fastpynuts

FastPyNUTS

A fast implementation of querying the NUTS - Nomenclature of territorial units for statistics dataset by location, particularly useful for large-scale applications.

Figure: Eurostat

Features

• fast querying of NUTS regions (~0.3ms/query)
• find all NUTS regions of a point or query user-defined NUTS-levels (0-3)
• use your own custom NUTS dataset (other CRS, enriched metadata, etc.)

Installation

pip install fastpynuts

FastPyNUTS requires numpy, shapely, treelib and rtree

Usage

Initialization and finding NUTS regions

The NUTSfinder class is the main tool to determine the NUTS regions of a point. It can be initialized from a local file containing the NUTS regions, or via automatic download from Eurostat.

from fastpynuts import NUTSfinder

# construct from local file
nf = NUTSfinder("PATH_TO_LOCAL_FILE.geojson")

# retrieve data automatically (file will be downloaded to or if already existing read from '.data')
nf = NUTSfinder.from_web(scale=1, year=2021, epsg=4326)


# find NUTS regions
point = (11.57, 48.13)
regions = nf.find(*point)                   # find all regions via a point

bbox = (11.57, 48.13, 11.62, 49.)           # lon_min, lat_min, lon_max, lat_max
regions = nf.find_bbox()                    # find all regions via a bbox

geom = {
    "type": "Polygon",
    "coordinates": [
        [
            [11.595733032762524, 48.11837184946995],
            [11.631858436052113, 48.14289890153063],
            [11.627498473585405, 48.16409081247133],
            [11.595733032762524, 48.11837184946995]
        ]
    ]
}
regions = nf.find_bbox()                    # find all regions via a GeoJSON geometry (supports shapely geometries and all objects that can be converted into one)


# filter for regions of specific levels
level3 = nf.filter_levels(regions, 3)
level2or3 = nf.filter_levels(regions, 2, 3)

Assessing the results

The NUTS regions will be returned as an ordered list of NUTSregion objects.

>>> regions
[NUTS0: DE, NUTS1: DE2, NUTS2: DE21, NUTS3: DE212]

Each region object holds information about

• its ID and NUTS level

>>> region = regions[0]
>>> region.id
DE
>>> region.level
0

• its geometry (a shapely Polygon or MultiPolygon) and the corresponding bounding box

>>> region.geom
<MULTIPOLYGON (((10.454 47.556, 10.44 47.525, 10.441 47.514, 10.432 47.504, ...>
>>> region.bbox
(5.867697, 47.270114, 15.04116, 55.058165)

• further fields from the NUTS dataset and the original input feature in GeoJSON format

>>> region.properties
{
    "NUTS_ID": "DE",
    "LEVL_CODE": 0,
    "CNTR_CODE": "DE",
    "NAME_LATN": "Deutschland",
    "NUTS_NAME": "Deutschland",
    "MOUNT_TYPE": 0,
    "URBN_TYPE": 0,
    "COAST_TYPE": 0,
    "FID": "DE"
}
>>> region.feature
{
    'type': 'Feature',
    'geometry': {
        'type': 'MultiPolygon',
        'coordinates': [
            [
                [
                    [10.454439, 47.555797],
                    ...
                ]
            ]
        ],
    },
    'properties': {
        "NUTS_ID": "DE",
        ...
}

Advanced Usage

# apply a buffer to the input regions to catch points on the boundary (for further info on the buffering, see the documentation)
nf = NUTSfinder("PATH_TO_LOCAL_FILE.geojson", buffer_geoms=1e-5)

# only load certain levels of regions (here levels 2 and 3)
nf = NUTSfinder("PATH_TO_LOCAL_FILE.geojson", min_level=2, max_level=3)


# if the point to be queried is guaranteed to lie within a NUTS region, setting valid_point to True may speed up the runtime
regions = nf.find(*point, valid_point=True)

Runtime Comparison

FastPyNUTS is optimized for query speed and result correctness, at the expense of more expensive initialization time.

A R-tree-based approach proved to be the fastest option:

Compared to other packages like nuts-finder, a large performance boost can be achieved

Tips:

• if interested only in certain levels (0-3) of the NUTS dataset, initialize the NUTSfinder using its min_level and max_level arguments
• if it's known beforehand that the queried point lies within the interior of a NUTS region, use find(valid_point=True)

For a full runtime analysis, see benchmark.ipynb

Contributors

• Colin Moldenhauer
• meengel