Adult American eels (Anguilla rostrata) are vulnerable to hydropower turbine mortality during outmigration from growth habitat in inland waters to the ocean where they spawn. Imaging sonar is a reliable and proven technology for monitoring of fish passage and migration; however, there is no efficient automated method for eel detection. We designed a deep learning model for automated detection of adult American eels from sonar data. The method employs convolution neural network (CNN) to distinguish between 14 images of eels and non-eel objects. Prior to image classification with CNN, background subtraction and wavelet denoising were applied to enhance sonar images. The CNN model was first trained and tested on data obtained from a laboratory experiment, which yielded overall accuracies of >98% for image-based classification. Then, the model was trained and tested on field data that were obtained near the Iroquois Dam located on the St. Lawrence River; the accuracy achieved was commensurate with that of human experts.

A diversified energy portfolio may include marine energy in the form of current energy converters (CECs) such as tidal or in-river turbines. New technology development in the research stage typically requires monitoring for environmental effects. A significant environmental effect of concern for CECs is the risk of moving parts (e.g., turbine blades) colliding with animals such as fishes. CECs are installed in energetic locations in which it is difficult to operate sensors to fulfill monitoring requirements for informing collision risk. Collecting data (i.e., about blade strikes or near-misses) that inform interactions of fishes with CECs is usually attempted using active acoustic sensors or video cameras (VCs). Limitations of low-light conditions or water turbidity that preclude effective use of VCs are overcome by using high-resolution multibeam echosounders (or acoustic cameras (ACs)). We used an AC at two sites to test its ability to detect artificial and real fish targets and determine if strike, near-miss, and near-field behavior could be observed. Interactions with fish and artificial targets with turbines have been documented but strike confirmation with an AC is novel. The first site was in a tidal estuary with a 25 kW turbine and water clarity sufficient to allow VC data to be collected concurrently with AC data showing turbine blade strike on tethered artificial fish targets. The second site was a turbid, debris-laden river with a 5 kW turbine where only AC data were collected due to high water turbidity. Data collection at the second site coincided with downstream Pacific salmon (Oncorhynchus spp.) smolt migration. Physical fish capture downstream of the turbine was performed with an incline plane trap (IPT) to provide context for the AC observations, by comparing fish catches. Discrimination between debris and fishes in the AC data was not possible, because active movement of fishes was not discernable. Nineteen fishes were released upstream of the turbine to provide known times of possible fish/turbine interactions, but detection was difficult to confirm in the AC data. ACs have been used extensively in past studies to count large migratory fish such as Pacific salmon, but their application for small fish targets has been limited. The results from these two field campaigns demonstrate the ability of ACs to detect targets in turbid water and observe blade strikes, as well as their limitations such as the difficulty of distinguishing small fishes from debris in a high-energy turbid river. Recommendations are presented for future applications associated with CEC device testing.

Hydroelectric power stations generate turbulent flow conditions, which represent a potentially significant hydraulic stressor to fish passing through the turbine system. A test facility has been developed using two randomly actuated synthetic jet arrays (RASJAs) of 25 independent submersible pumps to generate a turbulent flow field for biological dose-response testing. The novel elements of this approach include the ability to control the exposure duration within a test volume due to low mean flow velocity as well as the capacity to scale the turbulence levels as a function of pump capacity. Juvenile Chinook salmon (Oncorhynchus tshawytscha) were subjected to the turbulent flow regime with average turbulence kinetic energy per unit mass of $0.089{\mathrm{m}}^2/{\mathrm{s}}^2$ for periods of 2 min and 10 min. No significant loss of equilibrium or disorientation was observed after exposure for either duration at the level of turbulence achieved in this prototype. Further scaling of this approach is required to generate a complete dose-response relationship.

Throughout many areas of their native range, American shad (Alosa sapidissima) and other Alosine populations are in decline. Though several conditions have influenced these declines, hydropower facilities have had significant negative effects on American shad populations. Hydropower facilities expose ocean-migrating American shad to physical stressors during passage through hydropower facilities, including strike, rapid decompression, and fluid shear. In this laboratory-based study, juvenile American shad were exposed separately to rapid decompression and fluid shear to determine their susceptibility to these stressors and develop dose–response models. These dose–response relationships can help guide the development and/or operation of hydropower turbines and facilities to reduce the negative effects to American shad. Relative to other species, juvenile American shad have a high susceptibility to both rapid decompression and fluid shear. Reducing or preventing exposure to these stressors at hydropower facilities may be a potential method to assist in the effort to restore American shad populations.

Juvenile Pacific Lamprey and American eels were used for laboratory evaluations to determine potential effects from tag implantation. Telemetry technology has been identified as a way to obtain more detailed information on movement and behavior across a broader spatial scale than is possible with other known technology. The purpose of this method is to provide a detailed step by step instruction on tag implantation for both lampreys and eel. For laboratory studies using actively migrating juvenile Pacific Lamprey (120-160 mm), we determined that the presence of the tag did not alter the swimming ability between tagged and untagged Individuals or have any significant tag loss (<3%). Similar results were determined during laboratory testing of Yellow phase American Eels (113-175 mm). No mortality occurred during a 38-day holding period and there was minimal tag loss (3.8%). The presence of the tag did not have any significant effect on the swimming ability or survival of tagged eels compared to untagged controls and there was minimal tag loss.

An automatic system that utilizes data analytics and machine learning to identify adult American eel in data obtained by imaging sonars is created in this study. Wavelet transform has been applied to de-noise the ARIS sonar data and a convolutional neural network model has been built to classify eels and non-eel objects. Because of the unbalanced amounts of data in laboratory and field experiments, a transfer learning strategy is implemented to fine-tune the convolutional neural network model so that it performs well for both the laboratory and field data. The proposed system can provide important information to develop mitigation strategies for safe passage of out-migrating eels at hydroelectric facilities.