Article Preview
Top1 Introduction
Air pollution is a serious health issue and is of great concern for Europeans as reported by the European Commission (2017). This has led to various efforts to monitor air quality. Historically this has been done with static stations equipped with complex and costly equipment. Additionally, these stations are usually located in cities where most population live and work. More recently, the advent of low-cost sensors (LCS) has given rise to networks of these sensors massively increasing the coverage of monitored areas, as demonstrated by the work of Becnel et al. (2019), Hasenfratz et al. (2015), and Lin et al. (2020). Despite lower accuracy and resolution, Morawska et al. (2018) have shown that low-cost sensors complement the more complex monitoring stations. Hasenfratz et al. (2015) have also shown that the data collected by these low-cost sensor networks can be used to construct highly detailed pollution maps. However, despite the cheaper deployment cost, there are still major population centers that are not monitored.
There are several tools and programming libraries, such as Breezometer (2020), IQAir (2021), PlumeLabs (2020), which can be used for air pollution estimation and visualization. In general, these libraries use data collected by air quality monitoring stations in their estimation models. In areas not covered by these stations, it is difficult to obtain an accurate air pollution map, given that estimation models that are not based on sensory information are rather incomplete. Lakhani et al. (2010), reviews different monitoring methods, all of them requiring in situ measurements of the target pollutant.
To improve the representation of air pollution maps, we propose the representation of knowledge-based pollution maps, where an environmental expert can represent pollution sources. To realize this, we developed a novel graphical tool that taps on exploiting explicit expert knowledge for solving the task at hand. Through this tool, experts can graphically and, thus, intuitively represent dynamic air pollution profiles. A profile can depend on weekday, hour, or any other parameter that the expert finds suitable (e.g., wind orientation). The expert can then associate a pollution profile to different map elements (e.g., building, road), thus specifying which map elements are the actual pollution sources of the associated pollution profile. Once the expert assigns a set of air pollution values to the pollution source profile’s parameters, users can visualize the corresponding predicted pollution diffusion pattern through a heat map.
The design of our tool was influenced by existing map-based applications. For instance, in Google Maps users can select two types of maps: one that displays polygons and polylines representing different features such as buildings or roads among other; a second that displays satellite imagery. On top of these two information layers, users can view additional layers (e.g., traffic, terrain elevation). Another example, Open Street Map, is a programming library that provides the user with different pictorial representations. The selection of a particular library to develop our graphical tool was based on the associated learning curve and map data availability.
This article is an extended version of a previous conference paper (Vital et al., 2021). The focus of this article is to introduce the concept of knowledge-based generation of pollution maps, to present a graphical tool and to assess its usability to represent expert knowledge. To achieve this goal, we performed a set of usability tests with 30 participants, of which 3 were environmental experts and 3 were environmental sciences master students. Participants performed a set of tasks and filled a usability questionnaire.