Soundscape evaluation and ANN modelling in urban open spaces.

There is an increasing public and academic interest in the environmental qualities of urban open spaces. The study in this thesis focuses on soundscape research in urban open spaces, which is within the paradigm of environmental psychology. It explores how to use the results of soundscape research in aiding the design process of urban open spaces with regard to the sonic environment. It is based on common notions that the acoustic aspect of urban open spaces should be considered in the same way as the visual dimensions. The determinant of a soundscape is subjective evaluations, which depend on two acoustic aspects; one is the sound nois~ scales and the other is the effects of various . sound sources. Based on data collected from a series of field studies and laboratory experiments, the subjective evaluations of sound-level and sound preference have been separately studied using statistical analyses, and the overall evaluations of sounds cape and acoustic comfort have been examined. In 9rder to provide a feasible tool to aid soundscape deigns, the study develops a modelling tool, namely artificial neural network (ANN), to present the subjective evaluations of potential users at the design stage. Based on the ANN models, soundscape maps can be produced. The results of statistical analyses suggest that various factors influencing the subjective evaluations of sound level, sound preference and acoustic comfort are different in terms of a variation among case study sites and noticed sounds. Generally speaking, sound physical and psychological characteristics have the most influence on the subjective evaluations. The subjective evaluations of other physical environments are also much relevant to the soundscape evaluations, whereas socialldemographical and behavioural factors are insignificant although some relationships have been found for certain factors. In addition to giving useful guidelines and information to soundscape research and design, the results are also crucial in selecting input variables for ANN prediction models. For ANN model predictions, it is found that a general model for all the case study sites is less feasible due to the complex physical and social environments. Practical models for certain type of urban open spaces are more reliable. The performance of acoustic comfort models is considerably better than that of sound level models. It is also found that the key variables to determine the prediction performance of sound preference models are sound meanings and the sounds' physical and psychological characteristics. Furthermore, the prediction maps based on ANN models' outputsú have been successfully produced in presenting the potential users' appraisals of a soundscape in developing urban open spaces.