Historical evidence report · matched RAM calculations

The seabed choice often moves modeled transmission loss (TL) by several decibels. It does not move every path the same way.

The geology-informed seabed usually gives more transmission loss in the largest matched HA01 cohorts, but exact coherent endpoints can move in either direction. The current desired state uses the geology-informed branch for new RAM coverage and keeps the fixed branch as historical comparison evidence.

No single decibel correction can turn one treatment into the other.

Across 678 matched H01W1 paths at 63 Hz, the median geology-informed minus historical fixed difference is +3.21 dB, with a 5th to 95th percentile range of -5.69 to +14.57 dB. The full range is -21.02 to +28.77 dB.

What changed

Two seabed descriptions, one propagation model

Historical fixed seabed

One engineering tuple along the path

The established HA01 work used 1,668.5 m/s compressional speed, 1.813 g/cm³ density and 0.74 dB per wavelength attenuation. The tuple stays spatially constant and now serves only as a historical comparator.

Geology-informed seabed

Range-varying complete tuples

OceanData classifies each sample, OceanAcoustics supplies a complete class-consistent tuple, and RAMTools passes those varying values to RAM. This changes the propagation environment, not the ship source model. The desired-state contract uses this branch for new RAM coverage.

Compressional speed, HA01 population1,623.7 m/smedian; 1,606.9 to 1,647.5 m/s from the 5th to 95th percentile

Density, HA01 population1.823 g/cm³median; 1.782 to 1.879 g/cm³ from the 5th to 95th percentile

Attenuation, HA01 population0.777 dB/λmedian; 0.770 to 0.784 dB/λ from the 5th to 95th percentile

The desired-state choice is an execution policy, not a physical validation. "Geology-informed" names how the inputs were chosen. It does not mean that the local seabed has been measured or that the resulting TL is known to be closer to reality.

Matched calculations

The size of the change depends on the cohort and on where TL is sampled

Each row below keeps frequency, water, bathymetry, geometry, source depth, grid and output definition fixed within a pair. The cohorts remain separate.

Matched cohort	n	Plotted measure	Median difference (dB)	5th to 95th (dB)	Full range (dB)	Median range-summary difference (dB)
Juan paths to the H01W1 hydrophone, 63 Hz	678	Endpoint TL	+3.209	-5.689 to +14.574	-21.021 to +28.767
Selected paths at HA01 and HA10, 63 Hz	18	Endpoint TL	+4.852	-3.690 to +10.422	-25.048 to +11.004	+3.669
Juan 26 March dense track, 63 Hz	180	Endpoint TL	+1.863	-2.261 to +4.949	-14.909 to +12.587
Juan 26 March dense track, 80 Hz	180	Endpoint TL	+1.915	-1.948 to +5.898	-7.661 to +10.402
January 2025 H01W array reciprocal grid, 63 Hz	1,800	Final range sample	+10.247	-2.505 to +43.615	-22.982 to +105.862	+4.654
January 2025 H01W array reciprocal grid, 80 Hz	1,800	Final range sample	+10.776	-1.615 to +36.246	-21.983 to +148.930	+4.599
February 2026 holdouts and refinements, 63 Hz	600	Final range sample	+15.146	-0.313 to +36.260	-13.341 to +90.679	+6.356

Why endpoint tails get so large

The seabed can shift coherent peaks and nulls along range

RAM computes a coherent pressure field. A change in seabed speed, density or attenuation changes phase as well as average loss. At one exact range, moving a narrow null can produce a large signed TL difference even when the neighboring range average changes by only a few decibels.

Eighteen matched fixed and geology-informed 63 Hz path comparisons — The selected 18-path study includes full fields. Its exact endpoint differences range from -25.05 to +11.00 dB, while every trailing-1-km linear-power difference is positive and the median is +3.67 dB. The isolated -25 dB endpoint is a shifted coherent null, not a 25 dB broadband correction.

January 2025 H01W sampled geology-informed minus historical fixed TL at 63 Hz — The separate January ship-screening derivative samples the paired reciprocal fields at the nearest computed bearing and range. The 1,411 row differences have median +3.80 dB and a 5th to 95th percentile range of -5.20 to +13.67 dB. No interpolation or measured acoustic comparison is implied.

What this report covers

Completed computation is broader than the portable comparisons summarized here

Family	Matched state at report cutoff	Use in this report
Juan H01W1 month-matched direct paths	678 pairs at 63 Hz; the 80 Hz computation is also complete	Full 63 Hz portable table; 80 Hz execution is not used for a numerical claim because no equivalent portable comparison table is bound here
Juan 26 March dense track	360 matched hydrophone/minute/frequency pairs	Separate 63 and 80 Hz interactive tables
January 2025 H01W reciprocal grid	1,800 pairs at 63 Hz and 1,800 at 80 Hz	Both paired tables, plus separate 1,411-row screening derivatives
February H01W, H10N and H10S array base bearings	4,536 pairs at 63 Hz and 4,536 at 80 Hz	Completion is recorded; no new statistic is inferred without a portable paired summary
February holdouts and refinements	600 validated 63 Hz pairs	Paired endpoint explorer and range-summary statistics
Former paired HA03/HA08 atlas	Outside current desired RAM coverage	Excluded; any saved fields remain historical sensitivity evidence
HA03/HA08 artificial-water pilot	Quarantined because it used a uniform artificial water profile	Excluded from evidence and conclusions

The HA03/HA08 artificial-water pilot remains quarantined. None of its TL values enters this report.

What the comparisons establish

Seabed treatment materially affects modeled coherent TL. The typical change is a few decibels in several large HA01 summaries, but the signed result depends on path, frequency and sampling rule.

What they do not establish

They do not rank the two treatments against real pressure measurements, validate local sediment truth, justify a universal correction, detect a ship, support ship attribution, or test prediction.