v1.10.0: Advisories as issued, and six landfalls that were in Mexico

Two data corrections, and both were the same shape: a source read as if it answered a question it does not answer.

Every advisory NHC issued, from the product it issued. The 2015-2024 replay was built from ATCF a-decks, which are a database of forecast cycles, and a cycle is not an advisory: a special advisory issued off the six-hourly clock has no cycle to be built from. 22 were missing, and where one went out at a synoptic hour exactly it took the number of the advisory that followed it, so eight records held the wrong forecast. Each advisory is now read from its own archived forecast/advisory product, the archive holds 1,684 of them, and the build refuses to ship a hole: a gap in the numbering is either reported by name or an error.

Six landfalls the atlas counted were in Tamaulipas. The rule that recovers a landfall HURDAT2 never marked tested only whether the previous fix sat inside a US state, and Mexico is not one, so a storm that came ashore south of the Rio Grande and carried on north read as arriving from the Gulf. A centre already ashore cannot come ashore again, and AOML's own table settles the rest, because it marks seven storms as having made landfall over Mexico first. 759 landfall events become 753, 374 at hurricane strength become 370, and 590 storms are left where 595 were. AOML's scoring of this atlas rises from 93.1 to 94.1 percent precision with recall unchanged, so nothing it calls a landfall was lost.

Iselle's Hawaii landfall is the one HURDAT2 records. The exclusion that keeps Tamaulipas landfalls out of Texas had no western limit, so it covered Hawaii too, threw out a real landfall record and let the fallback put an interpolated one back three hours late.

The replay line names both clocks. An advisory is issued three hours after the forecast cycle it runs on, and a special one whenever it is needed, so two advisories in a row can carry one forecast with different positions. The panel says when the advisory went out and which cycle it came from.

This release also carries everything that had accumulated since v1.9.3, which is a great deal: the side-by-side storm comparison, the NHC storm-surge flooding layer, the 2008-2014 half of the advisory replay, continuous wind and pressure colouring, the cascade-layer fix that brought 127 dead map-styling rules back to life, and a long run of contrast, accessibility and gate corrections. The full entry is in CHANGELOG.md, which is too long for a release body.

Download

Neither build needs Node, npm or a clone, and neither talks to the network once unpacked.

Profile	Archive	Unpacked	Contains
core	hurricanemap-1.10.0-core.tar.gz (6.1 MB)	25.3 MB	The whole historical atlas: 590 storms, 753 landfalls, every panel and export
full	hurricanemap-1.10.0-full.tar.gz (489.8 MB)	525.9 MB	Everything in core plus the 1,697 archived NEXRAD radar frames

Both are built reproducibly, with entries sorted by name and timestamps and ownership pinned, so npm run dist:package on this commit gives the same SHA-256. Check a download against SHA256SUMS.txt beside them.

tar -xzf hurricanemap-1.10.0-core.tar.gz
sha256sum -c SHA256SUMS.txt
cd hurricanemap-1.10.0-core
python serve.py --port 8765