ADVANCING SINKHOLE IDENTIFICATION AND MAPPING IN KENTUCKY USING LIDAR AND MACHINE LEARNING

The original Kentucky karst sinkhole database was limited in accuracy because the database was compiled from outdated 1:24,000 scale topographic maps. Lidar(Light Detection and Ranging), which measures the Earth's surface using lasers, provides high-resolution and high-accuracy elevation data for improving sinkhole identification and mapping. We use a four-step process to map sinkholes from lidar data. The process involves creating a digital elevation model (DEM) using Lidar point clouds, extracting surficial depression features from the DEM, inspecting these features for potential sinkholes, and conducting field-checks for verification. To expedite the inspection of depression features, a trained neural network classifier is implemented, dramatically reducing the time for inspection. This project represents a continuation effort to update the sinkhole database in Kentucky using Lidar and machine learning. This work results in a fivefold increase in mapped sinkholes in an area encompassing Carroll, Gallatin, Grant, Henry, and Spencer Counties