Add rendering for natural=valley labels

[matthijsmelissen]

The following issue has been moved over from trac:

The following are all tags used for common natural features in Iceland (and probably elsewhere) on nodes (along with place=locality to get them to render). They're traditionally rendered as plain text on maps:

• natural=fjord
• natural=valley
• natural=peninsula
• natural=channel
• natural=inlet

The list is ordered roughly by how prominent each should be on the map. But of course these things can differ wildly.

[daganzdaanda]

natural=heath has been added in the meantime, see #780.
"valley" is not in the code, but #787 seems to say it is rendered? maybe it was once?
The name of a "bay" should be rendered now at >=14 (#199 removed the fill)

+1 for showing the names of the rest.

Good rendering will not be easy, I assume. Putting a name on a fjord or inlet should be possible, since watery areas are not usually full of other things that get in the way, But a valley, or a natural=ridge?

People use things like place=locality or place=region to give names to larger areas.
User maxbe has a way to render a name over a large area:
http://geo.dianacht.de/topo/ Go to the alps to see the names of the regions there.
Explanation: http://wiki.openstreetmap.org/wiki/User:Maxbe/Kartenversuch#Beschriftung_von_Gebirgen
Short explanation in English (pic): http://wiki.openstreetmap.org/wiki/File:Maxbe-stubaier-beschriftung_en.png

It looks really great, but my guess is, it can't be done with carto-css alone. Or am I wrong?

[matthijsmelissen]

#787 seems to say it is rendered?

I guess the catchall rendered it on lines and areas.

[Klumbumbus]

See also openmapsurfer as an example of nice rendering of strait (Kattegat, Skagerrak) and some other similar features. Also on the other zoom levels.

[23cpo]

I would also add the following place tags to the list:
place=sea - http://www.openstreetmap.org/node/305640275
place=ocean - http://www.openstreetmap.org/node/305640074

[sabas]

natural=cape http://wiki.openstreetmap.org/wiki/Tag:natural%3Dcape

[matthijsmelissen]

Reopened - only resolved partially.

[matkoniecz]

I think that it would be better to open separate tickets for separate tags. BTW, natural=bay is rendered and I removed it from list in the ticket.

[matkoniecz]

place=ocean - http://www.openstreetmap.org/node/305640074

It is a really poor idea. Borders of oceans (and even number of oceans) varies depending on sources (see for example http://en.wikipedia.org/wiki/Borders_of_the_oceans). Also, mapping ocean as node is absurd. At this scale manually selected labels are a better solution.

[imagico]

Also, mapping ocean as node is absurd

Please keep in mind we do not only map for the renderer - the ocean nodes provide useful and relevant information and nodes are the only sensible way to map those currently in OSM. And in a (non real time) renderer they can - when reasonably placed - even be used to calculate appropriate automatic label placement.

[matkoniecz]

natural=heath label is now rendered.

[matthijsmelissen]

The tag natural=peninsula has 59 occurrences, natural=channel has 34 occurrences, and natural=inlet has 9 occurrences. That's too little to render them.

[dieterdreist]

2015-03-22 1:45 GMT+01:00 math1985 notifications@github.com:

The tag natural=peninsula has 59 occurrences, natural=channel has 34
occurrences, and natural=inlet has 9 occurrences. That's too little to
render them.

I don't agree, these are typically very significant features and supporting
them will increase the overall cartographic quality. Also there exist much
fewer of these features in the world than of other features, so maybe those
numbers aren't that bad (refering to channel and peninsula).

[imagico]

Note natural=channel, natural=inlet, natural=fjord and natural=sound are all undocumented and it is not clear what exactly distinguishes them from natural=strait and natural=bay.

natural=fjord is the only one with more widespread and consistent use, could be defined as something like 'long and narrow bay formed by glaciers' and would make sense to be rendered together with natural=bay.

[eehpcm]

@jeisenbe

That's why it may be a problem to render just the label, oriented along the way. This could encourage mappers to draw "labeling geometries", based on how they want the label to look on openstreetmap.org, rather than based on a verifiable feature on the ground.

You appear to assume that many mappers would wish the label to look like anything other than the course of the valley. I think that most mappers, like myself, would like the label to follow the valley but accept small imperfections introduced by the rendering process. I think what is more likely if the label is non-oriented is that they will trim the valley length or course to try to get the label in the centre of the valley (essentially twiddling the line to get the same result as using a node).

For valleys that contain rivers or streams, usually that would be the lowest spot, so that could be a verifiable place to draw the natural=valley way. However, I notice in your examples that the valley does not exactly follow the waterway. I've seen this commonly in other valleys drawn as ways: the mapper often uses a simplified line, instead of following the thalweg (line of lowest elevation).

I suspect the other mappers did it that way for the same reason I did: that's what the wiki suggests. The river or stream will almost always follow the thalweg, but the valley is often a good deal wider than the waterway and the waterway meanders within it. This is why the wiki says "Note that valleys and waterways are not necessarily congruent. Rivers are often meandering. There may be multiple branches of the river, or no river at all. Therefore, you better ignore waterways in U-shaped valleys, and go for the middle line of the valley bottom instead." That is what I attempted to do, as best I could interpret aerial images and the OpenTopoMap layer.

Since natural=valley ways are not being drawn in a particularly consistent way, we would not want to contribute to this problem and encourage more "mapping for the renderer".

I am not convinced that they are being drawn inconsistently. I think it likely that the discrepancy is in your expectation that the valley follow the thalweg and mappers trying to map the centre of the valley as the wiki tells them to. It's easier to map the thalweg if the waterway is already there, just follow the waterway. It's more work to try and figure out the centre of the valley. And, right now, people most definitely aren't tweaking the valley in order to affect label placement because there are no labels.

If mappers start using natural=valley in a more consistent way, for example by attempting to follow the lowest ground in the valley (usually the same as the waterway, if there is one), we could reconsider using those geometries for labeling purposes.

I believe they already are using them consistently. I would have taken the easy way out and followed the watercourse had not the wiki persuaded me to put a little more effort in to do the job properly. But even if I'd decided to avoid the watercourse just because I didn't want to superimpose the valley on it, I'd have placed the valley close by and following a somewhat smoother path.

Nobody has abused the tag to place a label because there currently are no labels. Might they do that in the future in order to get around the renderer suppressing a label because of space restrictions? Maybe, but there are already many other opportunities for them to do things like that if they're so inclined.

[jeisenbe]

It's more work to try and figure out the centre of the valley.

I agree. From my experience trying to map some valleys here in Indonesia with linear ways, it's quite hard to decide on where the center of a wide valley should be. Is it the line that is halfway between the tops of the ridges on each side, or the line at the center of the flatish part of the valley, or something in between? Unfortunately this isn't defined.

Nobody has abused the tag to place a label because there currently are no labels.

The labels are shown in other styles like Opentopomap, which show contour lines and hillshading. It's more appropriate to show linear valleys differently in that case, since mappers will see if there is something clearly wrong, like a line that has been draw without regard to the actual shape of the valley to just get a certain label position.

[eehpcm]

@jeisenbe

Is it the line that is halfway between the tops of the ridges on each side, or the line at the center of the flatish part of the valley, or something in between?

It's an approximation. The best we can do using aerial imagery and OpenTopoMap. Complicated by the fact that the valley may contain a wood. But everything we do is an approximation. Look at the Afon Teifi just north of Cwm Morgenau. Two years ago it came from NPE and was a very crude approximation to the true course. Some time in the last 2 years I tweaked it, but I had no idea where the thalweg was so placed it roughly midway between the banks. Sometime after that I mapped the banks to show the true width. One day somebody might actually map the thalweg, but until then it's a better approximation than it was with NPE data. Comparing different aerial imagery shows transitional, rotational and scaling differences, so everything is approximate.

No two mappers would map the same valley exactly the same way. But most conscientious mappers would come up with approximations that were in broad agreement and that were good enough.

Unfortunately this isn't defined.

I think that may be a good thing. It would turn a difficult task (that's a reasonable approximation) to an impossible one (there's no way of knowing the exact centreline to that standard, so I can't map it).

The labels are shown in other styles like Opentopomap,

I hadn't realized that. I just checked Cwm Morgenau and can't see why the label is there twice. I don't think it's something I did wrong, it looks like OpenTopoMap had a brain fart.

It's more appropriate to show linear valleys differently in that case, since mappers will see if there is something clearly wrong, like a line that has been draw without regard to the actual shape of the valley to just get a certain label position.

I still don't see that as a huge problem. Sure, I could map a valley, be unhappy with the non-oriented label placement, remap it, check again, remap it,... until I was happy with the label placement. Or I could just add a place=locality and do a lot less work. I think the majority of valleys will be mapped as honest attempts to show the true valley, not to place labels. I also think that if standard carto renders a valley label no differently from that which would be shown by place=locality, more mappers will be tempted to just use place=locality because it's far less work. In fact, I think if valleys are shown with non-oriented labels, that makes it more likely mappers will tweak the valley (possibly a lot) just to get a better placement when they'd accept an oriented label that looked a little unaesthetic as simply representing reality.

[mboeringa]

I think the majority of valleys will be mapped as honest attempts to show the true valley, not to place labels.

I have seen few if any problems with mappers mis-using valley tagging on lines and polygons. I agree most are honest attempts to show the approximate location of the valley.

The main problem I see with current tagging sometimes is that people simply replicate the course of a river or main stream running through the the valley, which may not always be appropriate for the valley as a whole.

[jeisenbe]

Here are a couple of valleys that I mapped a year or two ago (though I'm not too happy with them now). Attempting to render a text label on the line does not work well:

• The valley line should closely follow the river, but this looks bad when rendered on the line.

• Here the label might be confused as belonging to the residential area.

Perhaps some combination of simplification of the line, plus stretching out the label to match the length of the way to some extent could work in another style, but probably not in this style which lacks a representation of the terrain.

[jeisenbe]

I considered rendering the valley label near the center of the way (or on the point). But this looks odd at high zoom levels, when the valley is large, though it is fine for small valleys.

@imagico, in this case would it be appropriate to stop rendering the labels for very long valleys which extend off the screen, as we would for the text label of a large area?

E.g:

  ST_Length(way)/NULLIF(SQRT(!pixel_width!::real*!pixel_height!::real),0) AS way_length,

(though it would be better to use scale_denominator)

  [feature = 'natural_valley'][zoom >= 12][way_length < 1200] {

I know we wouldn't want to show text labels larger or at an earlier zoom level without a representation of the line, but perhaps this is acceptable, since it only is removing the text label when the valley is too long for the usual screen size?

[eehpcm]

@jeisenbe

Here are a couple of valleys that I mapped a year or two ago (though I'm not too happy with them now). Attempting to render a text label on the line does not work well:

What other maps do (which may be a simple matter of programming here or very, very difficult) is offset the label so it doesn't overlap the watercourse. It results in a label that isn't positioned exactly, but gives a reasonable indication of the length and geometry of the valley.

[jeisenbe]

offset the label so it doesn't overlap the watercourse.

Generalization is a great idea for map legibility, but another very important goal of this style is mapper feedback, and for that it is important for the precise geometry to be shown. That's why we don't move overlapping roads along rivers. Also the processing would be somewhat computationally expensive, and the more complex code would be harder to edit and maintain

[imagico]

Independent of the questions if there is a consistent and verifiable mapping of valleys at the moment i would - like in case of polygons - be very much against labeling linear ways without providing feedback on the geometry and its extent. That would incentivize drawing of non-verifiable labeling geometries.

[jeisenbe]

As mentioned above, my suggestion is to render a horizontal text label at the center point of the natural=valley way only.

The only possible use of the line length would be to stop rendering valleys which were clearly much larger than the usual screen size. All would start rending at the same zoom level.

If this isn't feasible, we could only render labels on nodes. At least in that case it would incentivize using natural=valley instead of a place=locality node, and would get us a step closer to removing the rendering of place=locality.

[imagico]

As mentioned above, my suggestion is to render a horizontal text label at the center point of the natural=valley way only.

The only possible use of the line length would be to stop rendering valleys which were clearly much larger than the usual screen size. All would start rending at the same zoom level.

I see - i am doubtful about the usefulness of that regarding the goals of this style. Even if i imagine that tag was consistently used on linear ways in some form (which is purely wishful thinking) that would only lead to a meaningful labeling at a very small range of zoom levels and what range that is depends a lot on the circumstances. There are valleys that are in size of the order of a few hundred meters while there are also valleys of a few hundred kilometers. The range of sensible zoom levels for those two does not even overlap.

[jeisenbe]

there are also valleys of a few hundred kilometers

Fortunately, very few of these are mapped as ways in Openstreetmap. I haven't found any so far in the dozen areas that I have looked at.

There are valleys that are in size of the order of a few hundred meters

Most valleys which I have checked are in this range of a few hundred meters to a few kilometers.

Many areas have no natural=valley features mapped, e.g. most developing countries which I checked, and flat areas like most of the Netherlands and Delaware. Also most US states and Canadian provinces do not have any valleys mapped yet.

However, there are some mountainous and hill areas where they are commonly used.

Of the places I have downloaded, Burgenland in eastern Austria and Hautes Alpes in France have the most valleys, and fortunately they are fairly consistently used there. Some regions of Spain also have a number of valleys, though these are often very coarselly mapped with just 2 to 4 nodes, but I'm still happy to encourage tagging these, rather than using locality for all small unpopulated place names (very common in France and Spain).

[jeisenbe]

In these test images I've rendered the valley name label on the center of the line (or on the node) with landform-color (same as peaks/saddles), and rendered a thin landform-color line along the way, if there is one, to show the underlying data. This is not a proposed rendering, but a way to show what the data looks like. Testing at z14 since that would be the earliest to show these features (thoug z15 would be consistent with ridges, aretes and cliffs).

Hautes Alpes

• 26 ways, all except 1 with a name, no nodes
• Never added to the same feature as waterway=stream, though sometimes it follows a similar geometry

https://www.openstreetmap.org/#map=14/44.4583/6.5547

https://www.openstreetmap.org/#map=15/44.8117/6.7131

https://www.openstreetmap.org/#map=15/44.8117/6.7131

• 5 valleys - one name is blocked by the peak icon (northeast)
  

https://www.openstreetmap.org/#map=15/44.6942/5.8519

• Smallest one. This works at z15, but z14 is pushing it
  

== Burgenland ==

• 61 ways, no nodes, all but 2 with names.
• This appears to be a karst region, since many valleys lack a waterway feature.
• One of the forest polygons is broken in this extract; pretend the background is green.
• Obviously the test rendering doesn't work well with track roads.
  https://www.openstreetmap.org/#map=14/47.9775/16.8216
  

https://www.openstreetmap.org/#map=14/47.8852/16.5819

https://www.openstreetmap.org/#map=14/47.8828/16.5008]

https://www.openstreetmap.org/#map=14/47.4354/16.2099

https://www.openstreetmap.org/#map=14/47.3785/16.3751

https://www.openstreetmap.org/#map=15/47.3355/16.3519

== Asturias, Spain ==

• I also checked Extremadura, but there were no valleys there.
• Asturias has 5 ways with natural=valley, all named.

https://www.openstreetmap.org/#map=14/43.2623/-4.8033

At z16 lots of place=locality features are visible nearby

https://www.openstreetmap.org/#map=14/43.1975/-4.7749

• Valle de las moñetas is duplicated as a place=locality node, the intermittent waterway has a different name "riega de las moñetas"
  

https://www.openstreetmap.org/#map=15/43.1951704/-5.2233951

• This valley looks like a pass or saddle too
  

https://www.openstreetmap.org/#map=14/43.3719/-4.8662

• Somewhat roughly mapped
  

[jeisenbe]

There are a few valleys mapped in Tasmania, Wales and Hawaii, though many other English-speaking areas do not have any mapped yet.

== Tasmania ==

• 5 valleys, all mapped as ways but one, all named
• The ways are rather rough

https://www.openstreetmap.org/#map=14/-43.6010/146.8627

• Finally a node
  

https://www.openstreetmap.org/#map=14/-41.6632/145.9525
Marigold valley - rather small, perhaps it is a ravine?

https://www.openstreetmap.org/#map=14/-41.8186/146.3115

• Fun Biblical / Holy Land names in this area
  

https://www.openstreetmap.org/#map=14/-41.9200/146.1540

== Wales ==

• 3 ways, all named

https://www.openstreetmap.org/#map=15/51.9998/-4.6185

https://www.openstreetmap.org/#map=14/52.0590/-4.6242

• z14 is marginal for these 2 small valleys
  

== Hawaii ==

• No nodes, 42 ways, but only 30 named. The 12 unnamed ways are also tagged waterway=*, as are 8 of the named ways - these would not be rendered.
• Lots here are mapped in gullies which might deserve to have a waterway=intermittent mapped as well, based on the aerial imagery, though often the name includes "valley" rather than "gully" or "stream" etc.

https://www.openstreetmap.org/#map=16/23.0581/-161.9168

• These ways are double-tagged as valleys and intermittent streams, though the name includes the word "valley":
  

https://www.openstreetmap.org/#map=15/20.0296/-155.8133

• Small gulches which look like intermittent streams on aerial imagery (there are bridges on the road)
  

https://www.openstreetmap.org/#map=14/21.8934/-160.0998

• All these valleys are fairly nicely mapped. They should also have the intermittent streams mapped, though this has not been done yet. The name appears to refer to the valley not the stream.
  

[jeisenbe]

Besides the 2 really big valleys that I mapped here in Papua, the only really large valleys I have found so far are in Norway - partially they appear big due to the high lattitude. Here are there some that really should be shown at z12 or z13.

== Norway, Sogn og Fjordane ==

• One node, 35 ways, all named. One is also a waterway=stream.

https://www.openstreetmap.org/#map=14/61.4989/8.3040

• Mapped as a node, moderate sized per topo map
  

Most of the ways are 5 to 20 kilometers long (22/35), like these. The very longest is 38 but includes part of a fjord: only 5 are over 20km.

https://www.openstreetmap.org/#map=14/61.4749/8.2782

https://www.openstreetmap.org/#map=14/61.4825/7.9626

https://www.openstreetmap.org/#map=14/61.4640/8.0680

But there are 8 that are less than 5 kilometers long, (about 1/4) eg:
https://www.openstreetmap.org/#map=14/61.9903/6.4627

[jeisenbe]

=== Analysis of all natural=valley ways ===
I downloaded all 11,106 natural=valley ways with name=* tags - ouch, that was 49 megabytes.

2495 have a waterway= tag (mostly waterway=stream), but most of these are waterway=drystream (stream is the next most at 846). Also, 1547 of these are all in one small region of Russia(?) centered on 50.8797426, 40.1522567, most added by User:keder in 2016-17

No other combinations are very significant.

• 95% are shorter than 13 kilometers (only 579 have waylength > 12,000 meters)
• 90% are shorter than 8 kilometers
• 85% are shorter than 5 kilometers

The median natural=valley way is 1281 meters long, according to JOSM.

• 95% are longer than 150 meters
• 90% are longer than 300 meters
• 85% are longer than 400 meters

So a super-majority (70%) are between 0.4 kilometers and 5 kilometers in length, and 80% are 300m to 8km in length

The middle 46% are between 0.5 and 2.0 kilometers: only 2 zoom levels difference.

=== Conclusion ===

I think we can consider rendering these features by focusing on the majority that are a couple hundred meters to a few kilometers in length (and also the small number of nodes should be rendered).

Unfortunately I do not have any good ideas for a linear representation. Since these are not usually nodes, a point icon (like natural=saddle) would not be appropriate.

So I am planning to just render the name at the center, though probably in gray rather than landform-color.

[jeisenbe]

I believe a gray label is the best option. While I considered something similar to the landform-color text labels used for peaks and saddles, that is matching an icon.

As a text label only it might make more sense to match the place=locality, place=island and natural=cape labels, and natural=ridge / natural=arete which are gray.

Options:

1) Brown labels (based on landform-color)

2) Gray labels, oblique

3) Gray labels, standard (book-fonts)

[jragusa]

Grey label with standard font would be consistent with ridge names. I also notice that natural=cape is labelled in black colour, do we need to have two colours for natural features ?

[imagico]

So a super-majority (70%) are between 0.4 kilometers and 5 kilometers in length, and 80% are 300m to 8km in length

The middle 46% are between 0.5 and 2.0 kilometers: only 2 zoom levels difference.

It is natural for this kind of feature there are a large number of smaller ones and only very few larger ones. But that does not mean there are no larger ones. Like:

https://www.openstreetmap.org/way/374056124
https://www.openstreetmap.org/way/539764751
https://www.openstreetmap.org/way/706934555
https://www.openstreetmap.org/way/539763901

Overall i have strong doubts about the kind of labeling you suggest. The main problem with the current mapping of natural=valley is that there is no consistency in the geometries. The suggested labeling would not help changing that because the mapper will get feedback about having done something right (by there appearing a label) but that positive feedback will appear no matter how they draw the line. At the same time the benefit for the map usefulness seems rather small by placing a point label somewhere along the mapped geometry.

[jeisenbe]

So you would only be in favor of labeling these if we also show the line?

I haven't been able to find examples online of linear representations for valleys. Probably in modern maps it is assumed that shading and contour lines are used instead.

Any ideas on how it might be done?

[imagico]

So you would only be in favor of labeling these if we also show the line?

No, even if there was a good way to do that - mapping lacks the consistency for this to make sense.

It seems to me that most of the natural=valley geometries in the database were created to generate a certain result in maps which render labels for those. This is in particular cases where mappers draw the lines with a deliberate offset to place an offset label.

We could of course do the same and offer mappers the option to directly draw labels in our map. But i don't think this is the responsible thing to do.

Ultimately it is not within our mandate to tell mappers actively how they are supposed to map valleys. Therefore i think the only thing we can say at the moment is that the current way these are mapped does not work and we therefore can't support it.

[eehpcm]

I just noticed that OpenTopoMap renders valleys. I'm not saying it's the way to go (especially as I don't know how much computational load their approach requires) but it gives an example of how it could be done.

z=16
z=17

I think it looks acceptable. We might quibble over the text colour for consistency with other features, but apart from that I think it looks OK.

BTW, would a river gorge be considered natural=valley or should it be natural=gorge? If the latter, would code be added to render natural=gorge if code is added for natural=valley?

[Ircama]

I also considered this topic in the past and as far as I remember the reason why valley is not currently rendered in openstreetmap-carto is due to the lack of verifiability in the valley definition (as with other geographical features in OpenStreetMap). For instance, currently its data model does not support intersecting a house with a valley to check whether that house is inside a valley.

Discussions on data models are common to other geographical features in OpenStreetMap, where unfortunately geometry is not yet agreed and related arguments do not provide improvements.

Designing an acceptable geometry for valleys (e.g., whether a valley shall be represented through its thalweg or it is an area) is needed before considering its rendering here. At least this looks the spirit of the maintainers of openstreetmap-carto, the official OpenStreetMap rendering, possibly also as an incentive to improve arguable data models of some OpenStreetMap features.

Platforms like OpenTopoMap and Osmand, which are not official even if remarkable, decided to render valleys and many other geographic features based on current definitions, also in case they are more oriented to directly draw a label with the current renderers than describe the feature.

[eehpcm]

Designing an acceptable geometry for valleys (e.g., whether a valley shall be represented through its thalweg or it is an area) is needed before considering its rendering here. At least this looks the spirit of the maintainers of openstreetmap-carto, the official OpenStreetMap rendering, possibly also as an incentive to improve arguable data models of some OpenStreetMap features.

In this thread there have been many reasons put forward why valleys cannot be rendered, or should not be rendered, or must not be rendered. One of the arguments was along the lines of "Even if we could rigidly define how to map a valley as a way, there is no good way of rendering it because reasons."

I think OpenTopoMap has shown that it is possible to adequately render a valley mapped as a way, even when the label partially obscures the associated river.

Now we can move onto the arguments about accuracy and definitions (and even thalwegs, if you insist). To which i'll respond that everything we map is an approximation and we understand that when better aerial imagery, or better GPS traces or whatever come along then we refine that approximation. If we insist that we must not map something unless we can map it perfectly then we can all give up an go home.

[Ircama]

In this thread there have been many reasons put forward why valleys cannot be rendered, or should not be rendered, or must not be rendered. One of the arguments was along the lines of "Even if we could rigidly define how to map a valley as a way, there is no good way of rendering it because reasons."

I am generally in favor to see names of valleys rendered, together with other mountain features and I have tried to summarize the essence of the issue as I understand it.

[ppete2]

I wish to see valleys rendered too. Valleys are an important topological feature, which are in frequent use in talks, newspapers etc.

Btw: On the very top of this issue it's written "Fixed by #941" . Could this be deleted, cause the issue isn't fixed at all.

[dch0ph]

I suspect that rendering valley names will look odd in a non-topographic map like Carto.

It can work if you have contours / hill-shading. Here a Carto-derivative where valley names are only rendered at Z11 and Z12:
Z12:

Z13:

The valley name would interfere with the extra detail that appears at Z13.