Estimating turbulence intensity along the glide path using wind tunnel experiments combined with interpretable tree-based machine learning algorithms

Abstract

Wind fluctuations near airport runways can make aircraft landings risky. Consequently, an aircraft may veer off its glide path, missed approach, or crash. In this study, a scaled-down model of Hong Kong International Airport (HKIA) and the complex terrain in its vicinity were built in a TJ-3 atmospheric boundary layer wind tunnel to investigate wind turbulence intensity. Cobra probes were placed along the glide slope of the models' runway to compute turbulence intensity at various locations under different inflow wind directions, with and without surrounding terrain. Next, advanced tree-based machine learning algorithms, including Random Forest, Extreme Gradient Boosting, Adaptive Boosting, and Light Gradient Boosting Machine optimized via Bayesian Optimization, were used to estimate turbulence intensity along the glide path. The Bayesian optimized-random forest model outperforms all other models in prediction performance, as measured by MAE (0.521), MSE (1.046), RMSE (1.024), and R2 (0.934). Furthermore, according to SHAP analysis, "Terrain", "Distance from Runway", and "Wind Direction" significantly contributed to high turbulence intensity along the glide path. The most optimistic predictions for high turbulence intensity were the presence of complex terrain, a shorter distance from the runway, and an inflow wind direction between 125 and 200°.