---

name: segysak description: Read SEG-Y as labelled xarray datasets with SEGY-SAK, inspect header-defined survey dimensions and process seismic volumes lazily. Use for SEG-Y/xarray/Dask workflows; use segyio for exact low-level trace and header editing. license: MIT metadata: version: "1.0.0" author: Geoscience Skills skill_type: domain tags: '["SEG-Y", "Seismic", "xarray", "Dask"]' dependencies: '["segysak==0.5.4"]' complements: '["segyio", "xarray", "bruges"]' workflow_role: data-loading

SEGY-SAK

Map trace headers to labelled dimensions without losing the original survey geometry. Header byte assignments are evidence to inspect, not safe guesses.

Before opening

Inspect format/endian, trace count, sample count/interval, delay recording time, inline/crossline or CDP headers, coordinate units and scalar. SEG-Y sample intervals are stored in microseconds; the backend's vertical samples use milliseconds for ordinary time sections. A depth section needs an explicit depth convention. Confirm actual data units before labelling an axis.

Check dimension key uniqueness and missing traces. Sparse or irregular geometry may not form a complete rectangular cube. Keep the original trace index for exports. A dimension byte offset is one-based SEG-Y numbering (CDP ensemble is 21, inline is 189, crossline is 193); some upstream old examples use a wrong CDP offset. Use the actual file's header convention.

Labelled read

This fragment needs an existing, verified rectangular post-stack file. It uses the 0.5 backend API; old segy_loader tutorials describe a different interface.

# example: labelled-segy
import xarray as xr
import segysak  # registers accessors; the installed distribution provides the backend

with xr.open_dataset(
    seismic_path,
    engine="sgy_engine",
    dim_byte_fields={"iline": 189, "xline": 193},
) as seismic:
    inline_labels = seismic.iline.values.copy()
    crossline_labels = seismic.xline.values.copy()
    sample_coordinate = seismic.samples.values.copy()
    amplitudes = seismic["data"].load().copy(deep=True)

For large surveys, open with Dask chunks chosen for the access pattern and retain lazy operations until the selected result fits memory. Avoid .load() on a whole production cube. Inspect coordinate variables before .sel() and use .isel() only for integer positions.

Validation and exports

Compare several labelled traces against original trace samples using segyio, including the last sample and nonzero start time. Verify shape, orientation, missing-trace behavior and coordinate scalars. Round-trip any export and compare headers as well as amplitudes; xarray labels alone do not preserve every SEG-Y header or prove geographic CRS. Keep source hash, byte mapping and transforms.

Use the current documentation and backend source.