In this paper, the effectiveness of deep learning for automatic classification of grouper species by their vocalizations has been investigated. In the proposed approach, wavelet denoising is used to reduce ambient ocean noise, and a deep neural network is then used to classify sounds generated by different species of groupers. Experimental results for four species of groupers show that the proposed approach achieves a classification accuracy of around 90% or above in all of the tested cases, a result that is significantly better than the one obtained by a previously reported method for automatic classification of grouper calls.

Fish spawning aggregations (FSA) consist of the temporary gathering of a large number of fishes at a specific location to spawn. Monitoring of FSA is typically conducted by divers, but surveys are often restricted to a limited area and dependent upon sea conditions, thus our knowledge of FSA dynamics is extremely limited. Fisheries independent research strives for new technology that can help remotely and unobtrusively quantify fish biomass and abundance. Since some fish species, such as groupers, produce sounds during reproductive behaviors, Eulerian passive acoustic monitoring provides information when divers cannot access the FSA site. Fish sounds provide an innovative approach to assess fish presence and potentially their numbers during reproductive events. However, this technology is limited by the sound propagation range, hence the distance between the fish emitting sounds the hydrophone location. As such, this Eulerian monitoring approach implicitly creates a knowledge gap about what happens beyond the range of the recorders. Furthermore, the large datasets make the detection process by human ears and eyes very tedious and inconsistent. This paper reports on two innovative approaches to overcome these limitations. To facilitate fish call detections, we have developed an algorithm based on machine learning and voice recognition methods to identify and classify the sounds known to be produced by certain species during FSA. This algorithm currently operates on a SV3 Liquid Robotics wave glider, an autonomous surface vehicle which has been fitted to accommodate a passive acoustic listening device and can cover large areas under a wide range of sea conditions. Fish sounds detections, classification results, and locations along with environmental data are transmitted in real-time enabling verification of the sites with high detections by divers or other in-situ methods. Recent surveys in the US Virgin Islands with the SV3 Wave Glider are revealing for the first time the spatial and temporal distribution of fish calls surrounding known FSA sites. These findings are critical to understanding the dynamics of fish populations because calling fish were detected several kilometers away from the known FSAs. These courtship associated sounds from surrounding areas suggest that other FSAs may exist in the region.

The Loop Current (LC) is the dominant circulation system in the Gulf of Mexico. A long‐term prediction of the LC system (LCS) behavior is critical for understanding the Gulf of Mexico oceanography and ecosystem, and for mitigating outcomes of anthropogenic and natural disasters. In early 2018, the National Academies of Science, Engineering, and Medicine posed a challenge to the research community to develop systems that can forecast the movement of the LCS over longer periods of time than the current state of art. In this paper, a Recurrent Neural Network, the Long Short‐Term Memory (LSTM) network, is applied to predict the LC evolution and the LC ring formation. The LSTM model is trained to learn patterns hidden in sea surface height (SSH) time series. To reduce the memory demand owing to the use of high spatial resolution SSH data set, the region of interest is partitioned into nonoverlapping subregions. After partitioning, an LSTM network is trained to predict the SSH in each subregion. A smoothing function is then applied to reduce discontinuities of the SSH predictions across the partition boundaries, hence error propagation. It is shown that such a machine learning model is capable of predicting the LCS SSH evolution 9 weeks in advance within 40 km in terms of the LCS frontal distance errors. Furthermore, it is shown that the model predicted the timing and general location of eddy Darwin's shedding event 12 weeks in advance, and eddy Cameron's detachment and reattachment 8 weeks in advance.

A transfer learning approach is proposed to classify grouper species by their courtship-associated sounds produced during spawning aggregations. Vessel sounds are also included in order to potentially identify human interaction with spawning fish. Grouper sounds recorded during spawning aggregations were first converted to time-frequency representations. Two types of time frequency representations were used in this study: spectrograms and scalograms. These were converted to images, and then fed to pretrained deep neural network models: VGG16, VGG19, Google Net, and MobileNet. The experimental results revealed that transfer learning significantly outperformed the manually identified features approach for grouper sound classification. In addition, both time-frequency representations produced almost identical results in terms of classification accuracy.

The northeastern Caribbean Sea is under the seasonal influence of the Trade Winds but also of the Orinoco/Amazon freshwater plume. The latter is responsible for intensification of the Caribbean Current in general and of its eddy activity in the northern part of the Caribbean Sea. More importantly, we show in this study that the front of the freshwater plume drives a northward flow that impinges directly on the island of St. Croix in the United States Virgin Islands. The angle of incidence of the incoming flow controls the nature of the wake on both sides and ends of the island, which changes from cyclonic to anticylonic wake flow, with either attached or shed eddies. Using an off-line bio-physical model, we simulated the dispersal and recruitment of an abundant Caribbean coral reef fish, the bluehead wrasse (Thalassoma bifasciatum) in the context of the wake flow variability around St. Croix. Our results revealed the role played by the consistent seasonal forcing of the wake flow on the recruitment patterns around the island at the interannual scale. The interannual variability of the timing of arrival and northward penetration of the plume instead controls the nature of the wake, hence the regional spatial recruitment patterns.

Inertial measurement unit sensors (IMU; i.e., accelerometer, gyroscope and magnetometer combinations) are frequently fitted to animals to better understand their activity patterns and energy expenditure. Capable of recording hundreds of data points a second, these sensors can quickly produce large datasets that require methods to automate behavioral classification. Here, we describe behaviors derived from a custom-built multi-sensor bio-logging tag attached to Atlantic Goliath grouper (Epinephelus itajara) within a simulated ecosystem. We then compared the performance of two commonly applied machine learning approaches (random forest and support vector machine) to a deep learning approach (convolutional neural network, or CNN) for classifying IMU data from this tag. CNNs are frequently used to recognize activities from IMU data obtained from humans but are less commonly considered for other animals. Thirteen behavioral classes were identified during ethogram development, nine of which were classified. For the conventional machine learning approaches, 187 summary statistics were extracted from the data, including time and frequency domain features. The CNN was fed absolute values obtained from fast Fourier transformations of the raw tri-axial accelerometer, gyroscope and magnetometer channels, with a frequency resolution of 512 data points. Five metrics were used to assess classifier performance; the deep learning approach performed better across all metrics (Sensitivity = 0.962; Specificity = 0.996; F₁-score = 0.962; Matthew’s Correlation Coefficient = 0.959; Cohen’s Kappa = 0.833) than both conventional machine learning approaches. Generally, the random forest performed better than the support vector machine. In some instances, a conventional learning approach yielded a higher performance metric for particular classes (e.g., the random forest had a F₁-score of 0.971 for backward swimming compared to 0.955 for the CNN). Deep learning approaches could potentially improve behavioral classification from IMU data, beyond that obtained from conventional machine learning methods.