Coastal Cliff LiDAR Change Detection Pipeline

A modular Python pipeline for processing terrestrial LiDAR surveys of coastal cliffs. Ingests raw point clouds, removes non-cliff features (beach, vegetation), performs 3D change detection (M3C2), clusters erosion and deposition events, and aggregates results into spatiotemporal grids for analysis.

Paper: Mack, C.J., Maclay, M., Krier-Mariani, R., & Young, A.P. (2026). Integrated machine learning segmentation and 3D change detection for a scalable coastal cliff monitoring workflow. Computers & Geosciences, 106165. doi:10.1016/j.cageo.2026.106165

Pipeline

graph TD
    A[Raw Survey Data] -->|Step 0-1| B[Survey Inventories]
    B -->|Step 2| C[Crop to Study Area]
    C -->|Step 3| D[Beach Removal - Random Forest]
    D -->|Step 4| E[Vegetation Removal - CANUPO]
    E -->|Step 5| F[M3C2 Change Detection]
    F -->|Step 6| G[DBSCAN Clustering]
    G -->|Step 7| H[Spatial Gridding]
    H -->|Step 8| I[Grid Cleaning & Hole Filling]
    I --> J[Event Lists & 3D Data Cubes]

Loading

Requirements

• Python >= 3.9
• PDAL — point cloud cropping (install)
• CloudCompare — M3C2 and CANUPO (install)
• xvfb — required on headless Linux servers for CloudCompare steps

Setup

# Conda (recommended)
conda env create -f environment.yml
conda activate cliff-change-detection

# Or pip
pip install -r requirements.txt

Usage

Scripts in code/pipeline/ run sequentially (0–8). Each step transforms data for the next.

Step	Script	Description
0	0_make_survey_lists.py	Build initial survey inventory CSVs
1	1_update_survey_lists.py	Update inventories with new surveys
2	2_crop_files_parallel.py	Crop raw LAS to study area via PDAL
3	3_remove_beach_parallel.py	Remove beach points (Random Forest)
4	4_remove_veg_parallel.py	Remove vegetation (CANUPO via CloudCompare)
5	5_m3c2_parallel.py	M3C2 change detection (CloudCompare)
6	6_dbscan_parallel.py	Cluster erosion/deposition events (DBSCAN)
7	7_make_grids.py	Aggregate into spatial grids (10cm, 25cm, 1m)
8	8_clean_fill_grids.py	Apply cliff-top cutoffs and fill occlusion holes

Most scripts accept --location <name> or --all, and --n_jobs to control parallelism. Use --help on any script for full options.

# Example: process San Elijo through the full pipeline
python3 code/pipeline/2_crop_files_parallel.py --location SanElijo --replace
python3 code/pipeline/3_remove_beach_parallel.py SanElijo --n_jobs 5
python3 code/pipeline/4_remove_veg_parallel.py SanElijo --cc /path/to/CloudCompare
python3 code/pipeline/5_m3c2_parallel.py SanElijo --cc /path/to/CloudCompare
python3 code/pipeline/6_dbscan_parallel.py SanElijo --eps 0.35 --min_samples 30
python3 code/pipeline/7_make_grids.py SanElijo --resolution 25cm
python3 code/pipeline/8_clean_fill_grids.py SanElijo --resolution 25cm

Automated Daily Pipeline

python3 code/pipeline/run_daily.py             # Process locations with new data
python3 code/pipeline/run_daily.py --force-all  # Force reprocess all

Study Sites

Location	MOP Range
Blacks	520–567
Torrey	567–581
DelMar	595–620
Solana	637–666
SanElijo	683–708
Encinitas	708–764

Testing

pytest tests/pytest/

Citation

If you use this software, please cite:

@article{Mack2026,
  author  = {Mack, Connor J. and Maclay, Matthew and Krier-Mariani, Raphael and Young, Adam P.},
  title   = {Integrated machine learning segmentation and 3D change detection for a scalable coastal cliff monitoring workflow},
  journal = {Computers \& Geosciences},
  pages   = {106165},
  year    = {2026},
  doi     = {10.1016/j.cageo.2026.106165}
}

License

See LICENSE for details.