The Climate Model Trust Atlas · Europe + global preview

Click a city. See who earned the projection.

Every pin is a real Arasense run: one consistent method across Europe. Climate models are scored against observed local climate, then trust-weighted into a mid-century flood-driving-rainfall signal. Model agreement is shown for every number, because a projection without it is just a guess with confidence.

Side by side

Same method, every city. Ranked.

Trust-weighted mid-century change in maximum 1-day precipitation, SSP2-4.5. Click a row to inspect the city.

Global preview · new

The method is global. The honesty travels with it.

World cities, run through the exact same pipeline as Europe. One thing changes outside Europe: how strongly ground observations constrain the reference data. So every pin carries a reference-data confidence flag: we trust-score our own baseline, not just the models.

Preview tier. Global cities are screening-level, unranked, and deliberately few. The confidence flag is qualitative: it reflects how densely long-record ground observations constrain the ERA5-Land baseline in each region. Where it reads weak, the trust scores inherit that uncertainty; publishing the flag is the point. Mexico City is this tier's proof of process: first published as withheld, because one downscaled model carried a physically impossible 1,513 mm worst-rain-day baseline over its high-relief terrain. So we built an extreme-plausibility screen into the pipeline (any model beyond 5x the observed ERA5-Land level is excluded and named), re-ran the city, and published the honest result: a contested signal at 50% agreement. The pipeline now screens extremes as strictly as means.

Method & honesty

How to read this atlas

Ensembles

Two depths of analysis

Atlas cities are screened with a 5-model trust ensemble. Bologna additionally has a full 34-model CMIP6 deep dive, which is why its two numbers (+10.8% screening, +14.2% full ensemble) differ. Both are shown, on purpose.

Data

What's underneath

ERA5-Land observations score each model's local skill via the Aras Diagram; trusted models are weighted into the projection. Baseline 1995–2014, future 2040–2059, scenario SSP2-4.5, 50 km radius per city.

Coverage

Cities across Europe, more with every pilot

Every ranked pin was computed with the same pipeline on the same settings, so the ranking is comparable by construction. All 36 published numbers were re-verified under the extreme-plausibility screen on 5 July 2026: 35 matched within 0.02 points, the largest deviation was 0.1 points, and the only exclusion was Mexico City's documented artifact. Every re-run is recorded in the Drift Ledger, and each city has a full trust profile page.

Scope note. Atlas signals are screening-level evidence for prioritization and planning conversations, not a substitute for site-specific engineering studies. Screening ensembles are small by design; a full-ensemble deep dive is how a pin becomes decision evidence.

Your geography

Your region isn't on this map. That's the point.

A pilot engagement starts with your locations and ends with them pinned here: trust-scored, projected, and defensible.

Put your region on the map Read the case studies