A proposal (v2) to reduce this tag (initially, in the USA)
To see its geographic distribution, click taginfo on the Page, then click the "Map" tab. This tag is mostly found in North America, with light usage in South America, limited usage in Central and Southern Europe and very limited usage in Asia.
This tag is a vestige of older tagging in an era when "park" tagging was less-well defined. While its entry ostensibly clarified attributes of parks, today, its existence prolongs semantic blurring. A remedy seems to better tag as we reduce usage of this tag, eventually leading to its deprecation and elimination.
The Ideas section table is an ongoing conversation, it built up well but seems to have stalled towards completion. It identifies existing park:type=* tags and proposed new tagging, allowing park:type=* to be removed entirely. These are wet paint, chalk lines of present understanding, especially as the table grows and even moreso as we might recognize patterns in the data that could suggest how things be better structured. One early indication from this effort is that it is likely an ownership=* tag can be gained, often with value "municipal" when park:type=city_park, so don't delete the tag without wringing it of yielding its ownership value, should one be discernable from tags.
This proposal (v2) presents with two "initially seem easy" approaches: the first finds park:type=city_park and does some analysis. The second approach looks at boundary=national_park state by state and does some analysis. Further details can be discussed here, using the sketch of the older (v1) proposal (see Archive 1) to fill in any missing details. C'mon, step right up; the city_park task looks not-that-difficult. An OT query (with a California geocodeArea directive) shows isolation of the tag to north coast and north central coast areas. A simple rule or two of "get/set the ownership=municipal, operator=*, protection_title=*... keys correct for 2019 like this..." and we've accomplished something. (I find it interesting that the OT query data will likely experience a "tag migration" with a result of at least QL of way["ownership"="municipal"](area.searchArea); where that QL today yields two results in the San Fernando Valley, one of which has authorship of Adamant1; I make no judgements here about that, good or bad). We can systematically "smart batch" these together with good dialog and relatively small effort (reducing batch by batch), eventually deprecating this tag. (Because of solution-space size, this benefits well by breaking up into geocodeArea chunks in my opinion, it makes the data bite-sized, we might as well start with California). Stevea (talk) 20:20, 24 June 2019 (UTC)
An early "result" appears to be the realization that "tag migration" to ownership=* (in part) seems like it will be a result of this semantic-to-newer-syntax transition from 2009 to 2019 (and beyond, to eventual deprecation, should we get there). Stevea (talk) 20:53, 24 June 2019 (UTC)
If I were to "structure out loud" steps I might sketch to pick the city_park "low hanging fruit #1:"
1) Run this Overpass QL.
2) Click the Export button, then (from the Export dialog) download/copy as raw OSM data or data exported to JOSM.
3) Use JOSM to open the result, likely called export.osm in your Downloads folder. Or, if JOSM was already running, you might simply need to accept the imported data on a dialog or two.
4) In JOSM, perform a Find and in the dialog, search for leisure:park to "replace results" (default radio button). The 100 to 200 results can shrink with this find, you can flip to finding an exclusion set with Find dialog's Remove from selection radio button.
5) Results show a deep tagging structure, with inheritances from TIGER, SCCGIS, CPAD and more including many that are original data from Contributors (non-imported data). That means each should have its tagging examined individually.
6) When "more modern" tagging (say, including ownership=municipal) is deemed correct for each object (and data are otherwise "more fully correct"), delete the park:type=city_park tag. Repeat for all 100-and-something objects. I suggest starting near the Oregon border and working southward. Upload.
That's a somewhat tedious, human-fully-in-the-loop, one-by-one decision process that shouldn't prove too stunningly difficult, though perhaps somewhat time-consuming. (Careful deprecation often is). Of course, deciding whether to include (back in with the upload) the leisure=park tag shouldn't be too hard either, right? (These are ALL parks, right?) It seems most intermediate-skilled (or better) OSM Contributors comfortable with a repeat-loop JOSM workflow might be comfortable completing this task. (Then typing a few lines here to say how it went). Stevea (talk) 00:29, 26 June 2019 (UTC)