Predicting traffic noise using land-use regression-a scalable approach

Area of Science:

• Environmental Science
• Geostatistics
• Urban Planning

Background:

• Ubiquitous noise pollution negatively impacts human health and the environment.
• Limited spatial knowledge of noise distribution hinders effective management.
• Current noise mapping is resource-intensive and restricted to select areas.

Purpose of the Study:

• Develop a scalable geostatistical model for comprehensive noise mapping.
• Utilize publicly available data and advanced modeling techniques.
• Address limitations of current noise mapping approaches.

Main Methods:

• Employed a linear land-use regression (LUR) model with context-aware feature engineering.
• Utilized publicly available data for large-scale noise prediction.
• Conducted virtual field campaigns and spatial cross-validation for model assessment.

Main Results:

• A model with 21 variables explained 70.2% of road noise variability (R²=0.702).
• Achieved a mean absolute error of 4.24 dB(A) for general areas and 3.84 dB(A) for built-up areas.
• Generated continuous noise level predictions from 24 to 106 dB(A), filling gaps in existing maps.

Conclusions:

• The model necessitates over 500 stratified samples for representative noise mapping.
• Findings provide crucial, novel noise data for underserviced small communities.
• This approach supplements traditional noise monitoring methods.