Geospatial File Formats
A filename extension is not a data contract. A useful geospatial dataset couples geometry or cells with CRS, schema, identifiers, missing-value rules, encoding, and provenance. Format conversion can preserve some of these and silently weaken others.
Prerequisites
Complete vector and raster data. You should be able to state which metadata is needed to interpret each layer.
Learning goals
- Compare GeoJSON, GeoPackage, GeoParquet, Shapefile, GeoTIFF, and ASCII grid trade-offs.
- Distinguish geometry encodings such as WKT and WKB from feature containers.
- Validate CRS, schema, geometry types, missing values, and counts after round trips.
- Avoid assuming that extension, driver defaults, or successful reads prove semantic equivalence.
Choose by contract and workflow
| Format | Useful for | Important boundary |
|---|---|---|
| GeoJSON (RFC 7946) | small web-facing feature collections | WGS 84 longitude-latitude; text size and numeric precision matter |
| GeoPackage | portable SQLite container with vector tables, CRS metadata, and optional indexes | layer and geometry-column rules; a file can hold several tables |
| GeoParquet | columnar analytics and larger feature tables | reader/writer specification version, geometry encoding, CRS metadata, and row-group behavior |
| Shapefile | legacy exchange | multi-file dataset, constrained field names/types, weak modern schema support |
| GeoTIFF | georeferenced raster bands | transform, CRS, band metadata, masks, tiling, compression, and overview choices |
| ASCII grid | transparent tiny teaching grids | limited metadata; CRS commonly travels in a sidecar file |
WKT and WKB encode a geometry. They do not by themselves carry a feature's attributes, stable identifier, provenance, or necessarily its CRS. A CSV column of WKT is therefore not automatically a self-describing geospatial dataset.
Our GeoJSON fixtures use a project-owned foreign metadata member to make the teaching contract visible. RFC 7946 fixes coordinate semantics to WGS 84 longitude, latitude; production writers and readers should be tested against the exact standard and driver settings they use. Do not add a deprecated GeoJSON crs member and assume every reader will honor it.
Worked example
import json
from pathlib import Path
import geopandas as gpd
import shapely
path = Path("tests/fixtures/geospatial-data-science/places.geojson")
document = json.loads(path.read_text(encoding="utf-8"))
places = gpd.GeoDataFrame.from_features(
document["features"],
crs=document["metadata"]["crs"],
)
geometry = places.geometry.iloc[0]
missing_kind = int(places["kind"].isna().sum())
wkb = shapely.to_wkb(geometry)
print(f"features={len(places)} missing_kind={missing_kind} geometry={geometry.geom_type}")
print(f"wkt={geometry.wkt} wkb_bytes={len(wkb)} crs={places.crs.to_string()}")
Expected output:
features=5 missing_kind=1 geometry=Point
wkt=POINT (13.39 52.52) wkb_bytes=21 crs=OGC:CRS84
The example constructs the frame from local content so the verifier has no driver or network dependency. In an application, geopandas.read_file and GeoDataFrame.to_file use a GDAL-backed engine; inspect its supported drivers and pass an explicit driver where inference would be ambiguous.
Round-trip verification
Before and after a format conversion, compare at least:
- feature and null counts;
- stable identifiers and attribute types;
- CRS equivalence, axis convention, and units;
- geometry type, dimensionality, validity, emptiness, and bounds;
- precision or tolerance appropriate to the source measurement;
- layer names, band descriptions, NoData/masks, and time metadata where applicable;
- source hash, tool versions, options, and rejected records.
Topological equality is often more relevant than byte equality for geometry, but it can hide coordinate-order or precision changes. Conversely, byte equality is too strict across valid encodings. Choose checks from the intended semantics rather than from convenience.
Limitations and common mistakes
- Renaming a file does not convert its format.
- Driver defaults can change CRS representation, field types, precision, index columns, or polygon orientation.
- Missing CRS metadata is not permission to guess from plausible coordinate ranges.
- A successful read can still coerce unsupported values or omit auxiliary metadata.
- Cloud-optimized layouts and partitioning improve access patterns only when writers, object storage, and readers share the contract.
Practice
Design a round-trip checklist for moving a multi-layer GeoPackage to an analytics format and a web format. Identify which output cannot preserve multiple layers in one file and how you will retain provenance.
Next steps
Next, validate geometry validity and topology before measurement, overlay, or repair.
Primary sources
Checked 2026-10-18: IETF, RFC 7946 GeoJSON; GDAL, GeoJSON driver and GeoPackage driver; OGC GeoParquet, format specification; GeoPandas, reading and writing files.