Head to HEad--is post-processing that much of an advantage in trail mapping?

On May 12, 2008, at 4:18 pm, I conducted a test of a recreational GPS and a GPS whose data could be differentially corrected. (I assume here an understanding of GPS technology.  For a detailed explanation of how that technology works, see What is GPS? ).  The test equipment consisted of a Garmin Colorado Colorado 400t, a recently-released product with the ability to take position samples every second.  The differentially correctable GPS was a Thales MobileMapper Pro.

The objective of the test was to see to what degree of accuracy the two units could depict a terrain feature, here, a railroad bed that has been converted to a paved parkway.  The larger question is whether these types of GPS units are suitable to map roads, dirt roads, jeep trails, established trails, and traces.  All of these features are used in the burgeoning sport of trail (or fell) running.  This type of running has grown worldwide.

Notwithstanding this growth, the means of providing participants with meaningfully accurate maps has fallen far behind the needs of the participants.  In road races courses are marked with chalk, pylons, and other striking means, with crucial turns often manned by volunteers or police.   Not so in trail events.  In the early days of the sport map and compass were de rigueur, as were excruciatingly detailed trail descriptions.  Markings were few and far between.  Participants were self-selectedly hardy and self-reliant, but nonetheless still got lost on the courses.

Expectations have evolved in the last 30 years.  It is now common for trail races to be marked at regular intervals of flagging, chalk, or other markers.  Night races are marked with glowsticks.

Nevertheless, and despite race directors’ best efforts, runners still get lost.  The reasons for this, apart from the usual addling that accompanies a harried or hurried runner, are varied.  Vandals abscond with markers (or worse, mark the wrong trail).  Wildlife, especially the American elk, have been known to eat flagging.   Chalk wears out.   Race directors may not always anticipate where flagging is needed.  And volunteers, with the best intentions, may mark a course improperly.

The common solution to the problem of runners getting lost is to provide detailed maps.  A race can insulate itself from runners’ criticism if a detailed map is provided.  The runner is expected to use the map when in doubt.  One very famous American 100-miler, in which dozens of runners have become lost, boasts that not a single one of those runners was carrying the detailed maps and instructions the race provides.

Unfortunately, detailed maps are difficult to produce.  Terrain features such as trails change from year to year, rendering topographic maps partially obsolete.  Hand-drawing maps is tedious, time-consuming, and inaccurate. 

GPS technology provides a possible solution.  By creating a GPS “track” (a literal breadcrumb trail in which a “crumb” is dropped into the GPS memory every second), and then overlaying that track on a digital topographic map, one can readily create a detailed course map. 

But is GPS truly accurate enough to render maps that runners can use?  This experiment attempts to answer that question.

A corollary question, but equally important, is whether GPS tracks can effectively measure course distance.  In road races course distances are measured by a calibrated bicycle.  Such a method is impractical, and even illegal, on rough courses or in protected areas.  Some trail races have gamely attempted to measure trail with a calibrated wheel (measured against steel tape), but these results are generally somewhat short and very difficult to replicate over distances of more than about three miles because of measurer fatigue and wheel slip. (This has been the subject of other experimentation and not the subject of this paper.  The assumption here is that wheel measurement is impractical in a trail environment, and in any event cannot create a map.  Thus, the question is whether an accurate track measurement suitable for mapping can also be used to calculate distance in both an accurate and efficient manner).

The concern over using GPS to measure trail races is whether a GPS track can be a true rendering of the terrain features, and whether such a rendering can yield a correct distance.

The experiment in this paper does not seek to debate whether GPS distances are more accurate than wheeled distances.  Rather, the experiment is to determine how costly an investment is required to accurately render terrain features.   If those features are rendered accurately, the next step is to compare a GPS against a calibrated wheel course to determine the variation in distance calculation, if any.


Both GPS units were equipped with external antennae collocated on my hat.  The experiment began with good satellite lock for both units.  I commenced running on the parkway along the fixed yellow line separates the running path from the bike path.   The line was very convenient because it ensured that I followed the exact same path on the outbound as I did on the inbound.  The only variation was in the beginning and end of the test.  I began calculating the track from the parking lot, but finished on the parkway itself, near the parking lot. 

I chose the parkway because I was guaranteed a repeatable path without unusual obstacles, yet was in an environment very hostile to GPS—dense foliage, extreme cliffs on either side, and many twists and turns.  It was an ideal test setting, replicating in many ways a dirt trail in the mountains yet amenable to precise measurement technique.

The software I use to depict the track is AllTopo, by igage Mapping Corporation (www.igage.com).  This software is about 90% as robust as Arcview, but 90% less in cost.  It allows importation of any georeferenced tiff, thus my use of 1m resolution orthographs in the figures below (Utah County stupidly did not buy in to the state 25cm orthograph project, but I digress).  It uses a high resolution DEM that I have found accurate and easy to use.  The profiles and distances calculations in the figures below are generated by the software.

The overall figures below show the Colorado track (in red with pink track points) and the Thales track (in blue with turquoise track points).   From a distance the two tracks look similar,


Thales trackThales Track    Colorado TrackBoth tracks overlaid

but the close-up views below show both how close together the Thales inbound and outbound legs follow each other, and how erratic the Colorado track is, both to its inbound and outbound legs and to the Thales.  The Colorado error is often no more than 20-30 feet, but at one point deviates by 150 feet.


The test included one incredibly challenging terrain feature for GPS—an underpass under the highway.  The figure below shows the havoc this wreaked on both units.  Note how quickly the Thales recovered, while the Colorado stumbles along, almost reeling.  I took on this challenge for this very reason—how would GPS respond to such a signal loss?

The distance computation differences are as follows:

Thales overall:                   9.611 (Map)        9.637 (DEM projected)

Thales Profile

Colorado overall:              9.921 (Map)        9.962 (DEM projected)

Colorado Profile

The next measurement eliminates the underpass errata, and calculates the distance from a point where the Thales data converges east of the underpass to the terminus and back to that point again.

Detail of cut routes

  It thus proves a true inbound/outbound comparison:

Thales overall:                       4.800 (Map)        4.813 (DEM projection)

Outbound                           2.400 (Map)        2.393 (DEM projection)                

Inbound                            2.410 (Map)        2.404 (DEM projection)


Thales Profile

Colorado overall:              5.118 (Map)        5.144 (DEM projected)

Outbound                     2.531 (Map)        2.546 (DEM projected)

Inbound                      2.586 (Map)        2.597 (DEM projected) 

 Colorado Profile

I made similar splices on the other side of the bridge, with a corresponding splice at the route end to compare similar distances (before the legs diverged).



This comparison is significant, as the canyon is more open in these lower reaches:

Thales overall:                    N/A

Outbound                        2.174 (Map)        2.180 (DEM projection)                

Inbound                         2.174 (Map)        2.180 (DEM projection)

Colorado overall:                 N/A

Outbound                        2.183 (Map)        2.191 (DEM projected)

Inbound                         2.220 (Map)        2.227 (DEM projected)


For general mapping a recreational GPS is acceptable to roughly depict terrain features and routes, but is not acceptable for distance calculation or for detailed depictions of terrain features.  Upon attempted replication, recreational GPS, even with a powerful external antenna and the latest generation SIRF chipset, cannot produce sufficiently reliable data to plot extremely detailed courses or to render distance.  Rough terrain severely exacerbates the inaccuracy.  Errata generally exaggerate distance to such a degree that those distances should be considered an approximation only, especially in challenging terrain.  In contrast, where satellite acquisition is favorable, recreational GPS can come within 1-2% of distances yielded by post-processable GPS.  In ultra distances this degree of error is unacceptable, but should be acceptable to produce maps, especially where commercially available maps are substandard.

Objectively, the post-processed track stays very true to the terrain feature in the experiment, as depicted by the orthographs below.  It appears to be accurate enough to render precise distance calculation, in addition to accurate mapping.   The next step is to compare post-processed GPS data against a calibration course used for road races, an idea conceived of by Ross Zimmerman.  When that data is examined we will know whether GPS is the next tool for measuring trail races.



Thales shape file        Colorado Shape File    Thales .csv File (lists data for every track point, including satellite lock data, in Excel)