What you gain, and what it costs
LAZ compression is the rare case where the tradeoff is close to one-sided. You are not choosing between size and quality — LAZ stores exactly the same data as the LAS it came from — so the only real cost is a small amount of CPU time whenever the file is read.
| Typical LAS size | Expected LAZ size | Saved |
|---|---|---|
| 100 MB | 10 – 20 MB | ~85% |
| 1 GB | 100 – 200 MB | ~85% |
| 10 GB | 1 – 2 GB | ~85% |
Ranges reflect the commonly cited 7–20% figure for LASzip on real survey data. Your ratio depends on point density, how many attributes are populated, and the coordinate scale factors in the header — so treat these as an expectation, not a guarantee.
Why lossless means lossless here
LAZ does not approximate anything. It exploits the fact that neighbouring LiDAR points are highly correlated — consecutive coordinates differ by small amounts, classifications repeat, return numbers cycle — and encodes those differences with an arithmetic coder instead of storing each value in full. Decompression reverses the arithmetic exactly.
This converter verifies that claim rather than repeating it: the test suite converts LAS to LAZ and back and asserts the result isbyte-for-byte identical to the input — header, VLRs, EVLRs and every point record. How that is tested.
When to compress
- Archiving. There is no reason to store uncompressed LAS long-term. Same data, a fraction of the disk.
- Sending files to clients. A 12 GB LAS is awkward to transfer; a 1.4 GB LAZ usually is not.
- Cloud storage costs. An 85% reduction applies directly to the monthly bill.
- Not for a file you are about to read repeatedly in a tight processing loop, where the per-read decompression cost adds up. Keep a working LAS, archive the LAZ.
What is preserved
The original public header block, all user VLRs and EVLRs — including the coordinate reference system, whether stored as GeoTIFF keys or WKT — extra bytes, and every per-point attribute. The naive way to build a LAZ writer drops user VLRs and takes the CRS with them; this one does not.