Monitoring of urban pollution has been largely dependent upon relatively small numbers of stationary monitoring locations. Although there is certainly value in being able to provide high-quality data for a continuous period from a fixed monitoring station, these do not capture data from each street, junction, courtyard, building edge, or pedestrian route in a dense and complex city.
Pollution at street level can vary significantly. Buildings affect airflow. Emissions from vehicles vary in time and space. The wind speed, direction and air temperature experienced above the height of a building may not reflect those conditions experienced by the urban canopy.
Thus, it is possible for a sensor to provide accurate measurements for the point of the sensor installation at any given time, while missing other important pollution concentrations in nearby locations.
A sensor network can help improve the understanding of pollution concentrations in a city. This value is not only in the quantity of data it provides, but in the way the data can be acquired and interpreted to improve knowledge of changing local conditions.
This is the primary question examined in the DYCE project, looking at how a small set of deployable wireless sensor nodes can be used more intelligently to aid response to chemical release in outdoor environments of interest, namely an urban or industrial area.
What Is the Problem with Using Fixed Monitoring Stations?
Fixed monitoring stations are certainly necessary to assess urban air pollution. They provide long-term records, are required for compliance monitoring with legal limits, and provide data to allow changes in air quality to be assessed over an extended period.
However, in an urban environment, pollution does not have uniform concentrations. Levels of air pollution can vary widely over short distances, for example between streets with varying amounts of traffic, different building dimensions and varying prevailing wind direction.
A fixed monitoring station on the side of a main road, for example, does not necessarily represent the air pollution level in a nearby street with higher traffic. Furthermore, air quality monitors at rooftop level can provide useful data on flow conditions over a city, and may help to explain conditions at street level, but may not be representative of pollution levels there.
Part of the reason why street-level air pollution is so difficult to predict is that, whilst knowing what levels of pollutants are being released is important, it is also important to know how they are transported through an environment, mixed with clean air or trapped and recirculated.
Sensor networks can help to address this by providing a better evidence base. They can provide better understanding of spatial variation of concentrations in the street environment, shorter-term changes and localised pollution conditions.
Why More Sensors Do Not Necessarily Translate into Improved Evidence
It is natural to think that simply deploying more sensors will improve air quality data. However, the number of sensors is only one factor to take into consideration.
Data from a sensor network is unlikely to improve the evidence base if the sensors are not positioned in the most suitable locations. If a set of sensors are not positioned to observe the key transport pathways or plume characteristics, they may not provide sufficient evidence. Or, if the data quality provided by each sensor in the network is insufficient, the network may produce poor evidence.
This is why DYCE concentrated on collecting data from a limited set of wireless sensor nodes in an optimised way, and the use of data gathering and deployment planning tools to determine the best way in which to deploy and redeploy nodes in response to evolving environmental or chemical data.
This is an important policy issue in urban monitoring. The effectiveness of any monitoring network cannot be determined solely by the number of sensors involved.
Ultimately, the question that should guide a network’s assessment, as well as its future deployment, is this: does this sensor network capture the necessary data at the right locations and under the correct circumstances?
Adaptive Strategy for Evolving Environments
Pollution levels within cities tend to evolve rapidly. Changes to wind direction can transport a chemical plume from one road to another. A stretch of traffic congestion can generate a transient peak. A chemical release can migrate through a junction and enter an adjoining side street, or become temporarily trapped in a street canyon.
A static monitoring network is by no means able to track each and every shift that might occur. A more adaptable network can be implemented, in which sensor placement can be adjusted to reflect new data points, environmental factors, or the goals of the exercise at hand.
This is especially significant within the DYCE project as it relates to chemical releases. The network was conceived from the outset to consist of mobile sensor nodes that would allow for quick and ad-hoc deployment in response to local shifts in conditions or changes to the chemical composition of a release.
We see this idea as having broader significance for urban air quality. Ultimately, a monitoring network should be able to react according to its own knowledge base.
We are not necessarily arguing here that every air quality network must consist of mobile sensors. Rather, what we are saying is that network design should remain malleable. The network structure must respond to the characteristics of the environment, not simply administrative preference.
Tracing the Path of a Plume
Monitoring networks provide especially good value when we consider the movement of a plume that is not well mixed throughout a region of interest. In this case, the intention is not only to measure concentration, but to understand how a plume is moving.
Where are the plume sources located? In which direction is the chemical travelling? Is the material moving along the length of the street, crossing from one road to another, collecting along the base of a building, or dispersing above rooftops? Is a high reading an indicator of an extended plume, or is the material concentrated in a relatively small area?
These questions are closely linked to the value provided by tracer experiments in understanding how pollution moves through cities. Tracer work uses a known quantity of material to help reveal pathways and the way they may change depending on a particular condition of the environment.
A sensor network can be used to achieve a similar goal under more real circumstances by enabling an observer to see how the level of concentration can change at different points along a route and at different times.
The objectives are similar, but the approaches are different. Tracer experiments enable us to see how a material might move, while sensor networks enable us to see where material is moving under current conditions.
What Does It Mean to Be Representative?
Representativity is one of the more pressing issues that must be addressed in the context of urban monitoring.
Simply because a reading is taken in a representative location does not automatically mean that this location is actually representative of a greater area. Even in a simple road layout, a sensor can be located in a relatively isolated position that might be influenced by a nearby source, building corner, queue of traffic, recirculation zone or ventilation route.
This introduces the issue that, even if we consider the data to be reliable, they might actually be hard to interpret for policy makers. A sensor reading might be representative of the particular environment in which it is located and is therefore not necessarily representative of the surroundings in a given area.
A low level might not indicate that the concentrations at surrounding points will also be low, and a high reading might be representative of exposure in a particular area but not in an area of the wider city.
The real advantage comes through comparison: if several sensors pick up a similar pattern, it increases confidence that the pattern is indicative of the wider environment. On the other hand, if a sensor behaves differently from others, it could be highlighting something of interest at the local scale.
The value is therefore in the pattern.
Sensor Networks and Emergency Response
A second potential role is emergency response planning in the context of airborne material releases. During an incident, knowledge can be incomplete, uncertain and rapidly changing. Emergency responders may need a better sense of where the material is going, what areas are at risk, and, perhaps more importantly, where it would be useful to take more measurements.
The DAPPLE-HO project focused on short-range dispersion in the urban canopy and its relevance to emergency response planning. DYCE has extended this by considering how sensors could be deployed and redeployed after a chemical release.
The issue here is that it is risky to rely on only general assumptions about wind and source distance to inform urban emergency monitoring. The urban structure can channel material in ways that cannot necessarily be inferred from wider meteorology.
A sensor network capable of responding to local conditions may be better suited to supporting decision making in such scenarios.
Data Quality and Interpretation
Sensor networks also open up different issues regarding data quality and interpretation. It is well understood that having more data may improve one’s knowledge of the air, but only if that information is trustworthy, understood and appropriately interpreted.
Different types of sensors have different accuracies, response times, requirements for calibration, and levels of sensitivity to environmental variation. There are many situations where sensors are not intended for accurate measurements, or where a low-cost sensor with a fast response time may be useful in helping identify patterns, but where this will not be enough.
This does not mean they are useless, only that it is important to be clear about their intended role. A sensor intended for rapid assessment of conditions may have different requirements to a reference monitoring station used for compliance purposes in the long term, for example.
A sensor used to help understand where a plume is going will not necessarily need to provide measurements of the same quality as a sensor used to estimate an individual’s annual exposure.
The issue here is how to be explicit about what the network is and is not capable of showing to decision makers. In all situations, sensor data must be treated as evidence.
Combining Sensors With Models
The utility of sensor networks is likely to be enhanced if combined with modelling.
Models can be used to estimate conditions at unmonitored sites, assess which of multiple source locations might be responsible for a plume, explore plume movement patterns and provide an assessment of potential scenarios.
Sensors can provide measurements of what is actually happening in the world, which can be used to help improve, challenge or confirm model predictions.
It is an important relationship since there is an argument that neither sensors nor models are sufficient in their own right. Sensors are able to provide measurements of concentration only at a specific set of places, whereas models are able to estimate conditions at a specific point in space beyond measured points, but they are dependent on assumptions, input and initial conditions.
The DAPPLE “How” page describes how field work, wind tunnel and computational simulations were combined to help understand dispersion at street and neighbourhood scales.
A network of sensors can be integrated in this type of evidentiary framework. It may be possible to contribute to the amount of field evidence used to inform models during evaluation or operational interpretation. The dynamic network also enables us to use the model for moving sensors. There is a feedback cycle between measurement and prediction.
The Role of the Network in Urban Air Quality Policy
In this case, the sensor network should be designed in relation to the questions for policy.
A sensor network designed to measure long-term exposure will be different to a network designed to measure short-term peaks in air pollution concentrations. A network designed to support emergency management would differ from one designed to investigate the effect of transport changes, and one to investigate street canyons would differ from one designed to investigate neighbourhood background conditions.
We can help to design sensor networks more efficiently if we can ask a set of planning questions, including:
- What are the key exposure pathways?
- Are we interested in long or short-term exposure?
- Where are the strong street-scale variations likely to be?
- How will we factor in wind direction and urban form?
- What is the acceptable level of uncertainty for this decision?
- How will this data be related to modelling and reference network data?
As we have discussed, it is easy to collect a lot of data for a sensor network. What is harder to do is ensure that data is interpretable.
We can use a sensor network to show fine-scale features of air pollution concentrations. But we must be careful not to confuse high-resolution data with high-resolution certainty.
We might make a sensor network dense enough to generate detailed maps of concentrations, but those maps may be more confident than the underlying data. We might create maps of smoothly varying air pollution concentrations that are not true to the actual concentrations in the urban environment. We might measure a high concentration in one location, but have too little data to be sure that it is an indication of long-term exposure, a short-term peak, or just an instrument error.
To avoid this, we need to build sensor network design with quality control, calibration strategy, metadata, uncertainty assessment and the effective dissemination of these results.
We are not trying to create the best maps of concentrations possible; we want to produce evidence that can improve decision-making.
Improving the Evidence Base
By understanding the purpose of a sensor network, we can use that data to improve monitoring of air pollution concentrations in urban areas.
Sensor networks can give us spatial information that we cannot get from other sources of information. Sensor networks can help us identify plumes and short-term variation. They can support the network for emergency management, model evaluation and for assessing exposure to air pollution. They can make urban monitoring more flexible and relevant to local conditions.
To do this, we need to think carefully about how they are designed and used. The sensors are useful if they can answer the right questions, collect data at the locations that are most relevant, are sensitive to the conditions of the urban air pollution environment, and are supported by modelling, field measurements and knowledge of urban form.
The wider lesson from DAPPLE and DYCE is that urban monitoring of air pollution concentrations should not just be considered as a matter of locating some instruments around the town. Instead, we should focus on better understanding the movement of air in the urban environment and how this information can support decision making at the scale of relevance to exposure.
With better urban air quality monitoring, we can strengthen the evidence base. Sensor networks cannot get rid of uncertainty, but can help to bring it to light. We can measure that uncertainty, we can make it visible and we can use that evidence in urban air quality policy.
