Abstract:On-road air pollution exhibits substantial variability over short distances due to emission sources, dilution, and physicochemical processes. Integrating mobile monitoring data with street view images (SVIs) holds promise for predicting local air pollution. However, algorithms, sampling strategies, and image quality introduce extra errors due to a lack of reliable references that quantify their effects. To bridge this gap, we employed 314 taxis to monitor NO, NO2, PM2.5 and PM10 dynamically and sampled corresponding SVIs, aiming to develop a reliable strategy. We extracted SVI features from ~ 382,000 streetscape images, which were collected at various angles (0°, 90°, 180°, 270°) and ranges (buffers with radii of 100m, 200m, 300m, 400m, 500m). Also, three machine learning algorithms alongside the linear land-used regression (LUR) model were experimented with to explore the influences of different algorithms. Four typical image quality issues were identified and discussed. Generally, machine learning methods outperform linear LUR for estimating the four pollutants, with the ranking: random forest > XGBoost > neural network > LUR. Compared to single-angle sampling, the averaging strategy is an effective method to avoid bias of insufficient feature capture. Therefore, the optimal sampling strategy is to obtain SVIs at a 100m radius buffer and extract features using the averaging strategy. This approach achieved estimation results for each aggregation location with absolute errors almost less than 2.5 {\mu}g/m^2 or ppb. Overexposure, blur, and underexposure led to image misjudgments and incorrect identifications, causing an overestimation of road features and underestimation of human-activity features, contributing to inaccurate NO, NO2, PM2.5 and PM10 estimation.

What problem does this paper attempt to address?

### Problems the Paper Aims to Solve This paper aims to address the issue of predicting road air pollution based on Street View Images (SVIs) and machine learning methods. Specifically, the research team achieves this goal through the following points: 1. **Data Collection and Processing**: - Using 314 taxis to dynamically monitor four pollutants (NO, NO₂, PM₂.₅, and PM₁₀) in the central area of Guangzhou and collect corresponding street view images. - After cleaning and processing the data, 4,948,120 valid records were retained, and the data was spatially aggregated using a 200×200 meter grid. 2. **Street View Image Feature Extraction**: - Obtaining street view images for each aggregated location through the Baidu API and extracting image features using the Mask2Former algorithm. - An image quality recognition algorithm was used to detect low-quality images with blur, overexposure, or color distortion and classify them. 3. **Machine Learning Model Comparison**: - Comparing three machine learning algorithms (XGBoost, Random Forest, Neural Network) with the linear Land Use Regression (LUR) model to evaluate the performance of different algorithms in predicting the concentrations of the four pollutants. - The results showed that Random Forest performed the best in predicting all pollutants, followed by XGBoost and Neural Network, while the linear LUR model performed the worst. 4. **Sampling Strategy Optimization**: - Exploring the impact of different angles (0°, 90°, 180°, 270°) and ranges (100m to 500m buffer zones) of street view image sampling strategies on prediction errors. - The study found that using an average strategy within a 100m radius buffer zone could achieve the best prediction results, significantly reducing prediction errors compared to single-angle sampling strategies. 5. **Impact Analysis of Low-Quality Images**: - Identifying and analyzing 16.25% of low-quality images and finding that overexposure and color channel distortion were the main quality issues. - Filtering out low-quality images can significantly improve prediction accuracy, especially in the prediction of NO and PM₂.₅. Through these methods, the research team developed a reliable strategy that can accurately predict road air pollution levels over short distances, providing valuable reference and support for urban planners and policymakers.

How to predict on-road air pollution based on street view images and machine learning: a quantitative analysis of the optimal strategy

Quantifying and predicting air quality on different road types in urban environments using mobile monitoring and automated machine learning

Enhancing Urban Real-Time PM2.5 Monitoring in Street Canyons by Machine Learning and Computer Vision Technology

Portraying on-road CO2 concentrations using street view panoramas and ensemble learning

Urban Air Pollution Mapping Using Fleet Vehicles As Mobile Monitors and Machine Learning

Combining Google traffic map with deep learning model to predict street-level traffic-related air pollutants in a complex urban environment

Applying Traffic Camera and Deep Learning-Based Image Analysis to Predict PM2.5 Concentrations

Fine-Scale Prediction of Roadside CO and NOx Concentration Based on a Random Forest Model

Optimized air quality management based on air quality index prediction and air pollutants identification in representative cities in China

Using User Generated Online Photos to Estimate and Monitor Air Pollution in Major Cities

Prediction of Air Pollutants on Roadside of the Elevated Roads with Combination of Pollutants Periodicity and Deep Learning Method

Air Quality Index and Air Pollutant Concentration Prediction Based on Machine Learning Algorithms

Predicting Neighborhood-Level Residential Carbon Emissions from Street View Images Using Computer Vision and Machine Learning

High-resolution mapping of regional traffic emissions using land-use machine learning models

Urban Air-Quality Estimation Using Visual Cues and a Deep Convolutional Neural Network in Bengaluru (Bangalore), India

Prediction of High Resolution Spatial-Temporal Air Pollutant Map from Big Data Sources.

Prediction of PM2.5 Hourly Concentrations in Beijing Based on Machine Learning Algorithm and Ground-Based LiDAR

Characterizing and Interpreting the Spatial Variation of Traffic Pollution in Urban Non-Motorized Lanes Using Mobile Measurements

Development and application of a multi-scale modeling framework for urban high-resolution NO2 pollution mapping

High-resolution on-road air pollution exposure informed by taxi-based mobile monitoring sensors

Application of machine learning algorithms to improve numerical simulation prediction of PM2.5 and chemical components