Closed almccon closed 8 years ago
It looks like the landcover in the currently-deployed Terrain starts getting lighter as early as zoom 8!
d507142 helps a lot.
Although I'm disappointed to lose so much of the white background in urban areas. Ideally, the map would converge to white in urban areas, and some neutral brown/green color in non-urban areas. Perhaps if I bring in a blurred-out version of Natural Earth urban areas I can achieve that effect...
Here's what ne_10m_urban_areas
looks like at 100% opacity, no blur. It's promising raw material to work with:
On second thought, it kinda sucks:
Even applying a huge stroke and a huge blur, it doesn't really work.
Maybe we can find a better dataset for urban areas?
The blur and stroke don't look that bad. Also isn't there 2 urban data sets from natural earth? I think one might be more high res.
On Thursday, October 22, 2015, Alan McConchie notifications@github.com wrote:
On second thought, it kinda sucks: [image: screen shot 2015-10-22 at 22 oct 8 21 18] https://cloud.githubusercontent.com/assets/1212178/10683781/8586c600-78fa-11e5-8d29-2f99fb2ddd9b.png
Even applying a huge stroke and a huge blur, it doesn't really work.
[image: screen shot 2015-10-22 at 22 oct 8 22 39] https://cloud.githubusercontent.com/assets/1212178/10683793/ab44c072-78fa-11e5-8dc5-4de17a29934f.png
Maybe we can find a better dataset for urban areas?
— Reply to this email directly or view it on GitHub https://github.com/stamen/terrain-classic/issues/41#issuecomment-150447725 .
MFA, Design & Technology Parsons, The New School For Design
http://chrishenrick.com Github: @clhenrick https://github.com/clhenrick Twitter: @chrislhenrick https://twitter.com/chrislhenrick
“Whenever I see an adult on a bicycle, I no longer despair for the future of the human race.” — H.G. Wells
@almccon what about creating a 4th lc500mMODIS
layer that only contains values for urban areas? That way we could keep the opacity cranked up for urban areas at higher zoom levels.
This would mean pulling out all pixels = 13 for the new raster.
Interesting idea, @clhenrick. I was thinking something similar, like grabbing the higher-resolution landcover for North America (in use for OG Terrain) and just pulling the urban areas.
Also, I see that Natural Earth has a reference to a higher-resolution landcover that was released more recently... so I'm going down this rabbithole: http://iopscience.iop.org/article/10.1088/1748-9326/4/4/044003/fulltext/;jsessionid=8F3F52D64ACE430A9408B513147AAC35.c3.iopscience.cld.iop.org
Oh, and one more thing I need to add: pull some of the landcover polygons from OSM (like landcover=residential
, landcover=industrial
, etc...) and color them white, too. Those landcovers aren't mapped everywhere in OSM, but they're pretty good in some parts of the world.
It looks like the higher-resolution landcover referred to in that article is the 500m MODIS we're already using in the raster. So I think your idea of the urban-only raster layer will do the same thing. Trying that now...
@almccon cool, seems like gdal_polygonize.py could be used to extract the urban areas from landcover raster, see this gis stackexchange post
I don't really need to polygonize, if I'm already using a raster for the other landcover, I can just use a raster for these urban areas, too.
But here are some experiments with blurring the landuse from OSM: I used red for testing so I can see where I'm getting OSM data vs from the landuse raster:
And here it is with a bit of blur:
And finally if I make it white:
I think that helps get urban landuse under a lot of neighborhoods that need it. I wish we had more landuse tagging in OSM, but this will provide more motivation for that!
Added in dd7ecf294d8057c1b718a487a6e4d5930fc07184.
@clhenrick @mojodna be aware that this adds a new table in imposm, so you have to re-run your database creation in order to get this fix. :(
@almccon cool, do you think combining the OSM urban landcover with the NE urban areas might work? I've been trying to pull out just the urban areas from the MODIS data and haven't been successful just yet.
Yeah, right now I'm using both. The more the merrier.
On Oct 23, 2015, at 19:13, Chris Henrick notifications@github.com wrote:
@almccon cool, do you think combining the OSM urban landcover with the NE urban areas might work? I've been trying to pull out just the urban areas from the MODIS data and haven't been successful just yet.
— Reply to this email directly or view it on GitHub.
:+1:
@almccon I altered the color of the osm urban areas to be 7% K to match the MODIS urban area RGB value. Think we are reserving 0%K / white for snow & ice correct? Feel free to change it back if you think white looks better for urban.
Good idea, I think matching the off-white of the landcover is the right thing to do.
I'm going to leave this open until we test it out in more parts of the world. I'm particularly curious other US cities where there is patchy landuse data in OSM, but also European cities that probably have very good landuse. I was pretty aggressive in https://github.com/stamen/terrain-classic/blob/master/imposm3_mapping.json#L402-L417 grabbing any landuses that seemed remotely urban, but I'm not sure if it would result in too much white in well-mapped areas like Europe.
@almccon sounds good, I'll take a look at some European cities too. While working with an extract for Seattle I noticed an odd wavy band happening with the landcover data as well as some tile boundaries showing up, let me know if you are getting these too?
Whoa, trippy. I'll see if I can replicate.
I can't replicate that wavy band, but I can replicate the seams at tile boundaries. I'll look into what I can do about that.
I'm also noticing quite a jump in the way the colors are styled between zooms 8 & 9. Might be a combination of taking both opacity and blur from 0 to 0.7 and 20 respectively. I'll try adjusting it a little more subtler and seeing if it helps.
That's also when I start to notice the tile seams showing up:
I think it's zoom 9 where something especially strange is happening with the color. z10 seems to look more like zoom 8 again.
So I removed the blur, and it still looks okay. The raster-scaling:gaussian;
line causes a bit of a blur anyway. You can still see the raster grid, but it's not unpleasant in my mind:
But supposedly we shouldn't be having that problem with the seams: https://www.mapbox.com/blog/tilemill-seamless-blur/
It may depend on the version of mapnik we're using?
Meanwhile, I figured out why zoom 9 started to look bluish. Turns out I wasn't turning on the "land" until zoom 10, so when we slowly added transparency to the landcover raster starting at zoom 8 and 9, we started to see the ocean show through. And then at z10 the yellowish land would appear in the background, causing a sudden hue change.
Now that's fixed... although I still am not totally happy with the gradual yellowing of all the background as we decrease opacity, but I think it's probably just something we have to live with.
After #23 I think the landcover is looking really good at low zooms.
But when you reach the city scale, the colors are way too harsh.
I'm not exactly sure how I'm going to fix this, but I've got some ideas.