Application of land-use regression models to estimate sound pressure levels and frequency components of road traffic noise in Taichung, Taiwan

Abstract

Few studies have applied land-use regression to predict road traffic noise exposure, and there are few predictive models for different frequencies. This study aimed to measure 24-h average road traffic noise levels and to analyze the frequency components over one year to establish land-use regression models of noise exposure. Fifty monitoring stations were set up to conduct 3 measurements for A-weighted equivalent sound pressure levels over 24 h (Leq,24h) and night equivalent sound pressure levels (Lnight), as well as octave-band analyses, during the 2013–2014 period. Noise measurements were integrated with land-use types, road and traffic information, meteorological data and geographic information systems to construct land-use regression models. Leave-one-out cross-validation was performed to test the validity of the predictive models. The annual means of Leq,24h and Lnight were 66.4 ± 4.7 A-weighed decibels (dBA) and 62.1 ± 6.0 dBA, respectively. Octave-band frequency analyses revealed that the highest means over 24 h and at night were 61.4 ± 5.3 decibels (dB) and 56.7 ± 6.6 dB (both at 1000 Hz), respectively. The model-explained variance (R2) of the full-frequency noise was 0.83 for Leq,24h and 0.79 for Lnight. The R2 values for octave-band-frequency noise ranged from 0.67 to 0.88 for Leq,24h and 0.65 to 0.85 for Lnight, with the highest R2 at 250 Hz for Leq,24h and at 125 Hz for Lnight. The differences between the model R2 and the leave-one-out cross-validation R2 ranged from 5% to 15% for both Leq,24h and Lnight at all frequencies. In the validation, the root mean squared error was 2.09 dBA and 2.80 dBA for the full-frequency Leq,24 and Lnight, respectively, and ranged from 1.89 to 2.62 dB and from 2.51 to 3.28 dB for the octave-band-frequency Leq,24h and Lnight, respectively. This study observed that the annual means of the measured Leq,24h and Lnight in Taichung were both above 60 dBA and had the highest level at 1000 Hz. The developed land-use regression models of Leq,24 and Lnight both had good predictive capacity for the full frequency spectrum and within octave bands and can therefore be applied for epidemiological studies.