Landcover

[bdon]

Description
We want a landcover layer at low zooms from https://daylightmap.org/2023/10/11/landcover.html

Screenshots
Here are some rough previews:

@nvkelso should we put the 7 daylight landcover classes into https://tilezen.readthedocs.io/en/latest/layers/ or invent new kind values?

Barren / Snow / Crop / Shrub / Grass / Forest / Urban

[nvkelso]

The more OSM-y / Tilezen-y kind values might be (which already in Tilezen have both detailed and generic aggregations and segmentations), where (data value), blog label > proposed Tilezen kind value:

• (urban) Urban > urban_area
• (crop) Crop > farmland
• (grass) Grass > grassland
• (shrub) Shrub > scrub
  • note this is is slightly different than the remapping in queries.yaml, but is a good enhancement based on my experience the last 3 years
• (barren) Barren > barren (was initially desert, updated based on comments and further research)
  • based on the viewer link below, this is desert. An alternative value is barren
  • Tilezen already uses barren for low zoom consolidation of detailed classes, so fair game
• (trees) Forest > forest
• (snow) Snow > glacier (these are glaciers and permanent snow fields)

Then on zoom-in, the OSM data can be used instead, with the current OSM derived value moving to kind_detail, and these rollup values used for kind?

In the blog post another class was also mentioned, but the download doesn't have it:

• Water > ¿¿¿ wetland or water ???
  • based on the viewer link below, could be either...
  • based on the download and closer reading of blog post, this was either NE or OSM water used as a masking knockout, so we ignore it.
  • So if there isn't a polygon in the data layer, it's assumed to be water

See also:

• https://worldcover2021.esa.int/viewer (which has a legend for the original classes)
• https://github.com/tilezen/vector-datasource/blob/master/queries.yaml#L369C0-L450C0
• Separate landcover kinds from landuse layer into new layer tilezen/vector-datasource#2025
• Missing landcover "green" areas at low- and mid-zooms tilezen/vector-datasource#1905

[wipfli]

@nvkelso desert might be not ideal for esa's "Bare / sparse vegetation" because this landcover type is not only used in real deserts like the Sahara, but also wet mountain areas. For example, the Alps go from "Grass" to "Bare / sparse vegetation" to "Snow and Ice" with increasing altitude...

[daniel-j-h]

Hey folks I gave the Daylight landcover data a try even before we have it integrated properly. If anyone wants to follow along, here's how to get it into a .pmtiles file and the cevtor tiles styled on a Maplibre map.

Download the shapefile

$ aws s3 sync s3://daylight-openstreetmap/landcover/ . --no-sign-request

Convert to GeoJSON to check it out and to convert it to a .pmtiles file later on

$ ogr2ogr -f GeoJSON landcover.geojson low.shp

$ jq '.features[].properties.class' landcover.geojson | sort | uniq
"barren"
"crop"
"grass"
"shrub"
"snow"
"trees"
"urban"

Use tippecanoe to convert the GeoJSON file to a .pmtiles file

$ tippecanoe -o landcover.pmtiles -zg -n "Daylight Landcover" -N "https://daylightmap.org/2023/10/11/landcover.html" landcover.geojson

Host the .pmtiles on a static host; I have S3 + Cloudfront set up but your use-case might differ

aws s3 cp --cache-control 'public, max-age=30758400, immutable' --content-type 'binary/octet-stream' landcover.pmtiles s3://.../20231024/landcover.pmtiles

Add the source to a Maplibre map and style the landcover layer based on the landcover classes from above.

sources: {
  landcover: {
    type: "vector",
    url: "pmtiles://https://example.com/landcover.pmtiles",
    attribution: "<a href='https://daylightmap.org/2023/10/11/landcover.html'>Daylight</a>" },
    maxzoom: 8,
},

map.addLayer({
  id: "landcover",
  type: "fill",
  source: "landcover",
  "source-layer": "landcover",
  maxzoom: 10,
  layout: {},
  paint: {
    "fill-opacity": [
      "interpolate",
      ["linear"],
      ["zoom"],
      0, 0.5,
      10, 0
    ],
    "fill-color": ["case",
      ["==", ["get", "class"], "barren"], "#a48137",
      ["==", ["get", "class"], "crop"], "#efc52e",
      ["==", ["get", "class"], "grass"], "#afe57f",
      ["==", ["get", "class"], "shrub"], "#dbf4c2",
      ["==", ["get", "class"], "snow"], "#f1eff7",
      ["==", ["get", "class"], "trees"], "#1c4321",
      ["==", ["get", "class"], "urban"], "#f1eff7",
      "#ffffff"]
  },
}, "physical_line_waterway_label"); // before first label layer

Results look as follows - I'm not a designer tho, please no hate 😛

[nvkelso]

Our approach here will start small:

• Add a new layer called landcover (that will complement the existing landuse layer).
  • Relevant Tilezen issues include: Missing landcover "green" areas at low- and mid-zooms tilezen/vector-datasource#1905 and Add new landcover layer tilezen/vector-datasource#1885
• We'll use this data from Daylight here
  • Because this Daylight data doesn't change with each release
  • We'll converted to GeoPackage format, and
  • Posted to a public Protomaps web assets location to make use of that format's spatial indexing in Planetiler
• We'll do light mapping of the raw Daylight data values into the Tilezen values mentioned above.
  • This will also make it more future proof if new data is sourced later and that new source has slightly different value conventions.
• This will yield low zoom coverage only (eg to zoom 6/7/8 before it looks chunky / pixelated.
  • Styling should fade the layer out over those zooms.
  • Styling should also consider the z-order of these new landcover feature versus landuse.
• At this time we won't be moving any existing OSM related features in the landuse layer to the new landcover layer as that would be a breaking change.
  • We may consider doing that in a later major version release; see Separate landcover kinds from landuse layer into new layer tilezen/vector-datasource#2025 for details of which features could move over later.

Other data reviewed include:

• @wipfli's ESA worldcover polygons which goes to a higher zoom detail, but at cost of much larger data size.
• Esri/NationalGeographic/Microsoft's Sentinel-2 10m Land Use/Land Cover Time Series which has better classification algo (like for for shrubland in US desert south-west) and goes to 10 meters, but would require a similar rasterization approach that @wipfli took and have large data size.
• If Planetiler adds GeoParque support, that could unblock this. Otherwise the build data dependencies would overwhelm making some cuts of say Monaco or Switzerland.

[bdon]

Tileset part merged in tiles 3.5.0; TODO style additions