seismic-zfp

Python library to convert SEG-Y files to compressed cubes and retrieve arbitrary sub-volumes from these, fast.

Motivation

Reading whole SEG-Y volumes to retrieve, for example, a single time-slice is wasteful.

Copying whole SEG-Y files uncompressed over networks is also wasteful.

This library addresses both issues by implementing the seismic-zfp (.SGZ) format. This format is based on ZFP compression from Peter Lindstrom's paper using the official Python bindings, distributed as zfpy.

ZFP compression enables smoothly varying d-dimensional data in 4^d subvolumes to be compressed at a fixed bitrate. The 32-bit floating point values in 4x4x4 units of a 3D post-stack SEG-Y file are well suited to this scheme.

Decomposing an appropriately padded 3D seismic volume into groups of these units which exactly fill one 4KB disk block, compressing these groups, and writing them sequentially to disk yields a file with the following properties:

• Compression ratio of 2ⁿ:1 compression, typically a bitrate of 4 works well, implying a ratio of 8:1
• The location of any seismic sample is known
• Arbitrary subvolumes can be read with minimal redundant I/O, for example:
  • Padding IL/XL dimensions with 4, and the z-dimension depending on bitrate
  • Padding IL/XL dimensions with 64 and the z-dimension with 4 (16:1 compression)

Using IL/XL optimized layout

• Groups of 4 inlines or crosslines can be read with no redundant I/O
• A single inline can be read and with no additional I/O compared to the SEG-Y best-case scenario (provided at least 4:1 compression ratio)
• A z-slice can be read by accessing n_traces/16 disk blocks, compared to n_traces disk blocks for SEG-Y

Using z-slice optimized layout

• A z-slice can be read by accessing just n_traces/4096 disk blocks, compared to n_traces disk blocks for SEG-Y

2D SEG-Y Support

As of v0.2.4 support for compressing 2D SEG-Y files (INLINE_3D and CROSSLINE_3D always zero) is included. Compression and reading follows the same pattern as 3D files, but segyio emulation only provides the following attributes: trace, header, samples, bin & text. However an additional funciton read_subplane() is available for extracting horizontally and vertically contrained data.

4D (Prestack) SEG-Y Support

As of v0.5.0 regular prestack SEG-Y files (inline-sorted, with the same offsets present at every IL/XL position, identified by the offset trace header) can be compressed. The offset axis is stored between the crossline and sample axes, and ZFP compresses the data in 4x4x4x4 units, so the blockshape becomes (il, xl, offset, z) with a default of (4, 4, 4, -1). 4D SGZ files can be read through SgzReader (read_gather, read_offset, read_subvolume_4d), the segyio-like interface (gather, iline[il, offset], subvolume) and xarray, and can be converted back to SEG-Y.

Headers

The seismic-zfp (.SGZ) format also allows for preservation of information in SEG-Y file and trace headers, with compression code identifying constant and varying trace header values and storing these appropriately.

For further explanation of the design and implementation of seismic-zfp, please refer to publications.

NOTE: Previously the extension .sz was used for seismic-zfp, but has been replaced with .sgz to avoid confusion around the compression algorithm used.

Get seismic-zfp

• Wheels from PyPI with zgy and vds support: pip install seismic-zfp[zgy,vds]
• Wheels from PyPI without zgy or vds support: pip install seismic-zfp
• Source from Github: git clone https://github.com/equinor/seismic-zfp.git

Note that seismic-zfp depends on the Python package ZFPY, which is a binary distribution on PyPI built for Linux and Windows.

The optional dependency pyvds - requires openvds package from Bluware which is not open-source

The optional dependency of zgy2sgz has been replaced with pyzgy - a pure-Python alternative.

Examples

Full example code is provided here, and standalone utilities such as a tool to regularise irregular prestack SEG-Y are in tools. The following reference is useful:

Create SGZ files from SEG-Y, ZGY or VDS

from seismic_zfp.conversion import SegyConverter, ZgyConverter, VdsConverter

with SegyConverter("in.sgy") as converter:
    # Estimate the output file size before compression
    size_bytes = converter.get_output_size(bits_per_voxel=4)
    # Create a "standard" SGZ file with 8:1 compression, using in-memory method
    converter.run("out_standard.sgz", bits_per_voxel=4)
    # Create a "z-slice optimized" SGZ file
    converter.run("out_adv.sgz", bits_per_voxel=2, blockshape=(64, 64, 4))

with SegyConverter("in_prestack.sgy", min_offset=0, max_offset=16) as converter:
    # Prestack (4D) SEG-Y is detected automatically, blockshape is (il, xl, offset, z)
    converter.run("out_4d.sgz", bits_per_voxel=4, blockshape=(4, 4, 4, -1))
                  
with ZgyConverter("in_8-int.zgy") as converter:
    # 8-bit integer ZGY and 1-bit SGZ have similar quality
    converter.run("out_8bit.sgz", bits_per_voxel=1)

with VdsConverter("in.vds") as converter:
    # VDS compression is superior, but doesn't permit non-cubic bricks...
    converter.run("out.sgz", bits_per_voxel=4, blockshape=(8, 8, 128))

Convert SGZ files to SEG-Y

# Convert SGZ to SEG-Y
with SgzConverter("out_standard.sgz") as converter:
    converter.convert_to_segy("recovered.sgy")

Read an SGZ file

from seismic_zfp.read import SgzReader
with SgzReader("in.sgz") as reader:
    inline_slice = reader.read_inline(LINE_IDX)
    crossline_slice = reader.read_crossline(LINE_IDX)
    z_slice = reader.read_zslice(LINE_IDX)
    sub_vol = reader.read_subvolume(min_il=min_il, max_il=max_il, 
                                    min_xl=min_xl, max_xl=max_xl, 
                                    min_z=min_z, max_z=max_z)

Use segyio-like interface to read SGZ files

import seismic_zfp
with seismic_zfp.open("in.sgz")) as sgzfile:
    il_slice = sgzfile.iline[sgzfile.ilines[LINE_NUMBER]]
    xl_slices = [xl for xl in sgzfile.xline]
    zslices = sgzfile.depth_slice[:5317]
    trace = sgzfile.trace[TRACE_IDX]
    trace_header = sgzfile.header[TRACE_IDX]
    binary_file_header = sgzfile.bin
    text_file_header = sgzfile.text[0]

Including equivalents to segyio.tools

with seismic_zfp.open("in.sgz") as sgz_file:
    dt_sgz = seismic_zfp.tools.dt(sgz_file)

cube_sgz = seismic_zfp.tools.cube("in.sgz")

Plus some extended utility functionality:

with seismic_zfp.open("in.sgz")) as sgzfile:
    subvolume = sgzfile.subvolume[IL_NO_START:IL_NO_STOP:IL_NO_STEP, 
                                  XL_NO_START:XL_NO_STOP:XL_NO_STEP,
                                  SAMP_NO_START:SAMP_NO_STOP:SAMP_NO_STEP]
    header_arr = sgzfile.get_tracefield_values(segyio.tracefield.TraceField.NStackedTraces)

Open SGZ files with xarray

seismic-zfp registers an xarray backend, so SGZ files open lazily as a Dataset with a single data variable over dimensions (il, xl, z), or (il, xl, offset, z) for 4D files, with coordinates in line numbers, offset values and sample times. Only the part of the file covering a selection is decompressed.

import xarray as xr
with xr.open_dataset("in.sgz") as ds:
    inline = ds.data.sel(il=IL_NO).to_numpy()
    gather = ds.data.sel(il=IL_NO, xl=XL_NO).to_numpy()              # 4D files
    stack = ds.data.sel(il=IL_NO).mean(dim="offset").to_numpy()      # 4D files

Command Line Interface

A simple command line interface for converting from SEGY or ZGY to SGZ is also bundled with seismic-zfp. This is available using the command seismic-zfp.

Converting from SEG-Y to SGZ and back

seismic-zfp sgy2sgz <INPUT FILE PATH> <OUTPUT FILE PATH> --bits-per-voxel 4
seismic-zfp sgz2sgy <INPUT FILE PATH> <OUTPUT FILE PATH>

To see all options run the command seismic-zfp sgy2sgz --help

Converting from ZGY to SGZ

seismic-zfp zgy2sgz <INPUT FILE PATH> <OUTPUT FILE PATH> --bits-per-voxel 4

To see all options run the command seismic-zfp zgy2sgz --help

Contributing

Contributions welcomed, whether you are reporting or fixing a bug, implementing or requesting a feature. Either make a github issue or fork the project and make a pull request. Please extend the unit tests with relevant passing/failing tests, run these as: python -m pytest

Publications

Wade, David (2020): Seismic-ZFP: Fast and Efficient Compression and Decompression of Seismic Data, First EAGE Digitalization Conference and Exhibition, Nov 2020, Volume 2020, p.1-5

• Full-Text PDF (without EAGE formatting)
• EarthDoc Link (N.B. paywall protected)