Feasibility of using the CNOSSOS-EU road traffic noise prediction model with low resolution inputs for exposure estimation on an international scale
Regular paper
Imperial College London
Monday 1 june, 2015, 18:00 - 18:20
0.3 Copenhagen (49)
Abstract:
The EU-FP7-funded BioSHaRE project is using individual-level data pooled from several national-level
cohort studies in Europe (UK, The Netherlands, Norway) to investigate the relationship between road traffic
noise and health. A noise model based on the CNOSSOS-EU method was developed, and applied with
harmonised inputs, to estimate exposures to road traffic noise at address locations. Ideally the noise model
would be offered the highest resolution inputs on traffic flows/speeds, land cover etc. We deviated from
this ideal for two reasons: 1) detailed input data is not always available over whole countries, and 2) to
undertake harmonised exposure assessment we needed a set of input data that are comparable in spatial
resolution between the different countries. To assess the feasibility of this approach in terms of the loss of
model performance from using coarser input data, we applied CNOSSOS-EU with different combinations of
high- and low-resolution inputs (e.g. high resolution road geography with low resolution land cover etc.)
and compared noise level estimates with measurements of LAeq1hr from 38 locations in Leicester, UK.
Starting with a model using all of the highest-resolution datasets (Model A), we piecemeal introduced lower
resolution datasets over five further model runs, with the final model run (Model F) including all of the
lowest resolution data as intended for use in BioSHaRe. As expected, Model A (highest resolution datasets)
performed the best [R = 0.94 (p = .000); RMSE = 1.63 dB(A)]. Model F (lowest resolution datasets) also
performed reasonably well in terms of correlation [R = 0.75; p = .000)] but with relatively large model
errors [RMSE = 4.46 dB(A)]. For a sample of postcode (i.e. zip code) locations (n=721) in Leicester, between
Model A and Model F, 80.1% and 95.2% of exposure estimates remained in the lowest and highest of three
equal exposure categories, respectively.