Topic A3 — Uncertainty in planning results: findings
15 November 2026; all sources opened that day. Follow-up 1 of
A2-competitors.md section 7. Read
A2-competitors.md and B-data-sets.md first;
this file does not repeat them. There is no topic file; the question
came from Ivaylo's request. Bracketed numbers refer to section 8. In
section 8, "(full text)" marks a paper read as text extracted from the
PDF or page, "(passages)" a document searched by keyword and read around
the hits, "(page summary)" a page read through an automated summary,
"(extract)" a source read as a search extract only. "Judgement" marks an
assessment, not a sourced fact. "Not found" means searched and not
found. Compiled with AI (Claude); not yet re-checked by a person.
Length: about 2,800 words before the source list, with links removed.
This is over the two-page limit of research-brief.md,
because the request had five sub-questions.
Abbreviations not in glossary.md: LDEM = DEM made from
LOLA data; RoI = region of interest; LCIP and LCSP = longest continuous
illumination and shadow period; LQCIP = longest quasi-continuous
illumination period; RMS = root mean square; SfS = shape-from-shading;
PSR = permanently shadowed region; PGDA = NASA Planetary Geodesy Data
Archive.
Question: does any published work or tool carry DEM uncertainty into illumination, Earth-visibility, solar-power or siting results at the lunar poles?
1. Feeds
1.1 objectives; 1.4.1 State of the Art; 1.5 Technical Problem Areas and Risks; 1.6.1 programmatic compliance (no duplication). Also the scope decision on "power siting with confidence".
2. Key findings
- The method is published. NASA did it in 2021 for four South Pole sites. Barker et al. ran the illumination computation on the nominal LDEM and on each of 100 clones, at 1 m and 5 m above the surface, hourly from 1 January 2024 to 1 January 2026 [1]. Sites: Connecting Ridge (Site 1), Shackleton rim (Site 4), a peak near Shackleton (Site 7), de Gerlache rim (Site 11); each 16 km by 16 km at 5 m/pixel [1]. Quantities: average illumination (visible fraction of the Sun's disc), LCIP and LCSP. Outputs: per-pixel maps of the 1st or 99th percentile and the standard deviation over the clones, a table per RoI, box plots, and a ranking of RoIs [1]. The authors wrote that this "has not been done extensively before" [1].
- Size of the effect: large at 1 m, almost none at 5 m. Site 1, seven RoIs, RoI-averaged case. At 1 m the 1st percentile of average illumination is 67 to 71%, of LCIP 63 to 142 days; the 99th percentile of LCSP is 13 to 22 days. At 5 m: 83 to 89%, 119 to 231 days, 4 to 8 days [1]. The paper states that LDEM uncertainties have "an almost negligible effect" on illumination at 5 m at all four sites [1]. The nominal LDEM predicts better conditions than the mean or median of the clones, because a height error raises the horizon more easily than it lowers it [1].
- What NASA did not publish. The clones, DEMs and uncertainty maps are public; the illumination results per clone and the illumination code are not listed on the PGDA page [2]. The paper compares one simulated image with one NAC image (M139811097) to judge the LDEM visually; it does not check the illumination statistics or their spread against images [1]. The clone model excludes rocks and boulders, and craters under about 100 m are under-represented in the LDEM [1]. NASA presented the method to users in 2021 with the line "Uncertainties can be incorporated into trades" [3].
- ESA's own Lunar Lander study stated the requirement in 2012. De Rosa et al. wrote that illumination characteristics (area size, LQCIP duration) must be derived "with a required confidence level", linked to mission success probability and landing-site certification [4]. They described a 3-sigma worst-case model, rejected it as over-conservative, and named a Monte Carlo run over terrain heights as the ideal solution, then judged it impractical for computing cost [4]. They reported error modelling as ongoing work; its results: not found. Their inter-tool check gives a spread: under 4% difference in accumulated illumination at the five best points; about 44.7 days RMS difference in LQCIP between two tools at Connecting Ridge at 2 m [4].
- Visibility with DEM uncertainty is also published. Mahanti and Robinson (2014) ran 100 Monte Carlo runs per observer on a NAC stereo DEM and produced "probabilistic viewsheds" with false-alarm and miss probabilities, observer height 1.5 m [5]. The site is on the rim of South Pole-Aitken basin, not at the pole. The same for Earth visibility at the South Pole: not found.
- A 2026 NASA paper says the published terrain uncertainties are too small. Bertone et al. compared SfS DEMs with the LDEMs and found the LDEM clone-based height uncertainties underestimated by about 40%; the ASP SfS height-error output underestimated their ensemble-based uncertainty by a factor of 1.2 to 8.5 [6]. They gave no uncertainty on their illumination maps [6].
- Validation against dated images is common; validation of a stated uncertainty was not found. Agreement figures are in table 3.2. Every case compares one nominal prediction with images. No source compares an ensemble spread or a confidence statement with images (judgement on the gap; absence in the sources read).
- Power-siting studies handle terrain error by margin, not by a stated confidence. Fincannon (NASA Glenn, 2008) lowered the site by 50 m and raised the terrain by 50 m as a conservative case and called this "the most appropriate approach for power system design" [11]. In 2020 he used a 10 m height offset partly to avoid "model uncertainty/error" [12]. Bryant (JPL, 2009) chose 2 m towers because of the 5 m vertical accuracy of the radar DEM [13]. The Aalborg microgrid siting study takes illumination time series for 15 sites and tower heights of 10, 50 and 100 m; an uncertainty treatment does not appear in its abstract (extract) [19].
- A tool or service that returns a confidence, probability or error bar on an illumination or siting result to a user: not found, in any country. NASA offers Artemis users 5 m illumination products on request for a chosen period and height above the surface, and names solar arrays at 5 m as an example; the text does not mention uncertainty [16]. LunaRay names "solar asset emplacement" as a use and validates renders against orbiter images; it describes no output uncertainty [15]. European tools: section 3.3.
3. Evidence
3.1 Work that carries terrain error into a result
| Work; country | Result quantity | Site; height | How terrain error enters | Uncertainty stated on the result | Check against images | Code or outputs public |
|---|---|---|---|---|---|---|
| Barker et al. 2021, NASA [1][2] | Average illumination, LCIP, LCSP; RoI ranking | Four South Pole sites; 1 m and 5 m | 100 clones, full recomputation | Yes: percentiles, standard deviation, box plots | One visual image comparison of the LDEM; none of the spread | Clones yes; illumination outputs and code no |
| Barker et al. 2023, NASA [17] | PSR area and PSR size distribution | South Pole tiles | Two clones | Fractional error in recovered PSR area | Not for this step | Not stated |
| De Rosa et al. 2012, ESA with Astrium [4] | LQCIP, illuminated area | South Pole sites incl. Connecting Ridge; about 2 m | Planned: 3-sigma or Monte Carlo; not carried out in the paper | No; inter-tool differences only | NAC snapshots; registration about 5 m RMS at shadow borders | No |
| Mahanti and Robinson 2014, US [5] | Line-of-sight viewshed | South Pole-Aitken rim; 1.5 m | 100 Monte Carlo runs | Yes: probability per cell | No | No |
| Fincannon 2008, NASA Glenn [11] | Illumination for power sizing | South Pole; surface and towers | Fixed ±50 m worst case | No; one conservative case | No | No |
| Otten et al. 2017, US [14] | Sun-synchronous rover route | Lunar pole | Not propagated; DEM error of 50 to 100 m named as a mission risk | No; operational buffer instead | No | No |
| Zhong et al. 2026, China [7] | Cumulative relative illumination | Chang'e-7 candidate area near Shackleton; surface | Resolution only: 1, 5, 20 m | No; bias from resolution (mean 0.167, 0.213, 0.276) | Two NAC images, table 3.2 | No |
3.2 Reported agreement between predicted illumination and dated images
| Work | Images | Metric and figure |
|---|---|---|
| Zhong et al. 2026 [7] | NAC M121374298LE (21 February 2010), M184474020RE (21 February 2012) | Lit or shadowed per pixel: 86.6% and 89.9% agreement; intersection over union 0.63 and 0.53 |
| Bertone et al. 2026 [6] | About 380 NAC images | Structural similarity 0.83 ± 0.07 for LDEM renders, 0.91 ± 0.05 for SfS DEM renders |
| Mazarico et al. 2018 [8] | Long-exposure NAC images of Faustini PSR, August 2012, at 50 m/pixel | Correlation above 0.9 after fitting one scale factor; radiance of scattered light, not sunlit timing |
| De Rosa et al. 2012 [4] | NAC images at simulated times | Visual; shadow-border offsets about 5 m RMS; features under 40 m not captured |
| Zhang et al. 2023 (extract) [18] | NAC and Chandrayaan-2 images, Chang'e-7 area | Mismatch rate of DEM-simulated images "mostly around 10%"; uses mismatched shadow length to judge DEM height error |
| Gläser et al. 2014 (extract) [10]; Gläser and Gläser 2019 [20] | NAC images; one WAC image | Visual; no figure. Gläser et al. 2018 gives no uncertainty on its 2 m results; it reports an interpolation error of 0.1 m (1 sigma) for its north-polar DEM [9] |
3.3 European tools, flagged separately
ESA Coverage Tool and Astrium ICAT (2012): no uncertainty output;
requirement stated, see finding 4 [4]. TU Berlin and DLR
illumination and thermal model: none stated [9][20].
Spaceship FR, PANGU, LandSAfe, Milrem and Tartu: none, as in
A2-competitors.md table 3.1. A European tool with a stated
confidence on an illumination or siting result: not found.
Also searched, nothing relevant found: JAXA LUPEX site selection; Chang'e-7 site selection, where the paper notes that its DEM misses small craters (extract) [21]; chance-constrained rover planning, where the modelled randomness is faults, not illumination (extract) [22]; commercial illumination services.
4. What this means for the Type A scope
All points are judgements.
- Verdict on the wording of the gap. "No published method": false; Barker et al. 2021 is the method, with the same quantities, two heights and a site ranking. "No published tool that delivers it to a planning user": survives as "not found in a review of November 2026". "No European tool": survives as "not found", with the weakness that ESA stated the need in 2012. "First" or "new method": do not use.
- Precedent an evaluator can cite: Barker et al. 2021 [1] against novelty of the method; De Rosa et al. 2012 [4] as proof that ESA already knows the idea; NASA's on-request service [16] against "nobody serves planners". Cite all three in 1.4.1 before the evaluator does.
- The research question that the sources leave open is calibration: does the predicted spread contain what dated NAC images show? No source tests this. A Type A proof of concept can test it at one site (TRL 3 critical function).
- Two further open points give the activity content beyond repeating NASA: error sources outside the clones (rocks, craters under about 100 m, the 40% underestimate in finding 6), and a comparison of LOLA-based and SfS-based ensembles at the same site.
- Height decides the value. With the published clones, uncertainty matters at 1 m and is almost nil at 5 m. A confidence statement adds little for a mast of 5 m or more unless the wider error sources of point 4 change that result. Check this against the end-customer's asset height before fixing the scope.
5. Ready-to-paste sentences
Re-check each against its source before use. Define acronyms again in the section where a sentence is used.
- Barker and co-authors computed illumination on 100 statistical clones of each Lunar Orbiter Laser Altimeter (LOLA) elevation model for four South Pole sites [1].
- At Connecting Ridge, 1 m above the surface, the 99th percentile of the longest continuous shadow period was 13 to 22 days [1].
- The same study found that terrain height uncertainty had an almost negligible effect on illumination 5 m above the surface [1].
- The ESA Lunar Lander study of 2012 required illumination characteristics with a stated confidence level and named a Monte Carlo simulation as the ideal error model [4].
- Zhong and co-authors reported 86.6% and 89.9% pixel agreement between simulated illumination and two Lunar Reconnaissance Orbiter Camera images [7].
6. Open risks and unknowns
Contradictions with the findings files and
CLAUDE.md:
A2-competitors.mdfinding 9 says uncertainty is published "for terrain, not for planning results". That is wrong for illumination: Barker et al. 2021 publishes percentiles of average illumination, LCIP and LCSP per RoI and uses them to rank landing areas [1]. The sentence about reviewed tools in the same finding still holds.A2-competitors.mdsection 4 point 2 presents the gap as a siting result that "states how sure it is, and is checked against measured data". The first half exists as a published method. Only the delivery to a user and the check of the stated spread against images were not found.B-data-sets.mdsection 4 point 1 says NASA "showed that height uncertainty changes predicted illumination". This understates the precedent: NASA ran the full propagation. It also omits that the effect is almost nil at 5 m [1].B-data-sets.mdfinding 1 quotes 0.3 to 0.5 m RMS height uncertainty for the site LDEMs. Bertone et al. 2026 find these uncertainties underestimated by about 40% [6].B-data-sets.mdtable row "SfS DEMs" says "paper not opened"; it is now opened.A2-competitors.mdtreats De Rosa et al. 2012 as a three-tool cross-check only. The paper also sets the confidence-level requirement and outlines the Monte Carlo method [4].A2-competitors.mdready-to-paste sentence 4 is correct but narrow: the 0.9 correlation is for scattered-light radiance in one PSR at 50 m/pixel after a fitted scale [8]. Do not cite it as validation of sunlit or dark timing.CLAUDE.mddoes not conflict. "Confidence" inglossary.mdis per cell; a confidence on a siting result is a new use of the term and needs a glossary entry before use.
Unknowns:
- Whether NASA's on-request illumination products include clone-based uncertainty. The Data Book does not say [16]. Whether a non-US company can request them: unknown.
- The outcome of ESA's error-modelling work after 2012: not found (searched: paper [4], web search on the Lunar Lander study and on Argonaut site analysis).
- Internal practice at Voyager (LunaRay), Intuitive Machines, ispace, JAXA and CNSA: not public. "Not found" here cannot exclude in-house uncertainty analysis.
- Gläser et al. 2014, Zhang et al. 2023, Saha et al. 2023 and two further items were read as extracts only; the publishers' pages refused automated access. Open them in a browser before citing.
- The supplementary figures of Barker et al. 2021 (Sites 4, 7, 11) were not opened; figures for those sites are from the main text only.
- Cost of the computation at Type A scale: Barker et al. call it "computationally expensive" and used a NASA computing centre; no run time is given [1].
- Searches used English keywords. Chinese-, Japanese- and Russian-language work may be missed.
Decisions for Ivaylo:
- (a) Reword the gap as in section 4 point 1 and accept that the method is NASA's.
- (b) Whether the research question becomes calibration against NAC images (section 4 point 3).
- (c) Which asset height the proof of concept targets, given finding 2.
7. Suggested follow-up topics
- Asset heights of planned South Pole solar power systems (masts, lander decks, rover panels) and of the candidate end-customer's product. It decides whether terrain uncertainty changes a siting result at the heights users need (1.1, 1.6.2).
- Dated NAC image coverage of one candidate site across Sun azimuths, and how many independent lit-or-dark checks it gives per cell. It decides whether a calibration test of the spread is feasible in a Type A activity (1.4.4, 1.5).
- Error sources outside the LOLA clones: rock abundance, craters under 100 m, SfS ensemble spread. It decides whether the activity adds to Barker et al. 2021 or repeats it (1.4.1, 1.5).
8. Sources
Accessed 15 November 2026.
- Barker, Mazarico, Neumann, Smith, Zuber, Head 2021, "Improved LOLA elevation maps for south pole landing sites: Error estimates and their impact on illumination conditions", Planetary and Space Science 203, 105119 (full text, NTRS manuscript): https://ntrs.nasa.gov/api/citations/20205009660/downloads/Mike%20Barker_%20Paper_%20Improved%20Lola.pdf ; record: https://ntrs.nasa.gov/citations/20205009660
- NASA PGDA, "High-Resolution LOLA Topography for Lunar South Pole Sites" (page summary, whole page in three parts): https://pgda.gsfc.nasa.gov/products/78
- Barker et al., LEAG 2021 slides (full text): https://ntrs.nasa.gov/api/citations/20210020639/downloads/leag2021_MikeBarker_v01_STI.pdf ; Barker et al., slides for a terrain-relative-navigation workshop, undated (full text): https://lsic.jhuapl.edu/uploadedDocs/presentations/766-2.01_Barker_TRNworkshopBarkerLOLADay2.pdf
- De Rosa et al. 2012, "Characterisation of potential landing sites for the European Space Agency's Lunar Lander project" (full text, arXiv version; sections 2.2.1 and 2.4): https://arxiv.org/abs/1208.5587
- Mahanti and Robinson 2014, "Informed lunar exploration with probabilistic viewsheds", EPSC2014-620 (full text): https://meetingorganizer.copernicus.org/EPSC2014/EPSC2014-620-3.pdf
- Bertone, McKenna, Barker, Mazarico, Beyer, Petro 2026, "Enhanced Topography Models for Selected Lunar South Pole Regions with Shape-from-Shading", Planetary Science Journal 7(5), 118, 20 May 2026 (page summary; appendices not read): https://iopscience.iop.org/article/10.3847/PSJ/ae5b70
- Zhong, Shi, Yang, Yan 2026, "Impact of Digital Elevation Model Resolution on Illumination and Thermal Stability Modeling in the Lunar South Polar Region", Astronomical Journal 171(6), 369, 29 May 2026 (page summary): https://iopscience.iop.org/article/10.3847/1538-3881/ae6918
- Mazarico, Barker, Nicholas 2018, Advances in Space Research 62(11) (full text): https://pmc.ncbi.nlm.nih.gov/articles/PMC6398960
- Gläser, Oberst, Neumann, Mazarico, Speyerer, Robinson 2018, "Illumination conditions at the lunar poles: Implications for future exploration" (full text, NTRS manuscript): https://ntrs.nasa.gov/api/citations/20170007365/downloads/20170007365.pdf
- Gläser et al. 2014, Icarus 243, 78 (extract of the abstract; the record did not open): https://ui.adsabs.harvard.edu/abs/2014Icar..243...78G
- Fincannon 2008, "Characterization of Lunar Polar Illumination From a Power System Perspective", NASA/TM-2008-215186 (passages): https://ntrs.nasa.gov/api/citations/20080018474/downloads/20080018474.pdf
- Fincannon, "Lunar Environment and Lunar Power Needs", NASA Glenn, NTRS record of 2020 (passages): https://ntrs.nasa.gov/api/citations/20205002224/downloads/Lunar%20Environment%20and%20Lunar%20Power%20Needs_2.pdf
- Bryant 2009, "Lunar Pole Illumination and Communications Maps Computed from Goldstone Solar System Radar Elevation Data", IPN Progress Report 42-176 (passages): https://ipnpr.jpl.nasa.gov/progress_report/42-176/176C.pdf
- Otten, Wettergreen, Whittaker 2017, "Strategic Autonomy for Reducing Risk of Sun-Synchronous Lunar Polar Exploration" (passages): https://publications.ri.cmu.edu/storage/publications/2018/01/FSR_2017_Otten_66.pdf
- Hough et al. 2022, "DEMkit & LunaRay: Tools for Mission Data Generation and Validation", Space Imaging Workshop (full text; same as A2 source 28): https://seal.ae.gatech.edu/sites/default/files/2022-10/SIW22-22_0.pdf
- NASA, "Lunar Surface Data Book", M2M-30044, baseline, effective 18 March 2025, section 5.1 (passages): https://ntrs.nasa.gov/api/citations/20250002845/downloads/M2M-30044%20Lunar%20Surface%20Data%20Book%20Baseline_Final.pdf
- Barker et al. 2023, "A New View of the Lunar South Pole from LOLA", manuscript (passages; same as B source 7): https://ntrs.nasa.gov/api/citations/20230014419/downloads/BarkernewviewSTIupdated.pdf
- Zhang et al. 2023, "Analysis of Illumination Conditions in the Lunar South Polar Region Using Multi-Temporal High-Resolution Orbital Images", Remote Sensing 15(24), 5691 (extract; the page refused automated access): https://www.mdpi.com/2072-4292/15/24/5691
- Saha et al. 2023, "Optimal Sizing and Siting of PV and Battery Based Space Microgrids Near the Moon's Shackleton Crater", IEEE Access 11, 8701 (extract of the abstract; the Aalborg University PDF refused automated access): https://ui.adsabs.harvard.edu/abs/2023IEEEA..11.8701S/abstract
- Gläser and Gläser 2019, "Thermal modeling of airless bodies – lunar polar case", EPSC-DPS2019-1573 (full text): https://meetingorganizer.copernicus.org/EPSC-DPS2019/EPSC-DPS2019-1573-1.pdf
- "Selection of Landing Sites for the Chang'E-7 Mission Using Multi-Source Remote Sensing Data", Remote Sensing 17(7), 1121, 2025 (extract; the page refused automated access): https://www.mdpi.com/2072-4292/17/7/1121
- Lamarre, Malhotra, Kelly 2024, "Safe Mission-Level Path Planning for Exploration of Lunar Shadowed Regions by a Solar-Powered Rover" (extract): https://arxiv.org/abs/2401.08558