In order to more accurately estimate perception and health outcomes related to environmental noise exposure, indicators beyond long-term equivalent sound pressure levels are needed (such as statistical levels, number of events, psycho-acoustical indices, etc). Predicting these more advanced noise indicators is challenging, especially in the urban environment.
This Phd will rely on noise mapping codes combined with (simplified) dynamic traffic estimations as a basis for the physical modelling. After issuing the Environmental Noise Directive in Europe, efforts have been undertaken to standardize and develop traffic source power models and outdoor sound propagation models. The “CNOSSOS” model is currently the preferred method to make noise maps and to report them to the European Commission. But such maps should be considered as “strategic” and are thus not well suited to predict the effect of noise abatements or to predict sound pressure levels at micro-environments (e.g. at a shielded side of a building). And although such strategic models often lack the necessary accuracy, in the urban environment, they are yet computationally very costly. Note that better propagation codes are available like the physically more correct HARMONOISE model, but employing this model would blow up simulation times even more.
Machine learning techniques are increasingly becoming popular and are able to catch (highly) complex dependencies between sets of input parameters and specific outcomes in many applications. In this work, their suitability to mimic physically based simulations will be explored in the dense urban setting. In this way, super-fast urban noise mapping should become feasible. Given the clear and well-known physics behind the propagation problem, there is a specific interest in explainable/interpretable artificial intelligence (XAI).
The physical modelling will be based on open source software like NoiseModelling and the use of publicly available GIS data like Open Street Map. An idea to be developed is to what extent street categorization could serve as a proxy for traffic information. The latter is typical lacking in smaller urban streets. In order to link the exposure to health and perception effects, e.g. for applications to birth cohortes, such population based maps should be made over various periods of time, and even retro-actively. Another research question is to what extent such trained models for a specific city can be transferred to other cities.
We are looking for :
Apply via email to Prof. Van Renterghem (firstname.lastname@example.org) before August 8 th 2022.
Applications should include a CV, transcript of bachelor and master degree, and a motivation letter.