What people actually look for in a park

This atlas started as an argument about scores. Every park gets read through a Jane Jacobs lens and an ecological-comfort lens, and the interesting work was in the scoring. But the traffic told a different story. People arriving on a park page were not there to debate a vitality index. They were scanning for something concrete: is there a washroom, a splash pad, an off-leash area, a rink? Is this a place I can take the kids on Saturday? The score was answering a question most visitors had not asked.

So we changed what a park page leads with. The “What’s here”panel now sits at the top of every park, above the scoring narrative, listing the park’s recreation and facilities in plain language. It is a small change with an awkward consequence: the moment you promise to tell people what is in a park, the gaps in your data become the product. Two of them turned out to be serious.

The amenity vocabulary was too narrow

The City of Toronto publishes a Parks and Recreation Facilities inventory, and it is good, but it speaks a narrow vocabulary of about eleven facility types and misses features that were never catalogued. A park with a well-loved bench-lined promenade, a public artwork, and a viewpoint could come back nearly empty.

We widened the answer with OpenStreetMap. Every OpenStreetMap feature whose location falls inside a park is normalized into a shared vocabulary of 32 amenity types and merged with the City inventory, with each amenity tagged by where it came from: City, OpenStreetMap, or both. The two sources are combined, not one trusted over the other. The result: 1,818 parks now list at least one amenity, and 1,780 of them gained an amenity the City record did not have. The heavily-used destinations gain the most. Paul Coffey Park, Downsview Park, Centre Island, Nathan Phillips Square, and Christie Pits each pick up ten or more amenity types that were previously invisible on their page.

Then the map itself was wrong

Widening the vocabulary exposed a deeper problem. Amenities are attached to a park by asking which features fall inside its boundary, and the boundaries came from a single City dataset frozen in April 2022. Parks that opened later were missing entirely, and some older outlines were simply out of date. Biidaasige Park, the 31-hectare park in the Port Lands, did not exist in our data at all, so none of its amenities could attach to anything.

We reconciled every City boundary against OpenStreetMap. Where OpenStreetMap has a boundary that clearly matches and improves on the 2022 outline, we adopt it; weaker or ambiguous cases are set aside for manual review rather than changed blindly; and parks the City file was missing are added. The catalogue grew from 3,273 parks to 3,788. Of those, 884 boundaries were replaced with a better OpenStreetMap outline and 515 are parks we simply did not have before, Biidaasige among them. Because every score is computed on the boundary, scores moved for the parks whose shape changed, and the new parks entered the rankings for the first time. Wherever a boundary comes from OpenStreetMap, the park page says so.

What it adds up to

None of this started as a data-engineering project. It started with a plain observation about why people visit these pages. Answering their actual question, what is in this park, forced the data underneath to get more honest: a broader amenity vocabulary, a corrected map, and a visible note on every page about where each piece came from. The scores are still here. They are just no longer the first thing a park has to say for itself.

The mechanics of both fixes, the matching thresholds, the review queue, and the source tagging, are written up in the methodology.