Sediment inputs actually used
This report starts from accepted computation receipts. It leaves out sediment sources that the software could query but that supplied no retained input to these runs.
Geographic overview
Click a station marker to open its case. A violet ring means EarthByte refined a compatible Diesing class.
Four bounded cases
The counts below share one source-map cohort and therefore add cleanly. They should not be mixed with later execution cohorts.
HA01 · Southwest Australia
The Australian parameter grids supplied mud, sand and gravel fractions at every accepted query. The fractions were normalized and mixed into one bottom tuple at each position.
Union across the four areas
These products describe sediment. None directly measures the complete acoustic bottom tuple at the modeled coordinates.
Geoscience Australia interpolated MARS seabed samples onto 0.01° grids of mud, sand, gravel and bulk carbonate across Australia’s marine jurisdiction. We use the three texture fractions; carbonate remains a separate attribute.
Quality reading. Grid spacing is not observation density. The underlying archive is sparse offshore, and the published product has no per-cell uncertainty layer.
A 10 × 10 km equal-area prediction for ocean deeper than 500 m. It provides a dominant class, five class-probability surfaces and maximum probability. The classifier was trained on 10,438 records after filtering and duplicate reduction.
Quality reading. Spatial cross-validation reported 59.4% overall accuracy. Calcareous sediment and clay were strongest; rarer classes were much less reliable.
EarthByte’s 0.1° global dominant-lithology grid distinguishes 13 classes. It was trained from 14,399 quality-controlled surface or near-surface samples selected from more than 200,000 archive records.
Quality reading. Reported accuracy reaches 80% for major lithologies, not for every class or cell. The distributed dominant-class grid has no cell-level confidence field.
From geology to RAM
OceanAcoustics 0.1.25 freezes one coherent value set. RAM consumes attenuation in decibels per compressional wavelength.
Normalize the mapped texture fractions so their sum is one. Density and attenuation are ordinary weighted averages. Sound speed is averaged through slowness—the time taken per unit distance.
ρ = Σ fᵢ ρᵢc = 1 / Σ(fᵢ / cᵢ)αλ = Σ fᵢ αλ,iFor comparison in dB per metre: αm = αλ f / c. That conversion is explanatory; RAM receives αλ.
| Material | c m/s | ρ g/cm³ | αλ dB/λ |
|---|---|---|---|
| Gravel | 1,800 | 2.00 | 0.60 |
| Sand | 1,650 | 1.90 | 0.80 |
| Silt | 1,575 | 1.70 | 1.00 |
| Clay | 1,500 | 1.50 | 0.20 |
| Mud | 1,495 | 1.47 | 0.7151 |
Gravel through clay use the indicative engineering values in Jensen et al., Computational Ocean Acoustics, Table 1.3. Broad calcareous and radiolarian fine cover maps to silt; unresolved HA03 material maps to the labelled sand fallback.
The mud exception
Mud sound speed and density use the silty-clay means from Orsi and Dunn. Its attenuation preserves a 63 Hz mud-to-sand ratio on Jensen’s sand scale:
αλ,mud = 0.8 × (0.0598 / 0.0669) = 0.7151 dB/λThe sand reference follows Zhou, Zhang and Knobles. This is a transparent project bridge, not a universal empirical law for mud.
Interpretation boundary