Abstract
Urban air quality assessment faces significant challenges due to the high spatial variability of pollutants like Black Carbon (BC). While mobile monitoring using bicycle-mounted sensors provides high-resolution data, individual measurements are often unrepresentative "snapshots" influenced by transient emission events. This study utilizes an extensive dataset of over 350 monitoring runs in Antwerp, Belgium, to establish a rigorous methodology for urban air quality mapping. By employing a Monte Carlo-style subsampling approach, we demonstrate that achieving a reliable map with 25% uncertainty typically requires between 24 and 94 repetitions. The study further validates that data processing techniques, such as background normalization and trimmed means, are essential for improving the efficiency and stability of mobile monitoring campaigns.
Introduction
Urban air pollution is characterized by extreme micro-scale variability. Concentrations of pollutants like Black Carbon—a primary indicator of combustion-related emissions—can vary by a factor of ten within a single neighborhood due to traffic patterns and "street canyon" effects. Traditional stationary monitoring networks, while highly accurate, are too sparse to capture these local hotspots. Mobile monitoring platforms (e.g., sensors on bicycles or vehicles) offer a promising alternative by capturing data at high spatial resolutions (down to 20 meters). However, the temporal variability of urban environments means that a single mobile run may capture an outlier rather than the representative average. This research addresses the fundamental question: how many repetitions are required before mobile data becomes a reliable map?
Methodology
The researchers deployed the "Aeroflex" platform, a bicycle equipped with a micro-aethalometer for BC measurement, a GPS, and sensors for noise reduction. The study was conducted in Antwerp, Belgium, focusing on two distinct routes:
- Route 1: A 2 km loop with 256 repeated runs.
- Route 2: A 5 km loop with 96 repeated runs.
To evaluate the reliability of the data, the study used a Monte Carlo subsampling approach. The average of all available runs was defined as the "true" pollution level. Smaller subsets of runs were then randomly sampled to determine how quickly their average converged to this true mean.
Data Processing and Analysis
To mitigate the impact of extreme, one-off emission events (such as passing a high-polluting vehicle) and city-wide meteorological fluctuations, several data processing techniques were tested:
- Background Normalization: Adjusting mobile measurements based on a central stationary monitor to account for regional weather-driven fluctuations.
- Trimmed Means: Calculating averages after removing the highest and lowest values to reduce the influence of temporal outliers.
- Noise Reduction: Utilizing the Optimized Noise-reduction Averaging (ONA) algorithm to handle sensor-inherent signal noise.
Key Results
- Convergence Requirements: To map BC at a 50m resolution with a 25% uncertainty level, a median of 41 runs was required (ranging from 24 to 94 depending on street complexity). If the uncertainty requirement is relaxed to 50%, the number of required runs drops to a median of 8.
- The Impact of Processing: The use of trimmed means and background normalization significantly reduced the number of runs needed for a stable map. Normalization is particularly effective as it allows mobile data to focus on local spatial differences rather than city-wide temporal trends.
- Validation: When aggregated over multiple days, mobile data showed a strong correlation ($R^2 = 0.82$) with stationary monitors, confirming that mobile platforms can accurately capture both spatial hotspots and temporal trends if sufficient repetitions are performed.
Discussion and Conclusion
This study establishes that while mobile monitoring can reveal pollution gradients invisible to stationary stations, "one-and-done" measurements are scientifically risky. The high number of repetitions required—often dozens of runs—suggests that "opportunistic" sensing may be the most viable path forward. Attaching sensors to vehicle fleets that already traverse the city daily, such as buses or delivery trucks, could provide the high repetition counts necessary for reliable, high-resolution urban air quality mapping. Future work should expand this methodology to other pollutants like $PM_{2.5}$ and $NO_2$, which may exhibit different dispersion characteristics.
Summary updated