Dear Graphhopper team,
I am currently working on imporvements of the elevation model in graphhopper and to compare my results with the actual graphhopper version, I would like to know wich configuration for elevation data do you use? I tried locally using CGIAR, with calcmean = true. This configuration has a different result, than the one on the live server. Therefore I wanted to ask which dataset with which configuration do you use?
All the best,
Oh, nice! This would be really appreciated and welcome
Hmmh, it should be CGIAR with calcmean=true but maybe the data has changed somehow on their servers?
What do you mean with different result? Only the elevation or only the routes? And which are are you using?
I tried now several versions, but I was never able to get the same elevation profile in the right lower corner and sometimes also different routes.
Current MASTER branch:
SRTM, CALCMEAN=TRUE, CH=TRUE
CGIAR, CALCMEAN=TRUE, CH=TRUE
- CGIAR, CALCMEAN=TRUE
- CGIAR, CALCMEAN=TRUE, CH=TRUE
I was never able to reproduce the online demo version. Do you have the possibilty to check your configuration? That would be great, otherwise I will use the local outcome to compare my results.
All the best,
We use a rather recent master (not always the latest) and CGIAR with calcmean=true. With or without CH is identical (should be at least and verified via tests).
I already checked that. What I could do is to send you the CGIAR data we use (?)
Or maybe other config options like:
are important in your cases?
Thank you very much for the quick response. Aren’t you using the CGIAR data from this location http://srtm.csi.cgiar.org/SRT-ZIP/SRTM_V41/SRTM_Data_GeoTiff?
I will also try with prepare.minOnewayNetworkSize=200 and check whether this changes something.
I’m not 100% sure which host we really used at the time we downloaded the data initiallly, as this is cached now on our back ends since several months. But I expected the data to be identical through the different hosts and time (?)
I tried again with CGIAR, Calcmean and prepare.minOnewayNetworkSize=200, but I was never able to get the same elevation profile as the online version. For my thesis I’m comparing to elevation data from South Tyrol (Italy). They gave me a very accurate elevation model of the region. I try to improve the elevation model with mean and kalman filter on the ways. And if you look at the following results you can see, that I am not able to reproduce the outcome of the online version:
GH MSC CGIAR
RMSE for SouthTyrolElevation: 8.631204727586132
Cosine Similarity for SouthTyrolElevation: 0.9999666403758977
GH MSC CGIAR Calcmean
RMSE for SouthTyrolElevation: 8.84799678225662
Cosine Similarity for SouthTyrolElevation: 0.9999657691700661
GH MSC CGIAR Kalman
RMSE for SouthTyrolElevation: 6.960855043827352
Cosine Similarity for SouthTyrolElevation: 0.999978452513585
GH MSC CGIAR Mean
RMSE for SouthTyrolElevation: 9.332883974199092
Cosine Similarity for SouthTyrolElevation: 0.9999646049584412
GH MSC SRTM1
RMSE for SouthTyrolElevation: 3.9848866443887045
Cosine Similarity for SouthTyrolElevation: 0.9999935131060188
GH MSC SRTM1 Kalman
RMSE for SouthTyrolElevation: 3.002770289543429
Cosine Similarity for SouthTyrolElevation: 0.9999964396456205
GH MSC SRTM1 Mean
RMSE for SouthTyrolElevation: 3.2894877126635906
Cosine Similarity for SouthTyrolElevation: 0.999995838820838
RMSE for SouthTyrolElevation: 8.911796302280104
Cosine Similarity for SouthTyrolElevation: 0.9999697998964302
“Graphhopper Track” is the GPX file from the online server and “GH MSC CGIAR Calcmean” should have the same configuration as the online server. As you can see they have diffrent results, also in other tracks. I’m just interessted why they are different. For my work it is not so important. What do you think? Btw SRTM1 is the new void filled version.
I’m just interessted why they are different.
Sorry, I don’t know … one idea left: use the encoder list
bike2,foot,mtb,racingbike instead of just
Otherwise let us assume something changed in the CGIAR data and the API data is somehow outdated. I could try for a small area of your choice on my dev box to fetch fresh elevation data with my config here. Would this help you?
I really think that there is huge potential with smoothing - be it for the whole elevation data itself (low|high pass filters) or for one route (e.g. via kalman, while import to make it fast). Please have a look into issue 517, issue 330 or issue 553 where problems are shown and we discussed this topic and I see a big potential to improve it via methods you’ve probably already in place
Btw SRTM1 is the new void filled version.
Do you mean the one discussed in this issue?