Distance accuracy - GPS vs Computer w/measured course
 

All,

I am curious if anyone has any numbers that can explain travel distance accuracy in GPS.

For longer rides I normally ride with a 7" tablet running an app called GaiaGPS to record my trips. The bike computer is an Edge 1000 Garmin. For the longest time I used a chart with my wheel size to get a cal number (2187). I never really looked past it. The numbers never seemed "off" that much to me. However, what caught my attention was a ride I did a while back that was 52-53 miles. When I document my ride data it comes from the Edge 1000. This particular ride I made a video about it and used the data from GaiaGPS. Looking at the two sets the distance stood out to me. 51.7mi on the tablet and 52.8mi on the Edge 1000, so 1.1 miles different, or about 2%.

Most of what I use the GaiaGPS for is a track (which I know the Edge 1000 does also) and I can get to a zoom level on the 7" screen that lets me see local roads at a resolution that is also a wide enough zoom to get a broader view of the area. For a lot of environments I ride in that is really useful. If I am navigating on-the-go I will use the tablet every time then after figuring things out I may go to the Edge 1000 to drop a pin and route to it, sometimes not. I had a thread on here a while back about trying to get the Edge 1000 to route correctly. It still doesn't route how I would like it to, but I am learning to live with it.

Last year I was trying to dial in the wheel cal on my folding bike and did a 200ft measured course then counted the wheel turns. I don't think this was super accurate as the difference between the computer and GPS was noticeable still. I ended up splitting the difference by figuring out the percent error, dividing by 2, and adding that in to the bike computer.

About a week ago I decided to do the 200ft course again with the big bike - the same one I did the 52-ish mile ride on. I came out with the exact same wheel counts on 3 runs - 28.6. The cal number came to 2131.47. I rounded this up to 2132. The original number I had was 2187, so this is quite a bit less (2.6% different - bigger than the error calculated in the difference with GPS before).

Back to the question on distance accuracy with GPS - I have seen vertical accuracy change in the range of 100ft sitting still for 2-3 hours. From past experience I know the position (lat/lon) accuracy can change as well, but not to the extent vertical accuracy does. I would venture a guess that consumer-grade GPS probably doesn't even stack up to the base-level accuracy of industrial GPS. Also of note - the accuracy I am familiar with is not over distances. It is side to side positioning in parallel tracks and elevation (grade control for creating drainage slopes, cut/fill, etc). So course distance traveled over x amount of time is a whole other ball game Ive never dealt with.

Am I correct to state that the longer period of track recording the worse the accuracy is going to get - speaking of GPS accuracy here? Just like if I ride for 6 hours on an out/back ride and on the return, going over the same terrain, the GPS records an elevation difference of 70ft whereas if I do a shorter out/back ride the difference may only be 20ft? Or is there a percentage of accuracy that has been commonly seen, or accepted, as "expected" with respect to GPS accuracy over longer distances and time?

I guess another question - does the GPS calculate distance from the georeferenced (lat/lon attached to waypoints along) track or is it adding up based on speed or distance traveled?

If the positional accuracy is, say, within 12ft and we take that as a worse-case scenario as that whole 12ft - if we travel a distance of 20 miles is that 12ft (worst case, probably never would happen) attached to every georeferenced point along that 20 miles?

When I was in college in one of my classes we used an agricultural GPS that spit out NMEA data on RS232. We let a computer log the NMEA for a period of time, then ran the data through a spreadsheet to calculate the accuracy (PDOP, HDOP, VDOP included). All of that happens in the background on consumer GPS's and all we see is the result of those calculations. As I have seen between the Edge 1000 and Strava, even, is different displays of the data (the Edge 1000 system vs Strava, respectively) calculate the data results differently.

To the original point - is my 200ft measured course doing the job? Should I compensate a bit for the error still seen in the GPS results? Or should I find a straight clear parking lot somewhere to measure off an even longer course?

Another note is I have been using my front wheel for the sensor. I do realize the front wheel travels a further distance than the rear due to it always being on the outside of turns compared to the rear. I can see where that may make a more significant difference on a short course, however with my 3 runs being exactly the same reading I would think it would be close enough.

WOW,

I may not completely understand your post reference because I am not all that smart I will be honest. LOL

When I used to ride in groups we all use a different devices of same brands, different brands and compared numbers. What we find is even if we start our GPS at the same point, we rarely have the same distance. Over a 20-30 mile ride we would sometimes at worst be off by .2 of a mile. Usually we are at .1 or closer though. I don't think that is significant at all, close enough.

I use a wheel sensor with my Garmin 1030 and I feel like my numbers are pretty consistent and accurate being I ride the same routes a lot and usually very close.
I have only once rode with two different GPS brands at the same time, no sensor, using only GPS and after about 25 miles they were within .02 miles that one time.
I don't worry about .1 or .2 mile differences. But I don't have a need for it to be perfect.

Sorry if my ramblings are no help to your question......I really don't under GPS all that well or how it works I guess.

The GPS saves points at a slower rate than the wheel sensor. That means the path described by the GPS points will have "short cuts". It's also measuring horizontal distance (which is less than the distance that includes elevation).

The bicycle isn't traveling in a straight line either. The front wheel "wobbles" more than the rear but the rear is wobbling too. Even without the "wobble" the bicycle is passing over more distance than the GPS will measure.

The wheel sensor depends on an accurate measurement of wheel circumference (which depends on the tire. pressure, and load). The "best" circumference is determined with the roll-out method with an appropriate load.

The GPS determines lat/lon (on a model surface of the earth). It also notes the time. It can determine distance and speed using that.

The wheel sensor counts wheel rotations. With wheel circumference, the head unit can calculate speed and distance.

Both of these accumulate small distance measurements to calculate the total distance traveled. The small differences are distance from prior recorded point for GPS and a single wheel rotation for the wheel sensor.

https://support.garmin.com/en-US/?fa...623Z535geTx2e9

Our recreational GPS's aren't as accurate as TV will have you believe. To many things that interfere with that weak signal coming back from space. 33 feet is as accurate as you should hope for any one point. Multiply that by all the other points on your ride and that can add up. Your gps's have a read out that supposedly tells you what it's accuracy is from moment to moment, however the makers don't really tell us what this really means. For Garmin's it's been suggested that perhaps it means 50 percent of the results for that position came back within the radius reported for accuracy. However you are moving most of the time, so it's not getting many plots of that same spot to compare against.

Survey grade GPS's use a supplemental transmitter/s... I believe, that is/are located at a known position on the ground to help it more accurately digest the satellite information. What agricultural GPS's do I have no idea, but certainly it must be more along the lines of survey grade instruments.

So how accurate is your need? If you are doing studies that need survey quality, then you need to fork out the big bucks for survey type equipment. $10,000 might get you started.

As for the wheel sizing you did..... is there a wheel sensor on your bike that connects to both GPS's? If not, then wheel travel is considered more accurate. But only if calibrated correctly. I will either put a drop of white out or paint on my wheel and just ride it for two full turns or so, then measure the distance between first and last spot, divide by appropriate number and then convert to mm as my tape measure is not metric. Most devices I've seen want that number in mm of wheel circumference.

Don't know what you were getting at with the 200 foot measured course. Who measured it? Was it straight? If not, staying to the inside of turns might throw some inaccuracies into the mix if the measure was down the middle of turns, and enough turns are involved.

And then there is also where did you look at the data? On the device themselves or from data uploaded to website or software? Some websites and software will derive their own figures for distance from the log file data. Log data on most devices is at best 1 second apart. If your route involves sharp turns or doubling back, then corners may be missed or part of where you doubled back. Remember at just 13 mph you are going 20 feet per second.

For recreational stuff, your figures are good enough. Don't dwell on it too much.

The only way anyone could hope to determine accuracy with that kind of fidelity would be to ride a preset course, identically, over and over again at a fixed speed-- like riding on a selected line on a running track without deviating. Each lane on a IAF standard track is 8 meters longer than the lane inside it. So out on the road, taking a turn to the inside vs. taking it as wide as possible can add several meters of distance traveled in a single turn.

What really works is picking a certain means (data from a GPS computer, a wheel speed sensor, etc) and just going with that. Even with basic, consumer-grade equipment, the margin of error is going to be one or two tenths of a percent at the worst.

Note as well that the Garmin "might" have a somewhat more accurate antenna for determining location, then the tablet.

And as NJ stated, the Garmin and the tablet GPS "cut corners" in that the devices log position (i think of it as sampling rate essentially) points are about every second ?, (I believe). Thus if the course is reasonably curvy, all GPS devices without supplemental wheel sensors, will have the distance a bit short. This is especially true for mt. biking on twisty single track and is why a lot of folks using GPS devices for tracking will add a wheel sensor that uses the earths magnetic field to generate distance (over-riding the device GPS in the process). This yields a more accurate measurement. And FWIW, 2 buddies I ride with have Garmin as well. MIne's a 1030 with a wheel sensor, another's is a 1030 no sensor, 3rd is an Edge Explore, no sensor. We end up with 3 different distances for the same ride, always.

Consumer grade GPS accuracy is usually stated as being about 11 ft. or some such when all is working perfectly. I've never checked... Commercial grade used for surveying, aviation, maritime use also picks up supplemental "Differential" positioning data from ground transmitting systems which improves accuracy, though it doesn't matter for our purposes as much, as the position data sampling rate on a bike ride is really all that matters.

Vertical accuracy is a complete crap shoot as by nature, GPS is poor at determining altitude and is why the pricier units like the 1030 use a barometric sensor to help improve accuracy. But do a search of folks unhappy with what their assorted GPS units state was their elevation gained on a ride, it varies all over the map (no pun intended) and then Strava changes it anyway.

Since none of you is going to be riding the same path, the distances will always be different (unless, they were the same randomly). We really would need to know how different they are (see the actual numbers).

Garmin has an option to "correct" the elevation. It does that by default for units that don't have barometers (suggesting that Garmin thinks that barometric elevation is better than elevation derived just from GPS). Strava (it would seem) is applying the correction to everything. Since a major point of Strava is comparing results with other people, treating elevation gain the same way regardless of the device makes some sense.

There's no standard for how little a bump is reasonable to include in the gain measurement. So, it wouldn't be surprising that different devices using barometers would produce different devices. I had a Bontrager computer and an 800 that both used barometers and the gain was close (within 10%).

People often expect that the gain measurement is more accurate than it's reasonable to expect. It's also not as good a measurement of effort as many think (it doesn't account for gain).

I think people should consider the elevation gain as an estimate that might be +- 20% off of the "real" number (and no one really knows what the "real" number is).

It's only moderately interesting but the differences are sometimes about 1/2 mile in 25 or so. The route followed is essentially the same, it's a road group ride that stays together, but there are always slight variations in terms of road position, how somebody takes a corner, etc...so I assume enough variation to account for that discrepancy. As well I've the wheel sensor so my distance is always the longer one. I'll have to pay attention next time.

That's 2%. If you are using a wheel sensor and the others are not, then it's expected that your number would be a higher (by some amount) with an accurate wheel circumference setting.

Where the numbers come from is going to be different with each device and app. They all have their own programming on how to do that. I know that Garmin doesn't just accept the raw data the GPS spits out, it does some post processing for better results. An app running on a phone or tablet is using an API to request location from an operating system, at whatever its polling rate, and interpreting the answers however it's programmed to.

For me the most accurate is my wired computer, it measures with great precision the number of wheel revolutions. Least accurate is my GPS, the wireless computer comes in between.

But I have WAAS and Glonass turned off to save battery life on GPS so I intentionally reduced accuracy.

I measure circumference of the wheel by rolling it once along a tape measure and measuring it to nearest eighth of an inch. Then convert that to mm and input into the computer.

If it's set up properly and decent and digital, the wireless computer should give numbers that are almost the same as the wired computer.

Wireless computers can be subject to interference (including from certain lights). If the distance is less on a digital wireless computer, it's suggesting that the wireless computer is dropping/missing wheel rotations.

What cycling GPSes do, or can do....if you have a wheel sensor it'll do a very accurate rollout. I.e. count wheel rotations over a mile. Best of both worlds.

Side note GPS can be very accurate horizontally, but not so much vertically due to the whole trigonometry on the surface of a globe thing.

I don't think a 200-foot rollout is enough for a wheel sensor if you want the best precision. What I did when I used to use the Shimano flight deck wheel sensor while also wearing a Garmin GPS watch was to divide the indicated distance of the wheel-sensor result for a certain 25-mile ride by the 25.1 or whatever average distance I'd recorded many times with the GPS on that same route, then multiply the wheel circumfrence factor stored in the unit by that ratio and reset it to the closest settable value. I other words, I was calibrating my wheel sensor to the GPS results.

I did this under the theory that, recording intervals and whatnot aside, the GPS was going to give me a more or less constant error, while the wheel sensor is going to give you a rate of error per mile. If I record a 200-foot ride with the GPS it might be hugely off due to this constant error, but 30 or 40 feet error in a 25 or 50 mile ride or whatever is pretty darn accurate. Calibrating my Flight Deck wheel sensor using the GPS this way consistently gave me ride length results from the flight deck that were typically either the same as, or within .1 mile or so of, what I was seeing on my GPS, regardless of the ride length.

I've also messed around with "smart" recording on the Garmin versus "every second" recording. This was more of an issue with my older Garmin units that had more limited storage space. With my current devices, they have enough space that I just leave it at "every second," though I never really noticed much of a difference in my recorded ride distances with either method. I just like the idea of more granular recording data.

Anyhow, I would trust any GPS-measured distance for a long-enough distance route before I'd trust a wheel-circumfrence calculated distance based on a 200-foot wheel rollout. I'm sure someone could come up with a scenario where the GPS result would be worse, like a 200-mile ride through constant hairpin turns, but for normal, mostly straight riding on mostly level ground, my money's on the gps.

Thanks for all the info in the replies so far.

How I did the 200ft course was I went to a church and used a straight section of parking lines. I eyeballed the ends of the lines from the ground to make sure they were straight down the row. I started at one end and marked the center of the line at the end as my 0 point. Using a 100ft reel tape I marked the 100ft position. Then I moved the tape and started at the 100ft position then stretched it against the lines to the 200ft point and marked it.

The wheel was set up with 2 bands of flagging tape 180deg apart.

To roll the course I started with one of the flagging tape lines on the wheel at the 0 point. Started to ride as straight as I could at about 5mph and counted the flagging tape sound going around the wheel (only one of the tapes was making the noise I counted). When I got close to the end I was a bit past the 1/2 way tape hence the .6. If the 200ft point was right on the 1/2 way tape it would have been .5 fo a 1/2 revolution.

So there is some inaccuracy in the guesstimate of the .6 value, for sure.

There is also likely to be a bit of inaccuracy in the placement of the markings for the 0, 100, and 200ft points that accumulates.

So back to the course length question - by using a 100ft increment tape measure from the get-go there is going to be some cumulative error in every 100ft increment. So by measuring off 1000ft instead there is the chance of 9x the error whereas at 200ft there is only 1x (the error between the 100ft increments).

I suppose a surveyors wheel might be another way to measure greater than 100ft.

In the tape-around-the-tire method - that is not factoring in tire pressure while loaded and riding. That's why I won't use that method. I don't have a way to calibrate for tire pressure. If there was a way to know the change due to fluctuations in air pressure from the start (like calculating the velocity factor in coaxial cable as it affects attenuation of RF through the cable - the better the dilectric and less resistance the higher the velocity factor - less than 1, the worse the dilectric and more resistance the lower the velocity factor) then it would make more sense. However, the only way to get that error per tire pressure setting is to do an accurate rolling measurement with each pressure level - which is what I am already doing with my riding pressure/load from the start.

As to the GPS distance calculations rounding corners - I can see that. It totally makes sense.

As to the GPS distance not factoring in the "angular" distance of ascending and descending slopes (as opposed to the horizontal plane distance on level ground) - that I can see as well.

So the general theme here is the wheel rotation counts from a sensor are the most accurate method of calculating distance. So the question of "the most accurate" distance achievable comes down to the accuracy of the calibration of the wheel counts. And that comes back to all the sources of error I've outlined in the measured course I've been doing vs the tape-around-the-tire not taking in to account tire pressure and loading while riding.

I suppose thinking through all of the above - the 200ft course, for me, is "good enough" and it is more trust-worthy than any distance not obtained with a sensor.

Bike computers -- GPS, non-GPS, wired, wireless, smartphone apps, dedicated computers, magnet vs earth induction sensors -- remind me of my early lessons in photography and photometry: Use three different light meters, get four different readings and, if you ask three other photographers for corroboration, nine different opinions.

Or something like that.

Whenever I add a cycling device or try a new app, I'll usually run two or three different devices and/or apps simultaneously for a few rides until I'm satisfied with the new device/app. They all differ a bit, and all are close enough.

I've compared -- on the same rides -- my old iPhone 4s (with ANT+ dongle), newer Android phone (Bluetooth only), both running Strava and Wahoo Fitness, and a cheap but good wireless computer using front wheel magnet sensor (non-GPS, non-Bluetooth/ANT+), along with a new GPS computer that supposedly uses all global positioning protocols (no way to select, it chooses automagically) along with both Bluetooth and ANT+ sensors.

Yeah, they all differ a bit. Especially after uploading to Strava, which massages all our data using its own secret sauce anyway. But they're all close enough.

I've never been close enough to taking any significant KOMs for the few nths of a second or milliparsecs to matter. Maybe if I was within 1 second of a worthwhile KOM I'd be concerned. But that's never gonna happen, not even in my age group.

A few years ago there were occasions when my iPhone with Strava had significant glitches in distance and time/speed. Difficult to say whether it was only Strava's fault, only my phone's fault, or a combination of both -- maybe with some weather or other interference thrown in. Many of y'all probably remember getting those inexplicable straight-lines across open fields, through buildings, etc., when we'd lose GPS sync, then regain it a few miles or minutes later.

I'm not sure that doesn't happen anymore, but it appears some apps do a little intelligent guesstimating about our probable route and fill in the blanks. Cyclemeter was the first app I used that seemed to do that. If I ran both Strava and Cyclemeter, Strava might glitch and show a straight line across a huge pasture or middle of a highway until it regained sync, while Cyclemeter would reconstruct my actual route with remarkable accuracy -- possibly based on my history of riding the same or similar routes.

When that happens, our data can only be an approximation.

And even riding the same route repeatedly, varying no more than a few feet here and there depending on how wide I take a turn, or whether I needed to dodge a snake or skunk, there are minor variations but nothing really significant.

Same with my bike computer's built in thermometer and barometric/elevation sensors. They're kinda sorta handy, but mostly just confirm data available through other sources. Elevation tends to be essentially identical to Strava and Google. Ditto temperature. The main thing the bike computer's thermometer is good for is to show the effects of micro-climates, those sudden temperature changes we feel some days/nights, dipping into low spots, then rises, on roller coaster terrain -- especially along rivers or lakes, even dry creek beds that serve as conduits for cooler air. And both sensors can be thrown off by other variables -- direct sunlight on the bike computer, humidity and weather fronts that affect barometric pressure.

Still pretty impressive technology for $50 or less, which is all I paid for my multi-function GPS computer.

Segal's Law: A man with one watch always knows the time. A man with two watches is never sure.

In my case the wired one counts every revolution.

The wireless ones, the sender sends a digital packet of info to the computer every several seconds, not sure why but sometimes it seems to add revolutions. And on rare occasion when the bike is moving really slow, like walking my bike across the patio, it suddenly shows very high speed that makes no sense, when this happens it can add several hundredths of a mile in seconds. Maybe it has unreliable debounce circuitry? My wireless ones are VDO brand, a lower end model. At one time I put the speed sensor on the rear wheel, it still worked but I later moved it to the fork and that did not seem to fix occasional errors.

My dynohub wiring is on the right fork blade, twisted wire pair, the speed sensor is on left fork blade.

If my distance will be critical to know, I use my GPS, not my computer.

The error does not concern me, I was just responding to the question.

The wireless ANT+ sensors don't seem to have this problem.

I'm mildly obsessive about my wired computers' distance measurements, perhaps. When I put on a new tire, and again after it's worn a bit, I'll ride up to the top of a nearby ridge and use the surveyed mile markers to measure accuracy. Coasting downhill (so my pedaling doesn't contribute extra distance through swerving), I'll check the odometer against the mile markers for a mile or two (depending on which side of the ridge!), figure out a correction factor to the distance settings, and apply to the computer. It's like your 200' vs. one tire revolution, except on a larger scale. Then, if it was way off, re-check. If I'm within 0.5-1%, I'm golden.

Beware, you can't do this on just any road. Some county roads in Missouri and Kansas, I think they give the good old boys a case of beer, and send them out for a ride. Anywhere a beer can lands, they put another mile marker. U.S. highways are generally reliable in other places I've been, and a few states and even counties appear to have surveyed their roads with great care and accuracy.

I've measured a few walks with a cell phone; the results are usually laughable. If your tablet gets within 2% of the Garmin, it's better than my phone.

The distance markers should be one mile apart on horizontal ground, but if you were going down a hill, the elevation would add some distance too. If you recall Pythagorean Theorem from high school trig, ... ... ... But if the hill is quite shallow, only a percent grade, then no big deal, no correction should be needed.

I understand the desire for great precision here, I am a retired Geological Engineer, I worked with maps and measurements of the earth for a living. But as far as I am concerned measuring one revolution with a tape measure is plenty accurate. After all, if I am within one eighth of an inch when I measure the circumference of rolling my wheel for one revolution, that eighth of an inch error becomes irrelevant when it would take a three quarter inch error to cause a 1 percent error.

Garmins have an option to use distance as 2D or 3D, meaning you can count or not the length the hill adds. I think it's on by default?

These were being placed before GPS.

It seems the only practical way to place so many of them is rolling-out a mile on the road surface.

The Edges don't seem to have this option. It's possible they take elevation into account.

If they don't, that would be another reason the GPS distance would be shorter than a wheel sensor with an accurate circumference.

The numerous mile markers that are on many roads may not be "surveyed". That is, they might not be that accurate.

There are "measured miles" (used for calibrating speedometers) that are supposed to be accurate but these are only in a few places.