Saturday, February 12, 2005

Knowing the Distance - An Interview with Dr. David Dreyfuss

How do you know that a trail marathon will equal exactly 26 miles and 385 yards? Few things are as frustrating as clocking a PR, only to realize that everyone clocked a PR thanks to the course being two miles too short. But then there are folks like Redwood Trails, who will start many races by saying “you MUST start at this post, otherwise you'll be 90 feet short”. Curious to how they can get so accurate on a varied terrain, I tracked down the man behind the method, Dr. David Dreyfuss, who addresses “accurate measurement” like a true PhD from MIT, helping dispel rumors of GPS accuracy, and explaining what it really takes to measure a course to the very last inch.


(Dr. David Dreyfuuss, out on the trail)

Brief background on who you are and how you got into the trail business.

DD: Well, I've been a lot of things and worn a lot of different hats over the years. Roughly, I still think of myself as an experimental physicist, though I have a doctorate in Gas Dynamics (MIT, 1980) and have spent much of my recent professional career developing digital printers--laser printers and the like. I came out to California to try my hand at a start-up ventures in new color printing technology, helping launch Google Answers, and rebuilt the website and e-commerce business for Aqua Safaris, a Santa Cruz SCUBA shop. I'm also a board member and edit the quarterly national newsletter for the Viola da Gamba Society of America - and yes, I also occasionally perform on the instrument.

So how did you find Redwood Trails?

DD: I started my relationship with Redwood Trails by responding to an ad for aid station staff. I've never been much of a competitive runner; I can probably count the number of races I've entered on one hand, and I'm not particularly fast. However, I do have a lot of useful skills related to trail running. I have thousands of miles of trail and wilderness travel experience, hiking, backpacking and recreational running. I spent three summers leading backpacking trips as a Ranger at Philmont Scout Ranch. I also have a lot of experience orienteering, which is to trail running roughly as trail running is to road running--it's off-trail running through the woods with map-and-compass navigation between control points. So I already knew many of the trails that Redwood Trails uses for events, and the basics of scouting, navigating and marking a race route are kind of second nature.

It would seem with the advancement in GPS that measuring trails these days is pretty easy.

DD: GPS (Global Positioning System) has revolutionized many aspects of navigation, mapping, surveying, and position determination. A $100 handheld receiver can tell you your location in three dimensions (let's say, latitude, longitude and elevation) to about 10' accuracy, at least some of the time. With slightly more expensive equipment and some patience (to average out some randomness that exists in the measurement due to things like atmospheric effects, and to allow time for the satellites to move around a bit), you can locate a single position at least of couple of orders of magnitude better--good enough to match or exceed the accuracy of typical surveyors' instrumentation.

But while GPS continues to get better, it is still an imperfect technology. To get the 10' accuracy, you need what is called "differential" GPS--simultaneous measurement of a known location so that some sources of inaccuracy can be subtracted out. Right now, differential GPS is readily available in the US via a couple of extra satellites that continuously broadcast the measured location of a set of fixed ground stations, which in turn have well-known absolute locations. Unlike most of the GPS satellites, which are in relatively low polar orbits, these two extras are in geostationary orbits near the eastern and western horizon. Unfortunately, that means they are often out of sight behind a ridge. Without their assistance, the accuracy drops to more like 30' at best. And, although nominal accuracy is the same for horizontal (lattitude and longitude) and vertical (elevation) positioning accuracy, practical experience shows that the elevation accuracy and repeatability is usually worse than the horizontal accuracy).

There are about 24 satellites in the system right now. In principle, you can get a position fix as long as you can see at least three of them at any given time. With an unobstructed horizon, you can typically see at least ten, but if your view of the sky is partially obstructed by hills or even dense trees, you can easily lose sight of most of them. Accuracy drops as the number of visible satellites goes down, and also gets worse if the visible satellites are clustered too close together. Airplanes and boats can navigate quite accurately using GPS, because they almost always have a good view of the sky down to the horizon. Car navigation systems generally cheat by assuming that you must be on their known network of mapped roads and using dead reckoning by direction change and wheel rotation when GPS signal is lost (occasionally generating some amusing errors, when their best efforts fail!). In any case, such cheating is not useful if your goal is measurement instead of navigation! Trail users in mountainous and wooded terrain find GPS to be useful, but only intermittently. Tests on numerous Redwood Trails routes have shown that there are always portions of any interesting run where accuracy is poor, and occasional complete loss of usable signal is very common in most locations.


So, given that the technology is not really there and perhaps never will be what does work?

DD: All the "old-fashioned" measurement techniques! You can choose any number of specific tools depending on your goals and the accuracy you need or want--surveyors' transits, tape measures, odometers, pedometers, etc. If you're "lazy," you can try to estimate distance off a map or by using published or posted trail distances, and reading elevations from a topo map. Many trail events are, in fact, organized based solely on such estimates. Occasionally we find specific trails that have been accurately measured and are accurately mapped and posted. More often, we find that such measurements are approximate at best, and large errors are not uncommon! We encountered one example where an advertised marathon route was short by several miles.

We measure all of the routes we use ourselves. Our preferred instrumentation is a wheel rolling along the ground. Depending on the circumstances, we use either a "measuring wheel" or a bicycle, essentially counting revolutions or fractions of a revolution of the wheel. For elevation, we use a barometric altimeter. When we generate a profile for a new route, we typically record elevation every 500' and both elevation and distance at all major landmarks (trail junctions and so on). This can be done with pencil and paper, but more recently, we usually use a pocket tape recorder with a headset microphone which allows data recording almost without stopping. Barometric altimeters are accurate to a few feet as long as the weather is stable and a reference altitude can be found to zero out the day's weather conditions.

How can you ensure accurate measurements?

DD: As with any measurement, accuracy requires attention to detail. And, of course, "precision" does not guarantee "accuracy"! We use a measuring wheel which reports measured distances in feet, but not surprisingly, we never measure exactly the same number of feet for a given trail route. Trails often have rough surfaces, and it is almost impossible to roll a wheel along exactly the same detailed path on a repeat measurement. The accuracy of a measuring wheel also depends on the constancy of the circumference of the wheel. If the wheel acquires a layer of mud, its diameter increases. It may expand with increasing temperature. It may wear with use. If it has a pneumatic tire (as do most bicycles), the circumference also depends on tire pressure and rider weight. For best accuracy, one needs to check the calibration of the wheel against a known distance on a similar surface before and after the trail measurement.

So, do you use the same method that is required by the USATF to certify a course.

DD: Essentially yes! The USA Track & Field Road Running Technical Council establishes measurement standards and procedures that are used for all events for which distances are certified for record purposes. The process requires the use of a calibrated bicycle with a special counter mounted to the wheel. A calibration course of at least 1000' must be run four times before and after each set of measurements. A course must be measured twice along the shortest possible route (usually not much of an issue for trail runs). Repeatability must be better than 0.08%. A "safety factor" of 1.001 is used to ensure that the actual race distance is AT LEAST the advertised distance. If a new record is set for the distance, the course may need to be recertified for distance, and the safety factor helps ensure that the record will not be disqualified for being set on a "short" course. So next time you think you're running 26.219 miles for a marathon, if it's a certified course, you're likely running at least an extra 138 ft--more of you don't cut all the curves as tight as possible. Or think of it as 5 ft per mile--not all that significant.

We went through this process to certify the Bizz Johnson Marathon so that it could be used as a Boston qualifier. Most of our trail races have no real need for formal certification. Each is a unique race with special topographic challenges, and records are really only meaningful for that particular route. But Redwood Trails tends to be obsessive about accuracy anyway. In most cases we relax our measurement accuracy requirements by skipping the repeated calibration checks and simply double-checking the race-day distance measurement against our previous profile run. But worst case, our distances are still accurate to within better than a hundred feet on a 10K. (If you can feel that difference when you're running, let us know--we'll hire you!) And while the RRTC requires the use of a calibrated bicycle, we usually prefer to use a measuring wheel instead, both because it is allowed on all trails while bicycles are not, and because it can go essentially anywhere you can run or walk (or even climb!). We do use a measuring wheel with a large circumference (six feet--like a bicycle wheel but with a solid rubber tread on a steel rim--makes a very stable circumference) which we find does a better job of averaging out small trail irregularities than a smaller wheel. Curiously, Rolatape claims accuracy for their measuring wheels without special calibration of the same 0.08% as the RRTC targets for measurement repeatability!

How much time can trail measurement take?

DD: Well, as with any careful measurement, you can't rush the process, and you have to be prepared to back up and start over if something goes wrong or measurements don't repeat to sufficient accuracy. It took us the better part of two days to complete the certification measurements for the Bizz Johnson Marathon. We had to establish our own custom calibration course on a section of the Bizz Johnson Trail surface, measuring it with a 100' steel tape and correcting for temperature. Then the measuring bicycle got a flat about eight miles out, essentially requiring that we start over. After that we chose to do segments of no more than about seven or eight miles at a time complete with repeats and calibration checks before doing the next section, so that in the event of another "instrument failure," we would not lose too much data. But we never had problems with repeatability, typically about 0.02% even on the relatively loose dirt surface of the Bizz Johnson Trail.

What about marking the trails on race day?

DD: We try not to make trail running an exercise in navigation! Many runners are accustomed to road events where they are pretty much forced to follow the race route, and they often tend to "zone out" while running. On a trail run, they do have to pay at least a little attention to where they are going, but our routes are well-marked, complete with mile markers, and we double-check for vandalism immediately before start time. The mile markers are the main reason that we have to measure the trail one more time for race day--we can't leave permanent markers in place. In addition to mile markers, we typically post direction signs at all trail crossings and turns, generous wrong way signs along incorrect route choices, and orange ribbons along the correct route away from any trail junction or intersection. We also may station a "traffic cop" at any difficult intersections where runners may be taking multiple routes, either for different races or different legs of the same race. The marking process isn't fast either! It can easily take six hours or so to fully mark a half-marathon route even if one runs between marking locations. And, of course, there is unmarking to be done after the race as well!

What mileage do you use for race results?

DD: I presume you mean, "what distance do we use to calculate "pace" for a race?" We use the actual measured distance, of course (including the 1.001 "safety factor" if the race has a certified distance). Trail races are often advertised at "standard" distances (marathon, half marathon, 10K, 5K). Where possible, we try to adjust the start or finish location, or the length of an out-and-back dogleg to make race distances "right," but sometimes that just isn't practical, and the real distance may be a little longer. Our courses are never short, because we can always add a little extra distance somewhere if the natural trail length is short. If a race is long, we try to let runners know, both before and after the race, exactly how far they did go (though if they ignored all the wrong way signs, they're on their own!).

At Redwood Trails events you always have results available right away. I understand that you using your own timing software. Aren't there existing commercial software packages out there?

DD: Sure, there are existing commercial packages to manage race results. But as one of the most active race management organizations around, we tend to develop our own ideas about how things should work, and it can be problematic to impossible to customize someone else's package. I've been programming for some 35 years now, and had plenty of support from Eric Gould (who has award winning software design credentials himself), my wife (another MIT PhD), and my son Ethan who is now a freshman at Stanford. We wanted a robust, easy-to-use package that would allow us to manage our complete events from pre-registration through immediately available results at the event and same-day posting of results with an easy-to-use interface on the web site. It might seem that there's not much to the task of recording start and finish times for a couple of hundred runners, but it turned out to be a good deal more challenging than we had expected.

Thanks for the interview, David. Looking forward to the Palo Alto Vista Run on March 5!

Cheers,

SD

12 comments:

  1. Yeah, you have to look at the fine print on many advertised marathons. The Breckenridge "Marathon", for example, is only 24.5 miles.

    Dana

    ReplyDelete
  2. I wish there was a way to assign a "degree of difficulty" to a trail. How can you look at two 1/2 marathons and know which is the most difficult? Accurate distance and altitude change is one thing, but "steepest ascent", " trail condition", etc., can also factor in. I've run 10ks much harder than some 25ks, and I've never got the hang of looking at the maps to understand it.

    Have you ever thought about adopting a "degree of difficulty" scale?
    - Erin

    ReplyDelete
  3. But you still use the GPS for altitude measurements, right?

    ReplyDelete
  4. Man that sounds like a lot of work. How long does it take to certify the distance of a marathon? Have you ever done a 100-miler?

    ReplyDelete
  5. While I understand your desire to have “degrees of difficulty” posted for races, it's a concept that is difficult to implement in practice. There are, of course, a number of factors to consider. Some are clearly a property of the particular route: average altitude, total climb, steepest climb, number of steep climbs, trail surface, distance. Others are a property of the race day: wind, temperature, rain, sun, mud. And different runners respond differently to these challenges. Some adapt easily to cold or heat. Others may be more or less sure-footed on rough terrain. Some cope better with steep slopes or long uphill sections or may be better adapted to the altitude, or be better able to speed up on downhill sections. I think as race organizers, the best we can do is to provide good information about the particular event. Even a composite number like “total climb” is kind of slippery, because it depends on how fine-grained the elevation data you use is. And, of course, race promoters are always going to play up the “attractive” features of an event and play down things which might discourage entrants. One thing I can suggest, though, is to learn to look carefully at the scale of an elevation profile. Profiles may look very similar at a glance, but if one has 10 times the vertical scale range, you know you’re in for more climbing. And look for elevation changes over moderate distances—say how much climb over a mile. Most people can cope with short climbs without slowing down much, but a long steady climb is more of a challenge.

    —David Dreyfuss

    ReplyDelete
  6. Actually no, we don’t use GPS for altitude, except possibly as a starting reference to determine the offset for barometric altimeter measurement if we don’t have anything better. The accuracy of GPS altitude measurements is often significantly worse than latitude/longitude. Now, as it happens, the barometric altimeter that I use is built into a hand-held GPS unit, so you might mistakenly think I’m using GPS if you see me out measuring!

    —David Dreyfuss

    ReplyDelete
  7. Certifying a marathon distance is a day or two’s work depending on ease of access and such. It takes some time, certainly, but so do many of the other organizational tasks that go into putting on an event. And if you think that’s a lot of work, you should try preparing for a major orienteering event! An “A” meet is generally conducted on a newly drawn map (so that no competitors have an unfair familiarity advantage), which can take a few man-weeks to prepare!

    I’ve never done a 100-miler, but the process wouldn’t be any different. If you need to get it done fast, you might use multiple measurement teams doing different segments, but one or two people could do the whole thing over a few days.

    —David Dreyfuss

    ReplyDelete
  8. I assume all of your comments are with the understanding that truly post-processible GPS has been out of reach for the consumer (and most RDs). That is not the case. I have recently measured the Wasatch 100 and the Katcina Mosa 100K Mountain Challenge with a Thales PP GPS unit that sells for the ridiculously low amount of $1800. Plotted over a DEM the tracks from this unit on the treacherous Wasatch trails (which beat the snot out of my Rolatape) yields a repeatability of <0.02%. A wheel cannot touch that. Wheels are doomed.
    Katcina Mosa 100K, by the way, is the most accurately measured trail race in the country, having been differentially surveyed. Looks like the Redwood race has a bit of catching up to do. ;-)
    BTW, I play string bass with the Mormon Tabernacle Choir. What is with lower strings and anal retentiveness?
    Cheers,
    Phil Lowry
    www.phillowry.com

    ReplyDelete
  9. Mr. String Bass Player guy (Phil) is truly the BEST mapper in the world. Apparently he has been able to get "better" accuracy than Dr Dreyfuss' wheel with his Thales GPS. No setup for differential correction (because that would take too long). Thales website states "sub meter" real time accuracy. So let's say that Mr Bass Player guy can actually fix enough points along his trail to "sub meter" accuracy. What does he have? A trail that is accurately measured IN LOCATION not the distance or length. All GPS software use interpolation for the distance measurement inducing inaccuracies between the known fixed points. This is best case. Worst case is the trail twists and turns between points inducing more error. It would seem you would need something that follows the ground accurately... like maybe a wheel... to measure with a lower amount of error in DISTANCE (Which is what we are trying to measure anyway) Looks like Mr Bass Player has some catching up to do!

    ReplyDelete
  10. My data are based on a track point plotted every second and then making the same track again in reverse. The track points generally set out at one per footfall. They are later differentally corrected. Did b_upright@hotmail.com (since he was too humble to post his name) miss this, since I stated that the Thales was post-processible? It looks like he didn't read much of my other data on my website, either, but by his tone I don't think he was very interested in comparing data. Those of you who are, feel free to respond. Thanks!

    Phil

    P.S. If you don't like anonymous posters clogging the blog with insults, let Mr. b_upright know.

    ReplyDelete
  11. Sorry if came off as mean, Phil. It just seems like you were flaming the good Doctor with your arrogance. It also appears that you still miss the point. Your system is less accurate than his. We are measuring DISTANCE here. No one really cares where the Trail is (as long as you can find it). The question here is How Long is it?

    Let's look at the facts:

    1. You can measure very accurately every 1/2 second.

    2. And giving you the benefit of doubt, let's just say that you are able to set up a correction benchmark ahead of time and your accuracy is 1 cm (I doubt that you get this in real time).

    3. And lets say (since you said it was fast) that you do it at a walking pace of 3 miles per hour. Three (3) mi/hr = 4.4 ft/sec so you have a very accurate measurement every 2.2 ft.

    4. In order for your system to be more accurate than Dr. Dreyfuss (0.02% error) you must measure your two points between a section of trail that does not deviate (in elevation or curvature) over 0.00044 ft (0.02% = 0.002 * 2.2 ft = 0.00044 ft).

    5. With a MAXIMUM deviation of 0.00044 feet allowed in the trail from point to point, this works out to be a curve with a MINIMUM radius of about 11,000 feet. i.e. your trail cannot curve (or change elevation) with a radius of less than 11,000 feet. A highway designed for 70 mph typically has a minimum radius of about 1,800 feet for example. So your system won't be accurate enough for an Interstate Highway, much less a trail.

    6. "Post processing" has little if any benefit since the algorithm involved still interpolates between the two points but offers no additional actual measurements and therefore is only "a better guess" than a straight line. Overall your PP may provide a small decrease in error. Let's say 10-20%. Way below the estimated 2,000% change you must have to make you system viable.

    Sorry I posted anonymously but I just didn't want to take the time to fill out the form (I will try and do that today). Please send your regards to your buddy Mark. And please let me know if you need help with the math on this proof.

    See you at the next race.

    Best regards,
    Tim

    ReplyDelete
  12. Good stuff here. Would Dr. Dreyfuss be willing to share his experience how a fixed circumference measuring wheel that is calibrated on a flat road course does on a rough mountainous trail? Assuming the wheel is kept on the ground, with the many undulations over rocks and roots and waterbars, etc, can we expect the wheel to overestimate the distance by 1%, 2%?

    Also, how long are his calibration courses and how does he get a representative section of trail? It would seem that trail courses have wildly varying types of terrain, from smooth to very rocky, and using only one calibration course might not be representative for the whole trail.

    Thanks.

    ReplyDelete

I LIVE for comments! Please add your thoughts, let me know you stopped by, etc., and be thoughtful of others. Always best if you sign your name, of course.