Skip to content

MarineBioAcousticsRC/GOM_NeuralNetwork

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

13 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

GOM_NeuralNetwork

Neural‑network model for ship source levels (20–1000 Hz).

This repository provides the trained full neural network and example data used in the paper “Characterization and modeling source levels of commercial vessels in the North Atlantic” published in The Journal of the Acoustical Society of America (JASA). It allows users to estimate third‑octave monopole source levels (MSLs) from AIS and environmental variables without accessing the original training dataset.

Repository contents

Model: NN_Full.mat

Trained MATLAB neural network appearing as NNet_22Jun2025 in MATLAB.

Data:

  1. F_new.mat - Column vector of third‑octave band center frequencies from 20 Hz to 1000 Hz.

  2. MSL_new.mat - Matrix of third‑octave monopole source levels (dB re 1 µPa m). Columns correspond to the frequency bands in F_new, rows correspond to individual ships.

  3. All_full.mat - Matrix of AIS and environmental predictor variables. Rows correspond to ships, columns correspond to variables listed in Variable_Information.xlsx.

  4. Variable_Information.xlsx - Table describing each column in All_full, including:

    a. Vessel Type – Generic

    b. Length (m)

    c. Draft (m)

    d. SOG (knots)

    e. MMSI

    f. Vessel Type – Detailed

    g. Built (year)

    h. Direction of travel

    i. Draught Max (m)

    j. Draught Min (m)

    k. Ship Builder

    l. Engine Builder

    m. Width (m)

    n. Capacity – Gross tonnage

    o. Capacity – Deadweight (ton)

    p. Commercial Size Class

    q. Current direction

    r. Current magnitude (m/s)

    s. Current in u‑direction (m/s)

    t. Current in v‑direction (m/s)

    u. Wind Direction

    v. Wind magnitude (knots)

    w. Broadband MSL 20–1000

    x. Broadband MSL 5–10000

Usage notes:

Inputs: Each row of All_full corresponds to a single vessel instance, described by the AIS and environmental variables in Variable_Information.xlsx.

Important: Before using All_full as input to the neural network, remove the two broadband MSL columns (20–1000 and 5–10000). These are output metrics and were not used as predictors during training.

Outputs: The network returns a matrix of third‑octave source levels in dB re 1 µPa m for the 20–1000 Hz bands defined in F_new.

Any normalization, encoding, or preprocessing applied during training is assumed to be already embedded in the way All_full was constructed. Users should maintain the same variable ordering and units.

Citation

If you use this model or data in your work, please cite:

Johnson, K.H., ZoBell, V.M, Hodge, L.E.W., Soldevilla, M.S., Hildebrand, J.A., Frasier, K.E.; "Characterization and modeling source levels of commercial vessels in the Gulf of Mexico." J. Acoust. Soc. Am. 1 September 2025; 158 (3): 2250–2268. https://doi.org/10.1121/10.0039379

Contact

For questions about the code or data, please contact:

Katrina H. Johnson – kaj007@ucsd.edu

Scripps Institution of Oceanography / Scripps Machine Listening Lab

About

This repository provides the trained neural network and example data used in the paper “Characterization and modeling source levels of commercial vessels in the North Atlantic” published in The Journal of the Acoustical Society of America (JASA).

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors