Predicting the Rise of Ghost Forests in Eastern North Carolina

Since 2015, the Data+ summer research program has allowed small teams of students to explore the potential of data visualization and analysis. Now, its new sister spinoff initiative Climate+ has emerged as part of Duke’s Climate Commitment. It’s research projects continue to see students exercising technical problem solving but focus on combatting and examining the effects of a warming globe.

One effect is particularly evident in the Albemarle-Pamlico Peninsula, a four-hour drive east of Duke campus. Though the coastal region is almost entirely composed of natural wetland, here and there, only bare tree stalks stand where productive ecosystems once thrived. These stretches of death, known as ghost forests, are brought on by saline water intruding inland. With rising seas pushing back into freshwater rivers faster than before, they’re becoming a more common sight throughout North Carolina’s coastline and the entire eastern seaboard at large. “Detecting Saltwater Intrusion in Rivers Using Remote Sensing,” a 2024 Climate+ project, aimed to determine when and where this was happening.

It’s the composition of blackwater rivers in particular that makes it possible to visualize salinity.

When salt encounters the dissolved organic matter that makes the water dark, the two bind together, causing the river to lighten in color. However, determining the salinity of every blackwater river throughout the east coast is no easy feat.

By building on top of previous research conducted by John Gardner, a former PhD student in the Nicholas School of the Environment who is now an assistant professor of earth, marine, and environmental sciences at UNC, the project team had an idea of how they could use satellite imagery to determine river color. The initial steps were to isolate the data.

By learning how to use Google Earth Engine API in Python, the team pulled down the satellite data they needed. Using the National Hydrology Dataset, they first mapped out the rivers’ flowlines with the programming language R and then buffered these in Python to accurately include their full width. From here, any parts that weren’t river water—such as clouds, vegetation, and land—were removed with the cloud-masking algorithm. At last, the data from remaining pixels could be used.

As a mechanical engineering student and an aspiring energy engineer, team member Derek Zhang applied to the project because its aims overlapped with his hopes for a healthier planet. And while it didn’t look like a typical engineering workspace, “an engineering mindset helped me out a lot,” Zhang said, “both in how to approach a challenge and work as a team.” He worked mainly on processing and sorting data using Python. Though Data+ projects often involve coding, students of any major and grade can participate, meaning some learn to code in these projects before being taught in class.

This experience proved important in the future, as Zhang entered his later engineering classes with a wider understanding of the capabilities of Python. “I think much of what I learned was useful for this year,” he said. Yet, in addition to technical exposure, much of what he learned was intangible. “It opened my mind up about what resources I could use,” he said. Where before he tended to attack problems with “brute force” and persistence, his team members caused him to adopt more of an efficient mindset and be quicker to shift methods— an important skill when working toward a strict deadline.

By the end of the summer session, the team saw how their data lined up with real-world context. Over the course of a year, surface reflectance increased in the summer, when warmer temperatures caused more thermal expansion and higher sea levels, then lower in the winter. Certain events could be correlated, too—an algal bloom that occurred in 2023 showed up as a large decrease in a graph for North Carolina’s Chowan River. In addition, surface reflectance increasingly varies further downstream, supporting the idea that downstream areas are more vulnerable to changes in environmental conditions.

While the team acknowledged potential for error remains in the program, their conclusions demonstrate a proof of concept. The framework manually set up using past data could, with considerable time and effort, become a machine learning algorithm for prediction. Moreover, Zhang stated, “It wasn’t just salinity we could infer from the data, but a whole myriad of quality factors.” These include levels of chlorophyll-a and dissolved organic matter. Such information will likely prove valuable for climate resilience and mitigation efforts.

The onset of ghost forests doesn’t just reduce habitat for flora and fauna—it directly affects the availability of water and fish for the local communities that rely on them. Being able to identify these trends in the data can help communities decide where and when they might need to prepare.

Crystal Han is a first-year engineering student.