Geopandas Knowledge Patch

GeoPandas Knowledge Patch

When to use this skill

Load this skill when working on GeoPandas code involving:

• dependency compatibility and optional integrations;
• spatial indexes, unions, precision grids, or polygonal coverages;
• geometry repair, measured coordinates, or coordinate extraction;
• Arrow, Parquet, Feather, feature, file, or PostGIS I/O;
• pandas interoperability, grouping, missing values, or geometry-column state;
• overlays, point sampling, or interactive map legends.

Use the quick reference for common decisions; open the topic reference when exact API detail matters.

Reference index

Breaking changes and deprecations

Replace removed GeoSeries methods

Do not call GeoSeries.select; the method has been removed because supported pandas versions no longer provide its counterpart.

Replace:

left.geom_almost_equals(right)

with:

left.geom_equals_exact(right, tolerance)

The deprecated geom_almost_equals method is no longer available.

Do not rely on point ordering

sample_points does not promise x-coordinate ordering. Treat the result as an unordered collection of sampled points. If downstream code needs a particular order, sort explicitly according to that application's rule.

Expect DataFrame downcasting

Deleting the final geometry column changes a GeoDataFrame into a pandas DataFrame:

del gdf["geometry"]

Code that continues to require geospatial behavior must retain or create a geometry column.

Dependency compatibility

For an installation or CI failure, first compare the runtime and core packages with the required floors. Optional-package floors describe tested support; older optional versions might run but are unsupported.

Open compatibility.md for the complete core and optional dependency matrix.

Spatial indexes

SpatialIndex.query can return boolean results in dense or sparse form:

dense_hits = frame.sindex.query(
    queries.geometry,
    output_format="dense",
)
sparse_hits = frame.sindex.query(
    queries.geometry,
    output_format="sparse",
)

Sparse output requires SciPy. Choose the format deliberately so downstream code does not assume index-pair output.

Union and dissolve choices

Use a precision grid when coordinates need snapping during a union:

merged = frame.geometry.union_all(grid_size=0.01)
dissolved = frame.dissolve("group", grid_size=0.01)

Use the disjoint-subset algorithm when that method fits the input:

merged = frame.geometry.union_all(method="disjoint_subset")
dissolved = frame.dissolve("group", method="disjoint_subset")

Both controls are available through union_all and dissolve. See geometry-operations.md for related coverage and repair operations.

Polygonal coverage workflow

Validate the coverage before simplifying it, and inspect invalid edges when validation fails:

valid = frame.geometry.is_valid_coverage()
invalid_edges = frame.geometry.invalid_coverage_edges()
simplified = frame.geometry.simplify_coverage(0.5)

These operations are exposed on both GeoSeries and GeoDataFrame.

Geometry repair and measured coordinates

Choose the repair algorithm and collapsed-component behavior explicitly when they matter:

fixed = frame.geometry.make_valid(
    method="linework",
    keep_collapsed=True,
)

For measured geometries, use m, has_m, and include_m=True:

measures = points.m
has_measures = frame.geometry.has_m
coordinates = frame.geometry.get_coordinates(include_m=True)

Arrow-backed I/O

Pass pandas-conversion controls through to_pandas_kwargs when reading Arrow data:

frame = geopandas.read_parquet(
    "data.parquet",
    to_pandas_kwargs={"use_threads": False},
)

The option is accepted by from_arrow, read_parquet, and read_feather. It also applies when non-geometry Parquet columns contain list or struct data.

Open io-and-databases.md before handling property-less features, CRS-less masks, missing string values, or PostGIS geometry-column names.

pandas interoperability

Import the accessor module to expose GeoSeries methods on pandas.Series.geo:

import geopandas.accessors

buffered = series.geo.buffer(10)

Named aggregation works for geometry columns through GroupBy.agg.

Be careful around pandas-driven state changes:

• a sole np.nan is accepted as missing geometry during construction;
• GeoSeries.value_counts() keeps the CRS on its result index;
• inserting a new row with .loc preserves geometry dtype and CRS;
• deleting the last geometry column downcasts to DataFrame;
• missing values from the pandas string dtype are handled by WKT, WKB, and PostGIS conversion paths.

See pandas-integration.md for the exact scope of each behavior.

Overlay decisions

Identity overlays support matching non-geometry column names in both inputs. Empty inputs are also handled more gracefully when keep_geom_type is used.

Before adding workarounds for either case, reproduce the issue on the installed GeoPandas version and consult overlays-sampling-and-maps.md.

Reproducible point-pattern sampling

Point-pattern methods accept one sample size per geometry and an explicit random state:

sampled = geometries.sample_points(
    size=[10, 20],
    method="cluster_poisson",
    rng=42,
)

Use rng when repeatability matters. Never infer meaning from the generated points' order.

Interactive map legends

For categorical or boolean columns, provide custom labels through legend_kwds:

result = frame.explore(
    column="category",
    legend_kwds={"labels": ["First", "Second"]},
)

The labels are honored by GeoDataFrame.explore().

Diagnostic checklist

When behavior differs from expectations:

1. Confirm the installed GeoPandas, Python, and core dependency versions.
2. Identify whether the operation crosses into an optional dependency.
3. Check whether pandas object type, geometry dtype, active geometry, and CRS changed during the preceding operation.
4. For Arrow data, inspect non-geometry column types and conversion kwargs.
5. For overlays, test empty inputs, duplicate column names, and keep_geom_type separately.
6. For sampling, fix rng and avoid assumptions about point order.
7. Follow the matching topic reference before introducing a compatibility workaround.