Reading Lava Tube Skylight Acoustics for Entry Planning
The Information Gap at the Skylight Rim
A Marius Hills descent concept hits a familiar wall during entry planning: orbital imagery can tell the team the pit dimensions to within a few meters, but the geometry past the overhang is guesswork until a sensor physically drops through. The cost of acting on bad geometry estimates is asymmetric, because a wasted mobility budget cannot be recovered after sol-1 commitments. Haruyama et al. 2009's SELENE observation of the Marius Hills Hole pinned the pit at about 65 m wide and 80-88 m deep, and Kaku et al. 2017's SELENE Lunar Radar Sounder analysis showed echo patterns consistent with a large void extending from the pit. The radar gives a first-order void estimate, not the engineering-grade geometry an entry planner needs to choose a drop point, a tether path, or a rover egress profile.
Martian candidates share the same gap. The HiRISE image of an Arsia Mons lava tube mouth captures a pit roughly 50 meters across and 178 meters deep, and Diviner thermal measurements show lunar pits running about 100 K warmer than the surface at night. The thermal signal is evidence of a connected subsurface space, not a map. Photogrammetric modeling of lunar lava tube pits adds another data layer but still resolves only what the orbital cameras can see. Entry teams need the first inward-facing measurement before they commit to the descent.
This information gap drives concrete operational risk. A descent rover that drops a tether the wrong distance into a Marius Hills void wastes irreplaceable mission tether and may end up dangling in unsurveyed space; a CADRE-class hopper that picks the wrong egress angle from a Hadley Rille analog wastes hours of mobility budget that the analog campaign cannot recover. Each unknown that the rover discovers after committing forces the operations team to fall back on contingency plans that cost time, energy, and (in flight) sols of the bus's limited mission lifetime. A skylight quilt produced before the descent commits removes a substantial fraction of these unknowns at the cheapest possible point in the mission. Cave diving survey teams working on cenote entrance mapping face the same operational asymmetry: an entrance mismeasurement compounds through the dive plan in ways that are expensive to correct mid-traverse.
How EchoQuilt Builds an Initial Quilt From Rim Acoustics
EchoQuilt reads a skylight as the first patch of a cave quilt rather than as a discrete measurement. A passive microphone array placed at the rim records a spectrum of incidental sounds: Martian wind vortices funneling over the lip, thermal-contraction ticks from basalt after local sunset, rover-generated vibration propagating through the regolith. Each of these transient sources excites the void below and returns a reverberation fingerprint. The fingerprint encodes the low-order acoustic modes of the chamber, which IEEE work on inference of room geometry from acoustic impulse responses has shown are enough to recover bounding surface geometry at meter-scale accuracy for rooms of this size. The inference works without an active source because the natural variability of incidental skylight sources sweeps enough of the relevant frequency band over a few-hour recording window to constrain the modal solution.
EchoQuilt stitches that first inferred patch into the quilt and marks its uncertainty envelope honestly. A 65-meter Marius Hills pit has a dominant first-mode reverberation around 2-5 Hz, well inside the passive recording band, and the later-arriving reflections carry information about the sinuous rille structure extending away from the pit. By the time the entry team commits to a descent path, the quilt already contains a coarse geometry estimate of the first 50-100 meters of void beyond the rim, along with a confidence map. The rover can then pick its entry trajectory using the quilt rather than relying only on orbital data, and the operations team can plan a contingency egress around quantified uncertainty rather than around an unbounded unknown.
A companion post on skylight wind noise separation matters because Martian aeolian signal at the rim can swamp the reverberation features, and the classifier that separates them is the single most-tuned component of the entry-planning configuration.
The geometry-inference step also benefits from contextual priors loaded before the recording window opens. EchoQuilt accepts orbital photogrammetric models as a coarse prior on the visible portion of the pit, which constrains the inference to plausible chamber sizes from the start. SELENE radar sounding data provides a similar prior at depth, and Diviner-derived thermal-emissivity maps constrain the wall material composition that drives reflection coefficients. Loading the prior cuts the inference's initial uncertainty by roughly an order of magnitude in our analog tests, which means the first quilt patch the entry team sees is already useful for go/no-go decision making rather than a wide bound that has to be tightened by minutes of additional recording.
Advanced Tactics for Entry-Stage Quilt Reconstruction
Three tactical moves raise the utility of skylight acoustic data during entry planning. First, target a recording window that spans local dawn or dusk. Thermal-contraction events at these times produce repeatable transient sources that sweep the acoustic modes of the void, giving the inference model richer data than a steady-state wind spectrum alone. This is particularly useful at lunar skylights, where Diviner-documented thermal swings of roughly 100 K drive these events reliably.
Second, deploy a distributed array with at least three widely spaced microphones around the skylight rim rather than a single colocated cluster. A distributed aperture improves angle-of-arrival estimation for late-arriving reflections, which means the quilt's initial patches extend further into the void with tighter uncertainty. The CHILL-ICE habitat deployment in Surtshellir demonstrated that astronaut-placed nodes can cover a skylight perimeter in a single EVA, so the distributed configuration is EVA-realistic.
Third, log the quilt's first-pass geometry before the rover descends and version it against the post-entry quilt. The difference between the skylight-only estimate and the in-cave ground truth gives a reusable correction curve for future entries at similar pit scales. Over a small number of analog campaigns, this correction curve becomes the basis for descent planning at pits that cannot be entered directly, which is the regime most lunar and Martian skylights will remain in for years.
A fourth tactic that tends to surface only after a team's first analog campaign is staggered recording across diurnal cycles. The skylight's acoustic transfer function changes systematically with thermal cycling, atmospheric pressure variation (on Mars), and the daily wind pattern, and similar diurnal effects show up inside rille habitats deeper in the cave system. Recording at three to five different times of sol, then averaging the inferred geometry across those samples, suppresses transient measurement biases and tightens the patch's confidence envelope substantially. EchoQuilt's pipeline accepts a multi-window recording schedule and produces a per-patch confidence trace that flight reviewers can interrogate for systematic error.
Lunar concepts get an extra benefit: the lunar day is 14 Earth days, so a multi-window campaign at a Marius Hills analog can sample most of the relevant thermal phase space in a single Earth-week of recording at the analog site, which is a dramatic compression of the validation timeline.
CTA
EchoQuilt is opening capacity for flight concept teams planning skylight-first missions at Marius Hills, Arsia Mons, Hadley Rille analogs, or Mare Tranquillitatis scale pits. We configure the rim-mounted passive array, the initial-quilt inference pipeline, the orbital-prior loader for SELENE radar and Diviner thermal data, and the uncertainty reporting that flight review boards expect. Join the Waitlist for Planetary Analog Researchers to align a configured stack with your next analog skylight campaign at a Surtshellir, Mauna Loa, or Corona-class field site. JPL cave rover teams, NIAC Phase II PIs, ESA PANGAEA leads, and CMU autonomy groups with booked field windows in 2026 get priority access to configuration support and field engineering integration.