Methods and data

What this device hears, what the numbers mean, and what they do not.

What the PUC hears

A BirdWeather PUC is mounted on the pavilion at Cherrylake in Groveland, Florida. It listens continuously and sends short audio segments to BirdNET, a machine-learning model trained to identify birds by sound. When BirdNET is confident enough, it records a detection.

Everything on this site comes from that one device, in that one spot.

Why detections are not abundance

A detection is one identification, not one bird. A single mockingbird singing for an hour produces many detections. A quiet bird passing once produces one, or none.

So the counts here describe what this microphone heard, not how many birds are present and not which species is most common. Loud, frequent, daytime callers are over-represented. Comparing two species' counts is not meaningful.

Coverage, and the gap

The device has been listening since 2026-06-02. Its most recent detection was 2026-09-07. It has recorded 97 species across 86 days.

It has not yet heard a winter. A species missing from this data was not detected by this device in that window. That is not the same as being absent from the site, and it is certainly not the same as being absent from Lake County.

Known limitations

BirdNET misidentifies. The station has recorded a Clapper Rail once. Clapper Rails are a saltmarsh species; this is almost certainly a King Rail, which is freshwater. No identification here has been reviewed by an ornithologist.
Rare one-off detections are the least reliable. A species heard once is far more likely to be a misidentification than a species heard on many separate days.
The quiz only uses species with at least 5 detections on at least 3 separate days. That threshold exists so children are not taught a misidentification as a fact.
Nothing here shows that the landscape caused the bird activity. This device records what it hears. It cannot attribute a cause.

Download the data

species-days.csv species.csv

Both files begin with comment lines (prefixed #) carrying the method and limits statement above. Most tools skip them; please do not strip them when passing the data on.

Attribution is required: data from BirdWeather, identifications by BirdNET.

Sources and licences

What	Source	Licence
Detections	BirdWeather API	Attribution to BirdWeather
Identification	BirdNET	Attribution to BirdNET
Bird photographs and text	Wikipedia / Wikimedia Commons	CC BY-SA, credited on each image
Reference recordings	xeno-canto	Mostly CC BY-NC, credited on each recording
Bird diet and foraging traits	EltonTraits 1.0 (Wilman et al. 2014, Ecology 95:2027)	Published dataset
Bird habitat and taxonomy	AVONET (Tobias et al. 2022, Ecology Letters 25:581-597)	CC BY
Caterpillar and specialist-bee counts	NWF / Tallamy keystone plants, Ecoregion 8	Counts are per plant genus, not per species
Plant functions and sources	158 sourced claims, principally UF/IFAS, the Florida Wildflower Foundation and FNPS	Cited individually

How the plant section is written

It is not written. Each bird's plant section is generated by rule from published trait columns: what the bird eats, which plants supply that, and which of those plants are actually growing here. No sentence about any particular bird was composed by hand.

Sections are labelled by what kind of statement they are. Documented means it comes from a published dataset or a cited source. Reasoning means it is an inference joining two documented facts — likely, but not observed. Observed at Pavilion Garden means someone watched it happen and wrote it down; there is currently one such record.

Where a bird has no sourced plant connection, the section is empty rather than filled.

Every label on this site describes what this one device has heard since 2 June, not what the species does in the world.