Ascent calculation is inaccurateFORUMS HOME SEARCH FORUMS

Thank you John,

You have done a fantastic job with this site and I understand there is a limit to what can be realistically done calculating elevation from map.

Just to clarify what I was meaning,I am providing what seems to be a good example:

This route I have generated with plotaroute https://www.plotaroute.com/route/187663 is quite famous in the area. The route is a continuos climb with not a single meter going downhill over the entire length (13km). The starting point is at 154m above sea level, the end is at 814m above sea level.

The real elevation gain for this route is 814 - 154 = 660 m (source: google earth, in good agreement with plotaroute values and altimeter readings)

Plotaroute elevation profile for this route https://www.plotaroute.com/routeprofile/187663 shows a fragmented pattern with continuous up and down that do not exist.

The unfiltered elevation gain from plotaroute provides 1305 m  that is about twice the real elevation gain.

I believe the limit is provided by the maps used to generate the elevation profile.

You are right in saying that barometric GPS absolute elevation readings can be inaccurate especially if you do not calibrate the altitude with a known value at the beginning of the ride to eliminate drifts due to barometric pressure variations (sometimes I read negative altitudes at sea level). In any case the differential elevation gain remains quite accurate (e.g.  My GPS with barometric altimeter is quite in good agreement with the  calculated 660 m elevation gain of the above route, the deviation from  the 660 m is always less than 4% ).

I would suggest when uploading rides to plotaroute, to provide the option of recalculating the elevation data or to leave the data in the uploaded route.

Regards

Andrea

I'm afraid elevation data from GPS devices is not a reliable benchmark against which to judge whether ths source data is accurate, even if the device does have a barometric alitmeter.  When we analysed the route you provided at the start of this thread, your barometric GPS device was measuring 63m above sea level at points where the route was next to the sea. I've seen other examples of this too.  

Also, it doesn't matter how accurate the altitude measruement is if the device loses the GPS signal -  it doesn't just stop recording coordinates when the GPS signal is lost, it also stops recording elevation data, so this creates gaps in the elevation data, which leads to some climbs not being counted.  

Unfortunately no solution is perfect, but our source data does seem to tally quite closely with Google's data, so I'm fairly sure that differences in Total Ascent calculations between different applications and devices are largely due to size of the gaps between the data, with larger gaps giving a lower result.

Pleas bear in mind that this is a free service and there is a limit to what we can realistically do.

John

Hi John,

In my view the issue is  plotaroute  source data: the elevation profile is given by the overlay of the route coordinates an the elevation data in the map. Errors in the road's coordinates and in the elevation data contained in the maps (google earth or any other map source) will generate a deviation from the real altimetric profile. In other terms the issue is in the input data that plotaroute processes. The accuracy of barometric readings from GPS devices seems to be much  better than the accuracy of data taken from road plotted on charts. This is confirmed by the relatively good agreement of different  instruments from different manufacturers when riding the same route. 

Hi Olof - I've had a quick look at your Sa Calobra route, which has a total ascent of 914m on our calculations but 715m if you assume there are no downhills.  However, it does appear that there are some downhill stretches, at least according to our data but also according to Google's data.  

It's not really possible to see this clearly in Google Street View, as it's a 2D image, but you can see it in Google Earth.  If you download the route as a KML file you can open it in the Google Earth software and move your mouse along the line of the route to see the elevation readings at the bottom of the screen.  If you look at some of the parts of the route where our data shows a downhill stretch, say 2.7km or 6.3km, you should see that Google also reports this as a downhill stretch.  It is all these short downhill stretches that make the total ascent higher than the difference between the altitudes of the start and the summit.  These can be smoothed out in our route profile tool by using elevation readings that are further apart - so if elevations readings were taken every 120m instead of every 30m the total ascent would be 765m instead of 914m.  I'm not saying that our data is perfect, as no solution is perfect and there is a limit to what we can provide as part of a free service, but it does seem like our data is fairly consistent with Google's.

I'll will do some further investigation though, but it does take quite a lot of time to look into individual routes on a case by case basis and I'm conscious that time spent on this is diverting us away from working on other feature requests that people have asked for. 

John

Yes, Strava probably uses my barometric data from my GPS (Garmin Efdge 500) and yes, barometric data can be way of. What I try make a point of is that Plotaroute have some major bug in calculating elevation data, either based on data or algorithms. Take a look again at the route (https://www.plotaroute.com/route/186065 ). It is basically going straight up from sea level to 1936MASL and then back down again. There are a few small hills during the way, but nothing that adds 3000 meter of climb more (up to at thousand, yes maybe). The end result can never be 5000+ meter of ascent in total. Look closely at this screen shot of the profile Plotaroute shows for it (https://drive.google.com/file/d/0B9MeEBv9khGkRGQ2dTB0UElxVjQ ) and we get a hint where this comes from, the road seems to go up/down constantly when it is just a steady hill. (Again, it is obvious that there isn't 3000 meter extra of ascent to find on this route.) The problem is probably that Plotaroute doesn't do any smoothing whatsoever of the data. This results in the error that for hillier parts of the world makes the total ascent hugely incorrect. I have ridden these roads a lot of times and can assure you they do not go up/down/up/down as the profile says (it is easy to take a look at Google Street view for yourself to confirm if you wish).

Another, even clearer example: https://www.plotaroute.com/route/186461 This is the famous Sa Calobra-climb on Mallorca. It starts at five meters above sea level and tops out at ~715-720 m.a.s.l. and I would say it has 0 meters of downhill. Hence it has 720-5=715 of ascent, remember that regardless of where during the climb you climb those meters, there can never be more than (hight at end - hight at start). Lets say I'm wrong about the 0 meter of downhill and there actually is 10, this would lead to a total ascent of 725 meter (again, have a look at Google Street view if you wish to validate (or even better, go there yourself, I can recommend it! :) )) Plotaroute states 914 meters of ascent, ie more than 25 % too much.

I'm afraid it's not possible to compare Total Ascent figures from different sources, as they are calculated differently and also elevation data from GPS devices is not reliable (see my post below 20 March).  

I believe that Strava uses the elevation data from your GPS device if it has a barametric altimeter, so I would expect Strava and your GPS device to give very similar results as Strava is just reporting what your GPS device says, possibly wirth some minor smoothing.  Unfortunately GPS devices are not infallible but there is a tendancy to place unwaivering trust in their results, even though closer scrutiny of the raw data often shows big measurement errors and gaps. 

John

Hi John, happy you are look into this so thoroughly!  

I think the problem is elsewhere, look at this route: https://www.plotaroute.com/route/186065 According to the overview the total ascent is 5031m (see also this screen shot: https://drive.google.com/file/d/0B9MeEBv9khGkZF9aN21xMFBIVGM ) I have riden it and it is about 2900-3000 total ascent, see Strava track: https://www.strava.com/activities/417517234 but also make an estimate by yourself using tools. 

Hi Olof - We've had a few people contact us about this and have done quite a bit of analysis of raw data from GPS devices.  I'm afraid every time it turns out that the data from the GPS device was a long way out.  This has been true when the GPS device used a barametric altimeter too.

I've just recently looked into a sample route provided by George below.  I compared our own elevation data with the elevation data recorded by his GPS (with barametric altimeter) and also with data from Google.  The Google and plotaroute readings are almost identical but the GPS readings are a long way out (and not by the same amount either, so it is not just a calibration issue).

When it comes to calculating the Total Ascent figure, the GPS data was also flawed, mainly becasue of large gaps between the data, presumably where the signal had been lost. On plotaroute the distance between readings is a fixed interval of 30m.  The GPS file that we looked into had readings on average 41m apart but up to 2920m apart!  Scanning through the data I could see there were several instances where there are gaps of over 1km between readings.

So, the main reasons GPS devices give lower total ascent figures than we calculate on plotaroute is due to a combination of:

1. Inaccuracies in the elevation readings being measured by the GPS device
   
   
2. Long gaps with no readings
   
   
3. Irregular intervals between readings

Differences between the total ascent figures on our site and other sites are most likely due to different sampling intervals.  You can see the effect of this by changing the sampling interval in our route profile tool - increasing the distance between readings makes a big difference to the total ascent, as it effectively smoothes out lumps and bumps.  We sample readings quite frequently - every 30m for most routes but less frequently for longer routes.

Unfortunately no method is going to be 100% reliable but I think the main conclusions are that you can’t really compare ascent calculations from different sources as they are calculated differently and you can't rely on elevation readings measured by GPS devices.

John

The ascent/descent-data for routes are often of with up to 50%. Happy to help out with raw-data if needed.

Thanks George.  I'll delve a bit deeper and see if I can work out where the differences come from.  Do you have a sample file directly from your GPS device that you could send me (ideally a fairly short route)?  It would be helpful to see the raw data recorded in the file and then compare this with elevation data generated for the same route on plotaroute.  I'll email you directly with an address to send it to.

John

Garmin records raw data from the barometric sensor, there is no smoothing of the data when it is recorded. After uploading to Strava, the evelavtion differs (I presume that is when the smoothing happens) but the change is not significant, usually within 3-5% of the original reading. In saying that Strava is known to be wildly inaccurate when using their mobile app as that relies on the map elevation data rather than sensor readings.

I agree with Andrea that it is most likely the accuracy of the underlying data as the best data set available for the area where I ride has 10m resolution and the public version of it has 30m. I had a quick look at some ride profiles and it seems that on long steady climbs where the road is mostly cutting across elevation lines the data from Plot a Route is quite accuarate. On twisty undulating roads it is quite far out as per my original post.

From a quick bit a research it appears that other sources may indeed ignore some elevation readings within a certain threhsold, effectively applying smooting to the data collected. For example, Strava's site says

"A gain must exceed a threshold in order to be counted. The elevation data used for calculating gain is smoothed before elevation gain is calculated in order to reduce noise.  If you are riding or running in a very flat area, and none of the individual climbs on your activity exceed that threshold, it is possible that Strava will list a total elevation gain of 0ft, even if the elevation profile is not totally flat. This is an unfortunate side effect of the thresholds necessary to reduce noise on the majority of Strava activities."

I suspect Garmin does the same - if anyone can confirm this I'd be interested to know.  We don't ignore any readings or do any smoothing, so all lumps and bumps along the route are counted in our total ascent calculations, which would explain the higher figure.

John

Hi,

I believe plotaroute is processing correctly the data, the point maybe is in the accuracy of the elevation data and contour lines in the maps itself.

As an example if you are analazing data on a flat route in the mountains with steep gradients on each side of the road, a small error when plotting the road on the map (even few meters each side) would generate a large elevation change although the road is flat. When plotting in areas with contour elevation lines very close each other i.e. with considerable gradient, this error could become significant. 

Regards

Andrea

We'll certainly look into it a bit more but I think Garmin devices ignore readings with small elevation changes when calculating the total ascent due to assumptions about accuracy levels of the data, effectively applying smoothing to the data.  So although it may be collecting data every second, this data isn't necessarly all used in the calculation of the total ascent figure.  Also, by collecting data at time intervals, it will be at varying distance intervals, which creates another anonomly.  And if the time interval is very short (e.g. one second), then changes in elevation may also be very small and may be discounted in calculations.

It might be quite hard to get to the bottom of it without having access to both the raw elevation data and formulae used by different systems.  I'll try to do some research on it.

John

Thanks for the quick response John

Unfortunately I don't think that's quite right. My Garmin is set to record a point every second, so on an avrage hill climb that would be 3-5m apart (I am not the fastest climber out there). This is a far higher sampling rate than what is present in any publicly available mapset and should in turn result in the highest elevation reading. The reality is that Plot a Route comes in roughly 50% above that (and this is consistent for all routes I've created or looked at in this area).

So I still think there is an issue with the underlying elevation data provider and it would be great if that was addressed at some point in the future.

Hi George - glad you're enjoying using the site.  Unfortunatley there is no "right answer" when it comes to total ascent, as it depends on the sampling interval (see my post below on 9 Nov 2015).  Less frequent sampling of elevation readings will give a lower total ascent as it smoothes out lumps and bumps. I think we sample more frequently than most.  Neither figure is wrong - they just need to be interpreted in context, so you can't really compare figures calculated using different sampling intervals.

John

I was just looking at the same thing, I created a route here and then exported the GPX file. Plot a route estimates the altitude at 3300m, RWG estimates it at 2100m and actually riding it gives 2100m (Garmin 705 with barometric altimeter). I am guessing the underlying map data is a bit off.

Other than that this site is brilliant and has become my goto place for mapping out new routes.

George

Hi John,

OK I see what you mean. I dont know what the others are doing but I can surely try to find out.
Can I ajust this somewhere when I'm creating a route?

It though seems a bit strange to me that the total ascent in la Marmotte this year was 7310 mtrs (in reality)

AllanH

Hi Alan,

Unfortunately there is no correct answer when it comes to calculating the Total Ascent.  You can get a different figure each time by having different distances between the elevation readings, but that doesn't make each answer wrong.  Longer distances between elevation readings will have the effect of smoothing out bumps and dips, so tend to result in lower Total Ascent.  We take readings at regular 30m intervals by default (unless the route is very long), so probably more frequently than other sources you might be comparing to.  GPS devices can be notoriously innacurate when it comes to calculating Total Ascent.  Not only do they sometimes gather quite inaccurate readings, they also take readings at irregular intervals, so they are not good to compare with.  Sometimes a barometric altimeter can help with accuracy but when we looked at the example quoted below, where a Garmin device with a baraometric altimeter was used, it was recording elevation readings of 63m at sea level, so quite a long way out.

You can see how the Total Ascent changes as elevation readings are sampled less frequently by having a look at your route with our route profile tool - drag the Elevation Interval slider on the right to increase the distance between each elevation reading and see how the Total Ascent changes in the data at the top.

John