More than a year ago i introduced an approach to process waterbody data for the purpose of generating a generalized worldwide river network. This method required quite a lot of ressources in processing so i did not reprocess it since then and the data got a bit stale now, not to mention the various errors in both source data and processing.
Long story short – i recently reprocessed this and made some improvements in the processing for that, in particular:
- Both the US and Canada are now based on additional data supplementing the OpenStreetMap data used elsewhere (which is fairly bad in those countries). The sources used are the NHD and the NHN.
- I improved the preprocessing of the source data for better identification of artificial waterways and making it generally more robust, broken multipolygons in the OSM data are handled much better.
- The main processing is improved as well, dealing with branching rivers for example is now much more reliable in finding the main branch.
- I spend quite some time in fixing some key problems in OpenStreetMap before the processing, removing faulty connections between river systems, adding some key rivers that were missing etc.
- The actual generalization of the river network, the final step in processing, is redesigned, again to be more robust and also including better separation of river networks.
Overall occurence of the issues mentioned is much reduced and the result much better quality. There are still errors but most of them fairly small.