And Civil Protection's 2022 historical flood map, drawn three years before that storm, agrees with v1.1 at all six sites checked.

21.2 km² changed, 2.7% of the island. Still a screening baseline. Every miss stays logged.

: : argos-geo.org/map

Re-validated against the February 2025 storm: the ranking is untouched (Τραπεζάκι 97.5, Πόρος 96.9, Περατάτα 94.6). The fix shows in the medians: Livadi fields 10.0 to 12.8, Koutavos lagoon 5.9 to 7.6, Omala plain 23.0 to 27.5.

What changed: a topo-wetness ponding term over flat, low-lying ground, a capped override on the v1 blend. Pooled rain can lift a cell to "high" but never to "very high" without real upstream convergence. We tested a depression gate first and rjcted it: Omala's 39 m sinkholes are drainage, not ponds.

Yesterday's forensic: dev.to/ptzivras/the...

Open-access paper: doi.org/10.3390/app1...

Map: argos-geo.org/map

6/6

That is the philosophy ARGOS is built on. Our flood layer ranks every catchment on the island from open data, validated against the February 2025 storm. We have not run the Ianos validation yet. It is on the list, and every miss gets logged.
5/6

The small failures, the ones that block mountain roads and isolate villages, sat at the edge of what 10 m imagery can reliably see. Manual rapid mapping still beat the algorithms.

4/6

The paper also contains an experiment that matters for anyone building hazard models. Four automated satellite change-detection pipelines were tested against field surveys. Success rate: about 52 to 54 percent.
3/6

Assos village buried under ~1.5 m of debris. A debris flow at Fiskardo with a 400 m runout. Rotational landslides near Myrtos beach. And the Agkonas bridge, gone, on the Argostoli–Fiskardo road. The same corridor we published an evacuation routing forensic about yesterday.
2/6

Full forensic: dev.to/ptzivras/the...

argos-geo.org/map
github.com/argos-geo/ke...

29.0 km.
116 road segments.
Agia Effimia Marina.
The uncomfortable number survived.
The comfortable ones didn't.

Evacuation planning runs on numbers. Unvalidated numbers make plans into rumors with maps attached. Every model in ARGOS carries its validation with it.

The forensic: a dead routing table.

The real network in another table.

A cost column named cost_m that was minutes, not meters.

Mixed projections.

A row-order bug that was mine, not pgRouting's.

And a coastal road the router detours around for a reason we still haven't confirmed.

Exactly. If a layer can’t be chalenged by something outside the model, it can easily end up as just another pretty map that eventually becomes a maintenance chore.

The important caveat:

Detection points, not burned-area polygons.

Directional validation, not proof.

The layer remains a screening baseline, not a forecast.

Independent evidence matters because the goal isn't to make the model look right. It's to find out where it isn't.

argos-geo.org/map

71.5% of detections fall in class 3 or above, versus 47.3% of the island baseline.

The highest class holds 9.3% of detections versus 2.0% of island pixels, about 4.8x enrichment.

Against the EFFIS harness, 16 of 17 historical perimeters fall within 500 m.

The fire-response layer lets you check every building against a 15-minute response time over the real road network.

ARGOS is still a screening system, not a crystal ball. That's why the workings matter.

“How do we know?” now sits one tap away on each layer.

argos-geo.org/map

The flood layer was tested against the February 2025 storm. Trapezaki, Poros, Peratata and Vlahata all fall in the mapped high classes, with Trapezaki ranking 97.5/100.

The wildfire model runs at 20 m and was checked against 24 EFFIS fire perimeters.

hydrogeological work on it already, but i dont want to let it derail the current ARGOS roadmap. It sure is a great later action item though! Combining the subsurface picture with the surface model would be fascinating!

Thats exactly the kind of rabit hole i was hoping someone would point toward :)).
Katavothres is actually one of the models biggest headaches. I am curious how feasible a subsurface model of the karst system would be with the data tha exists. There seems to be some really interesting

And the storm's worst flooding, at Livadi, was standing water on flat ground that a flow-accumulation model structurally cannot see.

Local knowledge caught it. v1.1 gets a pluvial term.

Screening baseline, not a hydrodynamic simulation.

Build Log #4 is live.
🔗 github.com/argos-geo/ke...