HULA EARTH / BIRD HABITAT HACKATHON
Build a model that predicts bird habitat suitability beyond Germany. Submit your predictions and explain your results. If you finish early, explore an optional extension.
Registration and preparation
THE MAIN CHALLENGE / 01
Task 1 is the main challenge for every team. Focus on a working model and a complete submission.
Train on German observations with AlphaEarth, WorldCover, terrain and historical OpenStreetMap features. Rank relative suitability for ten species at anonymous locations in neighboring Central Europe. Run inference on your own compute and submit predictions.
Start these only once your main challenge and presentation are ready. Choose either extension if you have time; both are optional.
Optional: choose an additional question about the German bird collection and explore it through two or three clear visualizations. Explain coverage, sampling and uncertainty.
Optional: test whether information learned from other birds improves your model. Compare with and without these signals on the same held-out data. A well-supported negative result is useful evidence too.
The benchmark ranks reported seasonal presence against matched recording opportunities. Backgrounds are not confirmed absences, and a high score does not establish occupancy or breeding.
RESOURCES / 02
A compact modeling package gets you running. Optional spatial inputs preserve both fine habitat detail and landscape context.
The supplied baseline runs on a CPU. Get a complete prediction pipeline working first, then use your remaining time for focused improvements and a clear explanation.
Choose your inputs →TIMING / 03
Download encrypted test inputs in advance. The key arrives at 13:30 for the 14:00 mentor round. Final predictions and presentation PDFs close at 17:30.
All times: Europe/Berlin, CEST / MESZ (UTC+02:00). Work: 11:00–17:30 (6.5 hours).
| German time | Program |
|---|---|
| 10:00 | Registration and coffee; starter code and participant guide open at 10:00. Environment checks only until 11:00. |
| 10:15 | Welcome and AI Nation introduction (Juliana Carneiro da Cunha Baumgartl, about 5 minutes). |
| 10:25 | Challenge brief by Hula Earth and Ames (maximum 10 minutes each), including the datasets. |
| 10:45 | Team formation, challenge/category confirmation and mentor introductions. |
| 11:00 | Hackathon modeling and analysis begin. Teams raise a hand for mentor help. |
| 13:00 | Lunch in the room; no fixed work break. |
| 13:30 | Test-key release via the Data page; decrypt preloaded inputs and begin final inference/submissions. |
| 14:00 | Mentor round: every team gets a check-in, including data loading, inference and submission questions. |
| 16:30 | Feature-freeze checkpoint and second mentor round; focus on final inference, figures, slides and delivery. |
| 17:30 | Deadline for predictions, nomination, method/code reference and presentation PDF. |
| 17:45-19:45 | Presentations: 15 teams, 5 minutes + 2 minutes Q&A. |
| 19:45-20:00 | Jury deliberation and final reconciliation. |
| 20:00-20:15 | Awards ceremony. |
| 20:15-20:45 | Dinner and networking. |
ASSESSMENT / 04
Your main modeling work can provide the insights and visualizations for judging: explain your experiments, model behavior and limitations. Tasks 2 and 3 are optional extensions, with no separate requirement or bonus. Any extension work is assessed within the same categories. The provisional public leaderboard is feedback; final performance uses your starred submission on the private split.