Artificial Intelligence is being applied to many areas of life, including forestry on the Colorado Plateau. At Northern Arizona University, a team led by Andrew Sánchez Meador is using AI models in conjunction with Light Detection and Ranging or LIDAR technology.
Often deployed from a plane, LIDAR shoots laser beams at the forest below, collecting the reflections in what’s called a "point cloud."
AI models designed to analyze visual data, called “convolutional neural networks,” are being taught to recognize what these arrangements of points represent, including identifying tree species and measuring forest structure.
One AI model training ground covers 3,000 acres of mixed coniferous forest on the Mogollon Rim south of Blue Ridge Reservoir. Located where the headwaters of the Salt and Verde Rivers rise, this is an important area for forest thinning, restoration and water quality management.
One of AI’s toughest tasks is identifying the seven tree species present. Even observers on the ground can find Douglas fir and White fir hard to distinguish, but AI is already accurate in separating them correctly more than 90% of the time.
After an AI model has been trained on one area, the researchers use a technique called transfer learning to test it on a new patch of forest to refine its accuracy and learn to identify new species.
Within the next 10 years, such AI models may make it possible to inventory forests even over remote terrain — much more quickly, cheaply and on a scale not previously possible with ground-based surveys.
This Earth Note was written by Diane Hope and produced by KNAU and the Sustainable Communities Program at Northern Arizona University.
