Monitoring of a Longleaf Pine Ecosystem Restoration Site Using UAS-Based Remote Sensing and Deep Learning Analysis

Integrated geospatial techniques, including UAS-Structure from Motion (SfM), UAS-LiDAR, high-accuracy GNSS receivers, and geospatial deep learning, were used to examine tree growth and survival patterns at a young longleaf pine ecosystem restoration site established in 2016 at Wormsloe State Historic Site near Savannah, Georgia. Direct comparison of SfM and LiDAR three-dimensional models showed that LiDAR produced slightly more accurate estimates of tree height. In recent 2023 UAS-SfM data, the neural network Mask R-CNN identified 3,369 longleaf pines with approximately 87% accuracy; median estimated tree height was 5.37 m. Time series results showed that steady pine vertical growth had begun by 2021, five years after planting. Multiple regression analyses with six environmental variables showed that the variables were better predictors of longleaf pine survival (R2 = 0.35) than height (R2 = 0.10). The results demonstrate the utility of advanced remote sensing and image analysis techniques for monitoring longleaf pine forests.