Constarium
← Search

Data · dataset · 2025

Passive acoustic monitoring observations of birds in Ukrainian Roztochya

Listed in GBIF

This dataset contains bird observations collected through passive acoustic monitoring in the Ukrainian Roztochya region.

Description

Data were recorded using autonomous acoustic recorders and analyzed using BirdNET, a deep learning-based species identification model. To ensure data quality, automated detections were randomly sampled and verified by human experts.

An optimal confidence threshold for species identification was determined using logistic regression following the methodology of Wood and Kahl (2024; doi.org/10.1007/s10336-024-02144-5 ). The aim is to provide observations that have confidence greater than or equal to the threshold, ensuring at least 95% identification accuracy. If a particular observation is only automatically identified (i.e., without human verification) but filtered according to the approach described above, it receives an 'unverified' status in the identificationVerificationStatus column.

Read the rest (4 more)

BirdNET confidence is indicated in the identificationRemarks column. Some records included in this dataset have been quality-controlled and represent verified presence observations. Such observations receive a 'verified' status in the identificationVerificationStatus column and the name(s) of the identifier(s) in the identifiedBy column.

Additionally, the identificationRemarks column includes a comment such as 'Confirmed by human expert(s). Original confidence: 0.38'. We do not upload all available observations to avoid 'duplication', i.e., hundreds of detections of the same species from a single day-location pair.

Observations are therefore aggregated to 1-hour intervals per location. This means that each of the 24 hours in a day from the same location may contain an observation of a particular species. Only one observation from a given hour is selected - the one with the highest confidence level (above threshold), or if there is an observation verified by a human expert, it is automatically preferred.

We think that hourly data is more useful in case somebody is interested in daily activity patterns. The full dataset can be viewed through an interactive tool - yurastrus.dev/en/pam .

Links

Topics

From keywords
Life Sciences
Provenance · 1 source records, 10 field assertions
SourceKeyLast seenRaw
GBIFaf2b5fb0-3177-4df2-9a5e-2d776ef5f0a410 d agoJSON v1
FieldAssertionExtractorEvidence
access_levelsource · gbifconnector:gbif@1.0.0
concepts[field].local:field:life-sciencesmapping · gbifconnector:gbif@1.0.0
created_datesource · gbifconnector:gbif@1.0.0
descriptionsource · gbifconnector:gbif@1.0.0/description
licensesource · gbifconnector:gbif@1.0.0
publication_datesource · gbifconnector:gbif@1.0.0
spatialsource · gbifconnector:gbif@1.0.0
temporalsource · gbifconnector:gbif@1.0.0
titlesource · gbifconnector:gbif@1.0.0/title
updated_datesource · gbifconnector:gbif@1.0.0