Knowledge of the distribution and quality of seafloor habitat is critical for the improvement of biomass estimates from fish resource surveys and spatially-based marine living resource management efforts. It is also critical to understanding the interactions of fish species with their environment and defining areas of special significance for the well-being of fish and other marine animals. In the economically- and ecologically-important Gulf of Alaska (GOA), our current knowledge of the distribution of habitat is incomplete and fragmentary. Current habitat classification approaches using underwater camera images allow us to assess sea floor characteristics with high degree of accuracy and detail. However, current coverage of large ecosystems such as the GOA with image-based methods is very low, reducing their usefulness for large scale inference. Acoustic remote sensing has the potential to greatly expand this sampling footprint, but represents an indirect, more abstract estimate of habitat metrics. 

This project will apply machine learning (ML) methods to associate image-based habitat classifications with spatially corresponding acoustic backscatter from the seafloor. This will involve the development of an ML classifier, which will be conditioned using a large training dataset from acoustic backscatter profiles and corresponding image–based habitat metrics from raw acoustic and image data that have been collected in the GOA over the past decade. The result would enable a classifier to be applied to the historic acoustic survey data set, which will enable full featured, high resolution mapping of important fish habitats in the GOA.

As part of this project, the Research Associate will explore various ML approaches such as supervised classification using features traditionally derived from the seafloor echoes and which are based on backscattering theory such as substrate roughness and hardness, as well as automated feature extraction (e.g. deep learning) directly from the raw acoustic return signal. It will also be possible to explore a broader range of habitat information that could be extracted from the acoustic data, and thus to provide benefit for more general fisheries and ecological studies. For example, important habitat metrics such as structural complexity (e.g. small-scale differences in vertical relief) or epibenthic fauna presence and density (e.g. sponge and coral cover) can be extracted from seafloor imagery and quantified, and thus may present a detectable signal in acoustic returns as well.

This project represents a research opportunity to explore the potential of the resolving power of ML to develop a critical link between acoustic and optic data sources and provide a critically important component for understanding the GOA marine ecosystem

key words

seafloor habitat classification, acoustics, underwater camera, machine learning, seafloor mapping

Citizenship:  Open to U.S. citizens, permanent residents and non-U.S. citizens

Level:  Open to Postdoctoral and Senior applicants