Database Reference
In-Depth Information
and landing directions. Therefore, we build 500-meter spatial buffers around the
SD-clusters, as shown in Figure
12.6
c. For an analysis by airports, irrespective
of the directions, we would do a second stage of clustering (after excluding the
noise) by only the spatial positions of the events and then build buffers around
the resulting spatial clusters.
In the third step of the analysis procedure, we aggregate the landing m-events
in space by the buffers and in time by 1-hour intervals. In the fourth step, we
visualize the resulting time series by temporal diagrams positioned on the map
display; two of them can be seen in the map fragment in Figure
12.6
c. They
show that the aircraft landed in the airport of Nice from the southwest almost all
times except for an interval in the middle of the day, when the landing direction
changed to the opposite. The exact times and values are displayed when the
mouse cursor points on an area.
Figure
12.6
d presents the map with the temporal diagrams for the Paris region.
We can see that the Orly airport and the northern runway of the Charles de Gaulle
airport have clear peaks in the morning and in the evening. It is a typical pattern
for airline hubs: a short period of time, during which many flights arrive and
take off, maximizes the number of possible connections. The southern runway
of the Charles de Gaulle airport is used with almost constant intensity during
the day. The remaining airports are used much less intensively and mostly in the
afternoon.
So far we have considered only the landings. To investigate the takeoffs, we
repeat the procedure. To extract the takeoff events in the first step, we use the
query condition that the altitude must be less than 1 km at the beginning of the
trajectory. The remainder of the procedure is similar to that for the landings.
To investigate the connections among the airports, we need to define the
airport areas so that they include both the takeoff and the landing events. We
join the sets of the takeoff and landing events, which have been previously
filtered by removing the noise after the SD-clustering. Then we apply clustering
by spatial positions, to unite the clusters of takeoffs and landings in different
directions occurring at the same airports. We build spatial buffers around the
spatial clusters to obtain the airport areas. In the third step (spatio-temporal
aggregation), we aggregate the trajectories by pairs of places (airport areas) and
time intervals (1-hour length). We use only those trajectories that have both
takeoff and landing events. As a result, we obtain aggregate flows (vectors) with
respective hourly time series and totals of flight counts.
To investigate the aggregates (Step 4: analysis of the aggregated data), we
visualize the total counts on a flow map. The aggregate flows are shown by
directed arrows with the widths proportional to the flight counts. By interactive
filtering, we hide minor flows (less than 5 flights) and focus on the short-
distance flows (less than 100 km distance). We see that there are quite many
flights connecting close airports, particularly in Paris. As explained by a domain
