| dc.description.abstract | Journey-time exposure to NO2 makes up a large part of the total exposure of an individual to air pollution as people tend to travel frequently and because they are located close to the source of the pollution during their trip. The aggregated exposure to air pollution over a route can be calculated when location data of the route is present as well as information about the air pollution concentrations that vary over the course of the day. However, location data is scarce and direct measurements cannot easily be carried out on a large number of individuals.
This research tests whether exposure simulations can be used to estimate the exposure. In order to do so, hourly air pollution maps for the city of Rotterdam are generated using land use regression models and a Python script is written to extract routes in this area from location data gathered by 160 inhabitants that carried a GPS sensor for one week continuously. The exposure over the resulting 509 routes is calculated and used as a reference dataset.
Three exposure simulation models are built that estimate the travel duration and the average concentration to get an estimate for the aggregated exposure: 1) the Static Location Model only uses the origin and destination to get an estimate for the concentration and the distance between them to get an estimate for the travel duration; 2) the Shortest Route Model simulates the shortest route over the road network to estimate the concentrations while the travel durations are estimated using the length of the route; 3) the Fastest Route Model simulates the fastest route to estimate both the concentrations and travel durations. The results show that Fastest Route Model performs best in estimating the exposure as well as the average concentrations and travel durations with R2’s of respectively 0.35, 0.83 and 0.14. The concentrations are accurately estimated because routes are predicted with a high certainty (50% of the routes has 71% or more overlap with the route from the reference dataset); while travel duration is difficult to predict as it depends on external factors such as traffic intensity which is not modelled here. When model run times are an issue, the Static Location Model provides a good alternative as the exposure estimates have an R2 of 0.19 and a run time of only 2 seconds while the Fastest Route Model runs for 253 seconds. The Shortest Route Model has no advantages over the other models and did not return a route for 34% of the cases.
The model’s sensitivity to changes in air pollution concentration over the course of the day is assessed by running the routes at minimum and maximum concentrations (i.e. 3:00 am and 7:00 am). The resulting exposures at minimum and maximum concentrations show a strong correlation (R2 = 0.98) with exposures at maximum concentrations being 1.8 times higher than those at minimum concentrations thus showing that the model is very sensitive to changes in the concentrations.
Improvements of the models are tested by adding additional route options and by supplying the travel duration to the simulation models. The results showed that adding extra route options did not result in a higher performance as the estimates for the exposures as well as for the travel durations and concentrations were less accurate. However, supplying the travel duration to the simulation models resulted in a considerable increase in performance of all models as this was the component that was most difficult to estimate.
In conclusion, this research shows that route simulation can be used as an alternative for GPS data as the simulated routes are very similar to the actual routes. However, the travel duration remains difficult to estimate; therefore, only an indication for the journey-time exposure can be given. But when the travel duration is supplied to the model or when these estimates can be made with higher accuracy (e.g. by including traffic models), exposures estimates can be made with high certainty. | |
