if SELIS makes its own terrain. Lit, shadowed or saturated state per cell: not yet counted.
NAC controlled mosaics (LROC PDS). B finding 6, table 1 m mosaics of Artemis candidate regions and of 84 to 90°S, grouped by Sun direction. Absolute position 5 to 20 m. Public domain. Supporting. Orthoimage layer for the 3D scene and a visual check. Too coarse in position for a cell-by-cell test at 1 to 5 m.
LROC WAC polar illumination movie (LROC PDS). B2 table 3.2 8,170 wide-angle frames of the South Pole at 100 m/pixel, September 2009 to December 2011. Public domain. Supporting. Measured cross-check of illumination over time, 20 times coarser than a 5 m analysis.
NASA illumination and Earth-visibility maps (PGDA, PDS). B finding 6 Average illumination and Earth visibility over 18.6 years, at 60 m (85 to 90°S), 120 m and 240 m. Public domain. Supporting. Benchmark for SELIS illumination and visibility layers. Averages, not time series.
SPICE kernels (NAIF). B table Positions and orientations of Sun, Earth, Moon and spacecraft, and frames and time. No licence or fees; commercial products allowed. Core. Geometry for time-dependent illumination, Earth visibility and user-defined relay orbits (baseline steps 4 and 5).
ShadowCam (KPLO, PDS4). B finding 7; B2 table 3.2 Images at 1.7 m/pixel inside permanently shadowed regions (PSRs) and DTMs controlled to LOLA. Sunlit terrain saturates; DTMs carry artefacts of 0.5 to 1 m. No DTM at Connecting Ridge, Shackleton or Malapert. Supporting, only if PSR terrain is in scope. The only imaging data inside PSRs.
Diviner PCP and PRP (LRO, PDS). B finding 4 PCP: bolometric brightness temperatures at 240 m, 80 to 90°S. PRP: thermal-model fit to first-year data (temperatures, depth to stable ice) on a mesh of about 500 m (estimate). Public domain. Supporting. Coarse thermal context. The ice-depth layer is a model output, not a measurement, and needs a class that says so. No thermal analysis finer than 240 m is possible from the data.
Diviner rock abundance (LRO, PDS). B finding 4 Rock abundance and soil temperature maps, ±60° and 70°S to 70°N only. Not usable. Does not reach the pole (baseline step 6 lists it).
Kaguya SLDEM2013 and TC products (JAXA DARTS). B finding 3, table Terrain from the Kaguya Terrain Camera; polar gaps filled with MI and LOLA data; offsets about 10 m horizontal, 5 m vertical. JAXA pages conflict on commercial use. Not needed. LOLA covers the role at the pole, without the licence conflict.
SLDEM2015, WAC GLD100 (PDS, USGS). B finding 3 Merged LOLA-Kaguya DEM to ±60°; LROC wide-angle DEM to ±79°. Public domain. Not usable. Do not reach the pole.
MADNet 1 m DTMs (UCL). B2 table 3.2 Deep-learning DTMs at seven sites, 20 to 54% coverage each. Release planned; released product and licence: not found. Not usable now. Possible later input if released under open terms.
Chandrayaan-2 OHRC and TMC-2 (ISRO PRADAN). B finding 5 Images at about 0.3 m and 5 m. ISRO copyright; non-commercial scientific use only. Not usable in a commercial site package without ISRO permission.
ESA Planetary Science Archive (for example SMART-1). B finding 5 CC BY-NC 3.0 IGO; commercial use needs ESA authorisation. Polar coverage not checked. Not usable commercially without ESA authorisation.
Hydrogen and radar data (LEND, Mini-RF, LAMP). B table note Listed by a NASA workshop as low resolution or ambiguous. Not checked further. Not assessed.

Conclusion: what can realistically be achieved with the current data

What the data support. One site, most likely Connecting Ridge, with Malapert Massif as fallback. At that site the archives allow the following:

All of these inputs are public domain or CC BY 4.0, so commercial reuse is permitted. This matches a TRL 3 proof of concept: each critical function can be shown on real data and checked against independent measurements (judgement).

What the data do not support. The data do not support the following parts of the RFI draft and the CLAUDE.md baseline:

Centimetre-scale rocks and microtopography can only be synthetic. Ice depth is a model product. The validation images near the pole date mainly from 2009 to 2011, and few new ones will come before a Type A start.

What this means for the proposal (judgement). The data favour a scope built on traceable illumination and visibility, not on new terrain resolution. They limit how far the full layer list of the RFI draft can be measurement-grounded at the pole. The data-class rule is supported by the data themselves: most terrain cells at the best sites are interpolated or inferred, not measured. The same data are freely available to every competitor, so data access gives SELIS no advantage; any advantage must come from how the data are combined, classified and checked. Three points remain unproven and are risks: