Probabilistic Turbulence Prediction using Random Forests

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Wednesday, 20 January 2010: 10:45 AM
B204 (GWCC)
Zhengzheng Li, The University of Oklahoma, Norman, OK; and T. A. Supinie

Presentation PDF (128.5 kB)

The predicting turbulence problems for this AI contest containing measurement from aircraft, satellite, ground-based radar and numerical weather model. According to the peak_edr value, scenarios can be classified into two class types: moderate-or-greater turbulence or no turbulence. Random Forest is notorious being effective on weakly correlated features, thus is adopted in this study. Unlike normal Random Forests having each tree in the forest to vote, a new technique of combining results from the forest is proposed to generate probabilistic forecast, which is the contest requirement. Preliminary results show that the BSS value can reach 0.4 for the proposed algorithm.