I added the STL generation to the GEBCO extractor so I can now make objects rather too large to display online. This one of the Mariana Trench shows Challenger Deep, the deepest point in the oceans at nearly 11km deep. The place was first explored by Don Walsh and Jacques Piccard in the submerssible Trieste in1960 Here is some BBC coverage and a rather slow documentary covers the history of the discovery of ocean terrain and technonic plates. Unfortunately they stirred up so much sediment they couldn't see anything down there but they took a great selfie on this very brave voyage.
In this model, depths are exaggerated by a factor of 6. The deep is bottom left and Guam is the peak on the top edge.
The model is 300 x 600 points using the 1 arc minute GEBCO dataset. This turns into 719,996 triangular facets and an ASCII file size of 97Mb. This data compresses really well - to 5.8Mb. The binary file size will be a little more than 50 * 719,996 = 36Mb. I havent experimented yet to see how well the binary compresses. Sadly CAD software typically doesn't seem to support the zip format because that would be much better than binary. However Shapeways does support zipped ASCII. Here is the object on Shapeways
The added complication with this data is that all coordinates have to be positive so I have to offset the heights by the minimum depth and add a base depth to ensure its printable. [Correction - although the guidelines say that vertex coordinates must be positive, OpenSCAD SLT has signed coordinates and software like NetFab, Repetier Host and slic3r shift the origin on loading].
Scaling caught me out until I realised that with numpy the following expressions are not always equivalent:
because as the depths are all integers, the first results in integer depths after scaling, resulting in terracing, whereas the second form converts to floats.