The physics under “Earth Intelligence” is older and stranger than the dashboard makes it look. GEO — the Group on Earth Observations at earthobservations.org — frames its mission as connecting governments, academia and the private sector to “harness the power of Earth Intelligence” against climate change, biodiversity loss and disaster risk, and it pins that work to three named yardsticks: the United Nations 2030 Agenda for Sustainable Development, the Paris Agreement, and the Sendai Framework for Disaster Risk Reduction. Fine. But strip the convening language and what’s underneath is a measurement problem that has never fully closed: how do you take a spinning, oblate, wobbling body and turn it into numbers a person can build against?

Start with the wobble, because it sets the frame for everything. Earth’s rotation axis is tilted 23.44° to its orbital plane, and that axis itself traces a slow cone — axial precession. The full circuit takes roughly 25,772 years (the classical “Great Year”), not the ~2,351 sometimes quoted; that shorter figure is closer to the ~2,150 years the axis takes to drift across a single 30° constellation width, which is where the confusion usually creeps in. The rate is about 50.3 arcseconds per year — one degree every 71.6 years. It is the reason Polaris is only temporarily the north star, and the reason every long-baseline geodetic frame carries a date stamp. A coordinate without an epoch is a rumour.

Now the part that lands on your desk: turning that curved, dated surface into a flat map is a lossy choice, and the choice decides what you are allowed to measure. There is no distortion-free projection — you trade away angle, or area, or distance, and you pick which lie you can live with. Web Mercator (EPSG:3857), the projection under nearly every browser map tile, preserves angles and quietly inflates area toward the poles. At Zürich’s latitude that inflation is not academic.

←TODAY: Copernicus Sentinel imagery is still free, but after the EU’s move it is no longer instant — open Earth observation just became conditional. →3012: every wall in the city carries its own datum and epoch, so a facade measured in 2026 still resolves cleanly a millennium on. Fulcrum: a map is a projection choice before it is a fact, and the choice — which distortion, which epoch — decides what you are permitted to measure off it.

The news that makes this urgent: as Geoawesome reported, the EU has placed a 24-hour delay on parts of the Copernicus Sentinel feed — the Sentinel-1 (radar) and Sentinel-2 (optical) streams that European offices lean on for flood extent, land-surface temperature and change detection. Open EO didn’t close; it became conditional. For an office that quietly built a flood-risk or urban-heat workflow on “free, near-real-time Sentinel,” the tolerance you never tested is now exposed — the same failure mode that catches every frontier tool that ships on a demo and meets its real limits in production.

Atelier: The Büro exposure here is quiet but real — any site-analysis or AI-assisted change-detection pipeline resting on same-day Sentinel now has a latency it didn’t budget for, and “the map said so” is a weak position in a Baueingabe. This Monday, audit which of your site workflows silently assume live, free EO, and pin each one to an explicit source, projection and epoch — for Swiss work that means georeferencing to LV95 (EPSG:2056) on CH1903+, not to whatever Web Mercator tile the viewer defaulted to.

Hack: Measure how much a Web Mercator tile lies at your site’s latitude before you trust an area pulled off it. The scale factor is simply the secant of the latitude, so area — which goes as the square — is inflated by sec²(φ). Run the three lines and read the number for your own site:

import math
lat = 47.3769                        # Zürich, degrees N
k = 1 / math.cos(math.radians(lat))  # linear scale factor
print(round(k, 3), round(k * k, 3))  # -> 1.477   2.181

At Zürich a Web Mercator patch exaggerates ground area by ~2.18×; take a screenshot area straight off the browser map and you are off by more than double. Reproject to an equal-area or a local CRS (LV95) before any quantity leaves the drawing. The arithmetic is boring — which is exactly why it gets skipped.

There’s a lineage here PAZ has already mapped: the whole discipline of asking “which point is nearest to where” runs from John Snow’s 1854 Broad Street map through today’s Voronoi-based accessibility analysis (see our Voronoi — En Ingenieria concept panel). Earth Intelligence is that same question at planetary scale, and it inherits the same first duty — declare your projection, or your answer is decoration.

So do the boring thing before the beautiful one: open the one site workflow you trust most, find where its coordinates come from, and write down the CRS and epoch next to it. If you can’t, you don’t have a datum — you have a picture.

Source: earthobservations.org