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

Abstract

Spatially explicit urban air quality information is important for developing effective air quality control measures. Traditionally, urban air quality is measured by networks of stationary monitors that are not universally available and sparsely sited. Mobile air quality monitoring using equipped vehicles is a promising alternative but has focused on vehicle-level experiments and lacks fleet-level demonstration. Here, we equipped 260 electric vehicles in a ride-hailing fleet in Beijing, China with low-cost sensors to collect real-time, spatial-resolved data on fine particulate matter (PM 2.5 ) concentrations. Using this data, we developed a decision tree model to infer the distribution of PM 2.5 concentrations in Beijing at 1 km by 1 km and 1 h resolution.Our results are able to show both short-and long-term variations of urban PM 2.5 concentrations and identify local air pollution hotspots. Compared with a benchmark model that only uses data from stationary monitoring sits, our model has shown significant improvement with the coefficient of determination increased from 0.56 to 0.80 and root mean square error decreased from 12.6 to 8.1 μg/m 3 . To the best of our knowledge, this study collects the largest mobile sensor data for urban air quality monitoring, which are augmented by state-of-the-art machine learning techniques to derive high-quality urban air pollution mapping. Our results demonstrate the potential and necessity of using fleet vehicles as routine mobile sensors combined with advanced data science methods to provide high-resolution urban air quality monitoring.

Modelling calibration uncertainty in networks of environmental sensors

“…The problem of calibrating low-cost, large-scale, air pollution sensor networks is quickly becoming acute: A number of large scale deployments of low-cost air pollution networks have happened recently, including one running on 260 cars in Beijing (Zhao et al, 2021). The difficulty in field-calibration and bounding the accuracy of estimates from these deployments suggests that methods that can provide automatic low-cost recalibration across the network are becoming increasingly necessary.…”

“…However such access required site permissions and expertise, while the mobile low-cost sensors could visit the static stations simply by the motor-bike taxi drivers parking next to the sensor. This simplicity provided by permanently mobile low-cost sensors is discussed in Zhao et al (2021).…”

“…Many low-cost network sensors exist (e.g. Parmar et al, 2017;Rai et al, 2017;Nyarku et al, 2018;Feinberg et al, 2019;Zhao et al, 2021) but few perform regular co-location calibrations. Possibly this is because there are no well developed, easily used, computational tools to perform the in-field calibration discussed in this paper.…”

Quantifying the impacts of COVID-19 on Sustainable Development Goals using machine learning models

“…For example, Storm et al stated that the econometric analysis is not sufficiently flexible to capture the non-linearities, which were so common to the processes in environmental systems [15] . Compared with the traditional methods, machine learning approaches can better capture complex non-linear relationship between responses and predictors so as to show better accuracy [16] , [17] , [18] , [19] , [20] .…”

High-Resolution Urban Air Quality Mapping for Multiple Pollutants Based on Dense Monitoring Data and Machine Learning

“…The portable monitoring devices are not necessarily accurate due to cost and volume limitations, and often focus on a specific area rather than the whole city [14]. The vehicles equipped with sensors cannot guarantee the monitoring time (less observation at night) and are easily affected by human factors (forgetting to open) or operating environments (the wind in driving) [19,31]. Therefore, it is necessary to explore a method to steadily infer the variations in urban air quality at both a high spatial and temporal resolution.…”