Spatio-temporal Analysis of Citizen Science Air Quality Measurements

Methodology

Data Preparation

Description	Illustration
First of all, there are two data sets, one for 2017 and the other for 2018, provided for spatio-temporal analysis of citizen science air quality. In each data set, the sensors are captured by the geohash code. After acquiring the longitude and latitude by using R programming, we are able to draw the density geographic maps on P1 or P2 showed by the following pictures. In this task, I would prefer to do the analysis by separately going through the two data sets because it was noticed that the number of sensors included in each data sets are different and cannot be joined effectively during my initial exploration on the two data sets. In order to get the clearer whole picture on the air quality, it is better to take care of them separately.

Insights & Dashboard Design

Characterize the Sensors’ Coverage, Performance and Operation

Description	Dashboard Visualization
Coverage—Sensors are not only distributed over the Sofia City By taking advantage of Sofia topography data set, we are able to distinguish whether the geographic location of the sensors included in the data set are all in the Sofia City. From the view of the group of maps in the dashboard showed below, both the Original Sensors Coverage Maps in 2017 and 2018 are located over the Bulgaria country rather than only for the Sofia City. Hence, we need to filter out the geohash codes that are out of Sofia City by referring to the sofia_topo dataset which includes Sofia urban area and some areas nominally external to the city (toward Vitosha mountain, note large elevation numbers), although it has no particular effort to include entirety of Sofia Capital's area as per administrative boundaries. In the view of the whole Bulgaria country, there are major three areas that have higher concentration on P1 or P2 in these two years and Sofia city is the most serious polluted city in Bulgaria with a larger intense cycle on the deeper blue colour. The other finding is that by comparing the two Original Sensors Coverage Maps in 2017 and 2018, it is easily to notice that there are more sensors captured in 2018 than in 2017 based on the larger number of data points plotted on the maps. Besides, by comparing the two Sensors in Sofia maps in 2017 and 2018, the points are more intense in 2018 than in 2017 which illustrate a situation that the pollution is more series in 2018.
Performance—Not all the sensors perform well at any time The dashboard above provides more information on the timeseries for each sensor in 2017. The Density Map on P1 over the four month shows the concentration on P1 continuously became higher as the time went on. The sensors record hourly during the day on the measurement of different variables, so as for the performance on the sensors in 2017, we make use of the number of readings recorded by each sensor. The larger number of readings recorded on the time series, the better the sensors performed during this period. The Records Volume 2017 tell us how well each sensor worked in 2017.It shows that not all the sensors perform well at any time. The Trellis plot on the timeseries can give us a better view on how P1 changed in the period and make better comparation over the large number of sensors based on the different time series plots in one picture. Some sensors show the lines change more frequently with extreme higher values, while others may fluctuate at the lower level on P1.
Operation This picture not only shows us how the P1 and P2 change over the four months, but also tell us which part of the city has relatively higher readings than others. The density map illustrate that the middle part of the city is more seriously polluted by P1 and P2. Besides, by plotting the same kind of density map with the three meteorological variables (Humidity, Pressure and Temperature), it is obvious that all these three variables went higher as time went on. The same trend is also showed with the concentration on P1 and P2. The interesting thing is that the places with higher concentration on P1 and P2 showed higher number on the three meteorological variables (Humidity, Pressure and Temperature) than the other places. Now it is time for us to explore more details on the highly concentration pollution places.

Air Pollution Measurements

Which part of the city shows relatively higher readings than others?

Description	Dashboard Visualization
Sensors in Year 2017 Since December in 2017 is the month with higher pollution, we filtered the time period to December so that more easily to obtain the patterns on the discharge of the pollution. Firstly, we know that P1 and P2 have a strong positive correlation by exploring the relationship of P1, P2 on those data points recorded hourly. The Treemap tells us that the Geohash “sx8derj5kqf” has the highest pollution both on P1 and P2, so we keep this sensor only in the Calendar heatmap. By zooming into this specific location on air pollution, the differences in the relatively higher readings are time dependent because the concentration ranges on P1 and P2 become larger from Sep to Dec in 2017 and the average readings are extremely high during the days from 5th to 12th in December.
Sensors in Year 2018 From the calendar heatmap, the air quality seems to become better from Jan to Aug in 2018. The density maps also show the same trend on air pollution which become not that serious in Aug. The difference between 2017 and 2018 is that there are more sensors working in 2018 which illustrate a larger part of Sofia City than in 2017 have been facing with serious air pollution. However, the same phenomenon with 2017 is that the arears with higher readings are also have higher meteorological variables on Humidity, Temperature and Pressure than the other places in the city. The Treemap show us the detail on which specific location has the most pollution on P1 and P2. The most serious polluted area is the place with geohash “sx8d8vjerh”. Apart from this area, there are five more areas are also in serious air condition. This is also the difference between 2017 and 2018, since in 2017 there are only one area that has obvious air pollution on P1 and P2.