About the Author

Imagery Analysis Improves Oyster Reef Restoration Efforts in the Chesapeake Bay

By Lina Sorg

Although the Chesapeake Bay boasts some of America’s most valuable commercial fisheries, its fish and shellfish have experienced a stark decrease in diversity and productivity. Over the last century, disease, habitat loss, and copious amounts of commercial overfishing have contributed to the collapse of the native eastern oyster — the most common oyster on the eastern seaboard. This decline has wreaked significant ecological and economic consequences on the bay and surrounding areas. 

Oysters are natural filter-feeders, and a single adult oyster is capable of filtering 50 gallons of water per day. Robust, healthy populations effectively filter the water to reduce sediment levels and maintain water quality. They also provide food and habitat to other wildlife, prevent erosion and sea level rise, and contribute to the overall maintenance of a healthy bay ecosystem. Unfortunately, declining numbers of oysters are overwhelmed and choked by unmanageable amounts of sediment. In an effort to restore eastern oyster populations, researchers are attempting to construct artificial reefs and introduce them to natural waterways.

During a minisymposium at the 2018 SIAM Annual Meeting, currently taking place in Portland, Ore., Sofya Zaytseva of the College of William & Mary uses imagery analysis on the remaining reefs in the Chesapeake Bay watershed to model reef morphology and guide oyster restoration efforts. “The big idea here is pattern formation, particularly in marine communities,” she said. “Not only is pattern formation ubiquitous in these communities, it’s important in understanding how these communities organize, self-organize, and adapt to environmental change.” 

Oyster reefs comprise a bottom layer of sediment; a middle layer of shell, rubble, or another substrate on which live oysters can attach; and a top layer of the oysters themselves. Reef height, salinity/turbidity, location, and geometry and orientation are all important factors in both natural and artificial reef construction. “Do you just dump a bunch of shell and hope that oysters just come and attach, or do you do it in an organized way?” Zaytseva asked. “That’s the question of interest.”

Reefs are typically configured in three main ways, based on orientation to flow. String reefs run perpendicular to the water channel; patch reefs are small, round, and located far from channels in areas of low flow; and long fringe reefs run parallel to flow and are closest to the channel in high-flow areas. Zaytseva seeks to discover which initial artificial reef configuration yields the best results, in terms of reef expansion and persistence. She quantitatively classifies naturally-occurring oyster reefs and investigates whether (i) reefs do in fact exhibit the three aforementioned overarching configurations, and (ii) similar reefs are clustered with a non-random distribution.

Zaytseva begins by identifying two different study sites around the Chesapeake Bay. The first site hosts intertidal reefs, which are located in the rocky areas between low and high tide marks. This particular site sits on private land and is thus natural and undisturbed. She and her team used drones to collect detailed multispectral aerial images of the reefs, taken in an RGB color model. Subtidal reefs—submerged on the river bottom at all times—populate the second site. Zaytseva utilizes solar imagery data from the National Oceanic and Atmospheric Association (NOAA) to visualize these reefs and draw conclusions about the specific geometries of various reef types. These images are in grayscale. She uses supervised image classification for the RGB aerial (intertidal) images and texture analysis for the grayscale (subtidal) sonar images. She begins with the RGB intertidal images and calculates a texture layer to use in classification by computing standard deviation over a seven-by-seven window. Upon obtaining the necessary information on color and texture, Zaytseva performs supervised classification on the resulting four banded images, and uses training samples to represent the different habitat types. She delineates the images by removing small features, filling holes, and separating/merging nearby features; this reveals the particular types of reefs.

While one can employ conventional image classification techniques for multispectral aerial imagery, such techniques are not sufficient for grayscale sonar imagery. Thus, imagery analysis for the subtidal reefs is more challenging because the images contain only one spectral band and no color. After calculating a texture layer, Zaytseva applies an edge detection filter and separates the pixels into edge pixels using a gradient-based edge detector with a three-by-three sliding window. Post-processing techniques remove small features and fill the holes, ultimately revealing the different reef configurations as a series of polygons.

In the future, Zaytseva plans to assess classification accuracy based on additional field data. She wishes to calculate the attributes for area, shape, and orientation of primary axis in relation to the water channel before clustering the reefs into meaningful groups, classifying spatial distribution, and assessing overall composition (ratio of living and dead shells). She also plans to quantify the relationships between reef geometry, flow conditions, bathymetry, and chlorophyll concentrations. Ultimately, these imaging techniques help researchers better understand reef morphology and guide oyster restoration efforts; results can help scientists identify ideal configurations and locations of artificial oyster reefs to begin restoring population levels.

Lina Sorg is the associate editor of SIAM News