class, / is the vessel length, and /o= 300 ft (91.4m) is the 
reference length. 
Although the CTVs transited within a constrained range 
of speeds—their typical transiting speed— and had similar 
dimensions (Table II; see Fig. 9 below), we tested whether 
he functional speed and length relationships proposed by 
:he J-E model hold for this vessel class. To assess this, we 
applied the J-E speed and length scaling to the measured 
RNLs and subtracted the scaled RNLs from the unscaled 
RNLs. This difference analysis serves as a sensitivity test of 
‘he J-E scaling: If these parameters explained a substantial 
fraction of the observed variability, applying the scaling 
would lead to systematic and frequency-dependent changes 
in the RNL distributions. 
Further analysis was conducted using GAM and RF 
regressions to investigate how speed, vessel length, propul- 
sion type, and DCPA influence CTV SLs. The propulsion 
:ype of each CTV was derived from publicly available spec 
ification sheets. Both methods allow for analysis of the con- 
tribution of individual factors to the observed variability in 
RNLs (Breiman, 2001; Wood, 2017), with separate models 
created for each frequency band and for different combina- 
ons of explanatory variables to enable a systematic com- 
narison of model performance using adjusted R? to account 
for differences in the number of predictors between models. 
Ill. RESULTS 
A. Measured CTV RNL spectra 
Figure 4 shows the spectral probability density (SPD) 
of the measured RNL spectra for all CTV passages 
‘Merchant ef al., 2013). 
The bulker reference spectrum is included for context. 
The measured CTV RNLs exceed the bulker RNL over large 
parts of the spectrum, particularly between 100 and 500 Hz 
and above 1 kHz. Levels below 25 Hz are affected by flow 
noise contamination at the hydrophone (not removed by the 
30) 
170 
A 
& 1607 
L 
zw) 
5 150 
3} 
440 
var 
- = 90th 
= 75th 
— Median (50th) 
= 25th 
= — 40th 
—— Mean 
- = Bulker (200 m: 14 kn)| 
DC 
10 
109 10° 
Frequency (Hz) 
FIG. 4. Spectral probability density (SPD) of RNLs of all CTV passages, 
zompared to a bulker reference RNL spectrum (200m, 14kn) used as a 
spectral benchmark. 
3410 J. Acoust. Soc. Am. 159 (4), April 2026 
https:/doi.org/10.1121/10.0043324 
JASA 
S 400 
[ 
"0 
80 | 
zol 
60 
50 
40: 
30 
an 
& 
nn 
Frequency (Hz} 
N 
FIG. 5. Percentage of CTV passages for which the measured RNLs exceed 
he bulker reference spectrum in each frequency band. 
applied background noise correction), and propagation of 
these low frequencies is limited in the shallow water envi- 
ronment. The pronounced peak in the 20 kHz band is consis- 
tent with the emission characteristics of ultrasonic 
antifouling devices that may be installed on several CTVSs, 
although this needs to be verified and requires further 
ınvestigation. 
Figure 5 illustrates the percentage of analyzed CTV 
passages for which the measured RNLs exceed the bulker 
reference spectrum in each frequency band, confirming that 
exceedances predominantly occur around 125Hz and 
L5 kHz. 
In general, the measured RNL spectra (Fig. 4) exhibit a 
relatively low variability, but there are substantial differ- 
ances between individual vessels. The SPD plots for the 13 
individual vessels (see Fig. Sl in the supplementary 
material) demonstrate that the measured RNLs are broadly 
comparable in magnitude to the bulker reference spectrum. 
with all vessels exceeding it in specific frequency bands. 
The bulker spectrum is used here as a general reference for 
comparison. 
B. Parameter dependence 
To evaluate whether the speed and length scaling pro- 
posed by the J-E model (MacGillivray and de Jong, 2021) 
captures a substantial fraction of the observed RNL variabil- 
ity for transiting CTVs, we compared measured RNLs 
before and after applying the J-E scaling. Figure 6 shows the 
mean and median differences between scaled and unscaled 
RNL spectra. 
The generally small and weakly frequency-dependent 
differences indicate that the J-E speed and length scaling 
removes only a minor portion of the observed variability. 
This suggests that, within the limited speed and length range 
of the CTV dataset, RNLs are only weakly controlled by 
vessel speed and length alone. 
Spearman correlation coefficients were calculated to 
further investigate the relationship between RNLs and 
Basan et al.