A debate broke out recently, in a local running online discussion group, concerning how to measure a race course — and the likely precision, or lack thereof, for Global Positioning System readings taken along a route. My experience suggests that GPS pathlength estimates tend to be low by 5-10% on winding, twisty, wiggly trails like those which I most enjoy rambling along. But rather than speculate in a vacuum I decide to harvest a few fresh figures. The results tend to confirm my previous skepticism. Adapted from Brian Tresp's Running Blog sprintbare.com, my notes for the past dozen days of data gathering:

GPS & Leg Tests

(Friday 27 Feb 2004)
I went out today for my first jog since the GW B'day marathon, and except for some minor left knee ache all systems seemed fine throughout the 49 minute jaunt. The GPS (Garmin eTrex "Legend" model) that I carried said I covered 4.60 miles, but my estimate is more like 4.8-5.0 based on measured segments and past experience on the route. Along Rock Creek Trail between mile marks 2.25 and 1.25 the GPS distance was 0.94 miles; on the overlapping Marathon in the Parks segment between mile markers 23 and 24 the GPS registered 0.98 miles. I held the receiver carefully so that it got a good signal, and it never appeared to lose lock on the satellites. Its estimated locational accuracy was typically 17 feet. But as Brian and others have commented, GPS seems to be far short of the required accuracy to measure race courses with their twists and turns ...

GPS on Rock Creek Trail

(Sunday 29 Feb)
~13 miles in ~155 minutes today, gathering numbers along Rock Creek Trail and the Marathon in the Parks course — same Garmin eTrex "Legend" model as before — between Rock Creek mileposts 3 and 7 the GPS measured 0.93 + 0.91 + 0.88 + 0.86 and on the return trip the distances were 0.92 + 0.90 + 0.93 + 0.90 ... average of ~9% short. Between MitP mile 20 and mile 16 and return the corresponding numbers were 0.90 + 0.92 + 0.82 + 0.88 and 0.92 + 0.85 + 0.94 + 0.91 ... ~11% deficit. (But since the courses overlap the data points are not independent.)

Maybe my GPS is using a bad algorithm to estimate distance, or maybe there is something else going on (e.g., user error?!) ... the signals were strong throughout the run (except for brief intervals when the trail goes through tunnels under Connecticut Ave. and under the railroad tracks in Kensington) ... the GPS positional accuracy estimate that the unit displayed tended to fluctuate beteween ~17 and ~45 feet ... as an experiment, after the halfway point I increased the "track sampling rate" to once every 5 seconds — but although that made a somewhat smoother curve on the built-in map display it did not change the odometer reading errors significantly, as per above data.

So I remain mystified ... but I continue to speculate that my GPS pathlength measurements are low by 5-10% because the unit can't sense the wiggles in the course ... see RichardsonianExtrapolation and related notes on http://zhurnaly.com/ ...

Great Weather

(Monday 1 Mar)
"speedwork": 4+ "fast" miles this afternoon (~45 minutes), including a blazing 8:57 along Rock Creek Trail between mile marks 1.25 and 2.25 ... said 'hi!' to several babies in carriers and prams ... lots of couples, old and young, out strolling ... noisy squirrels scrabbling in the leaves, geese honking above ... (no GPS this time!)

More GPS

(Thursday 4 Mar)
Nice 5+ miles late this afternoon, ~10:30 pace, no walk breaks ... from home to Georgetown Branch Trail to MitP route, then back via Walter Reed Annex ... on the measured segments, the GPS again was quite low, in spite of strong satellite signals: its odometer claimed 0.96 mi. for MitP 24-23 and 0.90 mi. for MitP 23-22 (and I followed the correct route; the crushed stone on the detour east of the creek near the Beltway was great — no mud wallows there any more) ... I lowered the 'track sample' rate to 0.01 miles before starting but that had no apparent effect on the results ... superb weather ... frogs peeped loudly in the swamps near Rock Creek ... many cyclists ... team of young girls practicing lacrosse at Ray's Meadow Park ... little kids rushing to pet a shaggy dog near the play area ...

GPS on the GBT

(Friday 5 Mar)
8+ miles, ~93 minutes on the Georgetown Branch Trail ... unseasonably warm ... fire engines race past on Brookeville Road, sirens howl from near Meadowbrook Stables, and then more rescue vehicles come from the other direction via Jones Bridge ... at the turnaround an almost-full Moon rises between high office buildings in downtown Bethesda ... a bat flitters overhead at Columbia Country Club as twilight deepens ... the Moon plays peek-a-boo with the clouds (and the clouds soon win) ... a welcome wind rises as I make it home without twisting an ankle on the jumbled ruts in the first mile of the trail ...

The Georgetown Branch is relatively straight; more GPS tests, with 'odometer' measurements recorded at the half-mile markers (0.5, 1.5, 2.5, 3.5, and then the reverse) give results in miles = 0.97 + 0.95 + 0.98 + 1.03 + 0.93 + 0.93 for an average of ~4% low. I took the straight-line crossing of Connecticut Ave. along the train tracks, but followed the crosswalks in Bethesda. Satellite signals were strong, but obviously I lost them for ~0.2 miles in the tunnel under Wisconsin Ave. The unit apparently took that into account in estimating distance. I increased the "track sampling" rate to 1/second, without much apparent effect on precision.

W&OD GPS

(Sunday 7 Mar)
Big accident on Beltway at 0630 this morning as I set off for northern Virginia — at Georgia Avenue, 6 cars smashed up, surrounded by a similar number of police cruisers (lights flashing) plus an ambulance and a fire engine ... I circumnavigate and proceed west, full Moon hanging huge in front of me, to the W&OD Trail (caboose parking lot in Vienna) where comrades SA & CR join me ... pleasant 10+ mile jog starting before milepost 12 and proceeding to milepost 17 and back ... cool and clear, few other runners and cyclists at 7am but increasing numbers of them as we return ... average pace ~10:30 min/mi outbound, accelerating to ~10:00 min/mi returning (as a group of young ladies passes us and motivates us to go faster) ... beer and omelettes at the Vienna Inn help us recover afterwards ...

GPS measurements for first nine miles only (since when the display passes the 9.99 mi point it stops showing hundredths of miles): 0.97 + 0.97 + 0.99 + 1.00 + 0.99 to turnabout + 0.98 + 0.99 + 0.99 + 0.99 ... net average only ~2% low, much closer than typically seen along the Marathon in the Parks route, Rock Creek Trail, and Sligo Creek Trail ... probably so accurate because the W&OD is almost a straight line, with only gentle curves in this section ...

Brisk Jog

(Monday 8 Mar)
Winter comes back ... the robins are bobbin' and rockin' as they discover it isn't Spring yet and hop about, trying not to blow away ... ~6 miles in ~63 minutes on Rock Creek and Georgetown Branch trails, with timed segments MitP miles 22-23 = 9:54 and 23-24 = 10:01 — wish I could go that fast at that point of the marathon! Beach Drive is blocked off between Stoneybrook and Connecticut; I catch a whiff of sulfurous smell east of the water fountain at Old Spring Road ... is it a leaky natural gas pipeline? sewage overflow? swamp gas?

GPS batteries are depleted and finally die in the cold, but I get one measured mile (MitP 22-23) which the odometer records 8% low ... setting sun flares crimson beneath horizontal cloud bands as I approach home ...

(see also CoordinateCollection (19 May 2002), RockCreekTrail (31 May 2002), MarathonCoordinates (3 Oct 2002), MarineCorpsOrdnance (1 Nov 2002), AnacostiaTributaries (28 Jan 2003), TenLeagueLeyLines (23 Nov 2003), Lincoln Memorial (6 Jan 2004), ...)

TopicRunning - TopicScience - TopicPersonalHistory - 2004-03-09

(correlates: OperatorHeadSpace, WikiWord, GoodFortune, ...)