# Building a Repeatable Pipeline for a Citizen-Science Bird Survey

## Field Protocol

Our spring bird survey combined volunteer observations with a lightweight QA process to produce data that local conservation planners could actually use. Each participant walked a fixed 1 km transect between 06:00 and 09:00, recording every bird seen or heard within a 50 m band. We kept the protocol intentionally simple: no playback, no duplicate counts for birds clearly moving with the observer, and weather recorded before each walk.

## Data Capture

Observations were submitted through a mobile form that wrote directly to a shared CSV export. The most important design choice was requiring structured fields for species code, count, detection type, latitude, longitude, and effort duration. Free-text notes were still allowed, but they were treated as supporting evidence rather than the primary data source.

```yaml
survey:
  transect_length_m: 1000
  survey_window: "06:00-09:00"
  max_wind_beaufort: 3
  required_fields: [species_code, count, detection_type, lat, lon, observer_id]
```

## Validation

Before analysis, we ran automated checks for impossible coordinates, unusually high counts, missing effort time, and species reported outside their expected seasonal range. Flagged records were not deleted; instead, they were routed back to regional reviewers who could confirm the sighting, request more detail, or mark it as uncertain. This kept the pipeline auditable while avoiding silent data loss.

## Results

Across 84 completed transects, volunteers recorded 6,421 individual birds from 112 species. The strongest habitat signal came from wetland-adjacent routes, where Red-winged Blackbird, Marsh Wren, and Common Yellowthroat detections were consistently higher than on upland routes. Urban routes showed lower richness overall but produced repeated observations of nesting Chimney Swifts near older brick buildings.

## Lessons Learned

The survey worked because the technical system supported the field protocol instead of replacing it. A modest schema, repeatable validation rules, and clear reviewer notes were enough to turn volunteer sightings into a defensible dataset. For next season, we plan to add offline form caching and automated effort maps so coordinators can spot under-sampled areas before the survey window closes.