General News > GPS analysis, the DfT and the National Travel Survey…
February 15, 2012
Recently the Department for Transport (DfT) completed a pilot study to see if GPS could be used in the methodology for the National Travel Survey. Their conclusion was that GPS technology would not be feasible. A sample of GPS tracks are shown on this map…
Having worked on GIS projects for trip route recording and travel mode analysis we understand some of the difficulties that the DfT pilot encountered but we think they may have been hasty in their outright dismissal of GPS as a data collection tool…
Here I will discuss one example of the issues raised by the DfT report – distance travelled – before showing some more detailed sample maps.
Going the distance:
We configure our GPS devices to report their position every second. Of course there are occasional errors in the locations reported by the devices as they are dependent on things such as the number of detectable satellites in the sky at any one time and the proximity to tall buildings that may potentially reflect these satellites’ signals.
While the devices themselves give estimates of the positional accuracy each second, simply looking at plots of the tracks gives a good idea of the devices’ precision.
When the GPS devices are reporting reasonable general accuracy the second-to-second positional errors may be minimal but if they occur every second they can add significant inaccuracy to estimates of distance travelled.
The cumulative effect is nearly always to give an overestimate of the distance travelled.
Not going the extra mile:
This ‘extra distance’ error can be minimised…
Instead of using the distance between neighbouring points, we estimate the speed at each point from several nearby points [1]. We then have a measure of the speed at each second in time from which we can calculate distance travelled [2]. This measure of distance is considerably better than the raw ‘data point’ to ‘data point’ measure.
Sample maps of GPS data analysis – Trips in Newcastle
The maps below show data captured from some test trips in Newcastle to give a feel for the raw data and some of the difficulties of interpretation. The data points are shown as squares and are displayed on a five second cycle. The cycle starts with a small square showing the data point at the first second. The size of the square increases for each subsequent second until it returns to the original size at the sixth second. This time cycle helps visualise the direction of travel and differences in speed when points are close together.
This is a picture of GPS data for 22/04/2010. It shows
In this example bus, car and tram are not identified as separate methods of transport as no bus or tram route information was available at the time of processing making mode identification less reliable. By removing small, still and unreliable tracks we can improve the analysis to identify the road links actually travelled and improve the accuracy of time, speed and distance data collected.
Taking a closer look:
A closer look at the northern section of the map shows motorised transport and walking patterns. There are some spurious motorised sections (filtered out in a later process).
These sections have very poor horizontal accuracy as measured by the GPS device.
In Newcastle town centre the GPS tracks show entry by motorised transport (probably bus) at 12.53 walking around and leaving at 14.09 on motorised transport.
During the walking phase seen below there were some patches of poor GPS accuracy which has caused some mis-allocation of mode – the section showing “cycling” down Grainger Street was due to these inaccuracies causing spuriously high speeds getting through the data filters.
Also in the absence of bus stop data, a forced mode break has been inserted at the Grainger Street/St John Street bus stop. Identifying a point at which the travel mode has a potential for change – ‘stopping at a bus stop’ [3] would be one of many – is an important part of our analysis.
In short we found the analysis of GPS data challenging, but it does give important information that is different from travel diaries.
At Beacon Dodsworth we have implemented an on-line map-driven travel diary software system, TripTrax. We have used TripTrax in mapping the outdoor industry in the Netherlands and Austria alongside the use of GPS devices as a data collection method.
As a check we were asked to compare the results of both methods of GPS analysis and TripTrax for a sample of tracks. The overall statistics from the two methods gave reasonable correspondence but one of the things that the comparison highlighted was the impact of human error. More frequently than we would have expected, respondents forgot journeys they had made or described journeys they simply could not have taken. Whilst both data collection methods had their own advantages, our advice would be to use a mixture of both methods in order to get more reliable and robust data.
We feel that the Department of Transport are mistaken in asking for an ‘all-or-nothing’ report on using GPS for the National Travel Survey. Considering an approach that incorporates several different data sources and collection methods could give better results and be more cost effective.
Posted by Geoff Beacon, Chairman at Beacon Dodsworth.
All maps copyright ©Navteq and ©Beacon Dodsworth.
[1] The choice of these points can be modified by the device’s measures of accuracy.
[2] Knowing the speed at every second gives a good measure of the distance travelled.
[3] A ‘stopping at a bus stop’ test depends on a fuzzy mixture GPS derived parameters to decide when ‘at’ and ‘stop’ conditions are met simultaneously.
