Development and application of a robot-assisted computer vision system to map Great Lakes bottom habitats and biology

Science Center Objects

Lake bottom environments are critical zones of interface between geology and biological processes that support health ecosystems and human well-being.  Over the past thirty years, Great Lake food webs have become dominated by bottom dwelling invasive species and nuisance algae, that are poorly mapped and understood.  USGS is developing a suite of new technologies to map habitat, invasive mussels and fish, and nuisance algae in high resolution to meet the needs of lake managers.

Overview

Great Lakes bottom environments are critical zones of interface between geology, the physical processes that control morphology and bottom composition, and biological processes that support healthy ecosystems and human well-being.  Many challenging natural resource management issues are tightly coupled with the lake bottom, and for many of these we do not have sufficient information to support effective management responses.  For instance, round goby and quagga mussels are highly influential invasive species that are common to the bottom environments of all the Great Lakes except Lake Superior to 130 m depth, but their distributions and abundances are poorly documented.  Further, well-resolved maps of habitat types have been identified as a major data gap that impedes progress in habitat assessments such as those being undertaken in the Great Lakes Water Quality Agreement Annexes 2 and 7.  Finally, excessive growth of the nuisance algae Cladophora is tightly associated with the lake bottom environment, and improved methods are needed for understanding its distribution, biomass, and impacts.

New technologies are needed that can efficiently map and quantify bottom features of the Great Lakes to serve the needs of management agencies in the United States and Canada.  The current project proposes to combine three technologies—underwater robots, computer vision, and stereoscopic imaging—to develop methods to remotely characterize bathymetry, surface geology, round goby abundances, and Cladophora biomass. 

Goals and objectives

The goal of the proposed work is to demonstrate the utility of underwater remote sensing by developing a robot-deployed computer vision system capable of automatically quantifying lake bottom features.

This goal will be accomplished with multiple partners by pursuing the following objectives:

  • Implement a field sampling program to collect spatially extensive data across (a) a range of bottom habitat types, (b) a gradient of Cladophora productivity, and (c) a gradient of round goby biomass density.
  • Develop computer models to automatically classify bottom types, with special emphasis on substrates, Cladophora, and other submerged aquatic vegetation (SAV).
  • Develop computer models to predict Cladophora biomass from its volumetric occupancy of the water column.
  • Develop algorithms to identify round goby and estimate their biomass from pixel areas occupied.
  • Train partners in field data collection, image handling, and the application of algorithms to automate image processing and interpretation.
  • Publish the results of the research in peer-reviewed scientific literature.

Environment and Climage Change Canada Webinar Recording

Anticipated products

The following products and deliverables will result from this work in 2019.

  • High-resolution maps of bathymetry, bottom type, and Cladophora biomass for sampled areas in the form of geospatial datasets.
  • Open-source software for:
    • Bottom type classification
    • Cladophora biomass estimation
    • Round goby biomass estimation
  • Written tutorial for using the software
  • Training of field and lab personnel in the use of computer vision methods
  • Image libraries of labeled images for use in future model refinement
  • Interim report in March 2018 reporting on what was achieved up to that point.
  • Scientific publications