Statistical analysis of ambient sound levels from long-termoutdoor measurements

摘要

Sound levels and meteorological parameters have been measured over a year in arelatively quiet agricultural area with three distinct sources within two kilometers distance (a railwayand two main roads). An automated measurement system has been used for continuous monitoring.Frequency distributions of sound and wind data, sampled once a second and aggregated in periods ofseveral minutes each, were put into a database. By reducing sound level, wind speed and directiondata, measured once a second, to percentile levels (in 21 equally spaced cumulative frequencies fromminimum to maximum value) the total amount of data is reduced significantly.A graphical user interface makes it possible to select and present the data and linear relations betweendata (e.g. a difference between two parameters). It is possible to recognize the influence of each of thethree main sources and wind by statistical analysis. When a train passes the distribution of sound levels will expand on the 'loud' side, so the skewness of the distribution can be used to recognize a passing train. When road traffic is dominant the sound level (within the measurement period of several minutes) will be relatively constant and thus the distribution relatively narrow: a small standard deviation and symmetry of the distribution can be used to recognize road traffic.