Fig 1.
Map of three passive acoustic monitoring stations in the May River estuary, SC that were deployed from 2013 to present.
(Inset) The May River estuary (black circle) is in reference to the east coast of the United States. In the figure, 9M, 14M, and 37M are the locations of passive acoustic recorders locations, where 9M was located near the source, 14M was in the middle but also close to the source, and 37M was located at the mouth of the tidal river, near the intra-coastal waterway.
Fig 2.
Example photos of instrument frames with passive acoustic recorders as part of the Estuarine Soundscape Observatory Network in the Southeast (ESONS).
The left figure shows six recorders before deployment and the right figure shows a platform after deployment.
Fig 3.
Boat signal patterns identified in spectrogram in the May River estuary from 2017 to 2021.
A) Narrow burst broadband, B) wide burst broadband, C) variable broadband, and D) low-frequency signal. Panel A inset highlights details of fish activities at low frequencies (0~800 Hz).
Fig 4.
Deep scanning process.
Fig 5.
Snapping signals in raw audio data.
Fig 6.
Signal cleaning examples removing snapping shrimp snaps.
A) the original spectrogram diagram with snapping signals, B) the cleaned version of the spectrogram diagram. The color scheme in the spectrogram, changing from turquoise to light green, yellow, and red, symbolizes the increase of signal intensity (or loudness) of the signal over time at specified frequencies in the waveform.
Fig 7.
Examples of the time-domain signal energy analysis.
A) Narrow burst broadband, B) wide burst broadband, C) narrow burst broadband (multiple signals), and D) loud signal without boat activity indicative of fish chorusing. Each panel includes a figure of raw audio data in the time domain (row 1), a figure of instantaneous broadband (1–40,000 Hz) SPL values (row 2), and a figure of the spectrogram of the corresponding audio data (row 3).
Fig 8.
Possible boat signal features (raw data).
Fig 9.
Example neural network.
Table 1.
Deep scanning algorithm.
Table 2.
Detection accuracy.
Fig 10.
Sound pressure level diagram with peak difference, mean difference, and area size for vessel detection.
The dotted line represents the median SPL of the entire two-minute.wav file. A) SPL values for a narrow-burst signal in Fig 7A, 7B) SPL values for a wide burst signal in Fig 7B, 7C) SPL values for a multiple narrow bursts signal in Fig 7C, and 7D) SPL values for a signal without vessel noise in Fig 7D. The Area Size is equal to the sound exposure level (SEL) for each vessel detection. In the case of Fig 7D, this sound is most likely a fish chorus and would not be included in cumulative noise energy (SELcum) associated with vessels.
Fig 11.
Time series of vessel noise detections at station 37M in the May River, SC.