Geospatial data sources and API contracts

Reviewed 2026-10-01. Public STAC metadata searches were exercised; authenticated pixel downloads, commercial APIs, CDS jobs and Earth Engine exports were not.

Satellite, elevation and thematic catalogs

Do not use the retired SciHub DHuS endpoint or SentinelAPI(...scihub...) for new workflows. CDSE is a different service; STAC metadata discovery does not imply unauthenticated access to every linked product. Follow the selected asset's CDSE S3/OData access instructions rather than inventing a download URL.

STAC search

Each supports Item Search under /search and returns GeoJSON FeatureCollection with features and pagination links. Follow rel=next including its method/body; do not construct page numbers. GeoJSON bbox is west,south,east,north in degrees. Collection IDs and asset keys are provider-specific even for the same mission.

from pystac_client import Client

catalog = Client.open('https://earth-search.aws.element84.com/v1')
search = catalog.search(collections=['sentinel-2-l2a'],
                        bbox=[-122.5, 37.7, -122.3, 37.9],
                        datetime='2023-06-01/2023-06-30',
                        query={'eo:cloud_cover': {'lt': 20}},
                        limit=2, max_items=5)
items = list(search.items())
if not items:
    raise ValueError('No matching items')
asset_metadata = {k: a.to_dict() for k, a in items[0].assets.items()}

items() is an iterator that fetches further pages; item_collection() materializes the bounded result. limit is per-page, max_items is the client total. Scene cloud percentage is only a coarse filter. Record item IDs, acquisition times, asset checksums where available, scale/offset and the QA mask rule.

For Planetary Computer pass modifier=planetary_computer.sign_inplace to Client.open. ItemSearch itself is not iterable; use search.items(). The SDK's SAS token service is GET /api/sas/v1/token/{account}/{container}; it returns token and msft:expiry. SDK signing appends the expiring SAS query to Azure asset URLs. Do not log signed URLs or treat them as permanent identifiers. Public metadata access was tested without a subscription key; access policies and limits can differ for protected assets.

Sources: CDSE STAC, Earth Search, PySTAC Client, Planetary Computer SDK.

ERA5 through CDS

Configure CDS using its current personal-access-token setup and accept the chosen dataset terms. The URL is https://cds.climate.copernicus.eu/api; the old UID:key credential format and /api/v2 configuration are not the current recipe. Use the dataset download form's generated request for the exact variables/product. This illustrative request creates a remote retrieval job:

import cdsapi
client = cdsapi.Client()
client.retrieve('reanalysis-era5-single-levels', {
    'product_type': ['reanalysis'], 'variable': ['2m_temperature'],
    'year': ['2023'], 'month': ['01'], 'day': ['01'], 'time': ['12:00'],
    'area': [37.9, -122.5, 37.7, -122.3],  # north, west, south, east
    'data_format': 'netcdf', 'download_format': 'unarchived',
}, 'era5_temp.nc')

retrieve handles asynchronous processing and download, not pagination. Inspect actual NetCDF dimensions (valid_time versus time) and units before aggregation. CDS setup recommends current cdsapi; credentialed retrieval was not run.

OpenStreetMap, Overpass and Nominatim

OSMnx 2.1.1 uses ox.features_from_place(place, tags={'building': True}), ox.graph_from_place(place, network_type='drive') and ox.geocode_to_gdf(place). These call external services. Cache permitted results, bound areas and obey service policies. The public Nominatim service requires a meaningful identifying user-agent and at most one request per second; it is not a bulk or autocomplete endpoint.

import requests
query = '[out:json][timeout:25];way["highway"](37.700,-122.401,37.701,-122.400);out geom;'
response = requests.post('https://overpass-api.de/api/interpreter',
                         data={'data': query}, timeout=40)
response.raise_for_status()
data = response.json()
if data.get('remark'):
    raise RuntimeError(data['remark'])  # A timeout/error can accompany HTTP 200.
elements = data['elements']

Overpass bbox is south,west,north,east. It has no generic page parameter; subdivide large AOIs and reconcile OSM type/ID duplicates. This query is an illustrative public request, not executed in this refresh. See OSMnx, Overpass, Nominatim policy.

Geocoding and current weather

These are credentialed illustrative request contracts, not live test results. Read tokens from an environment variable or credential store; never embed them in a notebook, output URL or exception log. Use request timeouts and HTTP checks, then validate the provider's response status before selecting a location.

Sources: Google request/response, Mapbox v6, OpenWeather.

Raster quality and elevation downloads

The elevation package wraps SRTM acquisition and native GDAL/Make tooling. elevation.clip(bounds=..., output=...) downloads/crops; elevation.clean() cleans temporary cache products, not raster voids. Void filling needs a separately chosen method and QA. Never call clean(input, output) expecting a repaired DEM.

import numpy as np
import rasterio

def assess_data_quality(path):
    with rasterio.open(path) as src:
        data = src.read(masked=True).astype('float64')
        valid = data.compressed()
        valid = valid[np.isfinite(valid)]
        return {'valid_fraction': valid.size / data.size,
                'range': (float(valid.min()), float(valid.max())) if valid.size else None,
                'crs': str(src.crs), 'resolution': src.res}

This checks numeric support, not acquisition quality or product correctness. Elevation source documents its native dependencies.