Modeling Non-Linearity From Aircraft Noise Measurements With Data-Driven Methods

Abstract

This paper presents an approach for modeling the non-linear behavior of aircraft flyover noise using data recorded by a network of 41 Noise Monitoring Terminals (NMTs) surrounding Amsterdam Schiphol Airport. This approach leverages measurements of the acoustic metrics SEL and LA,max to train a regression-based Random Forest model, which incorporates a new feature for real-time wind direction correction, along with community noise levels. The model achieves a high predictive accuracy, with a standard deviation of residuals below 2 dBA for both SEL and LA,max. To enhance interpretability, both global and local feature importance analyses are employed, providing complementary insights into the influence of input features relative to each other. The resulting feature rankings for both SEL and LA,max are in line with physical expectations: apart from community noise levels, noise event duration and distance are identified as the most influential parameters for predicting SEL, while community noise levels and distance play dominant roles in predicting LA,max . These findings demonstrate that the proposed framework effectively captures the physical dependencies while maintaining interpretability and robustness.