v1.11.0: The atlas reads a coastline

Superseded by v1.11.1. Nine landfalls in these archives carry the wrong time, by up to 5 hours 24 minutes, and the release notes below overstate Danielle 1980's move as 155 km when it is 154. Download v1.11.1 instead.

v1.11.0: The atlas reads a coastline (2026-09-10)

Changed

• Three more landfalls the atlas counted were in Mexico, and 753 events become 750. These are the ones the last pass could not reach: an 1874 storm, a 1925 storm and Alice in 1954 all came ashore on the Tamaulipas coast and crossed into Texas over the Rio Grande hours later, so nothing about the fix before the entry gave them away. The build reads a coastline now, data/land-mask.json, and asks where the storm came ashore rather than where it was one fix earlier. Those three crossings are 76, 103 and 115 km from the United States; the furthest genuine one is Danielle 1980 at 18 km, whose coast is a barrier island the Census polygons leave out. Sweeping the limit over both basins, the atlas is identical anywhere between 20 and 75 km, the 1925 storm returns at 80, and at 10 three real landfalls disappear. AOML's own scoring is unchanged at 94.1 percent precision and 337 of 352 recalled, which is the check that nothing real was lost.
• Danielle 1980's Texas landfall moved 155 km east and nearly seven hours earlier, to 29.4N 94.72W at 23:17Z on 5 September, where its track first meets the coast. The record had it at 29.3N 96.3W inland, because the Census state polygons first place the storm in Texas six hours after Natural Earth has it ashore near Galveston. Only landfalls that are wrong by a whole fix move; the rest are already within one fix of the crossing.
• Texas and Louisiana are level on 104 landfall events now, where Texas led on 107.