# Building a Reliable Citizen-Science Bird Survey

Citizen-science bird surveys can produce valuable ecological data, but only when observations are collected consistently. Our survey asks volunteers to complete a 10-minute stationary count at a registered location, recording every bird seen or heard within 100 metres. Each submission includes the species, count, observation time, weather conditions, and an optional confidence rating.

## Designing the Field Protocol

A fixed-duration protocol reduces variation between observers and makes results easier to compare across sites. Surveys begin within three hours of sunrise, when most species are active, and are postponed during heavy rain or strong wind. Volunteers record all visits, including surveys where no birds are detected, because zero-count data is essential for estimating occupancy and detection probability.

## Capturing Structured Data

Observations are uploaded as CSV files or entered through a mobile form. The schema uses stable species identifiers from a published taxonomy rather than common names, which vary by region and language. A minimal project configuration looks like this:

```yaml
survey_duration_minutes: 10
observation_radius_metres: 100
required_fields: [site_id, started_at, species_code, count]
taxonomy_version: "2026.1"
```

## Automated Quality Checks

The ingestion pipeline rejects impossible timestamps, negative counts, unknown site identifiers, and species codes absent from the configured taxonomy. It also flags unusual records—such as 500 individuals of a typically solitary species—for expert review without deleting them. Duplicate detection compares observer, site, start time, and species, while preserving legitimate repeated surveys conducted on different dates.

## Accounting for Observer Bias

Experience, weather, habitat, and background noise all affect whether a bird is detected. We therefore store survey-level covariates and ask participants to report their experience using broad categories. Analysts can incorporate these variables into occupancy or abundance models, separating changes in detectability from genuine changes in bird populations.

## Publishing Reproducible Results

Validated records are released with versioned metadata, protocol documentation, and a machine-readable change log. Precise coordinates for sensitive species are generalized before publication, and volunteer identifiers are replaced with anonymous codes. This approach keeps the dataset useful for researchers while protecting wildlife and participant privacy.