Someone asked the other day whether the cross-platform, free and open-source GIS Whitebox GAT can handle watershed delineation from massive, regional-scale DEMs. They had a particular interest in the Nile River basin. Heck, if you’re going to go big, why not go huge, right? So I decided to give it a try. First off, I used the Retrieve SRTM Data tool to download the approximately 800 SRTM 3-arcsecond (~90 m) tiles that make up the Nile River basin. This required some experimentation because my first attempt at doing so hit the boundary of the basin and I had to give it a second try. The tool downloaded each of the tiles and mosaicked them into a single large DEM. The final DEM was 45,601 rows by 25,201 columns (a little over 1.1 billion grid cells) and was 4.28 GB in size. I then used the new Breach Depressions (Fast) tool to hydrologically pre-process the DEM by removing artifact topographic depressions and flat areas (i.e. cells with no downslope neighbours). I used a D8 flow algorithm to calculate flow directions, perform flow accumulation, trace the flowpaths issuing from Lake Victoria (the White Nile) and Lake Tana (the Blue Nile), and lastly, to delineate the watershed. The result was this map:
To fully appreciate this amazing map, you need to enlarge it. Just to put a bit of perspective on the scale of this analysis, take a look at this one:
All of Europe has an area of approximately 10,180,000 km2 and the Nile River basin has an area of 3,400,000 km2. That is truly vast.
For my initial attempt, the one in which I truncated the watershed, I used my 13 inch Macbook Pro (2.8 GHz dual-core i7, 16 GB RAM, SSD). When I expanded the area, I also moved to my workstation (3.0 GHz 8-core Xeon, 64 GB RAM, SSD) just to speed up the process a little. I even extracted a long-profile for the White and Blue Nile, although I should have converted the distance units to metres:
It was the longest river that I have ever plotted a long profile for; of course, it is the longest river so I guess you can’t get much larger than that! I probably should have extracted the river network using a dispersive flow algorithm like Tarboton’s excellent D-infinity, since there are places where the river bifurcates (i.e. the river course splits), even before the delta. Nonetheless, I’m quite pleased with the result. In fact, I was quite surprised at how well the river course, extracted from the 90 m resolution SRTM DEM data, matched a mapped Nile River shapefile that I located:
Leave your comments below and, as always, best wishes and happy geoprocessing.