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

Abstract

Land use regression (LUR) models are widely applied to estimate intra-urban air pollution concentrations. National-scale LURs typically employ predictors from multiple curated geodatabases at neighborhood scales. In this study, we instead developed national NO 2 models relying on innovative street-level predictors extracted from Google Street View [GSV] imagery. Using machine learning (random forest), we developed two types of models: (1) GSV-only models, which use only GSV features, and (2) GSV + OMI models, which also include satellite observations of NO 2 . Our results suggest that street view imagery alone may provide sufficient information to explain NO 2 variation. Satellite observations can improve model performance, but the contribution decreases as more images are available. Random 10fold cross-validation R 2 of our best models were 0.88 (GSV-only) and 0.91 (GSV + OMI)�a performance that is comparable to traditional LUR approaches. Importantly, our models show that street-level features might have the potential to better capture intraurban variation of NO 2 pollution than traditional LUR. Collectively, our findings indicate that street view image-based modeling has great potential for building large-scale air quality models under a unified framework. Toward that goal, we describe a cost-effective image sampling strategy for future studies based on a systematic evaluation of image availability and model performance.

National ground-level NO2 predictions via satellite imagery driven convolutional neural networks

“…In the spatial evaluation, model performance dropped significantly, with a daily R 2 CV score of 0.593 and an annual R 2 CV score of 0.549 (Supplementary Table S4). This result is consistent with many other LUR models, indicating that the issue of a lower spatial evaluation performance has yet to be solved (Park et al, 2020;Ghahremanloo et al, 2021;Qi et al, 2022).…”

Spatial Differentiation of PM2.5 Concentration and Analysis of Atmospheric Health Patterns in the Xiamen-Zhangzhou-QuanZhou Urban Agglomeration

“…In this paper, we invert the spatial and temporal concentrations of PM2.5 in the Fujian triangle using a land use regression model commonly used in academia [ 24 , 25 ]. The specific approach is: 34 monitoring stations in Fujian Province were selected ( Figure 2 ), and multiply scaled multi-scale buffer zones with different radii from 250 m to 16,000 m were established with these stations as the center of the circle, and the values of meteorology and elevation, rainfall, wind speed, population, and land use type ratio in the buffer zone were counted in GIS software and used as the independent variables; Using Spss software, a model with a fit higher than 0.90 was obtained by applying stepwise regression with the PM 2.5 concentration values of each station as the dependent variable in 2020.…”

Weather and the City: Machine Learning for Predicting and Attributing Fine Scale Air Quality to Meteorological and Urban Determinants

“…The importance of weather drivers varied between locations, where zonal wind speed at 500 hpa, relative humidity, and total precipitation were among the dominant factors. It is important to note that alongside machine learning, other empirical, statistical models such as land-use regression models have also been widely employed in air quality research. − …”