Radar tracks were generated using the MAX Avian Radar, while visual observations were collected by bird spotters. All data comes from Windpark Eemshaven in the province of Groningen.

## Files
The dataset consists out of 2 files: **train_data.csv** and **test_data.csv**. Each CSV contains two types of information: radar-derived data and observation-related data. Both files include the following radar-related columns:

- `track_id` — The unique identifier for each radar track
- `timestamp_start_radar_utc` and `timestamp_end_radar_utc` — The start and end time of the radar track in UTC
- `trajectory` — The actual radar track encoded in EWKB (Extended Well-Known Binary) Hex. Data consists out of a series of measurements containing 4 elements in order: `Longitude`, `Latitude`, `Altitude` (m), `Radar Cross Section` (dB/m2)
- `trajectory_time` — A list of floating numbers indicating the elapsed time in seconds since `timestamp_start_radar_utc`  a measurement in the `trajectory` was made
- `radar_bird_size` — Which size of bird the radar track could belong to (Large, Medium, Small bird, Flock) as output from the MAX Avian Radar
- `airspeed` — Average airspeed of the radar track (m/s)
- `min_z` — Minimum altitude of the radar track compared to the height of the radar (m)
- `max_z` — Maximum altitude of the radar track compared to the height of the radar (m)

Then, **train_data.csv** also includes observation-based columns used to validate the radar tracks. These are privileged columns, which are only available during training-time. The target labels for bird group classification are in the column `bird_group` . The following columns are included:
- `observation_id` — The unique identifier of each observation
- `primary_observation_id` — The unique identifier of each unique observation. (The same bird (flock) can consist of multiple radar tracks)
- `observer_position` — UPDATED: Most recent bird position in the radar track when the observer clicked the button to enter the species label to the track, encoded in EWKB. ~~The location of the observer encoded in EWKB (Extended Well-Known Binary) Hex as Longitude / Latitude / Altitude~~
- `observer_comment` — A comment left by the observer regarding the observation
- `n_birds_observed` — The amount of birds observed
- `bird_group` — The bird group of the radar track
- `bird_species` — The exact species of the radar track

Finally, a **sample_submission.csv** is provided. The sample submission has the following format:
```
track_id,Clutter,Cormorants,Pigeons,Ducks,Geese,Gulls,Birds of Prey,Waders,Songbirds
526781579,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0
526782173,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0
526782843,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0
526786562,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0
526787997,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0
526788673,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0
526788796,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0
526788552,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0
```

Here, each row corresponds to a `track_id`  (the unique identifier for each radar track in the test set) with a score between 0-1 for the class `Clutter` and each of the bird groups.