Where to Get Real Geospatial Data

A city planning intern needs a layer of every park in the county by Friday. Nobody on the team surveyed the parks. Nobody has to. The data almost certainly already exists somewhere free, and the real skill in this situation is not collecting the data, it is knowing where to look, what the license actually allows, and how to check the file is honest before a single line of analysis runs on it.

The previous lesson worked through GeoJSON in depth: its structure, its always-WGS84 coordinate order, and the traps in writing one by hand. This lesson is about where the GeoJSON, or the GeoPackage, or the imagery, actually comes from before an analyst ever opens an editor. A handful of sources cover the overwhelming majority of real GIS work, and each one has a personality worth knowing before reaching for it.

OpenStreetMap: everything, crowd-built, with strings attached

OpenStreetMap, almost always shortened to OSM, is a map of the world built entirely by volunteers editing a shared database. Roads, building footprints, shops, trails, even individual park benches get added by people mapping their own neighborhoods. Coverage is uneven: a mapper-heavy city like Berlin has extraordinary detail, while a rural stretch of road elsewhere might be a single unlabeled line. It is still, by a wide margin, the largest free vector dataset of the built and natural world that exists.

Two ways to pull data out of OSM matter in practice. The Overpass API answers a specific query, "give me every hospital in this bounding box," and returns just that. Geofabrik publishes bulk extracts, whole countries or regions packaged as .osm.pbf files, for anyone who wants everything in an area rather than one query's worth.

The license is the part that trips people up. OSM data is released under the Open Database License, or ODbL, which is not the same as public domain. ODbL requires two things from anyone who redistributes a database built from OSM, or a substantial extract of one: credit OpenStreetMap and its contributors, and if the derived database is shared publicly, share it under the same license. A web map that just displays OSM tiles with a visible attribution line is on solid ground. A company that scrapes OSM, adds its own delivery-time data, and sells access to the combined database without attribution or a compatible license is not.

Natural Earth: public domain, three honest scales

Natural Earth is a free basemap dataset built specifically for cartography, covering countries, states, rivers, cities, and physical features like coastlines and elevation shading. It comes in three fixed scales: 1:10m, the most detailed, meant for country or regional maps; 1:50m, a middle ground for continental views; and 1:110m, deliberately simplified for a map of the whole planet where fine detail would just clutter the image. Picking the wrong scale for the job is the most common mistake: a 1:10m coastline dataset zoomed out to show the whole Earth renders as an unreadable mess of tiny wiggles that were never meant to be seen at that zoom.

The license is the simplest one on this list. Natural Earth places everything in the public domain, no attribution required, no share-alike clause, free for any use including commercial products. That makes it the default choice for basemap layers in software, textbooks, and presentations where a clean, dependency-free background is the only requirement.

Overture Maps: newer, and built by consensus

Overture Maps Foundation is a more recent open data project, backed jointly by several large technology companies, that publishes a global dataset organized into themes: buildings, places (points of interest like shops and restaurants), transportation, administrative divisions, and a base layer of land cover and water. Where OSM is built one volunteer edit at a time, Overture blends several existing open and commercial sources, runs conflation and cleanup on top, and republishes the result on a regular release schedule.

Overture data is distributed as GeoParquet, a columnar file format built for analytics tools rather than the single small file a GeoJSON download usually implies, and it is released under a permissive license (CDLA-Permissive-2.0) that does not carry OSM's share-alike requirement. For a project that wants broad global building and places coverage without the licensing overhead of ODbL, Overture is worth checking before defaulting to OSM.

Government portals: authoritative, if the metadata is honest

Government agencies publish an enormous amount of geospatial data directly: data.gov aggregates United States federal datasets, and most cities and counties run their own open data portals for things like parcel boundaries, zoning, transit routes, and permit records. This is often the most authoritative source available, since a county assessor's own parcel layer is the actual legal record, not someone's best reconstruction of it.

The catch is that quality and licensing vary wildly by agency. Some portals publish clean, current GeoPackage or GeoJSON files with a clear public domain notice. Others still bury a Shapefile from several years ago behind three clicks, with no license statement, no metadata about how recently it was updated, and no email response when a question comes in. An analyst treats the label "official government data" as a starting point for trust, not a substitute for checking it.

Satellite imagery: free, if the account setup is patient

Two satellite programs supply the overwhelming majority of free public earth observation imagery. Sentinel, run by the European Space Agency's Copernicus program, includes Sentinel-2, which captures the whole planet's land surface in visible and infrared bands roughly every five days at 10 to 20 meter resolution. Landsat, run by the United States Geological Survey and NASA, has the longer history, with a continuous archive stretching back to 1972 at roughly 30 meter resolution, making it the standard choice for anything that compares change over decades.

Both archives are free and open, no fee for any scene, but both require creating an account before downloading: the Copernicus Data Space Ecosystem for Sentinel data, and USGS EarthExplorer for Landsat. Neither charges money. The friction is entirely in the sign-up form, not the license, and it is worth doing once rather than hunting for a mirrored copy of unclear provenance.

Checking a download before trusting it

A file finishing its download is not the same as a file being ready to use. Before any join, buffer, or overlay runs against new data, an analyst opens it and looks. A command-line tool called ogrinfo, part of the GDAL library, prints a dataset's structure without loading it into any GUI:

ogrinfo -so county_parks.gpkg parks

That short printout answers the questions that matter before a single line of analysis code runs. The geometry type confirms these really are polygons, not points mislabeled as parks. The feature count catches an obviously broken download, 214 parks instead of the 4 that would mean the file only half-downloaded. The CRS line, EPSG:4326 here, confirms the coordinates are plain WGS84 longitude and latitude, not something projected in meters that would need a different treatment before a distance calculation. The same check works from Python, which is often more convenient when the file is about to be used in a script anyway:

import geopandas as gpd

parks = gpd.read_file("county_parks.gpkg")
print(parks.crs)
print(parks.columns.tolist())

Two lines of output, and an analyst now knows exactly what they are working with instead of guessing. Skipping this step is how a project ends up debugging a join that fails for no visible reason, when the real problem was a CRS mismatch that a ten-second check would have caught before any real work began.

Choosing between an API and a bulk download

The last decision is how to actually pull the data down, and it is not always obvious. An API, like Overpass for OSM or a city's REST endpoint for its open data portal, answers one specific question against live or near-live data: every restaurant in this zip code, every permit filed this month. It is the right tool when the area of interest is small, the data changes often enough that freshness matters, or storing the whole dataset locally would be wasteful. The tradeoff is rate limits: most free APIs cap how many requests can run per minute, and a script that ignores this gets throttled or blocked.

A bulk download, a full country extract from Geofabrik or a full imagery archive from EarthExplorer, is the right call when the work needs a large area, needs to run repeatedly offline, or needs a dataset frozen at one point in time so results stay reproducible. The tradeoff flips: storage and update management become the analyst's problem instead of the API provider's. A city-scale analysis run once a week is usually better served by a small API call. A statewide land cover study run across dozens of test iterations is usually better served by one bulk download sitting on disk.

The next lesson turns from where vector data comes from to the grid side of the same question: raster formats, and how a satellite scene like the Sentinel and Landsat imagery mentioned here actually gets stored, tiled, and read without loading an entire continent into memory at once.