Supported Projections

vibeProj supports 24 coordinate projections. Each has both a NumPy/CuPy element-wise implementation and a fused NVRTC GPU kernel.

Projection table

Projection	Internal Name	EPSG Examples	Notes
Transverse Mercator / UTM	tmerc	32601–32760, 27700	6th-order Krueger series
Web Mercator	webmerc	3857	Spherical Mercator
Mercator (ellipsoidal)	merc	3395	Variant A/B
Lambert Conformal Conic	lcc	2154	1SP and 2SP
Albers Equal Area	aea	5070	Conic equal-area
Polar Stereographic	stere	3031, 3413	Variants A/B/C
Lambert Azimuthal Equal Area	laea	3035	Oblique/equatorial/polar
Oblique Stereographic	sterea	28992	Double projection via conformal sphere
Plate Carree	eqc	4087	Equidistant cylindrical
Sinusoidal	sinu	–	Pseudocylindrical equal-area
Equal Earth	eqearth	8857	Polynomial pseudocylindrical
Cylindrical Equal Area	cea	6933	EASE-Grid 2.0
Orthographic	ortho	–	Globe view
Gnomonic	gnom	–	Great circle navigation
Mollweide	moll	–	Equal-area world maps
Robinson	robin	–	Compromise world maps
Winkel Tripel	wintri	–	National Geographic standard
Natural Earth	natearth	–	Polynomial pseudocylindrical
Azimuthal Equidistant	aeqd	–	Distance-from-center
Geostationary Satellite	geos	–	Weather satellite view
Oblique Mercator (Hotine)	omerc	3375	Variants A/B
Krovak	krovak	5514	North-orientated variant
Eckert IV	eck4	–	Pseudocylindrical equal-area
Eckert VI	eck6	–	Pseudocylindrical equal-area

Using projections via EPSG codes

Most projections are resolved automatically from EPSG codes via pyproj:

from vibeproj import Transformer

# UTM Zone 31N
t = Transformer.from_crs("EPSG:4326", "EPSG:32631")

# Netherlands national grid (oblique stereographic)
t = Transformer.from_crs("EPSG:4326", "EPSG:28992")

# LAEA Europe
t = Transformer.from_crs("EPSG:4326", "EPSG:3035")

Using projections without EPSG codes

Some projections (orthographic, gnomonic, etc.) don’t have standard EPSG codes. Use them via the pipeline API directly:

from vibeproj.crs import ProjectionParams
from vibeproj.ellipsoid import WGS84
from vibeproj.pipeline import TransformPipeline

# Orthographic centered on Paris
params = ProjectionParams(
    projection_name="ortho",
    ellipsoid=WGS84,
    lon_0=2.35,
    lat_0=48.86,
    north_first=False,
)
src = ProjectionParams(projection_name="longlat", ellipsoid=WGS84, north_first=True)

pipe = TransformPipeline(src, params)
x, y = pipe.transform(lat_array, lon_array, np)  # or cp for GPU

Datum shifting

When the source and destination CRS use different geodetic datums (e.g., WGS84 vs OSGB36), vibeProj applies a Helmert 7-parameter transformation automatically. This covers ~80% of real-world cross-datum transforms with ~1–5 metre accuracy.

# Cross-datum: WGS84 -> British National Grid (OSGB36 / Airy 1830)
t = Transformer.from_crs("EPSG:4326", "EPSG:27700")
x, y = t.transform(-0.1278, 51.5074)
print(t.accuracy)  # "sub-meter"

Helmert parameters are extracted from pyproj’s EPSG database at construction time; the actual datum shift math runs on vibeProj’s own GPU kernels (or NumPy on CPU). Same-datum transforms have zero overhead – the datum shift code path is bypassed entirely when no Helmert is needed.

15-parameter time-dependent Helmert is also supported for sub-decimeter accuracy on modern datum pairs (e.g. ITRF to ETRS89). Pass an explicit epoch or let vibeProj resolve it from the source CRS coordinate epoch:

t = Transformer.from_crs("EPSG:4326", "EPSG:27700", epoch=2024.0)
print(t.accuracy)  # "sub-decimeter" when 15-param rates are present

Not yet supported:

• NTv2 / NADCON grid-based shifts — these provide sub-centimetre accuracy for national datums (e.g. OSTN15 for Great Britain, NAD27-to-NAD83 in North America). When the best available transformation for a cross-datum pair is grid-only, vibeProj emits a RuntimeWarning and falls back to projection math without a datum shift. Results may differ from pyproj by meters to hundreds of meters in these cases. Use pyproj or rasterio directly if you need grid-based accuracy.

If no Helmert transformation is available for a cross-datum pair (grid-only datums), vibeProj warns and applies projection math without a datum shift.

Known limitations

• Equal Earth (eqearth): Uses the spherical polynomial formula on geodetic latitude rather than converting to authalic latitude first. This means absolute metre values differ from pyproj by ~15%. Roundtrip accuracy is exact (forward and inverse are self-consistent).

• Oblique Stereographic (sterea): The double-projection through a conformal sphere introduces ~130m systematic offset from pyproj’s more rigorous method in the forward direction. The inverse conformal sphere conversion has a known accuracy limitation (~0.2 degrees). Roundtrip accuracy is sub-millimetre.

• Geostationary (geos): The inverse has limited accuracy for off-nadir points due to a simplified geocentric latitude conversion.

• Winkel Tripel (wintri): The inverse uses Newton iteration and converges to ~0.005 degrees rather than machine precision.