Wind turbine noise uncertainty quantification for downwind conditions using metamodeling.

The influence of the ground and atmosphere on sound generation and propagation from wind turbines creates uncertainty in sound level estimations. Realistic simulations of wind turbine noise thus require quantifying the overall uncertainty on sound pressure levels induced by environmental phenomena. This study proposes a method of uncertainty quantification using a quasi-Monte Carlo method of sampling influential input data (i.e., environmental parameters) to feed an Amiet emission model coupled with a Parabolic Equation propagation model. This method allows for calculation of the probability distribution of the output data (i.e., sound pressure levels). As this stochastic uncertainty quantification method requires a large number of simulations, a metamodel of the global (emission-propagation) wind turbine noise model was built using the kriging interpolation technique to drastically reduce calculation time. When properly employed, the metamodeling technique can quantify statistics and uncertainties in sound pressure levels at locations downwind from wind turbines. This information provides better knowledge of sound pressure variability and will help to better control the quality of wind turbine noise prediction for inhomogeneous outdoor environments.

Inverse modeling and joint state-parameter estimation with a noise mapping meta-model.

This study aims to produce dynamic noise maps based on a noise model and acoustic measurements. To do so, inverse modeling and joint state-parameter methods are proposed. These methods estimate the input parameters that optimize a given cost function calculated with the resulting noise map and the noise observations. The accuracy of these two methods is compared with a noise map generated with a meta-model and with a classical data assimilation method called best linear unbiased estimator. The accuracy of the data assimilation processes is evaluated using a "leave-one-out" cross-validation method. The most accurate noise map is generated computing a joint state-parameter estimation algorithm without a priori knowledge about traffic and weather and shows a reduction of approximately 26% in the root mean square error from 3.5 to 2.6 dB compared to the reference meta-model noise map with 16 microphones over an area of 3 km2.

Emulation of wildland fire spread simulation using deep learning.

Numerical simulation of wildland fire spread is useful to predict the locations that are likely to burn and to support decision in an operational context, notably for crisis situations and long-term planning. For short-term, the computational time of traditional simulators is too high to be tractable over large zones like a country or part of a country, especially for fire danger mapping. This issue is tackled by emulating the area of the burned surface returned after simulation of a fire igniting anywhere in Corsica island and spreading freely during one hour, with a wide range of possible environmental input conditions. A deep neural network with a hybrid architecture is used to account for two types of inputs: the spatial fields describing the surrounding landscape and the remaining scalar inputs. After training on a large simulation dataset, the network shows a satisfactory approximation error on a complementary test dataset with a MAPE of 32.8%. The convolutional part is pre-computed and the emulator is defined as the remaining part of the network, saving significant computational time. On a 32-core machine, the emulator has a speed-up factor of several thousands compared to the simulator and the overall relationship between its inputs and output is consistent with the expected physical behavior of fire spread. This reduction in computational time allows the computation of one-hour burned area map for the whole island of Corsica in less than a minute, opening new application in short-term fire danger mapping.

Meta-modeling for urban noise mapping.

Urban noise mapping generally consists of simulating the emission and attenuation of noise in an area by following rules such as common noise assessment methods. The computational cost makes these models unsuitable for applications such as uncertainty quantification, where thousands of simulations may be required. One solution is to replace the model with a meta-model that reproduces the expected noise levels with highly reduced computational costs. The strategy is to generate the meta-model in three steps. The first step is to generate a training sample exploring the large dimension model's inputs set. The second step is to reduce the dimension of the outputs. In the third step, statistical interpolators are defined between the projected values of the training sample over the reduced space of the outputs. Radial basis functions or kriging are used as interpolators. The meta-model was built using the open source software NoiseModelling. This study compares the proximity of the meta-model outputs to the model outputs against the reduced basis, the class of the kriging covariance function, and the training sample size. Simulations using the meta-model are more than 10 000 times faster than the model while maintaining the main behavior.

Assimilation of mobile phone measurements for noise mapping of a neighborhood.

Noise maps are a key asset in the elaboration of urban noise mitigation policies. However, simulation-based noise maps are subject to high uncertainties, and the estimation of population exposition to noise pollution generally relies on static averages over an extended period of time. This paper introduces a method to produce hourly noise maps based on temporally averaged simulation maps and mobile phone audio recordings. The data assimilation method produces an analysis noise map which is the so-called best linear unbiased estimator: it merges the simulated map and the measurements based on respective uncertainties so that the analysis map has minimum error variance. The method is illustrated through a neighborhood-wide experiment. 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. Two LA eq , 1 h maps are produced, corresponding, respectively, to a morning and an evening time slot. The analysis maps achieve a reduction of at least 25% of root-mean-square error. The a posteriori error variance of the maps are generally around 50% of the a priori error variance in the vicinity of the observed locations.

Kriging-based spatial interpolation from measurements for sound level mapping in urban areas.

Network-based sound monitoring systems are deployed in various cities over the world and mobile applications allowing participatory sensing are now common. Nevertheless, the sparseness of the collected measurements, either in space or in time, complicates the production of sound maps. This paper describes the results of a measurement campaign that has been conducted in order to test different spatial interpolation strategies for producing sound maps. Mobile measurements have been performed while walking multiple times in every street of the XIIIth district of Paris. By adaptively constructing a noise map on the basis of these measurements, the role of the density of observations and the performance of four different interpolation strategies is investigated. Ordinary and universal Kriging methods are assessed, as well as the effect of using an alternative definition of the distance between observation locations, which takes the topology of the road network into account. The results show that a high density of observation points is necessary to obtain an interpolated sound map close to the reference map.

Evaluation and calibration of mobile phones for noise monitoring application.

The increasing number and quality of sensors integrated in mobile phones have paved the way for sensing schemes driven by city dwellers. The sensing quality can drastically depend on the mobile phone, and appropriate calibration strategies are needed. This paper evaluates the quality of noise measurements acquired by a variety of Android phones. The Ambiciti application was developed so as to acquire a larger control over the acquisition process. Pink and narrowband noises were used to evaluate the phones' accuracy at levels ranging from background noise to 90 dB(A) inside the lab. Conclusions of this evaluation lead to the proposition of a calibration strategy that has been embedded in Ambiciti and applied to more than 50 devices during public events. A performance analysis addressed the range, accuracy, precision, and reproducibility of measurements. After identification and removal of a bias, the measurement error standard deviation is below 1.2 dB(A) within a wide range of noise levels [45 to 75 dB(A)], for 12 out of 15 phones calibrated in the lab. In the perspective of citizens-driven noise sensing, in situ experiments were carried out, while additional tests helped to produce recommendations regarding the sensing context (grip, orientation, moving speed, mitigation, frictions, wind).