Airborne Laser Scanning point clouds - from LAZ to TileDB at scale

---

alsDB reads LAZ/LAS files via PDAL, stores them in a TileDB sparse array (locally or on S3-compatible object storage), and provides a full processing pipeline for forest structure and biomass products.

The package is dataset-agnostic — CRS, bounding box, and acquisition year are read directly from the LAZ file header, so any national or global ALS dataset works without custom parsers.

Features

Scalable ingestion	Parallel batch ingest; each tile becomes a TileDB fragment
Multi-temporal	X / Y / Year dimensions; repeated surveys stored and queryable independently
Ingestion manifest	Tracks CRS, bbox, point count and status; re-ingestion is a no-op by default
CRS-aware schema	Domain bounds selected automatically from tile CRS or global fallback
Local + S3 storage	Identical API for filesystem paths and s3:// URIs (tested on Ceph / RadosGW)
Zarr gridded output	CHM, DTM, DSM, AGB, gap fraction, LAI, and LiDAR metrics written directly to Zarr v3; no GeoTIFF intermediates, no mosaic step
Tiled processing	All products support tile_size / n_workers for large areas; parallel writes go to non-overlapping Zarr chunks
GEDI simulation	Full-waveform simulation and batch RH metric extraction at GEDI footprint scale
CLI	alsdb ingest and alsdb info commands

---

Installation

pixi is the only supported install method — pdal and python-pdal are conda-only packages and cannot be installed via pip alone.

git clone https://github.com/simonbesnard1/alsdb.git
cd alsdb
pixi install          # resolves all conda-forge + pip dependencies in one step
pixi shell            # activate the environment

Note: pip install alsdb will install the pure-Python dependencies but will fail to import without pdal present in your environment.

Dependencies

Package	Purpose
pdal / python-pdal	LAZ/LAS reading, HAG filter
tiledb	Sparse point-cloud storage (local + S3)
zarr	Zarr v3 store for gridded products
numpy / scipy	Array operations, rasterisation, peak detection
pandas / xarray	Query result formats
pyarrow	Parquet I/O for waveform batch results
pyproj	CRS parsing from WKT
rioxarray	CRS attachment on to_dataset() output (optional)
matplotlib / plotly	Visualisation
click	CLI

---

Quick start

import alsdb
from alsdb import ALSDatabase, ALSProvider

alsdb.setup_logging()

# Ingest
db = ALSDatabase(storage_type="local", uri="my_array")
db.ingest("path/to/tile.laz")                        # year, CRS, bbox from LAZ header
db.ingest_many(paths, max_workers=8)                 # batch — already-ingested files skipped

# Query
reader = ALSProvider(storage_type="local", uri="my_array")
df = reader.query_bbox(308_000, 4_688_000, 310_000, 4_690_000, year=2021)
print(reader.available_years())                      # [2019, 2021, 2023]

---

Examples

Runnable Jupyter notebooks are in the examples/ folder:

Notebook	What it covers
01_ingest.ipynb	Single tile, batch ingest, manifest, S3
02_processing.ipynb	CHM / DTM / DSM, gap / LAI, metrics, biomass (Næsset + sklearn), xarray, change detection
03_waveform.ipynb	Single footprint, batch simulation, 3-D waterfall, GEDI comparison
04_visualisation.ipynb	Point cloud plots, waveform plots, gridded product plots

Full API documentation and a conceptual guide are on Read the Docs.

---

CLI

# Ingest a tile
alsdb ingest tile.laz my_array

# Ingest to S3
alsdb ingest tile.laz s3://owner.bucket/als_array \
    --storage-type s3 \
    --s3-url https://s3.example.com \
    --s3-access-key KEY --s3-secret-key SECRET

# Show file metadata
alsdb info tile.laz

# Filter to ground + vegetation classes only
alsdb ingest tile.laz my_array -c 2 -c 3 -c 4 -c 5

---

Architecture

LAZ file
   │
   ▼
PDAL (readers.las)
   │  year / bbox / CRS ← LAZ header
   ▼
ALSTile.iter_chunks()           ← optional classification filter
   │  X, Y, attrs numpy arrays (1 M pts/chunk)
   ▼
ALSDatabase.write()
   │  TileDB sparse array  (X × Y × Year)
   │  ByteShuffle+ZSTD for float64 X/Y; DoubleDelta+ZSTD for int16 Year
   │  allows_duplicates=True  (multiple returns per XY)
   ▼
TileDB array (local  /  s3://)
   │
   ├── ALSProvider.query_bbox()          → pandas / xarray
   │
   ├── processing.chm / gap / biomass
   │      │  PDAL hag_delaunay + scipy binned_statistic_2d
   │      ▼
   │   ALSZarrStore  (Zarr v3, local / s3://)
   │      ├── 1m/   chm, dtm, dsm          (T × ny × nx) float32
   │      └── 10m/  gap, lai, biomass,
   │                h50…density            (T × ny × nx) float32
   │      store.to_dataset(resolution)     → xarray.Dataset (CRS-aware)
   │
   └── processing.waveform               → GEDI-like RH metrics

TileDB schema

Dimension	Type	Role
X	float64	UTM easting (m)
Y	float64	UTM northing (m)
Year	int16	Survey year

18 LAS attributes (Z, Intensity, ReturnNumber, Classification, RGB, …) are stored as TileDB attributes with ZSTD-9 compression. Spatial tile size defaults to 500 × 500 m; domain bounds are selected automatically per CRS.

Each ingested tile becomes a new TileDB fragment. Fragments are consolidated every consolidate_every tiles (default 50) to keep read performance healthy as the array grows.

---

License

EUPL-1.2 — see LICENSE.

© 2026 Simon Besnard, Helmholtz Centre Potsdam – GFZ German Research Centre for Geosciences.