eBird data will never be reliable for some species

My name is Tony. I am an eBird photo-review junkie. I have a problem. Is there a support group for my particular addiction?

I look at the recent photo feed nearly every day.

Oh, not as a whole. Yes, I do scrutinize the first 30 photos in the take, usually finding no identification mistakes (usually, but not always). However, what I am after in my dependence-derived daily dive into the search for misidentified subjects of photos submitted to eBird is focused on a very few species. One might say hyper-focused. For the most part, I review the photos of species that have proven to be difficult for the average Joe eBirder.

Quick! Which two-species ID conundrums come to mind when asked to name the most frequently-mistaken identifications from Canada through the United States and Mexico to some points farther south? I am certain that most experienced birders could guess the two combos I intended when devising this essay's title.

No, Trumpeter vs. Tundra is not a problem-child duo in my experience because I hate swans. I have never been great at them, so I do not look to parcel my agony over them to the various eBird reviewers that must deal with that headache {shudder}. Besides, those two species are a problem in a geographically relatively limited portion of my scope of relative expertise.

Okay, you got the next guess right... and it doesn't matter whether you guessed either the scaup dilemma or the problem duo that now involves species in different genera!

Sharp-shinned Hawk (Accipiter striatus) versus Cooper's Hawk (Astur cooperii)

For roughly three-quarters of the year, distinguishing between these two species of very similar forest/woodland accipitrine hawks is, seemingly, an unknown thorn in many eBirders' lives. "Unknown" because many eBirders confidently identify individuals to species, but then upload to their checklists good to great photos proving the identification was incorrect. Somewhat weirdly, the mistake is somewhat unidirectional, in that most ID mistakes of photographed birds identified as Sharp-shinned Hawk are actually Cooper's Hawks, but the reverse is far rarer. The subjects of photos submitted as Cooper's Hawk are, in general, more likely to be buteos than Sharp-shinned Hawk, specifically Red-shouldered and Broad-winged hawks. Northern Harrier is confused for Cooper's Hawk nearly as much as Sharp-shinned Hawk is, an aspect of hawk-identification I more than hinted at in my paper in Birding on harrier ID in North America, the paper in which I pointed out that Northern Harriers do not technically have white rumps, instead they have white uppertail coverts. (I would link to the paper at the American Birding Association, but it is behind a paywall.)

Although I have kept track of ID mistakes I have encountered while reviewing eBird photos, I have done so only in periods, so I do not know what my overall detected error rate in this duo is, but it is >7%, but less than <15%. During the aforementioned nine months of the year (September through May), I only very, very rarely have a day when I find no misidentified photos of either species.

You might ask, "What is the difference in summer?" In summer, Sharp-shinned Hawks are quieter or less... flamboyant, let's say, than their accipitrine cousins. In summer, they also disappear into areas in which few birders live and bird, with a huge percentage vanishing into the extensive boreal forest of Canada, an area hosting few resident eBirders. The species is also not nearly as readily detected in the breeding season as is Cooper's Hawk (personal observation), preferring more densely forested regions. The United States eBird data show this disappearance quite well. However, the summer in the Canadian eBird graph is not all that different from the US data, a similarity created by the Sharp-shinned's more retiring nature and habitat preference.

Greater Scaup (Aythya marila) versus Lesser Scaup (Aythya affinis)

As I noted for Sharp-shinned Hawk, above, in summer, most Greater Scaup disappear into the north country and become less available for most North American eBirders. Lesser Scaup, as a breeding species, has a much less expansive US range than does Cooper's Hawk, but it is still the nearly default scaup in the Lower 48 in summer. That means that I encounter misidentified subjects of scaup photos at that season much less frequently, a season when juvenile, worn adult female, or alternate-plumaged male Ring-necked Ducks become a minor confusion species for Lesser Scaup.

As for the accipitrine hawk duo, above, I check the recent photos of both scaup species nearly every day and, also like that example above, the error directionality is biased in favor of the "rarer" species. Perhaps more accurately, it is biased in favor of the more-wanted species. That is, I interpret the differences in error directionality of Lesser Scaup being more frequently misidentified as Greater Scaup than the reverse being due to desire. Conversely, many of the Greater Scaup images submitted as Lesser Scaup seem to come from areas in which there are relatively few birders and in which Greater Scaup are considered to be very rare, which may encourage unwarranted assumptions about the scaup in question.

As with the hawk example, I only very rarely have a review session in which I do not find Lesser Scaup misidentified as Greater Scaup. Conversely, I go many days at a stretch without finding Greater Scaup identified as Lesser Scaup. As I discussed in my July 2025 scaup-identification paper (in Birding, and behind the same paywall), the primary cause of scaup misidentification and the directionality of mistakes is sub-adult individuals with head shapes lacking the distinctive peaked crown of Lesser Scaup that many birders view as the primary distinguishing character from Greater Scaup. A secondary cause of misidentification is that at least some observers misunderstand the difference in wing pattern between the two species (and of which there is some slight overlap). Swimming scaup can often show variably extensive white on the wing, and this can (apparently) be interpreted as proving an identification of Greater Scaup. Unfortunately, the feather tract shown at those times is the secondaries, and both scaup have extensively bright white secondaries; there is no difference in that tract in the two species (Lesser, Greater).

However, the point of this essay is not the specifics of misidentification of scaup or of accipitrine hawks, but of the reliability of data housed within the eBird program. Yes, misidentification is a problem for all species, but, at least for species locally rare and for which the relevant eBird filter requires documentation, the review process catches most of the mistakes. Identification mistakes of locally common or regularly occurring species cannot be found and corrected unless observers upload photos of the bird(s). That means that some level of uncertainty exists for all eBird data. However, low or very low rates of identification errors for most such species do not provide a rationale for throwing the baby out with the bathwater; it is part of the noise expected in all estimations of biological populations. The problem is that those biological estimations depend on there being no undetected consistent, directional, and unmeasurable source of error.

Sure. One can go through scaup or accipitrine hawks' photos and measure those sources of error. However, those mistaken errors have already gone through two data-checking filters. First, in contrast to a sight record, the observer can study the photo(s) and, potentially, catch a field misidentification; the frequency with which this happens is an unknowable percentage of originally misidentified birds. Second, the eBird reviewers or others perusing uploaded photos can catch a misidentification. A third filter was added in April 2025: Merlin AI catches misidentified subjects when photos are uploaded. But what about sight records, which form the bulk of eBird data? These go straight into the database unless flagged as local or seasonal rarities, and there is no way to assess error rates in those data, although the rates must be higher, much higher, than that of reports supported by photos. I thus contend that scaup data and Sharp-shinned Hawk and Cooper's Hawk data in the areas of sympatry have unknowable and unmeasurable sources of error; of directional error of unknown proportion.

Although these examples are, far and away, the largest source of identification errors that I have detected in eBird photos, a few other species have high enough error rates in my experience that suggest those data are not reliable, not to be trusted to generate meaningful results except, possibly, at the grossest of scales, but also assuming that there is no change in percentage across time and space. However, if those error rates, those unknowable rates, change across time, then all bets are off. For example: If immature Lesser Scaup cause most of the problem in scaup ID, then in the time following summers of very low Lesser Scaup productivity (perhaps extreme drought in the prairie pothole region), one could guess that the percentage of misidentifications would be less than in years of greater reproduction. But at what magnitude?

Other species with high-enough photo-ID error rates as to cause great uncertainty as to the reliability of eBird data:

Spotted Sandpipers are frequently misidentified as Solitary Sandpipers (the reverse, much less so), and the problem is both directional and mostly seasonal, as it is unspotted Spotted Sandpipers that allow for the mistake.

One day, when reviewing Nashville Warbler photos in fall, I encountered less than a screen's worth of photos (i.e., ~20 photos) depicting Nashville Warblers, which included individual photos depicting individuals of five other species of warblers: Common Yellowthroat, American Redstart, Orange-crowned Warbler, Mourning Warbler, and one I do not recall. I had never before and have never since seen such a burst of errors, but the problem is persistent. However, it seems that if a warbler has eye rings or anything that can be construed as eye rings, the species will be misidentified at some point as Nashville Warbler. Although error rates are much lower than in scaup and accipitrine hawks, Common Yellowthroat, especially, is probably misidentified as Nashville Warbler frequently enough to cast doubt on the entire non-breeding data set of Nashville Warbler, again due to the directionality and unknown magnitude of the error source (I have only exceptionally encountered Nashville Warblers identified as Common Yellowthroats).

And then there is Passerellidae, peeps, gulls, terns, etc. I once encountered a Caspian Tern in a photo submitted as a Least Tern, and it was not due to a drag-and-drop error, as Caspian Tern was not listed on the checklist. Case closed.

Yes, eBird algorithms for detecting reliable observers may well find many observers that frequently make mistakes in identification and discard their data from consideration in analyses. However, observers who have had identification mistakes pointed out to them by eBird reviewers can easily distance themselves from those errors simply by deleting the species from the checklist and entering the correct species and uploading the photo(s), as opposed to the eBird-preferred data-correction method of moving the entry to the appropriate species. However, since eBird has not published (ever, as far as I can tell) the details of their data-selection criteria when reporting results for publication, there is no way for science to judge their published results, to know what the possible sources of error in those results might be.

{Thanks to Van Remsen for reviewing a previous draft of this essay.}