Worker sightings are noise

iNaturalist holds millions of ant observations, each with a photo, a date and a pin on the map. If you want to know when ants fly in, say, New Zealand, the obvious move practically suggests itself: pull every ant record, bucket them by month, and read the flight season off the histogram. Cheap, global, self-updating. We'd have loved for it to work.

It measures the wrong species' behaviour. Ours.

Nearly every ant a person photographs is a worker: wingless, sterile, and grounded for life. Workers are out on the pavement in March and in July alike, whenever the colony forages, which is most of the year in most places. So a histogram of ant records by month is a histogram of when observers and workers were both outdoors, which mostly tracks pleasant weather and school holidays. The flight season is in there somewhere, buried under a hundred times its own weight in lunch runs.

A dated worker photo does prove something, and something genuinely useful: a colony of that species lives near that pin. Presence is real evidence, and it's one-way. A record can confirm a species lives near you; a missing record can't prove it doesn't, since most colonies go unphotographed forever. That one-way rule is baked into how the forecast treats occurrence data.

But for the question that actually drives a flight forecast, when does this species send up its queens, a worker record carries nothing at all. The date on it belongs to the observer.

What would answer it is records of the winged castes, queens and males, photographed on the day they flew. Those exist, and iNaturalist even has a field for them. When we measured against the live iNaturalist API in August 2026, ant observations annotated as male broke down like this: about 11,500 research grade worldwide, another 42,000 stuck at "needs ID", where the community agrees it's a winged male but can't push the identification to species. Most of that second pile is genus-level agreement, entirely usable if you aggregate by genus, which is why our harvest takes both.

Then the numbers get thin fast as you leave the well-recorded world. New Zealand, a whole country with a real flying ant season, holds 62 research-grade male ant observations. Ever. Add the needs-ID pile and you reach about 250. You cannot date a national season for every genus from 250 photos, and we don't pretend to; the NZ pages say plainly which windows are estimates.

This isn't hypothetical for us. A reader near Atlanta wrote in because the app showed strong flight odds for species that barely occur there, and their diagnosis was right: stray occurrence records, a single photograph in some cases, had put species on the local list with the same standing as neighbours holding hundreds of records, and one stray could set the day's headline. The fix shipped in July 2026, and it was an honesty fix more than a modelling one: every species row now shows its nearby record count, thinly-recorded species are flagged as such, and a flagged stray no longer speaks for your whole day.

That's the whole recipe for a record that can date a flight, and it's exactly what the best datasets in this field share. Britain's Flying Ant Survey collected 13,394 analysable records by asking the public to report the swarms themselves, dated and timed, which is how Hart et al. (2018) could test weather triggers against a national picture. A century of museum specimens works for the same reason: a pinned queen carries her collection date, which is how Helms (2022) could show US mating dates shifting with latitude and elevation.

Sighting reports made in Dealate are shaped the same way, by construction: what you saw, when, where. A confirmed sighting near you feeds the same day's outlook, which is a thing no pile of undated worker photos, however vast, could ever safely do.

None of this is a complaint about iNaturalist, which is a marvel, and whose winged-ant records we lean on gratefully. It's a caution about a shortcut. The firehose has a date on every drop, and almost none of those dates mean what you need them to mean.

Hart, A.G., Hesselberg, T., Nesbit, R. & Goodenough, A.E. (2018). The spatial distribution and environmental triggers of ant mating flights: using citizen-science data to reveal national patterns. Ecography 41(6): 877 to 888. Source of the 13,394-record figure and the model for caste-aware, dated citizen science.

Helms, J.A. (2022). Climate, geography, and the mating phenology of ants. Insectes Sociaux 70(1): 119 to 125. Source for dated museum specimens resolving mating phenology.

The iNaturalist male-annotation counts were measured by us against the live iNaturalist API on 30 August 2026, across quality grades, worldwide and for New Zealand.