Phase II: Field validation of multibeam sonar zebra mussel detection
The first phase of this project (below) revealed sufficient differences in acoustic response of native and zebra mussels and the supporting sediment. This allowed researchers to develop an empirical approach to zebra mussel detection. Phase II of this project will test the use of this and other acoustic signatures to detect and map zebra mussel beds in the field, incorporating a larger range of variables, such as a greater range of mussel densities and substrate mixtures, water depths, and temperatures.
Phase I: Testing the utility of a swath mapping system
This study will test the utility of swath mapping systems such as multibeam sonar for detecting and quantifying the abundance of invasive mussels at a very large scale. Multibeam sonar can map tens to hundreds of square kilometers of river or lake bed in a single day from a moving vessel. There is a strong likelihood that mussels have a distinct acoustic response (echo) compared to their surrounding substrate. If so, this acoustic signature can be readily used to detect and map zebra mussel beds in any navigable waterway of sufficient water depth. This study will define the methodology needed to detect, distinguish and quantify mussels from a moving vessel by studying backscattering of sound by mussels and common mussel-supporting substrates.
The first phase of this study is designed to utilize multibeam sonar to distinguish among substrate, native mussels, and zebra mussels in a controlled laboratory setting. A second phase is planned to validate and develop methodologies for use in the field.
Current methods for detecting and quantifying zebra mussel populations relies on methods that can be very time-consuming and expensive, such as diving, video imaging, and veliger sampling in the water column. Detecting zebra mussel populations early significantly improves the possibility that quarantines can be put in place and treatment options implemented.
Lab experiments are complete as of January 2019. Using that data, researchers developed machine-learning-based substrate classifiers based on hypothetical situations of abiotic and biotic substrates. This information is put into models, which are trained over ten unique substrates: 1) sand, 2) mix sand-gravel (MSG); 3) gravel; 4) sand-supported A. plicata; 5) MSG-supported A. plicata; 6) gravel-supported A. plicata; 7) sand-supported D. polymorpha (low density); 8) sand-supported D. polymorpha (high density); 9) gravel-supported D. polymorpha (low density); and 10) gravel-supported D. polymorpha (high density).
Laboratory experiments were conducted to test the feasibility of using multibeam sonar to distinguish zebra mussel containing substrates. Acoustic backscatter data were collected in a two meter deep tank over sand, gravel, and mixed substrate containing high and low densities of zebra mussels and with native mussels using combinations of different sonar settings (frequencies and pulse lengths). Machine-learning was used to differentiate the acoustic backscattering signatures in a data-driven substrate classifier approach. Using these methods, we were able to classify substrate by size and mussel density. Classification errors decreased with more sonar settings. For minimum errors of less than 20%, 8 sonar settings are required, and for minimum errors of 10% or less for all substrates, 12 sonar settings. Each sonar setting corresponds to a separate boat survey of an area with a multibeam sonar in the field. This research will continue into a second phase, above.