Combining citizen science and deep learning for large-scale estimation of outdoor nitrogen dioxide concentrations.

Reliable estimates of outdoor air pollution concentrations are needed to support global actions to improve public health. We developed a new approach to estimating annual average outdoor nitrogen dioxide (NO2) concentrations using approximately 20,000 ground-level measurements in Flanders, Belgium combined with aerial images and deep neural networks. Our final model explained 79% of the spatial variability in NO2 (root mean square error of 10-fold cross-validation = 3.58 µg/m3) using only images as model inputs. This novel approach offers an alternative means of estimating large-scale spatial variations in ambient air quality and may be particularly useful for regions of the world without detailed emissions data or land use information typically used to estimate outdoor air pollution concentrations.

Extending the spatial scale of land use regression models for ambient ultrafine particles using satellite images and deep convolutional neural networks.

We paired existing land use regression (LUR) models for ambient ultrafine particles in Montreal and Toronto, Canada with satellite images and deep convolutional neural networks as a means of extending the spatial coverage of these models. Our findings demonstrate that this method can be used to expand the spatial scale of LUR models, thus providing exposure estimates for larger populations. The cost of this approach is a small loss in precision as the training data are themselves modelled values.

Predicting outdoor ultrafine particle number concentrations, particle size, and noise using street-level images and audio data.

Outdoor ultrafine particles (UFPs) (<0.1 µm) may have an important impact on public health but exposure assessment remains a challenge in epidemiological studies. We developed a novel method of estimating spatiotemporal variations in outdoor UFP number concentrations and particle diameters using street-level images and audio data in Montreal, Canada. As a secondary aim, we also developed models for noise. Convolutional neural networks were first trained to predict 10-second average UFP/noise parameters using a large database of images and audio spectrogram data paired with measurements collected between April 2019 and February 2020. Final multivariable linear regression and generalized additive models were developed to predict 5-minute average UFP/noise parameters including covariates from deep learning models based on image and audio data along with outdoor temperature and wind speed. The best performing final models had mean cross-validation R2 values of 0.677 and 0.523 for UFP number concentrations and 0.825 and 0.735 for UFP size using two different test sets. Audio predictions from deep learning models were stronger predictors of spatiotemporal variations in UFP parameters than predictions based on street-level images; this was not explained only by noise levels captured in the audio signal. All final noise models had R2 values above 0.90. Collectively, our findings suggest that street-level images and audio data can be used to estimate spatiotemporal variations in outdoor UFPs and noise. This approach may be useful in developing exposure models over broad spatial scales and such models can be regularly updated to expand generalizability as more measurements become available.