How do we know?
We keep the map, the rock and the fossil separate. Each can answer a different question about ancient water, and none gets to pretend it knows what the others do.
AI reconstruction — not a photograph · Shutterstock AI/Shutterstock.com · Shutterstock Standard License
Different evidence, different questions
Plate position, modelled sea cell, rock record and fossil occurrence are separate kinds of evidence. Read across them without treating them as interchangeable.
- 1. Plate position
- 2. Modelled sea cell
- 3. Rock record
- 4. Fossil occurrence
Where was this ground?
A plate model reconstructs a past position for a piece of ground. It does not establish a coastline, water depth or environment.
If the model cannot place the ground, we report that limit. We do not turn a missing position into a story about when the ground formed or arrived.
See a model limit in Los AngelesWhat does an ancient map reconstruct?
A palaeogeographic grid models land elevation and ocean depth. A cell below sea level is that model's answer, not a direct observation of a town's past.
Different reconstructions can disagree. We show the model and its limits instead of blending several answers into an apparently certain one.
Explore a labelled reconstructed EarthWhat does the rock record say?
Rock records describe geological units and their ages. An environment recorded for a unit is different evidence from a position reconstructed by a plate model.
The scale matters: a regional record is not a sample from every street. A page must say where its evidence comes from and what area it describes.
Compare the witnesses for KansasWhat was actually recorded nearby?
A fossil database records reported finds, their identities and their locations. We keep an occurrence's age range and the search area visible rather than turning them into one exact moment or place.
More fossil records of one kind do not make that kind a vote that settles an environment. A missing database record is not evidence that an animal or environment was absent.
Try reading a page like a scientist
- Start with the question. Is the page describing a model, a rock unit, a recorded fossil or a calculation? Those are different kinds of answer.
- Find the place and time limits. Nearby is not the same as exactly here, and an age range is not a single day.
- Open the named source. The evidence section shows the dataset or publication behind the answer and records the version or access date where available.
- Keep the qualification when you share the answer. Removing words such as reconstructed, around or nearby can turn a careful statement into a different claim.
Try this with a place you recognise. Ask what each witness supports before asking for one answer about the whole place.
How sure can we be?
We name the uncertainty. An honest unknown is useful. It tells you where this evidence stops, so you can ask a better next question. Reconstructions are models, not photographs of the past; missing data stays missing, and disagreement stays visible.
A labelled reconstruction helps you imagine a question. It cannot become evidence just because the picture looks convincing. The source belongs beside the claim, where you can check it.
For teachers
For school
- Plate position, modelled sea cell, rock record and fossil occurrence are separate kinds of evidence.
- A plate model reconstructs a past position for a piece of ground. It does not establish a coastline, water depth or environment.
- A labelled reconstruction helps you imagine a question. It cannot become evidence just because the picture looks convincing.
Classroom prompt
Choose one place page and sort each of its claims into the four witnesses: plate position, modelled sea cell, rock record or fossil occurrence. Which witness does the page lean on most, and which does it not use at all?
Extension: why is a missing fossil record not evidence that an animal was absent, and what would a scientist need to find before saying it was never there?
Ages: sorting claims into the four witnesses works from about nine; checking a source’s version and access date suits pupils of about twelve upwards. That range is our teaching choice for this page, not a curriculum claim.
Sources and model versions
Method guide prepared from the site's existing records. This is not a new data fetch or reconstruction; each example page keeps its own processing and access dates.
- EarthByte / GPlates
MULLER2019 v3.0. Plate model: Müller, R. D., Zahirovic, S., Williams, S. E., Cannon, J., Seton, M., Bower, D. J., Tetley, M. G., Heine, C., Le Breton, E., Liu, S., Russell, S. H. J., Yang, T., Leonard, J., and Gurnis, M. (2019), A global plate model including lithospheric deformation along major rifts and orogens since the Triassic. Tectonics, vol. 38, https://doi.org/10.1029/2018TC005462. Licensed CC BY 4.0.
- PALEOMAP PaleoDEMs (Scotese & Wright 2018)
PALEOMAP PaleoDEMs, Version 2. Scotese, C. R., & Wright, N. M. (2018). PALEOMAP Paleodigital Elevation Models (PaleoDEMS) for the Phanerozoic (Version 2) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.5460860
- Macrostrat
Macrostrat API v2. Peters, S.E. et al. Macrostrat: a platform for geological data integration and deep-time research. https://macrostrat.org, accessed on 2026-08-27 (Kansas example)
- Paleobiology Database
Paleobiology Database data1.2. Fossil occurrence data from the Paleobiology Database (https://paleobiodb.org), released under CC0 1.0.
Creative Commons CC0 (Public Domain Dedication) 1.0 Universal
Where next
The four witnesses are easiest to tell apart on a real place, and the places page lists every one this site has looked at. To see a modelled map standing on its own, labelled as a model, open the Ancient Earth pages; to see a name that comes from a quotation while the map beside it stays unnamed, try the ancient oceans.