Back to homepage


Hosting and accessing cloud optimized GeoTIFF files on AWS S3 - A practical guide using the EU-DEM elevation model

Published 2019-02-07 #geo #eudem #aws #gdal #geotiff #dem

Shameless plug ahead! This article was (probably) written in the course of building Geofolio, which is a project I've been working on that automagically generates factsheets for any place on Earth. It contains lots of maps and charts, all powered by open data. Check it out at or follow @geofolio on Twitter for news and updates.

There are lots of great geospatial datasets available online, but accessing this data automatically and at scale is often difficult, and manual access usually takes a bit of work to download and sort it all. You'll have to register for some silly data portal, pick out the tiles that overlap your area of interest, and then extract only the bits you're interested in.

The European Digital Elevation Model (EU-DEM) is one such example I encountered recently as part of a project I'm working on. The use case is that I would like to use GDAL in a task queue to automatically extract sections of the EU-DEM from different (user-defined) areas of interest all over Europe, and run some further analysis on the extracts.

To tackle these kinds of cases I've become fond of using cloud-optimized GeoTIFF files in combination with AWS Simple Storage Service (S3). Accessing the data can then be done on the fly using GDAL and its excellent virtual file systems.

This article explains how this works using the EU-DEM dataset as an example, but the approach is pretty similar for any type of large raster dataset. If you're just here for the data, this GDAL virtual file system path should satisfy your requirements: /vsis3/geofolio-public/eu_dem_v11/eu_dem_v11.vrt. But more on that later...


There are several advantages to hosting raster datasets using the cloud-optimized GeoTIFF/S3 tandem:

Downloading and preparing the EU-DEM dataset


The EU-DEM dataset is available from the Copernicus Land Monitoring Service. Of course they have a data portal for which you have to register to be able to download files. After that it's possible to select the tiles you are interested in, and it will give you a download link for an archive with tiles. Each tile is some 4000 by 4000 pixels with a spatial resolution of 25m.

EU-DEM tiling scheme

Tiling scheme of EU-DEM dataset

Because I like hoarding geodata I selected everything and the download file ended up being some 47Gb.

Investigate the data

After extracting all the archives there were a total of 27 tiles. Lets have a look what we've got using gdalinfo on one of the tiles:

$ gdalinfo eu_dem_v11_E40N40.TIF
Driver: GTiff/GeoTIFF
Files: eu_dem_v11_E40N40.TIF
Size is 40000, 40000
Coordinate System is:
Origin = (4000000.000000000000000,5000000.000000000000000)
Pixel Size = (25.000000000000000,-25.000000000000000)
Image Structure Metadata:
Corner Coordinates:
Band 1 Block=128x128 Type=Float32, ColorInterp=Gray
  Description = Band_1
  Min=-24.630 Max=2427.487 
  Minimum=-24.630, Maximum=2427.487, Mean=442.836, StdDev=360.914
  NoData Value=-3.4028234663852886e+38
  Overviews: 20000x20000, 10000x10000, 5000x5000, 2500x2500, 1250x1250, 625x625, 
             313x313, 157x157

There are a couple of things here we might want to address before considering the files optimized and ready for hosting and access via S3:

Optimizing the tiles

To implement the changes above, we can use a simple batch file to run a gdal_translate command against each of the EU-DEM tiles in our directory:

mkdir -p optimized
for file in *.TIF
   gdal_translate "$file" "optimized/$file" \
     -co TILED=YES -co BLOCKXSIZE=512 -co BLOCKYSIZE=512 \

Running this will create all the optimized files in the optimized subdirectory. There is also a utility in the GDAL sources called which runs a few checks to verify that the GeoTIFF is indeed optimized. Running this on your original files will also give some hints as to what can be improved.

Have a look at the GDAL docs on cloud optimized GeoTIFFs for more information.

Merging the tiles

There are now 27 tiles totalling 46Gb in size. These can be combined into a single virtual dataset with the gdalbuildvrt tool:

$ cd optimized
$ gdalbuildvrt eu_dem_v11.vrt *.TIF

And the eu_dem_v11.vrt file can be used from now on to read data seamlessly from all the individual tiles.

Sync everything to S3

We are ready to sync everything to an S3 bucket. I'm using my geofolio-public bucket, which has been created already and preconfigured as a 'Requester Pays' bucket. The AWS command-line interface (CLI) is used to sync the files to S3 and set the correct permissions:

$ aws s3 sync . s3://geofolio-public/eu_dem_v11 --acl public-read
upload: ./eu_dem_v11.vrt to s3://geofolio-public/eu_dem_v11/eu_dem_v11.vrt           
upload: ./eu_dem_v11_E00N20.TIF to s3://geofolio-public/eu_dem_v11/eu_dem_v11_E00N20.TIF

Then afterwards just to make sure they're actually there:

$ aws s3 ls s3://geofolio-public/eu_dem_v11/
2019-02-05 15:21:57      13197 eu_dem_v11.vrt
2019-02-05 15:21:57    3082221 eu_dem_v11_E00N20.TIF
2019-02-05 15:28:45  816689834 eu_dem_v11_E70N20.TIF

Other storage providers

While S3 works well for me because I do other things on AWS, there are many other storage providers out there for which a similar approach would work just as well. The only real requirement is that they allow HTTP range requests directly to the objects in the bucket. The GDAL /vsicurl/ virtual file system can then be used (instead of /vsis3/) to access the data.

Accessing the data

Working with 'Requester Pays' buckets

Because the EU-DEM dataset is in a 'Requester Pays' bucket you will need to authenticate with AWS in order to access data in the bucket. There are several ways to inform GDAL about your authentication details (refer to this page for more info), but for now we'll just add the credentials using --config options defined as --config <KEY> <VALUE>. There are also ways to store these credentials in a file or as environment variables, but I won't get into that here.

The options you have to provide to use to use 'Requester Pays' buckets are your own AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY. Additionaly, the AWS_REQUEST_PAYER option needs to be set to requester to confirm to AWS that you're aware of the fact that your account will be charged for data transfer out of the bucket.

Using the GDAL S3 virtual file system

The access pattern for the S3 virtual file system is /vsis3/<bucket>/<key>, so for our EU-DEM we can use /vsis3/geofolio-public/eu_dem_v11/eu_dem_v11.vrt as a valid data source (in combination with authenication details) in pretty much any GDAL command.

As a test we're going to fetch a DEM of Luxembourg. First check with gdalinfo that we can access the dataset at all and that our authentication and connectivity is working:

gdalinfo /vsis3/geofolio-public/eu_dem_v11/eu_dem_v11.vrt \
    --config AWS_ACCESS_KEY_ID <your_key> \
    --config AWS_SECRET_ACCESS_KEY <your_secret> \
    --config AWS_REQUEST_PAYER requester

If you see the usual gdalinfo output listing metadata, extent, coordinate system, and a bunch of files also using the /vsis3/ prefix, then everything should be working.

A few examples

Now lets fetch our Luxembourg elevation model using gdal_translate with a bounding box specified with the -projwin and -projwin_srs options:

$ gdal_translate /vsis3/geofolio-public/eu_dem_v11/eu_dem_v11.vrt luxembourg.tif \
    -projwin_srs EPSG:4326 \
    -projwin 5.6 50.2 7.0 49.4 \
    -co COMPRESS=LZW \
    --config AWS_ACCESS_KEY_ID <your_key> \
    --config AWS_SECRET_ACCESS_KEY <your_secret> \
    --config AWS_REQUEST_PAYER requester
Input file size is 282250, 216640
0...10...20...30...40...50...60...70...80...90...100 - done.

Which takes a couple of seconds and downloads an elevation model (2689 by 3707 pixels, 40Mb) of Luxembourg into luxembourg.tif:

Luxembourg DEM

Much larger areas such as all of the Czech Republic (22746 by 11279 pixels) are also possible, you're mostly limited by how long you want to wait for your download to complete:

Czech Republic DEM

It is also possible to use gdalwarp to download and reproject into a different coordinate system (in this case the Swedish SWEREF99TM):

gdalwarp \
    /vsis3/geofolio-public/eu_dem_v11/eu_dem_v11.vrt \
    sweden.tif \
    -t_srs EPSG:3006 \
    -te 458830 7290650 696048 7610650 \
    -ts 7400 10000 \
    --config AWS_ACCESS_KEY_ID <your_key> \
    --config AWS_SECRET_ACCESS_KEY <your_secret> \
    --config AWS_REQUEST_PAYER requester

Resulting in the following (7400 by 10000 pixels) file:

Northern Sweden DEM

Well, you get the idea... the possibilities are endless!

Thanks for reading! Get in touch via @kokoalberti for any questions or comments. I also post new articles there when they are first published.

Notes and References

  • [1] EU-DEM v1.1 is produced using Copernicus data and information funded by the European Union.
  • [2] I used a temporary EC2 instance for downloading the tiles, running GDAL commands, and syncing everything back to S3. For large datasets this is highly advisable and much faster than doing anything locally.