Over the past few decades, aircraft engine technology has improved considerably. High-bypass turbofan engines, the standard propulsion system for commercial aviation, have become significantly more fuel efficient and quieter. Despite this reduced noise footprint, annoyance caused by aircraft noise has not declined accordingly. Instead, aircraft noise complaints seem to increase over time, raising ethical concerns and limiting the potential growth of airports.

There is no objectively correct way to quantify aircraft noise annoyance because it is inherently subjective and cannot be measured directly. Individuals also respond differently to identical noise exposure. In this study, aircraft noise complaints submitted to Bewoners Aanspreekpunt Schiphol (BAS; Residents' Point of Contact Schiphol) were used as a proxy for aircraft noise annoyance.

Assuming a uniform response to aircraft noise across the population yielded no meaningful relationships, with coefficients of determination rarely exceeding 0.3. Therefore, the data were divided into four-digit postcode areas and analysed using mixed-effects models. More than 50 combinations of fixed and random effects were evaluated. Conventional noise metrics, such as Lden, were considered alongside newly developed exposure metrics, meteorological variables, and demographic characteristics. Complaint counts were modelled using a negative binomial distribution with a log-link function. Models were estimated in R and compared using the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC), balancing model fit and complexity.

The best-performing model included the natural logarithm of the number of flights heard within each postcode, calculated using a newly developed algorithm based on flight tracks and the Aircraft Noise and Performance Database; the average minimum slant distance between the postcode centroid and heard flights only; the average property valuation (WOZ value); an intercept; and postcode population as a logarithmic offset. The intercept and minimum slant distance were modelled with both fixed and random effects, whereas the number of flights and average property value were included as fixed effects only.

Model performance was evaluated using an 80/20 train-test split. Predictions at the individual postcode-day level remained inaccurate because complaint counts exhibited substantial variability. However, performance improved considerably when aggregated to annual postcode totals. The model achieved a mean absolute percentage error of 25.52% for annual complaint counts per postcode and only 1.12% error for the total number of complaints across all postcodes. The coefficient of determination between predicted and observed annual complaint counts reached 0.987.

Much of this improvement resulted from the varying intercepts, which substantially increased predictive accuracy but provided limited insight into the underlying causes of differences in complaint behaviour between postcodes. Instead, they effectively captured unobserved latent influences. Among the explanatory variables, average minimum slant distance was the strongest predictor, followed by the logarithm of the number of flights heard and average property valuation. The positive coefficient associated with property valuation further suggests that residents of higher-valued homes are more likely to submit aircraft noise complaints. ... Read more