Mass production of unvouchered [no specimens'] records fails to represent global biodiversity patterns

Daru, Barnabas H. and Jordan Rodriguez (2023 May 1st)
Mass production of unvouchered records fails to represent global biodiversity patterns.
Nature Ecology and Evolution 7 816–831.
Publisher official publication page (abstract available, full text paywalled):
Full text freely available from the primary author’s lab web page:

• Barnabas H. Daru from Stanford University: .
– Department of Biology, Stanford University, Stanford, CA, USA .

• Jordan Rodriguez from the Department of Biology, Institute of Ecology and Evolution, University of Oregon, Eugene, USA.

Abstract, quotation: "
The ever-increasing human footprint even in very remote places on Earth has inspired efforts to document biodiversity vigorously in case organisms go extinct.
However, the data commonly gathered come from either primary voucher specimens in a natural history collection or from direct field observations that are not traceable to tangible material in a museum or herbarium.
Although both datasets are crucial for assessing how anthropogenic drivers affect biodiversity, they have widespread coverage gaps and biases that may render them inefficient in representing patterns of biodiversity.
Using a large global dataset of around 1.9 billion occurrence records of terrestrial plants, butterflies, amphibians, birds, reptiles and mammals, we quantify coverage and biases of expected biodiversity patterns by voucher and observation records.
We show that the mass production of observation records does not lead to higher coverage of expected biodiversity patterns but is disproportionately biased toward certain regions, clades, functional traits and time periods.
Such coverage patterns are driven by the ease of accessibility to air and ground transportation, level of security and extent of human modification at each sampling site.
Conversely, voucher records are vastly infrequent in occurrence data but in the few places where they are sampled, showed relative congruence with expected biodiversity patterns for all dimensions.
The differences in coverage and bias by voucher and observation records have important implications on the utility of these records for research in ecology, evolution and conservation research.

I read this whole scholarly article when it was first published 1st May (2023).
First hand I notice the evidence for the same general points made in this article of serious problematic biases, in various online systems including this .

Within my professional work in 1994 I carefully and closely read this internationally published, high quality scholarly article reference :

• Margules Chris R. and Mike P. Austin (1994)
Biological models for monitoring species decline: the construction and use of data bases.
Philosphical Transactions of the Royal Society of London B34469: 75

Many more scholarly articles in this topic area I professionally read at this time too, and continue scholarly reading this topic ever since (when time).
• Chris Margules :
• Mike P. Austin – then CSIRO :
Etcetera .

This is really the typical “we need to kill it to be real science” approach that biologist so absurdly push. Obviously, the problem is not whether the observation is “vouchered” (what a fantastic euphemism for killing) but whether the observation was conducted on purpose to study the biology in a specific area, or for fun. That’s what causes the biases for accessibility etc…


What is your aim here?


I would disagree with this, though you should probably know that I’m a professional biologist. but… my work has been enabled in large part because of “vouchered” collections, usually ones that don’t kill the organism (i.e. a branch clipped from a tree). my study system (tarflowers) grow in a lot of inaccessible areas, or just areas that people don’t visit very much; historical collections have made the impossible possible for me in terms of sampling and simply being able to observe variation myself.
I think it’s pretty obvious that there is, and necessarily would be, massive bias towards easily observed organisms in easily visited places, when so much of our biodiversity data comes from platforms like iNaturalist. on the contrary, it speaks just to the need for more such data and for more people to take it. it won’t eliminate those biases (and others I’m not going to speak to right now), but it would help.
basically, there are a vast number of studies and realizations that have been enabled by voucher collections – and the knee-jerk response of “stop killing” is both false and inappropriate for many taxa. even with groups like birds that have been traditionally shot to produce “specimens”, a modern “voucher” might include blood and feather samples (paired with photos!) from a bird that’s been released again. with groups like trees and grasses, it’s trivial to take a voucher that doesn’t kill the organism. and don’t get me started on all the insects I’ve collected that were already dead on the ground, and would just have been stepped on by people otherwise.
accessibility is a whole other topic, probably one that plays into what I see as necessary mitigation of observation biases per above, but I’ve said my piece on vouchering now

1 Like

It’s too bad the article doesn’t discuss photos/sounds (only one return for a seach of the word ‘photo’. Photographs don’t provide the same benefits of a herbarium sample for instance, but they also have a lot of advantages as anyone can take them, for some species they can be very much diagnostic, they provide phenology data, they are almost always legal (unlike taking samples), they take much less money to store and curate, and they offer so many benefits over the likely alternative (field data in databases and notebooks with no way to verify at all). Yes it’s nice to have herbarium samples, but it’s very much possible to photo document 100+ research grade observations of plants on iNat in a day, and getting 100 herbarium samples per day per person is not usually viable and even if it was, there’s no way to store lots of people doing that. And while herbarium samples usually don’t cause much impact on populations and often don’t even kill the individual, that doesn’t hold true for animal collections. And some taxa like fungi are just hard to preserve in general.


This doesn’t seem surprising. Citizen scientists making observations on inaturalist generally don’t, and shouldn’t, have access to sensitive areas where they could cause damage to ecosystems. The more access there is to an area, the greater a chance that weeds will be introduced, native wildlife will be frightened or killed, and poaching will take place. Plus, it takes training to even notice some organisms, and people aren’t going to take photos of literally every wee thing on their woodland strolls.

Quite simply, using inaturalist observations to “represent global biodiversity patterns” would be foolish, and that isn’t how researchers are using them. It’s often for phenotypic, temporal, and presence/absence stuff.

Incidentally, while I took voucher specimens when I was an undergrad research assistant, now that I volunteer with “rare” plants, my instructions are to never take vouchers unless explicitly instructed to do so. It isn’t needed for the work they want done.


What is the “this” you are referring to here? Because the article doesn’t appear to be making that point at all. Rather, it is documenting the different sampling biases in vouchered (museum) records vs unvouchered observations (such as iNat and similar databases). It doesn’t argue that observations are not scientific, although it does note some of the applications that require vouchers.

The recommendations it makes are along the lines of improving/extending both vouchered surveys and unvouchered observations to address gaps in biodiversity documentation. Which is quite sensible (and not really surprising).

The title is a bit inflammatory, but otherwise this looks like an interesting analysis. I’ve only read it quickly though.


There is the strong tendency among scientists (I am one) to want to document anything that hasn’t been quantitatively documented before, even if it is obvious without being quantified. This is good and despite appearance, useful, in part because sometimes what seems obvious is wrong. However, this is generally accompanied by the decision to act like the obvious thing we are documenting was unexpected, in order to convince other scientists that one’s work is revelatory, fundable, and worthy of high status. This is useful to the individual researchers, but often results in ‘discoveries’ that were never in question.


I suspect this may have a lot to do with the difficulty of identifying many species without a voucher specimen. I personally take only photos of most taxa, but I have a specimen collection of Lepidoptera specifically, and I see this sort of trend very commonly- the taxa I’m photographing often get stuck at the genus or family level on iNat with no hope of making it to species, while I’m able to perform microdissections of my Lep specimens and take them all the way to species. And lo and behold, they very often turn out to be species with no or very few records at all on iNat. Even the common ones that people are photographing all the time simply can’t be identified from live photos alone in lots of cases, which leads to large chunks of biodiversity going unreported. For example, I’m sure a thousand people have photographed Pyla impostor at their lights all over the western USA, but without killing and dissecting a specimen, it’s literally impossible to know that that’s what they are; too many others are externally identical. So they sit at “Phycitinae sp.” forever. iNat currently has only 3 USA records of it, yet collectors can vouch for it occurring practically everywhere in the Rockies that has spruce trees. The number of common Lep species that I’ve submitted “state first” iNat records for is ludicrous, and it’s 100% because so few people are actually dissecting them to put names on them.

Here’s another example, the genus Chionodes in my state of Pennsylvania:
The two species with the top number of observations are not the most common ones in the genus in the state by a long stretch, but they’re the only two that are easily identified in live photos. There are also plenty more species here that aren’t reported at all because they’re never identifiable in photos. If you looked at a specimen collection of PA Chionodes and then at the iNat records of Chionodes, you’d come away with wildly different views of what the common species are and what the diversity looks like.

If you don’t want to kill the bugs, that’s fine; I get it, and I don’t expect everyone to be doing so. I don’t kill all the other taxa that I’d need to dissect to identify either. But even in an era of big data and millions of voucher photos, there’s something to be said for the value of collecting specimens, despite how in vogue it is to accuse specimen collectors of being backwards and murderous. And I think this is just another piece of evidence that there’s a benefit to the data collected from specimens. That doesn’t mean there isn’t value to photographic records too; clearly plenty of good science happens based on iNat photo records all the time. Anyone asserting either that photo records are useless or that specimen collecting is unnecessary and outdated is just ignorant of lots of the current scientific studies going on. And far too often, it seems we fall into these camps and dismiss everyone who collects data differently from us as “doing it wrong”, which is just sad.


I don’t think collecting is what causes the studies with vouchers to cover the earth more completely. Instead, natural history projects that have the money and institutional support to visit remote areas are usually projects that include collection of specimens. Amateurs and professionals who don’t collect are more likely to be people who don’t have the resources to get the middle of the Congo, for example. This isn’t too surprising. Pointing out the difference has value because it’s good to know what uses different kinds of data can be put to.


The limitations of iNat data has been discussed on this forum at length and the general conclusions of this article match the consensus in those discussions. There are no surprises in their analyses. The interesting part is the actual quantification of bias and discussion of what that means for research.

1 Like

High quality and nuanced, evidence, based, scholarly critical thinking, discussions and depth of humility.

Your aim here?

Video by Barnabas H. Daru (Stanford University) the primary scholar–author :
(eg. for more visual emphasis learners here):

Uhmmm… oh, never mind.

1 Like

Yeah! not false humility nor pretentious humility (of course nor over-confidence, nor arrogance, nor fallacious delusional God–eye–view humanly–impossible moral judgementalism ).

Made an account specifically to reply to this.

As a researcher, I’ve both used and seen inat used by others for various purposes. It’s generally NOT to “represent global diversity patterns”, however. I use it primarily to locate additional populations of poorly-documented bamboos in Bolivia so that I can travel to these locations and search for them. There just aren’t that many herbarium specimens, and many of them are missing important parts, such as culm leaves. I’ve seen other use it to examine phenology and phenotypic variation across the range of a species. Some characteristics don’t hold up well in voucher specimens. For example, in pinesaps, the plants turn black when they dry. In a recent talk at the Botanical Society of America conference, researchers talked about how they used inaturalist to get an idea of color variation and its relation to bloom time. Later collections used for phylogenetic analysis suggest pinesap is actually several species that vary by color and bloom time. Without inaturalist, it would have been much harder for these researchers, as, again, vouchers just don’t show color. I also have friends who have used it to locate previously unknown populations of rare insects. inaturalist does seem like a really useful tool, just not for what the researchers were doing here. I do of course make voucher specimens of the plants that I’m working with in the field. But vouchers aren’t necessary for records to be “useful”.


On the one hand, the conclusion about the irrelevance of observational data to the objective biodiversity distribution seems to me quite correct, although rather obvious and trivial. Of course, data from iNat and similar platforms have significant sampling bias. And this should be taken into account when using these data.

On the other hand, the second main thesis of the paper seems to me to have little validity. The authors compare patterns of observations with some “expected native biodiversity patterns”. Although there seems to me to be some contradiction in the very combination of the words expected (subjective expectation) and native (objective states).

But the more important question is what the authors’ “expectation” is based on. It is clear from the relevant section of the article that it is based on data from catalogs and atlases of floras and faunas. But what are these data themselves based on? I would even formulate it differently: what else can they be based on, except voucher records?

So, in my critical opinion, the only thing reliably shown in the article is that the collections data more correctly show patterns based on those same collections than other data. Unexpected, isn’t it? [I apologize, but that was sarcasm] But it’s an open question how relevant these “expected patterns” themselves are to objective patterns of biodiversity. I don’t see a clear answer to that in the article.

That said, the article does not discuss sampling bias arising from the collection of voucher specimens. Although it is also significant and well known to any experienced researcher. Moreover, the article has a reference to a paper by the first author that shows sampling bias in herbarium specimens.

So the authors’ conclusions seem at least a little contradictory to me. And certainly cannot be evidence that iNat data are useless or distort biodiversity patterns. They don’t represent them exactly, that’s true - as well as voucher records. And in some groups of organisms, they don’t represent them at all. But if the researcher doesn’t realize this, that seems to me to be the issue of the researcher, not the data or its source. When used knowingly and correctly, observational data are incredibly valuable, and sometimes there is simply no substitute for voucher records.


Good point, if you use vouchers as the gold standard of species distribution, you can’t be surprised if vouchers continue to show the same distribution patterns. To be honest i suspect the truth is somewhere in between. Vouchers focus on conserved land and known biodiversity hotspots, whereas ‘private’ land, areas which receive little scientific attention, and urban areas will be underrepresented in the vouchers. You can’t really say much about distributions and populations without systematic (like a grid) or randomized sampling. Sometimes that occurs but it doesn’t seem to be what comprises the majority of vouchers.


Yes, of course – a more or less “natural” pattern can only be obtained by systematic non-selective study of biodiversity (grid and so on). It is also obvious that the issue is that such projects require a lot of resources and time. And are only realized for some groups in a limited area. For example, the Atlas Florae Europaeae project has been going on for decades and remains incomplete. As far as I know, there are similar projects for birds. But most groups and areas are completely uncovered by such studies. And I doubt that this may change soon.

So the conclusions of the article would be more valid and objective if voucher data vs. grid mapping data vs. observation data were compared. But then the article would not have had such a global scope. And probably would not have been published in a journal of such a level. But the published comparison looks (to me) like voucher data vs. “expected pattern” vs. observation data, where the standard for comparison is likely to be knowingly similar to one of the sets being compared. At least IMO, without evidence to the contrary (that all three sets are independent and none are derived from another) the results of the comparison cannot be a proof of anything. I see no such evidence in the article.


Floras and faunas are based on expert assessment of records (vouchered or not, although historically mostly vouchered). This is not just semantics, the application of expert judgment is critical.

All records, vouchered and otherwise, are biased. Taxon experts have the training and experience to evaluate these biases, and correct for them in their monographs. Range maps are usually presented as more-or-less continuous areas on a map, despite the underlying data being point records. The expert’s judgment allows them to fill in the unsampled gaps, based on their understanding of the environmental conditions in the gap, and the biological requirements of the taxon in question.

Species A may only have a handful of records from remote boreal forest locations, but their distribution map includes large areas without records that are geographically contiguous and ecologically similar. Species B may have records relatively close together, but the distribution map excludes ‘islands’ of unsuitable habitat (high elevations in the mountains, unsuitable bedrock etc).

This isn’t a perfect system of course, but it’s not so clearly circular as you suggest.