Wind turbines pose a well-documented risk to bird populations, particularly in regions with high bird activity or along major migration corridors. To mitigate this risk, radar systems are being used to keep track of the bird presence near the wind farms. These systems collect thousands of data points each day, which are currently interpreted manually to support operational decisions, including whether turbines should be temporarily shut down.

Seasonal turbine curtailment has become a relatively common mitigation measure, especially during peak migration periods around March and September, and has already prevented the deaths of hundreds of birds. However, as reliance on wind energy grows and wind farms continue to scale in size and capacity, shutdown criteria are becoming progressively stricter. This is further complicated by the fact that radar data provides no information about bird species, leaving operators with limited context and forcing conservative, often costly, decisions.

TNO, Waardenburg Ecology and RWE identified this limitation and responded by collecting and labelling roughly 5000 radar recordings of birds flying near coastal wind turbines, aiming to improve current protection measures. The AI Cup builds on this work by inviting participants to create an AI model that can classify bird species using only radar data and to design a responsible and practical way to implement that model. Participants will reflect on how the model fits into wider operational systems and propose mitigation strategies that could result from its use.

A working radar-based species classification method could shift wind farm wildlife protection from broad, conservative measures to precise, evidence-driven strategies. It would help operators determine when curtailment is truly necessary and acquire data for future mitigation and conservation strategies.

In the model performance track of this AI Cup, participants will attempt to develop an AI model capable of identifying bird species using radar data. For the implementation track, participants are asked to create an implementation proposal. The goal is to propose an AI-based bird species classification system that supports bird conservation within wind farm environments via the use of radar data.

The proposed AI digital system forms the bridge between two existing operational domains: Data Collection & Labelling and Mitigation & Conservation Strategies. It connects raw and labelled radar data to actionable outputs that can support the protection of birds in wind farm operations.

Data collection & labelling covers all methods and processes required to generate both training data and real-time inputs for the digital system. This includes real-time radar measurements, field-based labelling efforts, and other complementary data sources. The system processes these inputs to produce predictions such as bird species classifications linked to radar tracks. These outputs can then be used to inform and trigger  Mitigation & Conservation Strategies, ranging from short-term operational measures to longer-term planning decisions, with the objective of reducing risks to bird populations, particularly for species with a significant ecological impact.

Your task is to write a implementation proposal of a maximum of 2000 words that shows how your proposed digital system would work in practice, considering the data collection & labeling and mitigation & conservation strategies associated with its long and short-term implementation.

Your proposal should form a narrative that covers the description of the entire digital system, responsible design, data acquisition, training strategy and how your system outputs connect to meaningful mitigation & conservation actions. You are free to highlight the aspects you believe make your proposal strong, but balance the amount of detail such that you can cover all the points mentioned. Research well and use your judgment to determine what is essential for your concept and how to communicate it.

You are advised to think critically, justify your decisions, and show awareness of limitations and trade-offs. Creative solutions are highly rewarded, as long as you prove that it has feasible potential to improve the impact of the system.

To also take into account

1. Radar-focused constraint. The proposed implementation must be highly focused on radar data for species identification and may be complemented by other publicly available data. Additional sensing modalities on sight (e.g. cameras, acoustic sensors, humans) can only be proposed for data labelling in order to train the AI model and to validate predictions.
   
   
2. The focus of this proposal is on conceptual clarity, system-level reasoning, and plausibility. You are in the conceptual design stage and therefore you are not expected to provide highly detailed engineering specifications, cost/market analyses, or regulatory approvals.
   
   
3. Sufficiently supported claims are more rewarded than unsupported ones. Claiming model performance improvements via the use of a certain technique/data must be supported.
   
   
4. You are encouraged to use your findings from the model performance track to support your claims.

Beyond addressing the instructions outlined above, your implementation proposal will be evaluated according to the following criteria:

Deliverable guidelines

The implementation proposal should be delivered as a report without the possibility of adding an appendix.

• The report should have a maximum of 2000 content words (titles, names, date, etc, do not count).
• You are encouraged to make use of pictures, but keep it as a report and not as a slide presentation.
• The submission should be in PDF format of a maximum of 10 MB.

Submission

Submissions for the implementation is now open, but will close at the implementation proposal deadline on the 17th of March at 11:59 pm CET. You are allowed to submit multiple times in behalf of your team, only the last submission will count