Proposal talk:Tag:natural=plateau

From OpenStreetMap Wiki
Jump to navigation Jump to search

Areas

I would like to see this rendered as an area. It is hard to depict something like the Cumberland Plateau or even sub sections of it with a node. Thoughts? -- Harrija 17:21, 4 August 2019

I mainly created this proposal to discourage mapping plateaus as areas, as explained in the main text of the proposal.. A feature like the Cumberland Plateau does not have well-defined boundaries, and therefore mapping it with an area geometry will have problems with Verifiability - there's no way for a mapper to confirm where to put the borders. Note that the Wikipedia page for the Cumberland Plateau mentions that there is no clear border between this and the Allegheny plateau: "The terms "Allegheny Plateau" and the "Cumberland Plateau" both refer to the dissected plateau lands lying west of the main Appalachian Mountains. The terms stem from historical usage rather than geological difference, so there is no strict dividing line between the two" --Jeisenbe (talk) 11:50, 5 August 2019 (UTC)
I'm not sure I understand your objections to mapping plateaus as areas just because the borders will be 'fuzzy'. Isn't the placement of a node at a point that you've judged by eye to be at the centre going to be just as fuzzy, and just as unverifiable? And, fundamentally, isn't mapping a two-dimensional feature onto a single point just replacing one kind of fuzziness with even greater vagueness?
At the very least, even a rough area outline will allow renderers to place a name at a, perhaps more accurate, geometric centre of a plateau, rather than at a point that you've subjectively judged to be the centre. It would also allow for renderers to adjust font size etc. based on the surface area of the feature.
Furthermore many named plateaus are weird, elongated or multi-lobed shapes that don't have an obvious centre. To make renderers position names sensibly would require multiple 'plateau' nodes - a case of mapping for the renderer. Even a rough area outline would allow renderers to make their own more elegant decisions about where to place the names.
I understand the problems arising from vast areas where the fuzziness could be of the order of tens or hundreds of kilometres - similar to the problem we have with mapping oceans and mountain ranges etc. But I reckon that the majority of plateau features are more manageable. For example the area I'm working in, the Grand Canyon, there are numerous plateaus that are not sharply defined mesas or buttes, but are otherwise discrete and smaller than a few square kilometres. They are known by local hikers and appear on the USGS maps, but there is no way currently to map them into OSM other than an awkward place=locality tag.--Tomthepom (talk) 05:21, 2 March 2020 (UTC)
I recommend you add natural=plateau nodes for these features, or consider natural=mesa if they are bordered by cliffs.
Mapping plateaus as areas is not verifiable, because if you ask 4 different mappers to draw the geometry, you will get 4 quite different lines. In contrast, a natural=mesa is surrounded by cliffs, so it can be clearly mapped as an area if we define the area as "the flat place inside of the cliffs". There isn't a benefit to asking mappers to draw areas for ill-defined features like mountain ranges, valleys and plateaus, because it is not clear if the area should include the surrouding slopes. Should it go all the way to the watershed (ridgeline) or all the way to the waterway? Both ideas are sometimes used for the same plateau, valley or mountain range. If you look at the USGS maps, they are not attempting to map an area or boundary for the plateau, but are just putting a label near the center.
I know it seems like mapping these features as an area would "make things easier" for renderers, but we don't map for the renderer. A renderer who wants to show these features can choose the appropriate label size and orientation by hand, or it is also possible to use a DEM (digital elevation model, for example the free and public SRTM data, used to make topo maps) to determine the approxiate shape and extent of the plateau, valley or mountain range by an algorithm. For example, Opentopomap already uses DEM data to orient natural=saddle icons in the correct direction. There are also other databases like Natural Earth if database users want hand-placed labels for large natural features, and of course USGS data is public domain.--Jeisenbe (talk) 23:52, 2 March 2020 (UTC)
I don't want to enter into the discussion about large and fuzzy area mapping in OSM; and anyway much of the debate there covers political, technological and practical issues that don't really apply to the small scale plateaus I'm concerned with.
As I understand it your main objection to using closed ways to represent plateaus is unverifiability. However, verifiability is not black and white and should only be applied, as the wiki article states 'as far as is reasonably possible'. Not everything in OSM is verifiable to sub-metre precision. Things like creeks, washes and wadis, as well as desert pistes or tracks, the central lines of valleys, and the boundaries between forest and grassland are intrinsically fuzzy to tens and sometimes hundreds of metres. The placement of the ways to represent them is a judgement call and no two mappers will agree. But a map without them wouldn't be a map.
The same I think would apply to smaller, discrete plateaus where the fuzziness is manageable. Even if you don't agree and prefer more precision, mappers will have the option to draw a 'lowest common denominator' boundary which 100% of mappers, and the locals who live there, will agree that the area inside this is the 'Foo' plateau. This boundary will, in principle at least, be verifiable.
I still don't understand how a plateau node would be any more verifiable than a closed way - if your four mappers can't agree on the boundaries then they certainly won't agree on the 'centre' of those boundaries. Local knowledge would be of no help either, we're not talking about an isolated township with an obvious centre but uncertain border. Instead you're replacing a fuzzy outline with a completely abstract and even more fuzzy 'centre' with no physical presence on the ground.
Yes, I would hope that in the near future some smart algorithms with maybe a touch of AI will be able to use DEM data to figure out the extent of plateaus, and for this finding a plateau node somewhere inside that boundary would be useful. However, the resulting created boundaries will be at least as fuzzy as anything mappers can come up with - if humans can't decide where the precise edge of a plateau is then a machine won't be able to make that decision for us.
There are a couple of points where I strongly disagree with you. USGS maps are not simply putting a name near the centre of a plateau. They give an indication of the size and shape of plateaus by using different font sizes, orientation of lettering, curving and spreading out of names and repeating the name where the plateau is lobed or not obviously contiguous. Admittedly they are a very fuzzy representation of fuzzy underlying data, but they do convey far more information than just a name and centre point.
Secondly, mapping a physical area, even a fuzzy one, with an OSM closed way and letting renderers decide how to represent the object and its name is obviously not 'mapping for the renderer'. On the contrary; mapping an area feature with an OSM node is, at best, a placeholder until better data is available and, at worst, label painting.--Tomthepom (talk) 05:27, 3 March 2020 (UTC)
It has long been the practice in Openstreetmap to use nodes to map things which do not have a verifiable extent: for example, aerodromes which are not fenced, place=village, place=sea, natural=volcano, etc. In these cases "as far as is reasonably possible" means mapping the approximate center with a node, rather than trying to map the ill-defined outside. In practice, Openstreetmap is improved by an iterative process, where each mapper fixes things that they notice are "wrong". For a plateau, if a mapper sees that the node is in the wrong place, they will move it to what they consider the correct location. Imagine that 1000 mappers look at the node for a plateau you mapped. Perhaps the first few will move it. Then over time, if the location is verifiable, mappers will no longer move it very often, because the node will be located in a spot that everyone agrees is close enough to the center of the plateau. If you take the history of all the locations of the plateau node, they will converge on a certain place as the center of the plateau.
But in contrast, if you ask 100 mappers to draw a closed way around the plateau, you will get several different groups, which do not converge in one consensus: some mappers will include all of the slopes down to the waterways or bottom of the valley around it, or up to the ridgetop. Others, as you suggest, will only want to include the relatively flat part of the plateau, and some will put the line in between these extremes. You will end up with areas that vary greatly in size, likely by a power of 2. This also happens if you try to map valleys and mountain ranges, since no one can decide how far up the valley goes, or how far down the mountain range extends. As you suggest, the best idea is the map the smallest possible area, but even in this case the polygon will be ill-defined, since most plateaus gradually transition into gentle slopes on many sides, and many plateaus have areas that are only partially flat or are hill or valleys inside of the plateau - do you exclude those areas with a complex multipolygon relation?
In contrast, mapping the approximate center of the plateau is much simpler and provides all of the 100% verifiable information: the name=* and approximate location, and looking at the topography (for a hand-labeled map) or a DEM (for an algorithmic label placement) will make it easy to make an appropriately sized and oriented label. It is absolutely not necessary to map it as a closed way or multipolygon to get a good label in rendering, or to properly represent the toponym in the database. --Jeisenbe (talk) 09:09, 3 March 2020 (UTC)
I think you might be surprised to learn how much of OSM is not strictly verifiable.
Take tree cover for example. Just like plateaus, there is no single, verifiable definition of what a 'wood' or forest is. Tree density? Vegetation cover? Average tree height? Different agencies and satellite vegetation surveys have their own, different definitions. This is not a problem where there is a sharp edge to a wood, but in most wilderness areas the line between wood and shrub or grassland is can be very diffuse, and different definitions can mean differences of kilometres. This vagueness at the edges is often greater than many of the plateaus I see. Indeed there are numerous cases where I have had to make a 50/50 decision about whether an area of many thousands of hectares is or is not a natural=wood at all.
And yes I've had discussions with local mappers whose judgement is different from mine, and different from each other. Every mapper who looks at the same area will map the treelines differently. This is inevitable; there cannot be iterative consensus because, like valleys and plateaus, there is no objective, universal definition of where the outlines should be.
But we make do, use our best judgement and try to make the map as accurate as possible, but no more accurate than reasonable. Your standard of strict verifiability would require the removal of millions of hectares of natural=wood polygons and would mean the only tree cover we can map are domesticated woods and woods bordered by roads.--Tomthepom (talk) 14:05, 4 March 2020 (UTC)
The tag natural=wood developed "organically" so the definition was not precise. The wiki page calls natural=wood a "Tree-covered area" and says to "draw an area to match the shape of the woodland" - so my standard is to use the tag for "areas which are mostly (>50%) covered by tree canopy". I believe this definition is used by many mappers, and it is verifiable. There will still be some border cases, and it would be ideal to develop a sub-tag to describe the canopy density of a woodland or scrubland area more precisely, but if I were making a proposal for natural=wood, that's what I would suggest as a verifiable definition. --Jeisenbe (talk) 23:28, 4 March 2020 (UTC)
"It has long been the practice in Openstreetmap to use nodes to map things which do not have a verifiable extent...". Well, yes and no. The examples you give; aerodromes, villages and volcanos, are objects with some kind of 'centre' which can be reasonably represented by a node. The one exception, place=sea, is a contentious issue and in my opinion is an egregious example of label painting.--Tomthepom (talk) 14:05, 4 March 2020 (UTC)
Also see place=continent, place=ocean, place=bay, place=strait (which were almost always mapped as nodes until 2018), and use of other place nodes like place=county / place=district when there are no good sources for administrative boundaries, and place=locality - I've used place=county or place=district for areas in Indonesia where I know that the district is in a certain area, but there are no correct, public sources for the boundaries (probably the government doesn't know where they are either). --Jeisenbe (talk) 23:20, 4 March 2020 (UTC)
"areas which are mostly (>50%) covered by tree canopy"
This may be appropriate to your areas of interest and agreed upon by local mappers. But it's far from universal - e.g.for most of the United States I use the Multi-Resolution Land Characteristics Consortium (MRLC) definition of;
"areas dominated by trees generally greater than 5 meters tall, and greater than 20% of total vegetation cover"
As you can see, this definition is drastically different from yours. Not necessarily better or worse - the reason I use it is simply that it's the first publically available open data source I could find. The MRLC is a group consisting of the USGS, the Bureau of Land Management, the US Forest Service, NASA etc. but their definition is just as arbitrary as any other. It's not the only definition, not even in the States, and not even by those agencies - individual agencies use different definitions for different maps, depending on purpose. In the UK, the National Forest Inventory uses a figure of >20% tree crown cover (a specific version of canopy cover), while the EU definition of forest is >10% cover.
The problems are even more profound. Canopy density is a measure of leaf ground cover, while the definition of forest or wood is a demarcation of tree population density. There are huge problems with using one to approximate the other. It's always been a fudge, but we have to use it for practical reasons - we can't realistically count trillions of individual trees by hand and the technology doesn't exist yet to do it by satellite. When it does, and when that data becomes widely available, definitions of wood and forest will rapidly move to a far more verifiable trees per square metre.
This conversation is fascinating but I fear it's gone off on a tangent from natural=plateau ! If you wish to continue I propose we move it to somewhere more appropriate, perhaps the natural=wood discussion page, or one of the OSM forums.
"place=bay, place=strait etc."
Definitely a debate for another forum. All I'll say is - "place=continent, place=ocean" - ugh, the horror! :) --Tomthepom (talk) 03:14, 5 March 2020 (UTC)
It is mentioned in the proposal, but where the plateau has distinct cliff lines surrounding it, it's pretty clear what the extent is and so can (and should be mapped as an area). https://www.openstreetmap.org/relation/12108346 is an example of one I just did. --Aharvey (talk) 13:36, 29 December 2020 (UTC)
See natural=mesa instead for tablelands which are surrounded by cliffs. --Jeisenbe (talk) 07:31, 30 December 2020 (UTC)

Dolines

The name natural=plateau might be misunderstood to be used also for tablelands that are surrounded by mountains, i.e. are lower than their environment. These are frequent in limestone areas, when sinkholes (dolines) are filled up with gravel and soil. The most popular on Creta are Omalos, Lassithi, and Messara, but they include thousands of minor "οροπέδιο" (tablelands), e.g. Niato. OSM Wiki recommends natural=sinkhole + sinkhole=doline for them, although, as you describe, they are currently tagged as place=locality or not at all, truly because for a non-geologist they are simply "flat land". Also, avoiding useless discussions, I would extend the definition to nodes and areas and leave the decision to the tagger. --GerdHH (talk) 12:42, 6 October 2019 (UTC)

A case could also be made for natural=valley if the flat area is surrounded by higher ground on almost all sides. For example, I currently live is a large valley where the flat land is 1600m elevation, surrounding by 2000 to 4000 meter mountains on all sides (except for a narrow gorge where the river exits). The whole eastern half is a karst formation too. But this is called the "Baliem Valley" (Lembah Baliem) and it's mapped as a natural=valley. We also have many individual sinkhole features which are tagged natural=sinkhole. --Jeisenbe (talk)


Small plateaus

natural=plateau can be used to tag a small plateau--but how small? Would it be appropriate to map a small flat area along a slope on a mountain? In the Alps many of such areas have a name, hence they are usually tagged with place=locality, but I'm after some more specific tag.

Flat areas are in important feature in the Alps, since these mountains have always been inhabited and exploited for grazing. Finding a small flat spot was nice, because animals and men could rest there, or it was possible to build a little house. This explains why almost every single flat spot here in the Alps have a name given many years ago. Currently people have abandoned the mountains, and those areas are no longer used. They have kept their name though, and are a significant natural feature like peaks or saddles.

I'm linking some pictures to give you an idea of what I mean:

--solitone (talk) 17:02, 2 May 2022 (UTC)