Extended Data Fig. 5: Alternative seascape classifications and predictions.

The results of 4 classification approaches are shown: K-means, Agglomerative, Spectral, and Gaussian mixture clustering. Seascape classes from agglomerative clustering are then projected globally, using 10 supervised learning algorithms from the scikit-learn Python module: AdaBoost, Decision Tree, Gaussian Process, Naive Bayes, Nearest Neighbors, Neural Network, Quadratic Discriminant Analysis (QDA), Support Vector Classification with linear kernel (SVM-linear), Support Vector Classification with Radial-basis function kernel (SVM-RBF), and Random Forest. The overall predictive accuracy of algorithms was calculated as in Extended Data Fig. 4 and the results are shown in the table above (F1 score). Examples of acoustic seascape projections are shown for the most accurate algorithms: AdaBoost, Decission Tree, and Random Forest. The method chosen for the present study was Agglomerative clustering and Random Forest projection.