The world carve's residual is a missing maritime boundary — coasts and land borders ask the same question with opposite answers #186

New issue

Open

opened 2026-08-20 09:13:58 +00:00 by art-bot · 0 comments

art-bot commented

2026-08-20 09:13:58 +00:00

Collaborator

The irreducible bit

background_land is Natural Earth land minus the countries we draw ourselves. Two questions
have to be answered at the edge of a country, and only one dataset can answer them:

At a coast, Natural Earth claims land where the real coastline says sea (it does not
resolve the Gulf of Corinth at all). We must carve that, or it draws as land in the sea.
At a land border, Natural Earth's line sits a few hundred metres to a few kilometres
off the national one. We must not carve past it, or the neighbour's land draws as sea.

Both look identical in the data: "Natural Earth says land here, and the country we imported
says nothing." The Gulf of Messenia and a strip of Norway are the same query result. Nothing
distinguishes them except knowing where the country's territory actually ends including its
waters — a maritime boundary.

Measured residual at 1:10m (R146): pl=1 930, fi=627, gr=515, se=374, is=0 km².

Things already tried and rejected, with numbers

Carve with the national mask (the original design): borders correct, but 26 000 km² of
Natural Earth land left inside our countries — grey wedges in every gulf. #184.
Extend the mask with WFD coastal waters (Greece, #176): the Gulf of Corinth window went
2 429 → 40 km², and the map still showed a 1 km line straight through it, because coastal
water bodies reach ~2 km and Natural Earth's error in an open gulf is tens of kilometres.
Carve with Natural Earth admin-0 (#184, current): coasts exact, borders off by the
generalization error.
A protection band along foreign borders: measured at several widths and both scales; it
trades the two errors roughly 1:1. At 1:10m with a 2 km band, pl unfilled drops
1 930 → 626 km² while 3 034 km² of Natural Earth land returns inside Poland. Rejected.
Inverting the background (base = land, draw Natural Earth water instead): makes land
borders structurally correct — there is no subtraction at a border at all — but moves the
same ambiguity to the coast, where the gulfs come back as land. It does not remove the need
for a maritime boundary; it changes which error is free. Worth revisiting only alongside
this issue, not instead of it.

What would actually close it

A per-country polygon of land + territorial waters, so the carve is "our territory" rather
than "Natural Earth's guess at our territory". Then:

Gulf of Messenia: inside Greece's territory → carved → sea. Correct.
Norway strip: outside Sweden's territory → not carved → Norway's land draws. Correct.

Candidate: Marine Regions (VLIZ / Flanders Marine Institute), which publishes a Union of
EEZ and land dataset — one polygon per country covering land plus its maritime zone, and an
eez_12nm territorial-seas layer. CC BY 4.0. It is a third party rather than a national
agency, which is a real question against R1 — but it would sit in the same slot Natural Earth
already occupies: a global reference dataset for the background, not a country's own data.
That call is worth making explicitly rather than by default.

Worth checking first, before adopting a third party:

Whether the land half of their union is any better than Natural Earth 1:10m at the
borders that matter. If it is derived from the same generalized sources, it buys nothing on
the border side and only helps the coast.
Whether a national maritime boundary is available per country from the country itself —
Greece publishes WFD coastal water bodies (too narrow, measured above), but several states
publish full territorial-sea limits, which would keep the no-third-party property.

Also here

Poland is the outlier at 1 930 km², three times the next country, and its border with
Belarus and Ukraine renders clean — so the area is probably concentrated somewhere specific
(the Vistula Lagoon and the Kaliningrad boundary are the obvious suspects) and may be Poland's
own mask rather than Natural Earth. Localise it before assuming it is the same problem:

-- where is Poland's R146 area?
SELECT ST_AsText(ST_Centroid(g)), ST_Area(ST_Transform(g,3035))/1e6 AS km2
FROM (SELECT (ST_Dump(ST_Difference(ne.geom, m.geom))).geom g FROM ...) x
ORDER BY 2 DESC LIMIT 20;

#184 fixed the world carve by subtracting Natural Earth's own admin-0 polygons instead of our country masks, and #185 cut the residual another 4× by moving from 1:110m to 1:10m. What is left is irreducible with the data we currently have, and it is worth writing down why. ## The irreducible bit `background_land` is Natural Earth land minus the countries we draw ourselves. Two questions have to be answered at the edge of a country, and only one dataset can answer them: - **At a coast**, Natural Earth claims land where the real coastline says sea (it does not resolve the Gulf of Corinth at all). We must carve that, or it draws as land in the sea. - **At a land border**, Natural Earth's line sits a few hundred metres to a few kilometres off the national one. We must *not* carve past it, or the neighbour's land draws as sea. Both look identical in the data: *"Natural Earth says land here, and the country we imported says nothing."* The Gulf of Messenia and a strip of Norway are the same query result. Nothing distinguishes them except knowing where the country's territory actually ends **including its waters** — a maritime boundary. Measured residual at 1:10m (R146): `pl=1 930`, `fi=627`, `gr=515`, `se=374`, `is=0` km². ## Things already tried and rejected, with numbers - **Carve with the national mask** (the original design): borders correct, but 26 000 km² of Natural Earth land left inside our countries — grey wedges in every gulf. #184. - **Extend the mask with WFD coastal waters** (Greece, #176): the Gulf of Corinth window went 2 429 → 40 km², and the map still showed a 1 km line straight through it, because coastal water bodies reach ~2 km and Natural Earth's error in an open gulf is tens of kilometres. - **Carve with Natural Earth admin-0** (#184, current): coasts exact, borders off by the generalization error. - **A protection band along foreign borders**: measured at several widths and both scales; it trades the two errors roughly 1:1. At 1:10m with a 2 km band, `pl` unfilled drops 1 930 → 626 km² while 3 034 km² of Natural Earth land returns *inside* Poland. Rejected. - **Inverting the background** (base = land, draw Natural Earth water instead): makes land borders structurally correct — there is no subtraction at a border at all — but moves the same ambiguity to the coast, where the gulfs come back as land. It does not remove the need for a maritime boundary; it changes which error is free. Worth revisiting only alongside this issue, not instead of it. ## What would actually close it A per-country polygon of **land + territorial waters**, so the carve is "our territory" rather than "Natural Earth's guess at our territory". Then: - Gulf of Messenia: inside Greece's territory → carved → sea. Correct. - Norway strip: outside Sweden's territory → not carved → Norway's land draws. Correct. Candidate: **Marine Regions (VLIZ / Flanders Marine Institute)**, which publishes a *Union of EEZ and land* dataset — one polygon per country covering land plus its maritime zone, and an `eez_12nm` territorial-seas layer. **CC BY 4.0.** It is a third party rather than a national agency, which is a real question against R1 — but it would sit in the same slot Natural Earth already occupies: a global reference dataset for the *background*, not a country's own data. That call is worth making explicitly rather than by default. Worth checking first, before adopting a third party: 1. Whether the *land* half of their union is any better than Natural Earth 1:10m at the borders that matter. If it is derived from the same generalized sources, it buys nothing on the border side and only helps the coast. 2. Whether a national maritime boundary is available per country from the country itself — Greece publishes WFD coastal water bodies (too narrow, measured above), but several states publish full territorial-sea limits, which would keep the no-third-party property. ## Also here **Poland is the outlier at 1 930 km²**, three times the next country, and its border with Belarus and Ukraine renders clean — so the area is probably concentrated somewhere specific (the Vistula Lagoon and the Kaliningrad boundary are the obvious suspects) and may be Poland's own mask rather than Natural Earth. Localise it before assuming it is the same problem: ```sql -- where is Poland's R146 area? SELECT ST_AsText(ST_Centroid(g)), ST_Area(ST_Transform(g,3035))/1e6 AS km2 FROM (SELECT (ST_Dump(ST_Difference(ne.geom, m.geom))).geom g FROM ...) x ORDER BY 2 DESC LIMIT 20; ```

art-bot self-assigned this

2026-08-20 09:13:58 +00:00

art-bot referenced this issue

2026-08-20 09:14:35 +00:00

world: carve Natural Earth with Natural Earth, not with our own country masks (#184) #185

art-bot referenced this issue from a commit

2026-08-20 09:48:40 +00:00

docs: write down R146, which conf/world.yaml was already citing

art-bot referenced this issue from a commit

2026-08-20 18:21:59 +00:00

conf/gr, docs: review fixes to water_bodies — the projected km², and the lake that is mostly not Greek

art-bot referenced this issue from a commit