Parking Lot Dataset: GIS Data for Spatial Analysis and Site Selection

What Is a Parking Lot Dataset?

A parking lot dataset is built from geospatial polygon data: a series of coordinates tracing the outer boundary of a lot, delivered as Well-Known Text (WKT), GeoJSON, or a shapefile depending on the provider. A polygon by itself is just a shape. What makes it useful as parking lot GIS data is the metadata attached to it: a unique identifier for the lot, its calculated area, and a link, often a placekey or similar ID, back to the point of interest it serves.

This is a different dataset from two categories people often search for interchangeably. Building footprint data maps the outline of a structure itself, not the surface around it; SafeGraph’s Geometry product and its companion post on building footprint data cover that distinction in more depth. Parking occupancy datasets, meanwhile, are computer vision datasets built from camera images of individual spaces, useful for training models that detect whether a space is full or empty right now, but not GIS polygons you can drop into a map. Parking lot polygon data sits between the two: GIS ready like a building footprint, but describing pavement instead of structure.

How Parking Lot Polygons Are Created

SafeGraph builds parking lot polygons in two stages: AI detection from satellite imagery, followed by manual cleanup of the edge cases the AI gets wrong on its own. The initial detection step trains a model to recognize parking lots across satellite imagery covering the entire U.S. The results vary widely, so cleanup matters as much as detection.

Fragmented Lots: Merging Pieces AI Splits Apart

AI models are generally good at recognizing that a stretch of pavement is a parking lot, but not always good at matching where its edge actually ends in reality. A single physical lot can get split into several disconnected polygons that should have been merged into one shape.

To fix this, SafeGraph filters using a minimum area threshold of roughly 100 square meters, about the footprint of six standard parking spaces. Lots below that threshold are typically fragments of a larger lot rather than standalone parking areas, so they get merged instead of counted separately. The same logic runs in reverse for oversized features: some large parking lot polygons turned out to be several smaller, separate lots joined together incorrectly, and those get split back apart.

Donut Polygons: Removing False Holes

The AI sometimes draws parking lot polygons with holes in them, known as donut polygons, when it mistakes a parked car, a tree’s shadow, or another object on the lot surface for a gap in the pavement. SafeGraph calculates a shape metric for each polygon that compares its area to its perimeter; a low ratio can signal a bad shape, though not always, so it functions as a flag rather than an automatic rejection.

(Left) AI incorrectly marking small holes outlined in yellow. (Right) Result after applying our threshold value.

Because the AI tends to misidentify small holes more often than large ones, SafeGraph also measures the size of each hole inside a polygon and filters out the ones below a set threshold, while keeping holes that are spatially real, for example an apartment building sitting in the middle of its own surrounding parking lot.

Simplifying Jagged Geometries

Drawing a clean straight edge is another common AI failure point. A straight line only needs two points, but AI-generated polygons often include extra, unnecessary coordinates that create jagged, stair-step edges instead of clean lines. SafeGraph applies geometry simplification to smooth these edges, producing polygons that are easier to visualize, store, and process, and that take up less space for customers working with the data at scale.

(Left) Jagged lines created by AI without simplification. (Right) Cleaner lines after applying simplification.

Where to Get Parking Lot Data: Sources Compared

Four types of sources cover most parking lot data needs, and each trade off coverage, format, and cost differently.

City open-data portals are the cheapest way to get parking lot data for a single city, but formats, update frequency, and even what counts as a “parking lot” vary from one planning department to the next, so they rarely scale past a handful of markets. OpenStreetMap is free and global in principle, but coverage depends entirely on whether volunteers in a given area have mapped it, so dense parts of a metro can sit next to unmapped gaps a few miles away. Research datasets like PKLot are built for computer vision: labeled camera images of individual spaces for training occupancy models, not polygon geometry, so they answer a different question than where a lot’s boundaries actually sit. Commercial, nationally consistent data such as SafeGraph Parking Lots is built for teams that need the same schema and refresh cadence across every market at once.

Use Cases

Parking Lot Data for Site Selection

Parking-to-building ratio is one of the simplest screening metrics in site selection: a location with too little parking relative to a building’s square footage can cap foot traffic, regardless of how strong the location looks on paper otherwise. Comparing a prospective site’s parking area to its building footprint, and to nearby competitors, lets analysts flag locations for deeper review before spending time on a physical site visit. Pairing parking lot polygons with building footprint data lets teams calculate that ratio automatically across thousands of candidate sites at once.

Parking Lot Data for Urban Planning

City planners use parking lot coverage to gauge whether an area has enough surface parking to support the points of interest nearby, or whether it is over-supplied with pavement relative to demand, a question that comes up often as cities revisit minimum parking requirements. Aggregating parking lot polygons by neighborhood or district gives planners a factual baseline for that conversation instead of a guess.

Parking Lot Data for Mapping and Accessibility

Parking polygons extend a basic map beyond building outlines and street lines by showing how someone actually approaches a POI on foot after parking. That detail matters for accessibility planning, wayfinding apps, and any product that needs to represent more than a single point on a map.

Parking Lot Data for Risk Assessment and Visit Attribution

Parking lots are impervious surfaces, and impervious surface coverage is directly tied to stormwater runoff. The EPA identifies parking lots, alongside roads and rooftops, as one of the most common impervious surface types, and notes that streams show consistent, measurable changes in channel stability once effective impervious cover in a watershed passes roughly 10 percent. Mapping parking lot density at scale helps risk and insurance teams estimate flood exposure for a property or portfolio before it becomes a claim.

Parking lot polygons also improve visit attribution. Combining parking lot boundaries with mobile location or connected vehicle data helps distinguish a customer parked at a specific store from a device merely passing nearby, which matters most in shopping centers where several POIs share one lot. SafeGraph’s guide on connected vehicle data and parking lot polygons walks through this in more detail.

What to Look for in Parking Lot Data

What separates the best parking lot GIS datasets from a liability comes down to five criteria.

• Accuracy: whether polygons are drawn tightly to the actual pavement or rely on rough approximations that overstate or understate lot size.
• Refresh frequency: parking lots change less often than store openings and closings, but new construction and lot reconfigurations still happen, so a static, years-old file loses value quickly.
• POI linkage: a polygon without a reliable link to the place it serves is far less useful, since most parking lot use cases start with a specific business or property, not an unlabeled shape on a map.
• Coverage: national, consistent coverage matters more than a handful of well-mapped test cities once a use case needs to scale.
• Delivery format: compatibility with existing GIS or data warehouse tooling (WKT, GeoJSON, shapefile) determines how much engineering work it takes to actually put the data to use.

Closing Thoughts

Parking lot data fills a gap that neither building footprints nor point-based POI data can cover on its own: how much paved space surrounds a place, and whether that space works for the people trying to use it. Whether you’re screening sites, modeling flood risk, or refining visit attribution, the value comes down to clean geometry and a reliable link back to the POI it serves. The comparison above should make it easier to pick a source that matches your coverage and accuracy needs, free or commercial.