Where to look now
Version 4, October 2026, starts from how fossils actually turn up. You need three things at once: the right rock (the marine limestones and shales the ichthyosaurs were buried in), that rock bare and eroding (not under alluvium or sagebrush), and ground you're allowed to walk. So v4 searches only the Triassic marine formations on the Geologic Map of Nevada (1:250,000, formation by formation) plus a kilometre around them. It measures the ground on a 30 m grid with USGS 3DEP elevation and Sentinel-2 summer imagery, including a carbonate index (limestone darkens one of Sentinel-2's infrared bands). Then it asks which of that ground looks most like the places ichthyosaurs have already been found, keeping the targets on BLM land and out of Berlin–Ichthyosaur State Park.
Forty places to look
| # | Where | Rock | Nearest known place | Slope | Land | Close-up |
|---|
T01–T30 are in the formations ichthyosaurs come from; T31–T40 are other marine Triassic rock, a longer shot. Every target is at least 2 km from a known place and 1.5 km from the next target. Each close-up is a 3 km square: Sentinel-2 on the left, a 10 m hillshade on the right, the mapped Triassic units outlined. The full list with every measurement is targets.csv, and the code is tools/v4/model.py.
The earlier model (v1–v3)
Two hundred and forty million years ago central Nevada was the floor of a warm sea, and ichthyosaurs, dolphin-shaped marine reptiles up to the size of a bus, swam in it. Their bones are in the limestone of ranges like the Humboldt, Augusta and Shoshone Mountains, and paleontologists have recorded them at a handful of places. PaleoWave learned what the ground looks like at those places and scored other ground by how alike it looks. Tap any dot.
How far to trust it
I built PaleoWave early on, and when I came back to it this fall I re-checked its numbers from the files in the repository. Three things it claimed held up less well than they looked. Here's what changed, with the arithmetic in tools/check.py so anyone can rerun it.
v3's fifty places
The current model (v3) splits the north (the Prida and Favret rocks, north of 39.5°N) from the south (Luning and Gabbs), and scores terrain on elevation, slope, aspect, ruggedness and topographic position: whether a spot sits in a basin or on a ridge. Bones end up where the sea floor was, and they come to light where erosion is stripping it; basins and lower slopes score better than ridge crests. Treat these as places to go and look, not as predictions.
| # | Where | Score | Setting | Nearest known | From mapped Triassic | Close-up |
|---|
Score is the v3 random forest's probability, 0 to 1. "From mapped Triassic" is the distance to the Triassic layer v3 used, which held only two of the state map's Triassic units and left out the Luning and Gabbs; version 4, above, uses every marine Triassic formation. The full list, with every column, is paleowave_v3_top50.csv; GPX and KMZ files for a GPS are in outputs/.
Close-ups
For twenty of the targets I pulled 3DEP elevation at about 15 m and drew three panels each: a hillshade, the topographic position (red for ridges, blue for valleys) and slope. The star is the target. They're the quickest way to see whether a high score is sitting in the kind of place bones get exposed.
Nevada's ichthyosaurs
Nevada's ichthyosaur rocks span most of the Triassic, and they're not one formation but several, from different times and places on what was the western edge of Pangaea.
Two corrections to the old README: Luning and Gabbs are Late Triassic, not Middle Triassic equivalents of the Prida; and Shonisaurus popularis, though huge, isn't the largest Triassic ichthyosaur known. Larger ones have since been described from Canada and Britain.
If you go
Look, photograph, report. Don't collect.
Vertebrate fossils on federal land (most of this country is BLM) can only be collected under a permit, under the Paleontological Resources Preservation Act. Collecting at Berlin–Ichthyosaur State Park isn't allowed at all. If you find bone, photograph it in place with something for scale, note the coordinates, and tell the BLM field office. A find that stays in the ground with good notes is worth far more to science than one in a backpack.
This is remote, hot, roadless country. Carry more water than you think, tell someone your route, and don't count on a phone signal.
What's next
Version 4 did most of last month's list: it scores the whole outcrop, tests on held-out places, and uses the formation-level geology. What it needs now is more known places, because sixteen is very few to learn from. The exact locality data museums hold for the Augusta Mountains specimens would help most; I've asked the Cincinnati Museum Center, and the Field Museum is next. And it needs boots: nobody has walked a v4 target yet. If you do, I'd love to hear what you find.