← Deployment 1·← Deployment 2·Species list·How the model works

How these surveys were run

Each report on this site is one deployment: a stretch of scheduled recording from a microphone on a deck in Kirkland, WA, scored afterwards in a batch. This page is the equipment and the settings behind all of them, so no report has to repeat them.

Recording

An AudioMoth running as a standalone SD-card logger — no network, no live classification. It wakes for its scheduled windows, writes a WAV per window, and sleeps.

Recorder
AudioMoth · Basic firmware 1.12.1 · one unit across every deployment
Audio
48 kHz mono · gain medium
Timestamps
Written in UTC by the recorder; every chart on this site converts to America/Los_Angeles

Scoring

Recordings are analysed in a batch afterwards with BirdNET-Analyzer. What a detection is, and what it does not establish, is its own page.

Analyzer
BirdNET-Analyzer 2.4.0 · GLOBAL 6K v2.4 checkpoint
Threshold
Minimum confidence 0.25
Candidates
The 260 species plausible here across the year

Every deployment uses these same settings, which is what makes the runs' numbers comparable at all.

What each report shows you

The audio clip on every charted species is that species' highest-confidence detection, cut straight from the source WAV — so any identification on the site can be checked by ear. Photographs are reference images from Wikimedia Commons, not pictures of the birds recorded here.

What changed between deployments

Amber marks a value that is not the same across runs.

Deployment 1Deployment 2
DatesJul 23–24Jul 24–Aug 2
Scheduled window (UTC)11:00–15:00, 23:00–06:0012:00–04:00
Same, in Pacific4a–8a, 4p–11p5a–9p
Hours of audio11 h137 h
48 Hz DC blocking filteroffon
Detections3,70351,282

One capture setting was not held constant. The AudioMoth's 48 Hz DC blocking filter — a high-pass that removes the sub-audible offset and rumble below roughly the bottom of the piano — was off in deployment 1 and on in deployment 2. It sits far below the frequencies BirdNET keys on, so the effect on identifications should be small, but it was not measured. Treat the two runs' species lists as directly comparable and their detection rates as only roughly so.

The scheduled windows do not line up either, which is why the two reports keep their own hour-of-day charts rather than pooling into one: only the dawn and evening hours were recorded by both, and an hour backed by a single morning does not belong on the same axis as one backed by nine.