Environment
Microclimate offset
The station against the nearest National Weather Service report. The PUC sits in shade behind a restricted-airflow vent, so a persistent offset is expected — and the offset is the interesting part, not the absolute reading.
Air quality — relative only
A BME688 returns one resistance for total VOC content. There is no particulate channel, so this is not comparable to the PM2.5-driven figure in a weather app, and eCO2 is a model output rather than an NDIR measurement. Read it against its own baseline, never as an absolute.
Canopy phenology
Chlorophyll absorbs blue and red and transmits near-infrared, so the NIR-to-clear ratio steps up at leaf-out and collapses at leaf-fall. This only works because the station is shaded.
Detections against ambient noise
A raised noise floor suppresses detections. This is a confounder, not a finding — the correction belongs on counts before any temporal trend is read from them.
Known limitations
Read before interpreting anything above. These are not disclaimers — they are the difference between a reading and a claim.
eAQI, eVOC, eCO2 are relative only From a BME688, a metal-oxide gas sensor returning a single resistance for total VOC content. No particulate channel, so not comparable to a PM2.5-driven AQI; eCO2 is a model output, not an NDIR measurement. No absolute air-quality number is published here.
Light is uncalibrated The AS7341 sits behind a window in fixed geometry at home, so comparison over time is valid and absolute irradiance is never claimed. On outings the geometry changes, making trailside light a relative openness index at best.
Sensors lag They sit behind restricted-airflow vents, which smooths transitions. Frontal passages read late.
Temperature is enclosure temperature The BME688 sits inside a powered case and warms itself, so it reads consistently above outside air — often by several degrees on USB power. Compare it against its own history, or against the NWS offset above, rather than treating it as the temperature outdoors.
Counts are not abundance Totals reflect vocal activity, analyser sensitivity and ambient noise — not how many birds are present.
Confidence is not calibrated probability BirdNET confidence is a network output. Use it for thresholds, never for false-positive rates.
Ambient noise suppresses detections Apply the detectability correction before reading any temporal trend.
Lens fouling Pollen, dust and snow show as a sustained drop in clear-day maxima. A rolling 14-day drift check to flag it is not yet implemented.