Skip to Main content Skip to Navigation

Estimation de la pollution sonore en milieu urbain par assimilation d’observations mobiles

Raphaël Ventura 1
1 ANGE - Numerical Analysis, Geophysics and Ecology
Inria de Paris, LJLL (UMR_7598) - Laboratoire Jacques-Louis Lions
Abstract : Noise pollution is a major environmental health problem, and cities are trying to define strategies intended to avoid, prevent or reduce its harmful effects. This requires the determination of the population's exposure to environmental noise in cities, which can be done through noise mapping. Until recently, noise mapping has been carried out mainly through numerical simulation. However, simulation-based noise maps are subject to high uncertainties, and often depict a static average over an extended period of time, without distinction between the different days of the year. This motivates to monitor noise pollution where and when people are exposed. The microphones embedded in dwellers mobile phones are natural candidates for this purpose. The mobile application Ambiciti, developed at Inria, informs people about their exposure to noise and air pollution, and allows us to gather dwellers data, under their consent, through so-called crowdsensing. This observational data is distributed in space and time and hence conveys information that is complementary to simulation data, and can make up for the latter's shortcomings. In this thesis, we propose data assimilation methods that allows one to merge prior noise maps issued by numerical simulation with observational mobile phone-acquired noise data. The result of this merging is an analysis that is designed to have minimum error variance, based on the respective uncertainties of both data sources, that are to be evaluated foremost as precisely as possible. The mobile measurements quality is a major issue to be addressed. It depends on the observable frequencies and intensities, the period of averaging, the environmental conditions, etc. We run a performance analysis that addresses the range, accuracy, precision and reproducibility of measurements. Conclusions of this evaluation leads us to the proposition of a calibration strategy that has been embedded in Ambiciti. Data assimilation addresses the problem of the estimation of an unknown state vector, using a prior state vector and observations. These components shall be compared and combined, based on the uncertainties that one has assumed for each of these components. We address this estimation problem on two different scales. A first method relying on the so-called "best linear unbiased estimator" is proposed. It produces hourly noise maps, based on temporally averaged simulation maps and mobile phone audio data recorded at the neighborhood scale. A systematic study of the errors associated with both the simulation map and the observations (measurement error, temporal representativeness error, location error) is carried out. The method is illustrated through an experiment where two hourly analysis maps are produced, achieving a reduction of at least 25% of root-mean-square error, when compared to the prior maps. A second method that produces maps at the city scale is developed. It leverages the crowd-sensed Ambiciti user data available throughout the covered city, in order to correct the prior simulation map in a global fashion. First, the observations set must be filtered and pre-processed, in order to only select the ones that were generated in adequate conditions (i.e. outdoors, with the phone held in the user's hand, etc.) and account for measurement biases. The assimilation method relies on the construction of an error model, where each point of the prior map is associated with a certain category defined by multiple parameters : prior noise level, distance to the closest road, estimated road traffic on the closest road, etc. A correction is applied to all map points depending on their class within the established categorization. The correction is computed for each category based on the departures between observations and prior levels that lie within this category, the respective assumed uncertainties for the noise model and the observations, and the size of the category. The method is evaluated using temporally averaged simulation maps, by comparing levels captured by fixed microphones of an urban noise monitoring network and the values of the simulated and analysis maps. This evaluation shows a reduction of the error for all considered periods of the day, and a general decrease of the error distribution spread.
Complete list of metadata

Cited literature [70 references]  Display  Hide  Download
Contributor : Vivien Mallet <>
Submitted on : Wednesday, October 31, 2018 - 3:19:10 PM
Last modification on : Wednesday, December 9, 2020 - 3:09:35 PM
Long-term archiving on: : Friday, February 1, 2019 - 5:03:20 PM


Files produced by the author(s)


  • HAL Id : tel-01910084, version 1


Raphaël Ventura. Estimation de la pollution sonore en milieu urbain par assimilation d’observations mobiles. Mathématiques [math]. Sorbonne Université, 2018. Français. ⟨tel-01910084⟩



Record views


Files downloads