# Marine-mammal noise exposure

> One page of the phonometry documentation, as Markdown. Index: https://docs.jmrp.io/phonometry/llms.txt

Canonical: https://docs.jmrp.io/phonometry/underwater/marine-mammal-exposure/
Language: en
Alternate: https://docs.jmrp.io/phonometry/es/underwater/marine-mammal-exposure/index.md
Version: phonometry 4.0.0
Install: pip install "phonometry[full]"
Author: José Manuel Requena Plens (https://orcid.org/0000-0003-1250-6212)
Published: 2026-07-30
Updated: 2026-11-16T21:36:10Z
Standards: ISO 18406:2017
Verified: 4 conformance checks (ISO 18406:2017) https://docs.jmrp.io/phonometry/reference/conformance/#underwater-acoustics-iso-18405-17208-18406
License: MIT (https://github.com/jmrplens/phonometry/blob/main/LICENSE)
Summary: Marine mammals: Southall et al. (2019) audiograms, NMFS 2024, NMFS 2018 or Southall weightings, TTS and injury criteria, pile-driving exposure.

---

A marine mammal does not hear every frequency equally, so an underwater noise
assessment cannot compare a broadband level against a single number. Regulatory
practice instead **weights** the spectrum with a filter shaped like the hearing
group's sensitivity, sums the weighted energy, accumulates it over the whole
activity and compares the result against a published onset criterion: the
exposure at which a **temporary threshold shift (TTS)** in hearing sensitivity
begins, and the higher one at which the shift becomes permanent, historically
**PTS** and, since the 2024 guidance, **auditory injury (AUD INJ)**. This page
covers the three pieces the library provides for that chain: the hearing curves
themselves, the weighting functions and criteria of the current guidance, and
the end-to-end assessment of a piling campaign.

Levels are in dB re 1 µPa (sound pressure), dB re 1 µPa²·s (sound exposure) or,
for the two in-air carnivore groups, dB re 20 µPa and dB re (20 µPa)²·s. The
underwater reference conventions themselves are in
[Underwater acoustics](https://docs.jmrp.io/phonometry/underwater/underwater-acoustics/).

## Hearing groups (and why the names collide)

Every current criteria set sorts marine mammals into hearing groups and gives
each one a filter. The **group codes are not portable between guidance
versions**, and the collision is a real trap:

| Animal group | Southall 2019 | NMFS 2024 | NMFS 2018 | Audiogram |
|---|---|---|---|---|
| Baleen whales | LF | LF | LF | no ($F_1$ unpublished) |
| Sperm, beaked and most delphinid whales | HF | HF | **MF** | yes (`"HF"`) |
| Porpoises, *Cephalorhynchus*, *Kogia* | VHF | VHF | **HF** | yes (`"VHF"`) |
| Phocid seals in water | PCW | PW | PW | yes (`"PCW"`) |
| Otariids in water | OCW | OW | OW | yes (`"OCW"`) |
| Sirenians | SI | — | — | yes (`"SI"`) |
| Phocid seals in air | PCA | PA | — | yes (`"PCA"`) |
| Otariids in air | OCA | OA | — | yes (`"OCA"`) |

Read the column you are working in. `auditory_weighting` and
`exposure_criteria` take the code of the `guidance` you select;
`group_audiogram` always takes the **Southall** code, because the audiograms are
Southall's. So `group_audiogram(f, "PW")` raises even though `"PW"` is a valid
NMFS 2024 weighting group (ask for `"PCW"`); `auditory_weighting(f, "PCW")`
raises under the default guidance for the mirror-image reason.
`hearing_groups(guidance)` lists the codes a version defines, and passing a
code from the wrong version raises rather than silently returning the wrong
filter.

```python
from phonometry import underwater

print(underwater.hearing_groups("nmfs-2024"))
# ('LF', 'HF', 'VHF', 'PW', 'OW', 'PA', 'OA')
print(underwater.hearing_groups("nmfs-2018"))
# ('LF', 'MF', 'HF', 'PW', 'OW')
```

## Group audiograms

`group_audiogram` evaluates the band-pass fit of Southall et al. (2019),
Equation (1), after Finneran (2016):

$$
T(f) = T_0 + A \log_{10}\!\left(1 + \frac{F_1}{f}\right) + (f/F_2)^{B},
$$

with $f$ in kilohertz. `normalized=False` (the default) uses the Table 2
parameters, fitted to the absolute median behavioural thresholds;
`normalized=True` uses the Table 3 refit on thresholds normalised to each
individual's best value. The article publishes **no fitted audiogram for LF
cetaceans** (no audiometric data exist and $F_1$ is never printed), so that
group is deliberately absent rather than reconstructed by guesswork.

```python
import numpy as np
from phonometry import underwater

freqs = np.logspace(2, 5.3, 400)
audiogram = underwater.group_audiogram(freqs, "VHF")
print(audiogram.best_frequency, audiogram.best_threshold)
audiogram.plot()   # threshold vs frequency (needs matplotlib)

print(underwater.AUDIOGRAM_GROUPS)
# ('HF', 'VHF', 'SI', 'PCW', 'OCW', 'PCA', 'OCA')
```

![Left, the hearing threshold against frequency from 100 hertz to 200 kilohertz for the seven Southall 2019 audiogram groups, in-water groups solid and the two in-air carnivore groups dashed, each with its best-sensitivity point marked and a note that no fit is published for low-frequency cetaceans; right, the three-branch killer-whale audiogram over 0.5 to 80 kilohertz with its branch breaks marked, its 39 decibel minimum at 22.4 kilohertz, the 51.2 decibel value the third branch gives at 50 kilohertz and the 50.5 decibels the second branch would have given there](https://docs.jmrp.io/phonometry/media/marine_mammal_audiograms.svg)

*The groups are tens of decibels and more than a decade of best frequency apart:
a porpoise hears best near 109 kHz at 48 dB re 1 µPa, a sirenian near 14 kHz at
60 dB. The gap at the left of the panel is the missing LF fit, and it is the
reason the weighting section below has a group the audiogram section does not.
On the right, the two markers at 50 kHz are 0.7 dB apart and that is the whole
point of the branch trap. An `AudiogramResult` draws itself with
`audiogram.plot()`.*

**Show the code for this figure**

```python
import matplotlib.pyplot as plt

fig, (ax_g, ax_o) = plt.subplots(1, 2, figsize=(13.5, 5.6))
f_all = np.logspace(2.0, np.log10(200e3), 700)
for group in underwater.AUDIOGRAM_GROUPS:
    res = underwater.group_audiogram(f_all, group)
    ax_g.semilogx(res.frequencies, res.threshold,
                  "--" if res.in_air else "-", label=group)
    ax_g.plot([res.best_frequency], [res.best_threshold], "o")
ax_g.set_ylim(-20.0, 170.0)
ax_g.set(xlabel="Frequency [Hz]", ylabel="Threshold [dB re 1 uPa]")
ax_g.legend(ncols=2)

f_orca = np.logspace(np.log10(500.0), np.log10(80e3), 600)
orca = underwater.orca_audiogram(f_orca)
ax_o.semilogx(orca.frequencies, orca.threshold, label="orca_audiogram")
ax_o.plot([orca.best_frequency], [orca.best_threshold], "o")
ax_o.plot([50e3], [underwater.orca_audiogram(50e3).threshold[0]], "s")
# What the second branch would have returned at 50 kHz, from the printed fit.
ax_o.plot([50e3], [242.9 * 50.0 ** -0.7578 + 0.5643 * 50.0**1.076],
          "o", markerfacecolor="none")
ax_o.set(xlabel="Frequency [Hz]", ylabel="Threshold [dB re 1 uPa]")
ax_o.legend()
plt.show()
```

For a species rather than a group, `orca_audiogram` implements the killer-whale
curve of Wensveen & Van Roij (2007) as printed in Ainslie (2010),
Equation (11.159), a three-branch power law over 0.5 to 80 kHz. Its minimum is
39.0 dB re 1 µPa at 22.6 kHz, and 51.2 dB re 1 µPa at 50 kHz. That second
value needs the **third** branch: evaluating the second one there returns
50.5 dB instead, which is why both published points are pinned by the tests.

```python
from phonometry import underwater

print(underwater.orca_audiogram(50e3).threshold[0])   # 51.20 dB re 1 uPa
```

That value is the hearing threshold $HT$ in Ainslie's orca-versus-salmon
example. An echolocating orca is an active sonar: it pays the propagation loss
twice, so with a source level $SL = 198.2$ dB re 1 µPa²m², a salmon target
strength $TS = -29.0$ dB re m² and its own hearing threshold in place of a
detection threshold, the hearing-limited figure of merit is
$(SL + TS - HT)/2 = 59.0$ dB re m²: the maximum *one-way* loss it can afford.
The [sonar equation](https://docs.jmrp.io/phonometry/underwater/underwater-propagation/#sonar-equation)
those terms come from, and `detection_range`, turn that figure into a range.

## Auditory weighting and exposure functions

An audiogram and a weighting function look alike and answer different questions,
and substituting one for the other is a common error in assessments. An
**audiogram** is a detection threshold, measured behaviourally or by evoked
potential; it answers questions about audibility, masking and detection range:
the orca-versus-salmon figure of merit above is exactly that kind of question. A
**weighting function** is fitted to *susceptibility* data, the frequency
dependence of temporary threshold shift, and it is the only one of the two that
may be applied to an exposure before comparing it against a TTS or injury
criterion. The two curves are not the same shape, because the frequency at which
an animal hears best is not always the frequency at which it is most easily
damaged, and an inverted audiogram is not a legitimate weighting function. That
distinction also resolves the asymmetry between these two sections: LF cetaceans
have a weighting function derived from modelled susceptibility, and no fitted
audiogram, because no audiometric data exist for baleen whales.

All three current criteria sets use the same generic band-pass filter (NMFS
2018 Equation 1, Southall et al. Equation 2):

$$
W(f) = C + 10 \log_{10} \frac{(f/f_1)^{2a}}
                       {\left[1+(f/f_1)^2\right]^{a}\left[1+(f/f_2)^2\right]^{b}} ,
$$

with $f$ in kilohertz. $C$ is fixed by putting the peak of $W$ at 0 dB, so the
companion **exposure function** $E(f) = K + C - W(f)$ has its minimum at the
weighted TTS-onset threshold $T_\mathrm{w} = K + C$. Below $f_1$ the filter falls at
$20a$ dB/decade and above $f_2$ at $20b$ dB/decade.

Only the parameter table changes between versions, so the version is an
explicit argument and is carried on the result:

- **`"nmfs-2024"`** (the default): NOAA Fisheries *Updated Technical
  Guidance* v3.0, October 2024. It supersedes the 2018 revision, sets $b = 5$
  for every group, adopts the Southall group names and replaces "PTS onset"
  with "auditory injury (AUD INJ) onset".
- **`"nmfs-2018"`**: the 2018 revision v2.0, still cited by assessments
  already in flight.
- **`"southall-2019"`**: the peer-reviewed criteria, numerically identical to
  NMFS 2018 on the five shared groups and adds sirenians and both in-air
  carnivore groups.

![Auditory weighting functions of the five NMFS 2024 in-water hearing groups: low-frequency cetaceans peak near 1.4 kHz, high-frequency cetaceans near 11 kHz, very high-frequency cetaceans near 27 kHz, phocid pinnipeds near 5.6 kHz and otariid pinnipeds near 7.8 kHz, each falling steeply outside its passband](https://docs.jmrp.io/phonometry/media/marine_mammal_weighting.svg)

*Five band-passes of the same algebraic form, and choosing the wrong one costs
more than the criterion itself does. At 1 kHz (where a pile drive or a ship
puts most of its energy), the LF filter is at −0.03 dB and the VHF filter at
−33.84 dB, so the same measured spectrum yields two weighted exposures 34 dB
apart before any threshold is applied. The −6 dB passbands barely overlap
(96 Hz-15 kHz for LF against 5.9-113 kHz for VHF), and the AUD INJ thresholds in
the legend run from 159 to 193 dB, so group and criterion have to be quoted
together or the number means nothing.*

**Show the code for this figure**

```python
import matplotlib.pyplot as plt
import numpy as np
from phonometry import underwater

freqs = np.logspace(1, 5.4, 700)
fig, ax = plt.subplots()
for group in ("LF", "HF", "VHF", "PW", "OW"):
    res = underwater.auditory_weighting(freqs, group, guidance="nmfs-2024")
    crit = underwater.exposure_criteria(group, guidance="nmfs-2024", impulsive=True)
    ax.semilogx(res.frequencies, res.weighting,
                label=f"{group} (AUD INJ {crit.injury_sel:.0f} dB)")
ax.set(xlabel="Frequency [Hz]", ylabel="Weighting amplitude W(f) [dB]",
       ylim=(-75, 5))
ax.legend()
ax.grid(True, which="both", alpha=0.3)
plt.show()
```

```python
from phonometry import underwater

res = underwater.auditory_weighting(1000.0, "LF", guidance="nmfs-2018")
print(res.weighting[0])           # -0.06 dB, the published Appendix D value
print(res.weighted_tts_onset)     # Tw = K + C

params = underwater.weighting_parameters("OW", guidance="nmfs-2024")
print(params.a, params.b, params.f1_khz, params.f2_khz, params.c_db)
```

**Note**

NMFS 2024 prints $C = 1.37$ dB for the otariid in-water group and states in the
same table's footnote that the value should be 1.36 dB. The library implements
1.36 (recomputing $C$ from the peak of $W$ gives 1.3643) and keeps the printed
digit as `c_db_as_printed`. Southall's Table 7 carries four peak-SPL misprints
corrected by the journal's own errata, and the corrected values are the ones
implemented. Both are recorded in
[Errata found in published sources](https://docs.jmrp.io/phonometry/reference/errata/).

![Top left, the exposure function E(f) of the five in-water hearing groups against frequency, each with its minimum marked as that group's weighted TTS onset; top right, the low-frequency cetacean weighting function under the 2024 and 2018 guidance showing the steeper high-frequency skirt that b equal to five produces; bottom, the impulsive onset criteria of all five groups as grouped bars, weighted sound exposure and unweighted peak side by side](https://docs.jmrp.io/phonometry/media/marine_mammal_exposure_functions.svg)

*The exposure function is what a measured band level is actually compared
against: a threshold spectrum whose minimum is the group's weighted TTS onset
$T_\mathrm{w} = K + C$. The 2024 revision set $b = 5$ for every group, which steepens the
high-frequency skirt from 40 to 100 dB/decade and therefore *discounts* energy
above $f_2$ far more heavily than the 2018 tables did. The bars show why the
group matters more than anything else in the chain: the porpoise group's
weighted TTS onset is 24 dB below the low-frequency cetacean's, and its peak
criterion 20 dB below.*

**Show the code for this figure**

```python
import matplotlib.pyplot as plt

f_w = np.logspace(1.0, np.log10(400e3), 800)
groups = ("LF", "HF", "VHF", "PW", "OW")
fig = plt.figure(figsize=(13.5, 9.6))
gs = fig.add_gridspec(2, 2, height_ratios=[1.0, 0.9], hspace=0.32, wspace=0.24)
ax_e, ax_w, ax_c = (fig.add_subplot(gs[0, 0]), fig.add_subplot(gs[0, 1]),
                    fig.add_subplot(gs[1, :]))

for group in groups:
    res = underwater.auditory_weighting(f_w, group, guidance="nmfs-2024")
    exposure = np.asarray(res.exposure_function)
    ax_e.semilogx(res.frequencies, exposure, label=group)
    ax_e.plot([res.frequencies[int(np.argmin(exposure))]], [exposure.min()], "o")
ax_e.set_ylim(135.0, 265.0)
ax_e.set(xlabel="Frequency [Hz]", ylabel="E(f) = K + C - W(f) [dB re 1 uPa^2 s]")
ax_e.legend(ncols=5)

for guidance in ("nmfs-2024", "nmfs-2018"):
    res = underwater.auditory_weighting(f_w, "LF", guidance=guidance)
    ax_w.semilogx(res.frequencies, res.weighting,
                  label=f"{guidance} (b = {res.parameters.b:g})")
ax_w.set_ylim(-60.0, 5.0)
ax_w.set(xlabel="Frequency [Hz]", ylabel="Weighting W(f) [dB]")
ax_w.legend()

idx = np.arange(len(groups), dtype=float)
for offset, attr in ((-1.5, "tts_sel"), (-0.5, "injury_sel"),
                     (0.5, "tts_peak_spl"), (1.5, "injury_peak_spl")):
    ax_c.bar(idx + offset * 0.2,
             [getattr(underwater.exposure_criteria(g, guidance="nmfs-2024",
                                                   impulsive=True), attr)
              for g in groups], 0.2, label=attr)
ax_c.set_xticks(idx, groups)
ax_c.set_ylim(120.0, 245.0)
ax_c.set(xlabel="Hearing group", ylabel="Onset criterion [dB]")
ax_c.legend(ncols=4)
plt.show()
```

## Onset criteria

`exposure_criteria` returns the published TTS and injury onset criteria of a
group. Impulsive noise carries a **dual metric**: a weighted sound exposure
level *and* an unweighted ("flat") peak sound pressure level, and the criterion
that produces the larger **isopleth** governs: the isopleth being the contour
on which the criterion is exactly met, so the metric that pushes it furthest
from the source is the one that sets the mitigation zone.

```python
from phonometry import underwater

crit = underwater.exposure_criteria("VHF", guidance="nmfs-2024", impulsive=True)
print(crit.injury_label)        # 'AUD INJ'
print(crit.tts_sel, crit.injury_sel)          # 144, 159 dB re 1 uPa2 s (weighted)
print(crit.tts_peak_spl, crit.injury_peak_spl)  # 196, 202 dB re 1 uPa (flat)
print(crit.source)
```

The impulsive criteria of the five in-water groups, under the default guidance,
are the numbers most readers come here for:

| Group | TTS SEL (weighted) | AUD INJ SEL (weighted) | TTS peak (flat) | AUD INJ peak (flat) |
|---|---:|---:|---:|---:|
| LF | 168 | 183 | 216 | 222 |
| HF | 178 | 193 | 224 | 230 |
| VHF | 144 | 159 | 196 | 202 |
| PW | 168 | 183 | 217 | 223 |
| OW | 170 | 185 | 224 | 230 |

The SEL columns are dB re 1 µPa²·s and the peak columns dB re 1 µPa; the in-air
groups PA and OA come from the same call and use 20 µPa references. The
non-impulsive criteria drop the peak columns entirely and sit on their own TTS
values plus 20 dB (LF 177 → 197, VHF 161 → 181), so they are not the impulsive
numbers with a column removed.

The tables are internally consistent in ways the library pins as tests: the
non-impulsive injury level is always TTS + 20 dB, the published weighted TTS
onset is always the rounded $K + C$, and in Southall's impulsive table the SEL
criteria run TTS + 15 dB and the peak criteria TTS + 6 dB, which the table
above satisfies row by row.

## A worked pile-driving assessment

![The exposure-assessment configuration: a pile driven by an impact hammer, a calibrated hydrophone at the ISO 18406 range of 750 metres where the band sound exposure level is measured, the receiving animal further out at range R with the propagation loss between them, the three processing steps from per-band exposure through the auditory weighting to the accumulation over strikes, and a plan view of the two isopleths of the dual metric with the larger one governing](https://docs.jmrp.io/phonometry/media/diagram_marine_mammal_exposure.svg)

Percussive pile driving is the canonical impulsive case. The chain runs from
the recorded strike to the verdict:

1. `strike_sel_spectrum` splits the single-strike sound exposure of the record
   into fractional-octave bands (the band energies sum back to the broadband
   `single_strike_sel` of
   [Underwater acoustics](https://docs.jmrp.io/phonometry/underwater/underwater-acoustics/#3-pile-driving-sound-iso-18406), by
   Parseval);
2. `weighted_exposure` applies $W(f)$ band by band, sums the weighted energy,
   accumulates it over the number of strikes (the ISO 18406 $+10\log_{10} N$) and
   compares the result with the criteria, unweighted peak SPL included.

### What the recording has to cover

The weighted sum runs over the bands it is handed, so **the record must span the
hearing group's passband**. The NMFS 2024 upper transition frequencies $f_2$ are
26.6 kHz (LF), 129 kHz (HF), 186 kHz (VHF), 68.3 kHz (PW) and 43.8 kHz (OW), and
the generalised hearing ranges reach further still (LF to 36 kHz, VHF to
165 kHz). As a rule the sample rate has to be about 2.5 times the highest band
edge that matters, so the 48 kHz record below (bands resolved to roughly 20 kHz)
supports LF, is marginal for OW, and truncates HF, VHF and PW badly. A
porpoise assessment needs a few hundred kilohertz.

`weighted_exposure` cannot detect a truncated spectrum: it weights what it is
given, so a short record simply reports a lower cumulative SEL and no warning.
The unweighted peak half of the dual metric has to come from the same record
with no clipping and headroom to spare, which is why ISO 18406 makes dynamic
range a survey requirement: a saturated capture reads as a compliant peak.

```python
import numpy as np
from phonometry import underwater

# One recorded strike (here a synthetic 200 Hz decaying burst). 48 kHz is
# enough for the LF group used below and for nothing above it.
fs = 48_000
t = np.arange(int(0.2 * fs)) / fs
strike = 50.0 * np.exp(-t / 0.06) * np.sin(2 * np.pi * 200.0 * t)

spectrum = underwater.strike_sel_spectrum(strike, fs, fraction=3)
peak = underwater.peak_sound_pressure_level(strike)

res = underwater.weighted_exposure(
    spectrum.band_sel, "LF", frequencies_hz=spectrum.frequencies,
    guidance="nmfs-2024", impulsive=True, n_events=3000, peak_spl=peak,
)
print(res.unweighted_sel, res.weighted_sel, res.cumulative_sel)
print(res.sel_margin, res.peak_margin)       # positive means the criterion is exceeded
print(res.exceeds_injury, res.exceeds_tts)
res.plot()   # weighted spectrum against the criteria (needs matplotlib)
```

Read those six numbers in order, because each one is a step. The record's
unweighted single-strike exposure is **135.7 dB re 1 µPa²·s**; LF weighting takes
it to **133.5 dB**, costing barely 2 dB, because a 200 Hz hammer sits inside the
low-frequency cetacean passband. Accumulating 3 000 strikes adds
$10\log_{10}3000 = 34.8$ dB and gives **168.3 dB**: that term, not the
weighting, is what usually decides the outcome. The margins are then read
against the criteria: `sel_margin` = **−14.7 dB** against the 183 dB auditory
injury criterion, so `exceeds_injury` is `False`, but the cumulative level clears
the 168 dB TTS onset by three tenths of a decibel, so `exceeds_tts` is `True`.
The flat peak of this record is 153.8 dB re 1 µPa, `peak_margin` = **−68.2 dB**,
nowhere near the 222 dB peak criterion.

Those are the magnitudes of a *synthetic* strike, and they are quiet. Measured
percussive piling at the standard 750 m position runs to peak levels of order
200 dB re 1 µPa and single-strike exposures of order 175-180 dB re 1 µPa²·s, so
a real pile approaches the low-frequency injury criterion of 183 dB while
staying far below it once VHF weighting is applied. A result far outside those
ranges is a units or calibration error rather than an unusual pile.

![Left, the single-strike sound exposure level of the recorded strike in one-third-octave bands, peaking at 200 hertz; centre and right, the same 3000-strike campaign assessed for low-frequency cetaceans and for very high-frequency cetaceans, each showing the unweighted and weighted band exposures, the TTS and auditory-injury criteria lines and the cumulative level, with margins of minus 14.7 and minus 49.7 decibels](https://docs.jmrp.io/phonometry/media/marine_mammal_assessment.svg)

*The same campaign, judged twice. For a baleen whale the weighting costs 2 dB
and the cumulative level lands between the two criteria; for a porpoise the
weighting costs 59 dB and the same piling sits 50 dB below its injury criterion.
That is the whole point of weighting, and the reason a single unweighted
cumulative SEL is not an assessment. `res.plot()` draws either panel.*

**Show the code for this figure**

```python
import matplotlib.pyplot as plt

fig, axes = plt.subplots(1, 3, figsize=(16.5, 5.4))
spectrum.plot(ax=axes[0])
for ax, group in zip(axes[1:], ("LF", "VHF"), strict=True):
    verdict = underwater.weighted_exposure(
        spectrum.band_sel, group, frequencies_hz=spectrum.frequencies,
        guidance="nmfs-2024", impulsive=True, n_events=3000, peak_spl=peak)
    verdict.plot(ax=ax)
    ax.set_title(f"{group}: cumulative {verdict.cumulative_sel:.1f} dB, "
                 f"margin {verdict.sel_margin:+.1f} dB")
plt.show()
```

The operational question a campaign asks is not "does 3 000 strikes exceed the
criterion" but "how many strikes may be driven, and which group binds first".
That is one loop over `weighted_exposure` with a rising `n_events`.

![Weighted cumulative sound exposure level against strike count from one to ten thousand on a logarithmic axis, one line per hearing group, with each group's TTS and auditory-injury criteria drawn as dotted and dashed horizontal lines in the matching colour, and the low-frequency cetacean curve crossing its TTS onset at 2820 strikes](https://docs.jmrp.io/phonometry/media/piling_campaign_accumulation.svg)

*Every curve is a straight line on a log-N axis, because accumulation adds
$10\log_{10}N$ and nothing else: the group decides the intercept and the strike
count decides the rest. For this 200 Hz hammer only the low-frequency group
reaches a criterion at all, at 2 820 strikes, and no group reaches its injury
criterion within 10 000. The curve is a stationary-receiver worst case: the
library does not move the animal.*

**Show the code for this figure**

```python
import matplotlib.pyplot as plt

counts = np.unique(np.round(np.logspace(0.0, 4.0, 90)).astype(int))
fig, ax = plt.subplots(figsize=(11.5, 6.4))
for group in ("LF", "HF", "VHF", "PW", "OW"):
    curve = [underwater.weighted_exposure(
        spectrum.band_sel, group, frequencies_hz=spectrum.frequencies, guidance="nmfs-2024",
        impulsive=True, n_events=int(n)).cumulative_sel for n in counts]
    line, = ax.semilogx(counts, curve, label=group)
    crit = underwater.exposure_criteria(group, guidance="nmfs-2024",
                                        impulsive=True)
    ax.axhline(crit.tts_sel, color=line.get_color(), linestyle=":")
    ax.axhline(crit.injury_sel, color=line.get_color(), linestyle="--")
ax.set(xlabel="Number of strikes N",
       ylabel="Weighted cumulative SEL [dB re 1 uPa^2 s]")
ax.legend(ncols=5)
plt.show()
```

### From a criterion to an isopleth

A margin is not the regulatory output; a **distance** is. An isopleth is the
locus on which a criterion is exactly met, so the answer an assessment owes is a
radius: the range beyond which the accumulated weighted SEL no longer exceeds
its criterion, and separately the range beyond which the flat peak no longer
exceeds its own, with the larger of the two governing.

The bridge from this page to that radius runs through the propagation guide.
Take the band SEL at the range it was measured at (750 m, per ISO 18406)
and propagate it to a trial range with `propagation_loss` at each band centre,
re-run `weighted_exposure` on the propagated spectrum for the strike count of
the campaign, and read off the range where `sel_margin` crosses zero. Then
repeat for the flat peak, where only the single-trip loss applies.

```python
def band_loss(r):
    """Per-band propagation loss at range r, in 30 m of water."""
    return np.array([float(underwater.propagation_loss(
        r, float(f), law="practical", transition_range=30.0).pl[0])
        for f in spectrum.frequencies])

reference = band_loss(750.0)          # the range the strike was measured at
for r in (750.0, 800.0, 1000.0, 3000.0):
    out = underwater.weighted_exposure(
        spectrum.band_sel - (band_loss(r) - reference), "LF",
        frequencies_hz=spectrum.frequencies, guidance="nmfs-2024", impulsive=True, n_events=3000,
        peak_spl=peak)
    print(int(r), round(out.cumulative_sel, 1), round(out.tts_margin, 2))
# 750 168.3 0.27   <- exceeded
# 800 168.0 -0.01  <- the TTS isopleth is here
# 1000 167.0 -0.98
# 3000 162.2 -5.76
```

For this campaign the low-frequency TTS isopleth sits at about **800 m**, and
there is no injury isopleth at all: the cumulative level is 14.7 dB below the
183 dB criterion even at the measurement range. The peak isopleth does not exist
either, for the same reason: this is exactly the case where the *larger*
isopleth rule has nothing to choose between, and the assessment reduces to one
radius.

Two consequences follow, and both belong in the report. The governing metric can
**change with range**, because the accumulated SEL carries the $+10\log_{10}N$
term and the peak does not, so the two isopleths shrink at different rates. And
a cumulative number means nothing without two facts attached: the range at which
the strike was measured or modelled, and the **accumulation window**: 3 000
strikes means one pile driven inside that window, conventionally 24 hours in the
regulatory guidance, not a whole campaign.

## What this guide covers

- **Covered.** `group_audiogram` and `audiogram_parameters` implement the Southall et al.
  (2019) group audiogram (Equation 1) with the Table 2 absolute and Table 3
  normalised fits; `orca_audiogram` implements the three-branch killer-whale
  audiogram of Ainslie (2010) Equation (11.159). `auditory_weighting` implements
  the band-pass weighting and exposure functions for the three selectable
  guidance versions (NMFS 2024 v3.0, the default; NMFS 2018 v2.0; Southall et
  al. 2019), with the parameter tables addressable through
  `weighting_parameters`. `exposure_criteria` returns the published TTS and
  injury (PTS / AUD INJ) onset criteria, impulsive and non-impulsive, including
  the unweighted peak-SPL half of the dual metric. `weighted_exposure` runs the
  assessment: weighting a band spectrum, accumulating it over events and
  reporting the margin against every applicable criterion. `strike_sel_spectrum`
  supplies that band spectrum from a recorded pile strike.

- **Not covered.** There is no audiogram for LF cetaceans, because Southall et al. never print
  its $F_1$ parameter. Behavioural-disturbance criteria (the step-function and
  dose-response thresholds used for harassment take estimates) are out of scope:
  only the auditory-effect criteria are implemented. The library does not choose
  a hearing group or an accumulation period for you: both are assessment
  decisions, made per species and per activity. Nor does it model the animal's
  movement relative to the source, so the cumulative SEL it reports is the
  stationary-receiver worst case. It also does not check that the supplied
  spectrum spans the group's passband: a record that stops short of $f_2$ is
  weighted as given and under-reports the exposure. Propagation from source to
  receiver belongs to [Underwater sound
  propagation](https://docs.jmrp.io/phonometry/underwater/underwater-propagation/).

## See also

- [Underwater acoustics: radiated noise and pile driving](https://docs.jmrp.io/phonometry/underwater/underwater-acoustics/):
  the ISO 18405 reference levels and the ISO 18406 pile-driving exposure that
  feed this page.
- [Underwater sound propagation](https://docs.jmrp.io/phonometry/underwater/underwater-propagation/):
  the propagation loss and figure-of-merit machinery that turns a criterion
  into a range.
- API reference: [`underwater.bioacoustics.weighting`](https://docs.jmrp.io/phonometry/reference/api/underwater/weighting/) and [`underwater.bioacoustics.audiograms`](https://docs.jmrp.io/phonometry/reference/api/underwater/audiograms/).

## References

- **Ainslie, M. A. (2010).** *Principles of sonar performance modelling*. Springer. https://doi.org/10.1007/978-3-540-87662-5
  The orca audiogram (Equation 11.159, after Wensveen & Van Roij 2007) and the orca-versus-salmon worked example.
- **Finneran, J. J. (2016).** [*Auditory weighting functions and TTS/PTS exposure functions for marine mammals exposed to underwater noise*](https://apps.dtic.mil/sti/citations/AD1026445) (Technical Report 3026). SSC Pacific.
  The band-pass filter form and the audiogram equation that both criteria sets adopt.
- **International Organization for Standardization. (2017).** [*Underwater acoustics — Measurement of radiated underwater sound from percussive pile driving*](https://www.iso.org/standard/62407.html) (ISO 18406:2017).
  The single-strike sound exposure (Formulae 3–4) that strike_sel_spectrum splits into bands and the +10 log10 N accumulation over strikes (Formula 9) that weighted_exposure applies; the broadband metrics are on the Underwater acoustics page.
- **International Organization for Standardization. (2017).** [*Underwater acoustics — Terminology*](https://www.iso.org/standard/62406.html) (ISO 18405:2017).
  The symbols and references the guidance documents adopt for their criteria (Lp,0-pk and LE,p re 1 µPa and 1 µPa²·s).
- **National Marine Fisheries Service. (2018).** [*2018 Revision to: Technical Guidance for Assessing the Effects of Anthropogenic Sound on Marine Mammal Hearing (Version 2.0)*](https://www.fisheries.noaa.gov/s3/2023-05/TECHMEMOGuidance508.pdf) (NOAA Technical Memorandum NMFS-OPR-59). National Oceanic and Atmospheric Administration.
  The Table 3 weighting parameters, the Table ES3 PTS onset thresholds and the Appendix D worked example (W at 1 kHz for the five hearing groups) used as the numeric oracle of the weighting section.
- **National Marine Fisheries Service. (2024).** [*2024 Update to: Technical Guidance for Assessing the Effects of Anthropogenic Sound on Marine Mammal Hearing (Version 3.0)*](https://www.fisheries.noaa.gov/s3/2024-11/Tech_Memo-Guidance_-3.0-_OCT-2024-508_OPR1.pdf) (NOAA Technical Memorandum NMFS-OPR-71). National Oceanic and Atmospheric Administration.
  The current guidance and the library default: Table 5 weighting parameters, Table ES3 AUD INJ criteria and the Navy Phase 4 Table A.E-2 impulsive TTS thresholds.
- **Southall, B. L., Finneran, J. J., Reichmuth, C., Nachtigall, P. E., Ketten, D. R., Bowles, A. E., Ellison, W. T., Nowacek, D. P., & Tyack, P. L. (2019).** Marine mammal noise exposure criteria: Updated scientific recommendations for residual hearing effects. *Aquatic Mammals*, *45*(2), 125-232. https://doi.org/10.1578/AM.45.2.2019.125
  The group audiograms (Equation 1 with Tables 2 to 4) and the weighting and threshold tables (Tables 5 to 7); its errata in 45(5), 569-572 corrects four peak-SPL values of Table 7.
