Landseed · Closure

What has this field actually done?

We mapped every paper we could find in one corner of conservation science — counting animals from camera-trap photos — sorted them by the method each one uses, and counted what nobody has done yet.

6 of 8
ways of counting animals from camera traps have never been automated. Every study still measures by hand.
SCR · REM · N-mixture · REST · spatial-count · TTE
only CT-DS, index-RAI has automated work — 6 papers

That is the kind of statement this is built to make: specific, counted against a population you can see, and wrong in a way somebody could prove in a minute. Every claim, with what would disprove it →

The map

Every paper we found, by method and by what it contributes.

Each row is one way of estimating how many animals live somewhere. Each column is what a paper does with that method — builds it, tests it, uses it, or reviews it. Click any row to see the papers themselves.

Methodbuilds ittests ituses itreviews it papersautomatedlabelled right
SCR 6628596 160 0 77%
2014 builds it Spatial Capture-recapture
2014 reviews it Spatial Capture-Recapture
2013 reviews it Spatial Capture-Recapture
1992 builds it Capture-Recapture Models
method not named in the paper 1915296 69 1 100%
REM 146103 33 0 100% ?
CT-DS 16511· 32 5 33% ?
index-RAI 47105 27 1 57% ?
N-mixture 1161· 18 0 75% ?
REST 724· 13 0 100% ?
spatial-count 631· 10 0 100% ?
TTE 52·· 7 0 100% ?
“·” means we found nothing of that kind. “?” means that score rests on fewer than six examples and should not be read as a measurement.

One paper against the map

Where a single manuscript sits in everything above.

Automated Parameter Estimation for Camera Trap Density Models Using Computer Vision-Enhanced Distance Sampling

298
papers in the field using a named method
14
of them this paper cites
5%
of the field it engages with
6
methods it could be first to automate
Low engagement is not a criticism. A paper about one method has no obligation to cite a whole field. The number is here so that “you haven't engaged with the literature” becomes something a reader can check instead of an opinion.

How

Three steps — and no AI in the part that decides what exists.
01
Find the field. A written-down list of searches pulls in every indexed paper on the topic. Then anything cited by enough of those papers is pulled in too — which catches the classics that no search phrase names.
02
Follow the citations. Papers citing the same work are having the same conversation. Nothing here is compared to a search term, which is why it surfaces work you didn't know to look for: a search can only return what resembles your question, and a gap resembles nothing.
03
Sort, then count. Each paper is labelled with the method it uses and what it contributes. Counting labelled things gives a table where a zero means something real — nobody has done this — out of a number you can see.
Why a table and not a ranked list? Nobody can check whether a paper is correctly ranked seventh. Anyone can check whether a paper is about distance sampling. We built the ranked version, measured it twice, found it carried no signal, and threw it away.

How much to trust it

The limits, up front rather than in a footnote.
59%
of the field we can prove we found
79%
of works real authors cited that our method recovers
88%
of papers labelled with the right method
0
papers read in full — this is metadata, not reading
This is not a complete census, and we can say by how much. We built the collection one way, then went looking for it a completely different way. Of 140 papers that second search says belong here, we hold 83 — 59%. Both searches only see English-language papers with a DOI, so the true figure is lower still.
It does not read papers. Everything comes from titles, abstracts and citation links — no full text, no PDFs. So every statement is about what the literature contains, never about what any single paper says inside.
Nobody who knows this field has checked it yet. Every number was measured against held-out data or independent labellers. None of it has been reviewed by a working camera-trap ecologist — the largest caveat on this page.

Where the data comes from

Openly licensed throughout, including our own sources.
398
documents with a licence recorded
34%
freely republishable
9
publisher policies to read to clear the rest
crossref openopenalex opens2 restrictedunpaywall openanthropic-claude open
Naming a paper is not republishing it. A citation is a fact and can always be shared; reproducing a paper's text cannot. The build fails if any reaches this page.