From the Piedmont to Pamlico Sound, the Neuse River runs 275 miles and is home to many endangered species, including the piping plover, loggerhead turtle and Carolina madtom — a freshwater catfish. But when big storms hit, it can drastically change the shape and composition of a river, which is why water managers look to conserve and protect areas around the river. Hurricane Florence in 2018 caused one bluff to retreat nearly 65 feet. However, knowing where to start along the nearly 300-mile river can be challenging. Jessica Richter, a recent graduate of East Carolina University, looked at the Neuse River and created a map to log what it looks like now and identify areas at risk of rapid erosion.
“I was drawn to the Neuse River because as sea level rise intensifies and extreme storms become more frequent or intense issues like erosion are going to become increasingly prominent,” Richter said.
Her research was jointly funded by the North Carolina Sea Grant and North Carolina Space Grant, which requires the use of data from NASA and the National Oceanic and Atmospheric Administration to conduct research within North Carolina’s coastal areas and watersheds. Richter used machine learning to classify small parcels of land along the river. It tackled two questions: Could machine learning produce accurate maps, and if so, could the maps be used to identify areas in need of preservation?
“The big question at the end then is if these methods of machine learning ensembles and image analysis combined together, can they produce an accurate, usable map of an estuarine shoreline?” she said.
To create the maps, she used satellite pictures of the Neuse River and a machine learning technique known as object-based image analysis, which classified groups of pixels as either a swamp, marsh, natural bluff, low- or high-sediment bank, or as a modified shoreline.
“I was able to combine the outputs of five different individual machine learning algorithms in order to make a final prediction," Richter said. "From that we’re able to take the majority vote and make that designation for a particular object or area as a marsh.”
Richter ran five different algorithms — called a machine-learning ensemble — when looking at each part of the river’s shoreline. One algorithm might say "that is a marsh” while another might say “it's a low-sediment bank,” whichever category had the most votes would win. Richter says the maps are 76% accurate.
“Obviously there’s room for improvement, but overall, I was happy with the results that we had gotten,” she said.
Having a map that takes a close look at the Neuse River’s shoreline helps to not only compare the area before and after a storm, but it also helps environmentalists, engineers, and homeowners know where to prioritize their efforts.
“For example, knowing that a bluff existed in one location on one map before a storm, and then seeing in that same location years later after another storm – maybe it’s water or another type of shoreline – communities can better track where erosion is happening in the Neuse River estuary and prioritize mitigation or restoration efforts in these areas,” Richter said.
Richter’s maps are different than other maps. Currently, the National Oceanographic and Atmospheric Administration regularly creates regional maps that categorize 30-meter tracts of shoreline. Richter’s maps are hyper localized, making them ideal for use by local officials in resiliency efforts.
“This information helps Craven County and the Croatan National Forest as they determine the best approaches to re-stabilize private property and recreational sites, respectively, to those two organizations, just to be more resilient to storms like Hurricane Florence.”
Richter’s research showed that machine learning can produce accurate maps, opening the door to regularly mapping and tracking the Neuse River as well as other estuaries.
“By figuring out that machine learning can accurately predict shoreline classes in the NRE, it makes it more feasible to produce high-resolution shoreline maps on a more frequent basis and give shoreline managers and shoreline communities more information that they can even better inform future mitigation and restoration efforts.”
