Data-Driven Windsurfing Performance Evaluation Using Field Tests and Olympic Racing Data

• A multi-input full-state surrogate model (8 inputs → 1 output) has been developed using the co-Kriging regression algorithm, effectively fusing multi-fidelity datasets: field-test measurements and the Velocity Prediction Program (VPP) simulation data.
• The model delivers excellent prediction accuracy with RMSE < 1.5%.
• Input sensitivities can be quantified directly via the model's Jacobian, providing clear guidance on how to adjust the sailing inputs to accelerate it.
• A double-input surrogate model (true wind speed + true wind angle) has been created for polar diagrams using a tailored co-polynomial algorithm.
• It successfully corrects and refines the original VPP polar diagrams by incorporating real-world racing data.
• The approach is directly applicable to open Olympic data from SAP Sailing, enabling visualisation and performance comparison between competing teams.