<aside>

Pipeline overview:

1. Download the 13 .las files from rclone into /scratch/jw8868/lidar_tmp/
2. Merge all files into one combined point cloud with laspy
3. Z-axis cleanup — estimate ground plane using a grid-based percentile approach, then remove points >1m above or below it
   1. Install deps — pip install --user laspy csf (sklearn + numpy already available)
4. CSF filter — classify ground vs off-ground, discard off-ground points
5. K-means clustering — run sklearn KMeans on the remaining ground points (K=5)
6. Save output — write the final clustered LAS file to /scratch/jw8868/ (or home) </aside>