Comprehensive Monitoring of Air Pollution Hotspots Using Sparse Sensor Networks

TL;DR

This paper tackles the limited spatial resolution of public air-quality sensors by marrying predictive interpolation with mechanistic dispersion physics. In New Delhi, an expanded low-cost sensor network enables reliable detection of both known and hidden hotspots, with Space-Time Kriging delivering high precision/recall under partial data. A GPU-accelerated Gaussian Plume model, built on an approximate emissions inventory, explains a substantial portion of transient hotspots and yields mechanistic insights into hotspot formation. Collectively, the work demonstrates that integrating data-driven prediction with physics-based modeling can guide scalable, policy-relevant air pollution management in cities with constrained monitoring infrastructure.

Abstract

Urban air pollution hotspots pose significant health risks, yet their detection and analysis remain limited by the sparsity of public sensor networks. This paper addresses this challenge by combining predictive modeling and mechanistic approaches to comprehensively monitor pollution hotspots. We enhanced New Delhi's existing sensor network with 28 low-cost sensors, collecting PM2.5 data over 30 months from May 1, 2018, to Nov 1, 2020. Applying established definitions of hotspots to this data, we found the existence of additional 189 hidden hotspots apart from confirming 660 hotspots detected by the public network. Using predictive techniques like Space-Time Kriging, we identified hidden hotspots with 95% precision and 88% recall with 50% sensor failure rate, and with 98% precision and 95% recall with 50% missing sensors. The projected results of our predictive models were further compiled into policy recommendations for public authorities. Additionally, we developed a Gaussian Plume Dispersion Model to understand the mechanistic underpinnings of hotspot formation, incorporating an emissions inventory derived from local sources. Our mechanistic model is able to explain 65% of observed transient hotspots. Our findings underscore the importance of integrating data-driven predictive models with physics-based mechanistic models for scalable and robust air pollution management in resource-constrained settings.

Figures (12)

• Figure 1: The adjoining plots show a sample of readings in a small neighborhood containing 7 of our low-cost sensors and 2 public stations on three consecutive days. The black lines show the readings from the closest public monitors, and the other lines show the readings from the rest of the sensors. While most of our sensors are in agreement with the public monitors, note that some of our sensors have recorded spikes in pollution levels that are missed by the public monitors.
• Figure 2: The map in Figure \ref{['fig:monitor-locations']} shows the placement of all the sensors for our study, 32 public (red), and the 28 low-cost sensors that we have deployed (blue). The size of individual points is proportional to the number of months they act as hotspots according to the definition presented in Section \ref{['sec:proof']}APH_paper.
• Figure 3: Figures \ref{['fig:ayanagar']} and \ref{['fig:safdurjung']} show the temporal correlation of readings of our deployed sensors (in blue) and nearest government sensors (in orange). These charts are representative of most of our deployed sensors.
• Figure 4: In the above figure, we first show the longitudinal analysis of hotspots (Figure \ref{['fig:long_hsps']}) and corresponding pollution data plot for one of the monitoring stations (Anand Vihar) over two years at 1 week resolution (Figure \ref{['fig:long_av']}). There is a clear elevation in both pollution levels and detected as well as hidden hotspots during the winter months. We observe the same trend in the remaining monitoring stations as well. During the months of winter, on specific occasions like Diwali (Figures \ref{['fig:dd_nn']}-\ref{['fig:dd_jh']}), the total pollution level goes beyond sensor's measurement capacity of 1000 $\mu g/m^3$.
• Figure 5: Top hotspot locations covered by our sensor network in New Delhi, their demographic and geometric information, and the number of months they behave as hotspots. The population and areas were sourced from geoiq. This is direct evidence that more than 150,000 people live in "pollution hotspot" neighborhoods that are not on the city government's map.