Machine learning based multi-index prediction of aviation turbulence over the Asia-Pacific

Abstract

In-flight turbulence is recognized as a major safety hazard for the global aviation system, capable of causing serious injuries to crew and passengers as well as structural damage to the aircraft. Its prediction at 1 to 2 days ahead mainly relies on numerical weather prediction (NWP) models, which suffer from various practical and inherent deficiencies. This paper applies the XGBoost algorithm in generating skillful aviation turbulence forecasts through optimal combination of a collection of conventional “turbulence indices” produced by an NWP model. Verification over a 1-year period against over 16000 aircraft pilot reports demonstrated consistent superior performance of the machine-learning based multi-index consensus (MIC) over the Asia-Pacific region, with median gains in skill score between 3% to 17% compared to constituent turbulence indices when taken individually.