high-res Tier-2 surface patches (HiRISE/LROC at ~2 m/px, matches NASA reference imagery)

[chipi]

Problem

Tier-2 surface patches on /mars and /moon are visibly low-detail compared to the public NASA reference imagery (HiRISE / LROC NAC sample-site maps, JPL traverse animations). User-reported on 2026-06-01 while comparing Orrery's Perseverance Tier-2 view against the JPL Perseverance Mars Rover Drive Path Animation and Mars Rock Samples: The Stories They Could Tell videos — our surface texture is grainy, NASA's is crisp at the same field of view.

The cause is mechanical, not artistic: we serve 2048² JPEG crops of the source HiRISE / LROC NAC products, regardless of how big the bounding box on the body actually is. At Perseverance's region (~14.79 × 17.74 km) that works out to ~7 m/px effective. HiRISE native is ~0.25 m/px, so we are throwing away about a 28× detail multiplier on the ground.

Current state

Planet	Source	Native res	Served crop	Served res on typical region
Mars	MRO HiRISE	~0.25 m/px	2048² JPEG	~7 m/px
Mars (regional)	MRO CTX	~5 m/px	2048² JPEG	~10 m/px
Moon	LROC NAC	~0.5 m/px	2048² JPEG	~5 m/px

Examples on disk:

• static/images/hotspots/mars/perseverance/tier2-hirise.jpg — 2048×2048, 1.3 MB
• static/images/hotspots/mars/perseverance/tier2-ctx.jpg — 2048×2048, 959 KB
• static/images/hotspots/moon/apollo11/tier2-lroc.jpg — 2048×2048

Until 2026-06-01 the TierContext attribution card mislabeled all planets as "LROC NAC ROI mosaic" — that's been fixed in the same change as this issue (Mars now reads "HiRISE detail patch · ~7 m/px (from HiRISE 0.25 m/px native)"), so the labeling is honest, but the texture is still 28× under-resolved versus source.

Target state

Match the apparent fidelity of NASA's reference imagery when zoomed all the way in (Tier-2 patch fully visible at camR ≈ 30.32, plus the flat-patch overlay at kmPerPx < ~0.01).

Suggested resolution targets:

• Detail tier (HiRISE / LROC NAC): ~1–2 m/px on the served patch (≈ 4× current). That's a 4K–8K crop per site.
• Regional tier (CTX / LROC WAC): ~5 m/px at the served crop (matches CTX native, ≈ 2× current). 4K crop.

That gets us within visual parity of the NASA videos at the user's current max-zoom — not full 0.25 m/px native, but the missing factor goes from 28× → 4×, which is no longer a fidelity tell.

Proposed approach

Multi-slice, modelled after the ADR-073 Layer A / Layer B pattern that we already use on the orbital body textures (2K → 4K LOD swap on /explore + /earth + /moon orbital). The Tier-2 patch is the surface-tier equivalent.

Slice 1 — fetch-pipeline upgrade

• Extend the existing HiRISE / LROC fetch scripts (scripts/fetch-assets/) to emit a higher-resolution variant per site (4K or 8K) into a parallel directory: tier2-hirise.4k.jpg, tier2-hirise.8k.jpg.
• Source product IDs stay the same — we're just changing the crop dimensions + JPEG quality.
• Per-site optimal pixel size depends on bbox: a small landing ellipse (Beagle 2, ~0.1 km²) wants different treatment from a traverse bbox (Curiosity, ~15 × 5 km).

Slice 2 — runtime LOD swap (surface Layer C)

• In SurfaceScene.svelte's Tier-2 patch material, mirror ADR-073 Layer B's camR-gated texture swap: start with the 2K patch, swap to 4K/8K when the user is committed to the patch (camR < SPHERE_TO_FLAT_CAM_R + ε).
• Hysteresis prevents thrashing on the boundary.
• Skip the swap on slow connections / prefers-reduced-data (existing infrastructure on /explore).

Slice 3 — flat-patch view consumes the same high-res asset

• SurfaceFlatPatch.svelte currently draws from the same tier2-hirise.jpg. When the high-res variant exists, flat-patch picks it up — this is where the biggest perceptual win lands (flat-patch is where the user "stands" in the patch and wants to see craters / rover tracks / drill marks).

Slice 4 — file size + bandwidth audit

• 12 Mars sites × 8K patches at ~5 MB each = ~60 MB of new static assets. Compared to current ~3 MB across the same sites, that's a 20× growth. Whether we ship it as part of the static bundle or behind a fetch-on-demand layer is the decision point.
• Decision input: current SW precache budget (see W6: responsive image variants + mobile-tier SW precache #276), Cloudflare static cache strategy, mobile-tier behavior.

Out of scope

• New sites (this is per existing site only).
• True progressive tiling (Web Mercator pyramid). Possible follow-up if 8K is still not enough, but 8K per Tier-2 patch is probably the simplest sufficient solution and matches what /explore does for orbital body textures.
• 2D fallback view textures (those are different assets at lower res by design).

Acceptance criteria

• At camR < 30.5 on /mars or /moon with a region-bounds site selected and high-res variant loaded, the served pixel density should be ≤ ~2 m/px on a typical landing region.
• TierContext card resolutionText updates to match the actual served res ("~2 m/px" instead of "~7 m/px").
• No regression on cold-cache load time at low zoom (the 2K patch must still load first, with the 4K/8K swap happening only when committed to the close-zoom path).
• No regression on existing e2e snapshots — high-res variant must be additive, never replace the 2K base.

Related

• ADR-073 — defines Layer B (2K→4K base-texture LOD on orbital bodies). This issue adds the surface-tier equivalent (Layer C, or extend ADR-073).
• W6: responsive image variants + mobile-tier SW precache #276 — responsive image variants + mobile-tier SW precache (file size budget input).
• /explore: upgrade all planet base textures to 4K with LOD swap (parity with /earth) #287 — /explore 4K texture LOD retrospective (the orbital-body precedent).
• v0.7 release gate — consolidated (audio + image pipeline v2 shipped; sensory moved to v0.8) #282 — v0.7 release gate (consolidated image pipeline work).

Signals

Reported 2026-06-01 via direct user comparison with two JPL reference assets:

• Perseverance Mars Rover Drive Path Animation (JPLraw on YouTube)
• Mars Rock Samples: The Stories They Could Tell (NASA JPL on YouTube)