Predicting Within-City Spatial Variations in Outdoor Ultrafine Particle and Black Carbon Concentrations in Bucaramanga, Colombia: A Hybrid Approach Using Open-Source Geographic Data and Digital Images

Abstract

Outdoor ultrafine particles (UFP, <0.1 μm) and black carbon (BC) vary greatly within cities and may have adverse impacts on human health. In this study, we used a hybrid approach to develop new models to estimate within-city spatial variations in outdoor UFP and BC concentrations across Bucaramanga, Colombia. We conducted a mobile monitoring campaign over 20 days in 2019. Regression models were trained on land use data and combined with predictions from convolutional neural networks (CNN) trained to predict UFP and BC concentrations using satellite and street-level images. The combined UFP model (R 2 = 0.54) outperformed the CNN (R 2 = 0.47) and land use regression (LUR) models (R 2 = 0.47) on their own. Similarly, the combined BC model also outperformed the CNN and LUR BC models (R 2 = 0.51 vs 0.43 and 0.45, respectively). Spatial variations in model performance were more stable for the CNN and combined models compared to the LUR models, suggesting that the combined approach may be less likely to contribute to differential exposure measurement error in epidemiological studies. In general, our findings demonstrated that satellite and street-level images can be combined with a traditional LUR modeling approach to improve predictions of within-city spatial variations in outdoor UFP and BC concentrations.

National Land Use Regression Model for NO2 Using Street View Imagery and Satellite Observations

“…Another study trained convolutional neural networks (CNN) on Google Maps satellite and street-level images, together with traditional LUR variables, to predict ultrafine particle (UFP) and black carbon (BC) concentrations. 33 Those results suggest that images may capture similar features as traditional GIS predictors while also allowing for the inclusion of higherresolution (i.e., street-level) features that traditional GIS variables lack. 37 Another notable trend in recent air quality modeling studies is the adoption of advanced machine learning approaches to improve model performance, including random forest, 39−41 gradient boosting, 42,43 artificial neural network, 44−46 and hybrid algorithms.…”

“…In recent years, innovative image data sources (e.g., street view imagery, high-resolution satellite imagery) have emerged as possible tools to capture hyperlocal characteristics of the natural and built environment. Image-based data may be promising for replacing or augmenting traditional LUR predictors , when enabled by imagery processing techniques (i.e., computer vision) and advanced modeling (e.g., machine learning) . For example, information (e.g., traffic, land use, built environment features) provided by traditional LUR predictors are also encoded in high-resolution digital images and can be extracted and quantified via advanced computer vision techniques.…”

“…As an emerging research topic, a small number of studies have explored using these new data sources to build empirical air quality models. ,,− For example, Lu et al compared national empirical models of six criteria pollutants using traditional LUR variables versus microscale variables (e.g., street view imagery, point of interest, local climate zones) . Similar model performance was achieved using the various combinations of variables, suggesting that microscale variables could be a suitable substitute for traditional predictors.…”

“…Similar model performance was achieved using the various combinations of variables, suggesting that microscale variables could be a suitable substitute for traditional predictors. Another study trained convolutional neural networks (CNN) on Google Maps satellite and street-level images, together with traditional LUR variables, to predict ultrafine particle (UFP) and black carbon (BC) concentrations . Those results suggest that images may capture similar features as traditional GIS predictors while also allowing for the inclusion of higher-resolution (i.e., street-level) features that traditional GIS variables lack .…”

Prediction of Short-Term Ultrafine Particle Exposures Using Real-Time Street-Level Images Paired with Air Quality Measurements

“…As computer vision techniques and data-driven methods have evolved rapidly in recent years, several studies have employed convolutional neural networks (CNNs) to investigate spatial variations in UFPs using unconventional data sources, such as high-resolution satellite images − and street-level images (e.g., Google Street View panorama images), − which can provide useful information when detailed land-use data are unavailable. These studies have shown that using features extracted from images in prediction models could effectively alleviate the challenge of estimating ambient air pollution in data-sparse environments and could be extended to global cities. , These studies employed images that are generally updated infrequently, which could hinder the discovery of the associations between short-term high air pollution exposures and the local urban context.…”