Skip to content
Switch branches/tags
This branch is up to date with master.

Latest commit


Git stats


Failed to load latest commit information.
Latest commit message
Commit time


Travis Appveyor PyPI Updates Python 3


The official documentation is hosted on readthedocs.



Segyio is a small LGPL licensed C library for easy interaction with SEG-Y and Seismic Unix formatted seismic data, with language bindings for Python and Matlab. Segyio is an attempt to create an easy-to-use, embeddable, community-oriented library for seismic applications. Features are added as they are needed; suggestions and contributions of all kinds are very welcome.

To catch up on the latest development and features, see the changelog. To write future proof code, consult the planned breaking changes.

Feature summary

  • A low-level C interface with few assumptions; easy to bind to other languages
  • Read and write binary and textual headers
  • Read and write traces and trace headers
  • Simple, powerful, and native-feeling Python interface with numpy integration
  • Read and write seismic unix files
  • xarray integration with netcdf_segy
  • Some simple applications with unix philosophy

Getting started

When segyio is built and installed, you're ready to start programming! Check out the tutorial, examples, example programs, and example notebooks. For a technical reference with examples and small recipes, read the docs. API docs are also available with pydoc - start your favourite Python interpreter and type help(segyio), which should integrate well with IDLE, pycharm and other Python tools.

Quick start

import segyio
import numpy as np
with'file.sgy') as f:
    for trace in f.trace:
        filtered = trace[np.where(trace < 1e-2)]

See the examples for more.

Get segyio

A copy of segyio is available both as pre-built binaries and source code:

  • In Debian unstable
    • apt install python3-segyio
  • Wheels for Python from PyPI
    • pip install segyio
  • Source code from github
    • git clone
  • Source code in tarballs

Build segyio

To build segyio you need:

  • A C99 compatible C compiler (tested mostly on gcc and clang)
  • A C++ compiler for the Python extension, and C++11 for the tests
  • CMake version 2.8.12 or greater
  • Python 2.7 or 3.x.
  • numpy version 1.10 or greater
  • setuptools version 28 or greater
  • setuptools-scm
  • pytest

To build the documentation, you also need sphinx

To build and install segyio, perform the following actions in your console:

git clone
mkdir segyio/build
cd segyio/build
make install

make install must be done as root for a system install; if you want to install in your home directory, add -DCMAKE_INSTALL_PREFIX=~/ or some other appropriate directory, or make DESTDIR=~/ install. Please ensure your environment picks up on non-standard install locations (PYTHONPATH, LD_LIBRARY_PATH and PATH).

If you have multiple Python installations, or want to use some alternative interpreter, you can help cmake find the right one by passing -DPYTHON_EXECUTABLE=/opt/python/binary along with install prefix and build type.

To build the matlab bindings, invoke CMake with the option -DBUILD_MEX=ON. In some environments the Matlab binaries are in a non-standard location, in which case you need to help CMake find the matlab binaries by passing -DMATLAB_ROOT=/path/to/matlab.


It's recommended to build in debug mode to get more warnings and to embed debug symbols in the objects. Substituting Debug for Release in the CMAKE_BUILD_TYPE is plenty.

Tests are located in the language/tests directories, and it's highly recommended that new features added are demonstrated for correctness and contract by adding a test. All tests can be run by invoking ctest. Feel free to use the tests already written as a guide.

After building segyio you can run the tests with ctest, executed from the build directory.

Please note that to run the Python examples you need to let your environment know where to find the Python library. It can be installed as a user, or on adding the segyio/build/python library to your pythonpath.


All code in this tutorial assumes segyio is imported, and that numpy is available as np.

import segyio
import numpy as np

This tutorial assumes you're familiar with Python and numpy. For a refresh, check out the python tutorial and numpy quickstart


Opening a file for reading is done with the function, and idiomatically used with context managers. Using the with statement, files are properly closed even in the case of exceptions. By default, files are opened read-only.

with as f:

Open accepts several options (for more a more comprehensive reference, check the open function's docstring with help( The most important option is the second (optional) positional argument. To open a file for writing, do, 'r+'), from the C fopen function.

Files can be opened in unstructured mode, either by passing the optional arguments strict=False, in which case not establishing structure (inline numbers, crossline numbers etc.) is not an error, and ignore_geometry=True, in which case segyio won't even try to set these internal attributes.

The segy file object has several public attributes describing this structure:

  • f.ilines Inferred inline numbers
  • f.xlines Inferred crossline numbers
  • f.offsets Inferred offsets numbers
  • f.samples Inferred sample offsets (frequency and recording time delay)
  • f.unstructured True if unstructured, False if structured
  • f.ext_headers The number of extended textual headers

If the file is opened unstructured, all the line properties will will be None.


In segyio, data is retrieved and written through so-called modes. Modes are abstract arrays, or addressing schemes, and change what names and indices mean. All modes are properties on the file handle object, support the len function, and reads and writes are done through f.mode[]. Writes are done with assignment. Modes support array slicing inspired by numpy. The following modes are available:

  • trace

    The trace mode offers raw addressing of traces as they are laid out in the file. This, along with header, is the only mode available for unstructured files. Traces are enumerated 0..len(f.trace).

    Reading a trace yields a numpy ndarray, and reading multiple traces yields a generator of ndarrays. Generator semantics are used and the same object is reused, so if you want to cache or address trace data later, you must explicitly copy.

    >>> f.trace[10]
    >>> f.trace[-2]
    >>> f.trace[15:45]
    >>> f.trace[:45:3]
  • header

    With addressing behaviour similar to trace, accessing items yield header objects instead of numpy ndarrays. Headers are dict-like objects, where keys are integers, seismic unix-style keys (in module) and segyio enums (segyio.TraceField).

    Header values can be updated by assigning a dict-like to it, and keys not present on the right-hand-side of the assignment are unmodified.

    >>> f.header[5] = { 10 }
    >>> f.header[5].items()
    >>> f.header[5][25, 37] # read multiple values at once
  • iline, xline

    These modes will raise an error if the file is unstructured. They consider arguments to [] as the keys of the respective lines. Line numbers are always increasing, but can have arbitrary, uneven spacing. The valid names can be found in the ilines and xlines properties.

    As with traces, getting one line yields an ndarray, and a slice of lines yields a generator of ndarrays. When using slices with a step, some intermediate items might be skipped if it is not matched by the step, i.e. doing f.line[1:10:3] on a file with lines [1,2,3,4,5] is equivalent of looking up 1, 4, 7, and finding [1,4].

    When working with a 4D pre-stack file, the first offset is implicitly read. To access a different or a range of offsets, use comma separated indices or ranges, as such: f.iline[120, 4].

  • fast, slow

    These are aliases for iline and xline, determined by how the traces are laid out. For inline sorted files, fast would yield iline.

  • depth_slice

    The depth slice is a horizontal, file-wide cut at a depth. The yielded values are ndarrays and generators-of-arrays.

  • gather

    The gather is the intersection of an inline and crossline, a vertical column of the survey, and unless a single offset is specified returns an offset x samples ndarray. In the presence of ranges, it returns a generator of such ndarrays.

  • text

    The text mode is an array of the textual headers, where text[0] is the standard-mandated textual header, and 1..n are the optional extended headers.

    The text headers are returned as 3200-byte string-like blobs (bytes in Python 3, str in Python 2), as it is in the file. The function can create a line-oriented version of this string.

  • bin

    The values of the file-wide binary header with a dict-like interface. Behaves like the header mode, but without the indexing.

Mode examples

>>> for line in f.iline[:2430]:
...     print(np.average(line))

>>> for line in f.xline[2:10]:
...     print(line)

>>> for line in[::2]:
...     print(np.min(line))

>>> for factor, offset in enumerate(f.iline[10, :]):
...     offset *= factor

>>> f.gather[200, 241, :].shape

>>> text = f.text[0]
>>> type(text)
<type 'bytes'> # 'str' in Python 2

>>> f.trace[10] = np.zeros(len(f.samples))

More examples and recipes can be found in the docstrings help(segyio) and the examples section.

Project goals

Segyio does not necessarily attempt to be the end-all of SEG-Y interactions; rather, we aim to lower the barrier to interacting with SEG-Y files for embedding, new applications or free-standing programs.

Additionally, the aim is not to support the full standard or all exotic (but standard compliant) formatted files out there. Some assumptions are made, such as:

  • All traces in a file are assumed to be of the same size

Currently, segyio supports:

  • Post-stack 3D volumes, sorted with respect to two header words (generally INLINE and CROSSLINE)
  • Pre-stack 4D volumes, sorted with respect to three header words (generally INLINE, CROSSLINE, and OFFSET)
  • Unstructured data, i.e. a collection of traces
  • Most numerical formats (including IEEE 4- and 8-byte float, IBM float, 2- and 4-byte integers)

The writing functionality in segyio is largely meant to modify or adapt files. A file created from scratch is not necessarily a to-spec SEG-Y file, as we only necessarily write the header fields segyio needs to make sense of the geometry. It is still highly recommended that SEG-Y files are maintained and written according to specification, but segyio does not enforce this.

SEG-Y Revisions

Segyio can handle a lot of files that are SEG-Y-like, i.e. segyio handles files that don't strictly conform to the SEG-Y standard. Segyio also does not discriminate between the revisions, but instead tries to use information available in the file. For an actual standard's reference, please see the publications by SEG:


We welcome all kinds of contributions, including code, bug reports, issues, feature requests, and documentation. The preferred way of submitting a contribution is to either make an issue on github or by forking the project on github and making a pull request.

xarray integration

Alan Richardson has written a great little tool for using xarray with segy files, which he demos in this notebook

Reproducing the test data

Small SEG-Y formatted files are included in the repository for test purposes. The data is non-sensical and made to be predictable, and it is reproducible by using segyio. The tests file are located in the test-data directory. To reproduce the data file, build segyio and run the test program,, and as such:

python examples/ small.sgy 50 1 6 20 25
python examples/ small-ps.sgy 10 1 5 1 4 1 3
python examples/ small.sgy

The small-lsb.sgy file was created by running the flip-endianness program. This program is included in the segyio source tree, but not a part of the package, and not intended for distribution and installation, only for reproducing test files.

The seismic unix file and were created by the following commands:

segyread tape=small.sgy ns=50 remap=tracr,cdp byte=189l,193l conv=1 format=1 \
suswapbytes < >

If you have have small data files with a free license, feel free to submit it to the project!



Import useful libraries:

import segyio
import numpy as np
from shutil import copyfile

Open segy file and inspect it:

filename = 'name_of_your_file.sgy'
with as segyfile:

    # Memory map file for faster reading (especially if file is big...)

    # Print binary header info

    # Read headerword inline for trace 10

    # Print inline and crossline axis

Read post-stack data cube contained in segy file:

# Read data along first xline
data = segyfile.xline[segyfile.xlines[1]]

# Read data along last iline
data = segyfile.iline[segyfile.ilines[-1]]

# Read data along 100th time slice
data = segyfile.depth_slice[100]

# Read data cube
data =

Read pre-stack data cube contained in segy file:

filename = 'name_of_your_prestack_file.sgy'
with as segyfile:

    # Print offsets

    # Read data along first iline and offset 100:  data [nxl x nt]
    data = segyfile.iline[0, 100]

    # Read data along first iline and all offsets gath:  data [noff x nxl x nt]
    data = np.asarray([np.copy(x) for x in segyfile.iline[0:1, :]])

    # Read data along first 5 ilines and all offsets gath:  data [noff nil x nxl x nt]
    data = np.asarray([np.copy(x) for x in segyfile.iline[0:5, :]])

    # Read data along first xline and all offsets gath:  data [noff x nil x nt]
    data = np.asarray([np.copy(x) for x in segyfile.xline[0:1, :]])

Read and understand fairly 'unstructured' data (e.g., data sorted in common shot gathers):

filename = 'name_of_your_prestack_file.sgy'
with, ignore_geometry=True) as segyfile:

    # Extract header word for all traces
    sourceX = segyfile.attributes(segyio.TraceField.SourceX)[:]

    # Scatter plot sources and receivers color-coded on their number
    sourceY = segyfile.attributes(segyio.TraceField.SourceY)[:]
    nsum = segyfile.attributes(segyio.TraceField.NSummedTraces)[:]
    plt.scatter(sourceX, sourceY, c=nsum, edgecolor='none')

    groupX = segyfile.attributes(segyio.TraceField.GroupX)[:]
    groupY = segyfile.attributes(segyio.TraceField.GroupY)[:]
    nstack = segyfile.attributes(segyio.TraceField.NStackedTraces)[:]
    plt.scatter(groupX, groupY, c=nstack, edgecolor='none')

Write segy file using same header of another file but multiply data by *2

input_file = 'name_of_your_input_file.sgy'
output_file = 'name_of_your_output_file.sgy'

copyfile(input_file, output_file)

with, "r+") as src:

    # multiply data by 2
    for i in src.ilines:
        src.iline[i] = 2 * src.iline[i]

Make segy file from sctrach



% Inspect segy

% Read headerword inline for each trace

%Read data along first xline
data= Segy.readCrossLine(Segy_struct,Segy_struct.crossline_indexes(1));

%Read cube

%Write segy, use same header but multiply data by *2
data = Segy.get_traces(input_file);
data1 = 2*data;
Segy.put_traces(output_file, data1);

Common issues

Creating a new file is very slow, or copying headers is slow

Quite often issues show up where someone struggle with the performance of segyio, in particular when creating new files. The culprit is often this code:

with segyio.create('new.sgy', spec) as dst:
    dst.header = headers

The code itself is perfectly ok, but it has subtle behaviour on some systems when the file is newly created: it is performing many scattered writes to a sparse file. This can be fast or slow, largely depending on the file system.

Possible solutions

Rewrite the loop to write to the file contiguously:

with segyio.create('new.sgy', spec) as dst:
    for i in range(spec.tracecount):
        dst.header[i] = headers[i]
        dst.trace[i] = traces[i]

If the file is modified copy of another file, without changing the trace lengths, it's often faster (and easier!) to first copy the file without segyio, and then use segyio to modify the copy in-place:

shutil.copyfile(srcfile, dstfile)
with as f:
    f.header = headers

ImportError: cannot open shared object file

This error shows up when the loader cannot find the core segyio library. If you've explicitly set the install prefix (with -DCMAKE_INSTALL_PREFIX) you must configure your loader to also look in this prefix, either with a ld.conf.d file or the LD_LIBRARY_PATH variable.

If you haven't set CMAKE_INSTALL_PREFIX, cmake will by default install to /usr/local, which your loader usually knows about. On Debian based systems, the library often gets installed to /usr/local/lib, which the loader may not know about. See issue #239.

Possible solutions

  • Configure the loader (sudo ldconfig often does the trick)
  • Install with a different, known prefix, e.g. -DCMAKE_INSTALL_LIBDIR=lib64

RuntimeError: unable to find sorting

This exception is raised when segyio tries to open the in strict mode, under the assumption that the file is a regular, sorted 3D volume. If the file is just a collection of traces in arbitrary order, this would fail.

Possible solutions

Check if iline and xline input parameters are correct for current file. Segyio supports files that are just a collection of traces too, but has to be told that it's ok to do so. Pass strict = False or ignore_geometry = True to to allow or force unstructured mode respectively. Please note that f.iline and similar features are now disabled and will raise errors.


Segyio was initially written and is maintained by Equinor ASA as a free, simple, easy-to-use way of interacting with seismic data that can be tailored to our needs, and as contribution to the free software community.


A library for reading and writing SEGY files in C with wrappers for Python and Matlab







No releases published


No packages published