Urban air pollution monitoring is of significant scientific, social and economic interest to countries across the globe. As a collaborative research to combine the knowledge in technology and environmental science, our aim is to examine the fine structure of pollutant distribution within defined urban environments, with focus on tall and narrow urban street canyons, characterized by long stretches of continuous buildings on both sides with little or no gaps in between. In our work we explore pervasive wireless sensor network (WSN) technology with Bracelet Carbon Monoxide monitors (Figure 1) as an alternative to the traditional approach for air pollution measurement.

WSNs are largely used in environmental monitoring applications, which compose of a significantly large number of low cost sensor nodes with radio capability. The sensor nodes are normally battery powered, and hence power consumption or prolongation of network lifetime is a major research challenge. In-network processing is encouraged to reduce the amount of data sensed or transmitted back to the base station[1], or to enhance the networking capability and for power conservation purposes.

With the goal of obtaining fine-grain data for scientific analysis and evaluation of our technological design, several experiments have been carried out both in a residential street in London, UK and a residential/ commercial street in Nicosia, Cyprus, two streets with very different characteristics. In this paper, our goal is not to discuss the scientific significance of the data, but to evaluate the system in the WSN perspectives particular to traffic air pollution monitoring application, which results in the definitions of design requirements for improvement and future research topics.

1. Technology Fit

WSN are best suited in applications where dense sensors are required in localized areas. In this application, the ventilation patterns of street canyon are explored through pollution dispersion modelling, the availability of high resolution temporal and spatial pollution data is a definite advantage for the evaluation of models. The wireless networking capability, which is absent in most traditional approaches, allows for ease of deployment when large number of sensors are to be deployed, as well as near real time availability of data. Moreover, new user policies and updates can be disseminated in the system during deployments.

The second advantage of system is the portability of the sensors. They can be carried by people to reflect true human exposure to traffic pollution, or attached to vehicles or bicycles to get comprehensive and dynamic views of a larger area. Again, the wireless capability means that the data can be collected, stored in the nodes and downloaded to a base station once in range without the user's intervention.

2. Choice of Sensors and Calibration Exercise

Electrochemical carbon monoxide (CO) sensors are used in the system. CO is a good indicator of traffic pollution and is inertia to sunlight and hence, was chosen in the first instance. In fact, in the experiments we observed a high correlation of CO concentration and the presence and flows of traffic, especially from motorbikes and high emission vehicles. However, CO disperses very quickly into the atmosphere and has less health implication than pollutants such as particular matters (PM). It is necessary to include other pollutants, i.e. NOx, SOx, Ozone and PM for a complete profile of pollution dispersion patterns.

With respect to sensor type, although electrochemical sensors are not as costly as some other gas sensors such as infra-red analyser, they still amount to >3 times of the other component costs and have a relatively short lifetime of typically 2 years. The reactive material depreciates and hence the sensitivity reduces over time. Recalibration is required to maintenance the accuracy of the measurements.

The deployed sensors were calibrated with CO concentration, temperature and humidity prior to the experiments. We have observed a discrepancy of the establishment of baselines in the calibration environment, where tested gases were maintained at controlled concentrations, temperature and gas flows, and the real world experiments. It was apparently that additional calibration is required to calibrate individual sensors against other possible environmental factors, which are time-consuming and costly, however, essential to ensure the accuracy of the measurements.

3. Data Handling Techniques

Currently, the sensors are sampled at 1Hz frequency and the data (Figure 2) are used in data analysis to extract representative features that are indicative to traffic pollution level. Such high frequency sampling is power exhaustive and may not be essential in the final system design. Part of the research objective is to find representations of the data that are less reliant on calibration accuracy whilst providing the requirement information, and to develop sampling techniques to preserve the essential features of the data whilst minimizing power consumption of the system (Figure 3).

Time synchronisation between sensor nodes is required for spatial correlation analysis. The results of spatial analysis can potentially give us information on the change of traffic conditions including volume and flows. Nodes that are measuring similar pollution levels can be grouped into clusters to reduce sampling, data processing and transmission requirements.

4. Network Strategy

As the application requires a combination of fixed and mobile sensors attached to people and vehicles, memory storage in sensor nodes are necessary to cope with conditions when the nodes do not have network coverage. The network is required to be self-organised and adaptive to these changing conditions. Currently, we can achieve >90% of data throughput with the implementation of Collect transport mechanism in the Contiki Operation System. It can be further improved with end-to-end reliability mechanism, control on transmission power to reduce inter-node interference and improvement on transport strategy.

Figure  SEQ Figure \* ARABIC 1: Bracelet Carbon Monoxide monitors deployed

Figure  SEQ Figure \* ARABIC 2: Carbon Monoxide data collected from the experiments in Cyprus

Figure  SEQ Figure \* ARABIC 3: 15 minutes averages of Carbon Moxide Data. Although the fine-grain data are very distinctive from each other, the averages of different series are surprising similar to each other.