CrsKit for Python

Coordinate reference systems, Well-Known Text and coordinate transformations, backed by the full EPSG catalogue. A thin binding over CrsKit, a C++23 implementation of the OGC Coordinate Transformation Services model, conformance-tested against the IOGP GIGS series.

pip install crskit[epsg]

Getting started

The [epsg] extra brings the EPSG catalogue with it, so init() needs no path — it finds the database on its own:

import crskit

crskit.init()

etrs89 = crskit.crs_from_epsg(4258)     # ETRS89, geographic 2D (latitude, longitude)
utm30n = crskit.crs_from_epsg(25830)    # ETRS89 / UTM zone 30N (easting, northing)

madrid = crskit.transformation(etrs89, utm30n).transform([40.416775, -3.703790])
print(madrid)                           # [440291.2843479216, 4474254.600145094]

Coordinates are given in the axis order the CRS declares — EPSG 4258 is latitude, longitude, not the other way round. Ask the system if you are unsure:

>>> etrs89.axes
[('Lat', 'North'), ('Lon', 'East')]
>>> utm30n.axes
[('E', 'East'), ('N', 'North')]

Transforming many points

transform_points() takes a NumPy array of shape (n, source_dimension) and crosses into C++ once for the whole batch, releasing the GIL while it works — the path to use for point clouds:

import numpy as np

points = np.array([[40.416775, -3.703790],      # Madrid
                   [41.385064,  2.173404],      # Barcelona
                   [37.389092, -5.984459]])     # Sevilla

transformation = crskit.transformation(etrs89, utm30n)
projected = transformation.transform_points(points)      # -> (3, 2) array

NumPy is optional: transform() works with any sequence of floats, and the module imports without it.

Coordinate reference systems

Build a CRS from an EPSG code or from Well-Known Text (WKT 1 and WKT 2 are both understood; the grammar is detected from the text):

crs = crskit.crs_from_wkt(open("layer.prj").read())

crs.name             # 'ETRS89 / UTM zone 30N'
crs.authority_code   # 25830, or None when the WKT carries no authority
crs.dimension        # 2
crs.remarks

Emit WKT in any of the flavours the library supports:

crs.to_wkt()                                        # OGC WKT 1 (the default)
crs.to_wkt(crskit.WktVersion.WKT2_2019)             # ISO 19162:2019
crs.to_wkt(crskit.WktVersion.WKT1_ESRI)             # ESRI dialect, for a .prj ArcGIS will read

Two systems compare by their mathematical definition, not by their name or authority: == asks "are these the same CRS for transformation purposes?", ignoring naming and axis order.

>>> crskit.crs_from_wkt(utm30n.to_wkt()) == utm30n
True
>>> etrs89.compare(utm30n)      # graded similarity, 0..100
0

Choosing between coordinate operations

Between two datums there is rarely a single transformation. EPSG defines twelve from ED50 to ETRS89, differing in accuracy, in the area they are valid in, and in whether they need a grid file. Choosing for you would mean silently deciding how wrong your coordinates are, so an ambiguous pair raises TransformationNotFoundError unless you say which one to use:

def for_mainland_spain(source_name, target_name, operations):
    for operation in operations:
        print(f"EPSG:{operation.code}  {operation.accuracy} m  {operation.area_of_use}")
    spain = [o for o in operations if "Spain - mainland" in o.area_of_use and not o.grid_files]
    return min(spain, key=lambda o: o.accuracy)

transformation = crskit.transformation(ed50, etrs89, select_operation=for_mainland_spain)

EPSG:15932   0.2 m  Spain - mainland and Balearic Islands onshore   needs SPED2ETV2.gsb
EPSG:1588    1.0 m  Norway - offshore north of 65°N; Svalbard
EPSG:1628    1.0 m  Gibraltar
EPSG:1632    1.5 m  Spain - mainland except northwest
EPSG:1650    2.0 m  France
EPSG:5040    5.0 m  Portugal - mainland - onshore
...

Each candidate tells you its code, accuracy in metres, area_of_use, grid_files and information_source — enough to choose by geography and not only by the smallest number.

Heights above sea level

A GNSS receiver gives an ellipsoidal height, which is not the altitude anyone means. Turning it into an orthometric height takes a geoid model. Pair a horizontal system with a vertical one and transform in 3D:

etrs89_3d = crskit.crs_from_epsg(4937)              # latitude, longitude, ellipsoidal height
utm_over_sea_level = crskit.compound_crs(25830, 5782)   # UTM 30N + Alicante height

crskit.transformation(etrs89_3d, utm_over_sea_level).transform([40.416775, -3.703790, 700.0])
# [440291.284, 4474254.600, 648.888]   <- 51.112 m of geoid undulation over Madrid

Or transform straight to a vertical CRS, and get the point back with only its height changed — EGM2008 (EPSG 3855) is the global model, so this works anywhere:

to_egm2008 = crskit.transformation(crskit.crs_from_epsg(4979), crskit.crs_from_epsg(3855),
                                   select_operation=lambda source, target, ops: ops[0])

to_egm2008.transform([27.988056, 86.925278, 8820.43])   # Everest, ellipsoidal height
# [27.988056, 86.925278, 8848.86]   <- the orthometric height everyone knows

When a transformation needs a grid

Datum shifts and orthometric heights often need a grid file (a geoid model, an NTv2 grid) that cannot be redistributed with the software. When one is missing, the error tells you exactly which file to obtain and for which EPSG operation:

try:
    crskit.transformation(source, target)
except crskit.GridFileNotFoundError as error:
    print(error.grid_file)           # 'egm08_25.gtx'
    print(error.searched_path)       # where it was looked for
    print(error.operation_code)      # the EPSG coordinate operation that needs it
    print(error.information_source)  # who publishes it

Put the grids in the data directory (init(..., data_directory=...), which defaults to the folder holding the EPSG database) and the transformation builds.

Where to get the grids

Grid files belong to the agencies that publish them, so they are not shipped with CrsKit. The two you are most likely to need:

EGM2008, the global geoid, published by the NGA — save the file under the name EPSG gives it, which is the name in the link:

• Und_min2.5x2.5_egm2008_isw=82_WGS84_TideFree (2.5′, 149 MB) — enough for most work.
• Und_min1x1_egm2008_isw=82_WGS84_TideFree (1′, 933 MB) — the finer grid.

These are mirrors of the NGA files, in the little-endian layout the reader expects; the originals are at earth-info.nga.mil.

Spain — the EGM08-REDNAP geoid, which is the model EPSG uses for heights over Alicante datum:

• EGM08_REDNAP.txt (4 MB) — peninsula and Balearics.
• EGM08_REDNAP_Canarias.txt (0.4 MB) — the Canary Islands.

The ED50↔ETRS89 NTv2 grids (PENR2009.gsb, BALR2009.gsb) come from the CNIG download centre. All of these are © Instituto Geográfico Nacional de España, mirrored under its CC BY 4.0-compatible licence; the originals live at datos-geodesia.ign.es.

data/grid-sources.json in the repository maps a grid name to where it can be obtained, so an application can offer to fetch it when GridFileNotFoundError is raised.

Every error the library raises derives from crskit.CrsError, so one except catches them all; AuthorityCodeNotFoundError, WktParseError, TransformationNotFoundError, DimensionMismatchError, CoordinateOutsideDomainError and UnsupportedFormatError let you discriminate when you care.

The EPSG database

The catalogue lives in its own package, crskit-epsg, which the [epsg] extra installs. It is versioned after the dataset, not after the library, so a newer EPSG release is pip install -U crskit-epsg — no new version of CrsKit needed.

init() looks for the database in this order: the path you pass, the DIGI21_EPSG_SQLITE environment variable, and then the crskit-epsg package. Any SQLite built from the official EPSG SQL scripts works — for instance one you build yourself with tools/epsg-sqlite for a version newer than the packaged one:

crskit.init("/path/to/epsg.sqlite", data_directory="/path/to/grids")

The data are the EPSG Geodetic Parameter Dataset (https://epsg.org), owned by IOGP and used under its Terms of Use, which ship inside crskit-epsg. IOGP provides them "as is" and does not endorse CrsKit.

Inside an application that embeds Python

The module links the CrsKit shared library rather than compiling it in. So when a host application that already uses CrsKit runs a script through an embedded interpreter, the loader binds the module to the library the application has already loaded, and both share its state: the EPSG catalogue and settings are the ones the application set up, and there is nothing left to initialise.

if not crskit.is_initialized():
    crskit.init()

Learning it

• TUTORIAL.md — the long way in, from "what is a CRS" to geoids and point clouds. Written for someone who knows Python and no geodesy; every output in it was produced by running the code. Also in español, français, italiano and deutsch.
• examples/ — runnable scripts: a quick start, WKT round trips, a million points with NumPy, choosing among candidate operations, and orthometric heights. The test suite runs them, so they cannot rot.

Typing

The package ships type stubs and a py.typed marker (PEP 561), so editors complete the API and mypy checks it, even though the module is compiled C++.

License

Apache-2.0. The EPSG dataset is IOGP's, used under its own terms; this product is not endorsed by IOGP.