Researchers Mahin Ganesan, Brendan Jalali, Tom Lever, and Nicholas Miller undertook a capstone project that researched ghost forest detection using satellite imagery.
Their project aimed to identify ghost forests across the East Coast of the United States using deep learning algorithms for image detection. Ghost forests are areas with more than 10 dead trees per hectare, and their locations are currently largely unknown.
For the study and preservation of coastal forests, their sponsors, researchers in the Department of Environmental Science at the University of Virginia, have taken an interest in developing a tool that will be able to identify ghost forests across the country. Their model, which predicts the locations of dead trees in ghost forests, is a key step in understanding the
mechanisms behind forest decline due to saltwater intrusion.
To predict dead trees, the team adapted an existing urban tree detection neural network to predict the centroids of dead trees in satellite images, achieving an F1 score of 0.795.
Overall, the goal of this project is for the members to participate in developing a data science pipeline to identify ghost forests, which will provide an understanding of where and why ghost forest hot spots may arise.
By knowing where ghost forests occur, governments and localities can push for legislation to keep these forests from dying, allowing them to continue protecting the surrounding coastal communities and ecosystems.
Researchers: Mahin Ganesan, Brendan Jalali, Tom Lever, Nicholas Miller
Sponsors: Xi Yang
Advisors: Heman Shakeri