Population Density Map Voxel Renders

A few weeks ago I noticed on various social media sites some creative and interesting renderings of 3D maps of countries showing their population density as colour-shaded bars extending vertically upwards proportional to the population density values, the original source of which was Alasdair Rae. They’re lit with a fairly low-angled soft light, simulating an early/late sun, which also allows the columns’ shadows to highlight magnitude as well.

I decided I’d try and see what would be required for Imagine to be able to create and render similar looking renders from scratch, using only Imagine and the original source data.

The original data was in GeoTIFF format, which is basically TIFF format but with specific metadata encoding geographic properties like projection and lat/long co-ordinates (which I ignored for the moment), and the data type is normally float32 or float64 (double). Imagine already had support for reading TIFF files in general, but not for reading values as full double-width (float64) floats, so I had to implement support for that.

I then created a Scene Builder plugin (a plugin in Imagine’s UI which can procedurally generate scenes based on input parameters) to generate geometry consisting of small cubes/cuboids based off the data values in the image files, using input image X position for the 3D X axis, input image Y position for the 3D Z axis and the population density values for the Y axis height. Values below a threshold would not generate any geo (i.e. in the ocean with no land).

I also had to merge multiple images together, as each source image represented smaller geographic areas, with boundaries across multiple images for some of the regions I was interested in. For this I wrote custom code as part of the Scene Builder plugin UI to do manual single file / batch merging based off index coordinates.

Functionality to then shade the materials (height falloff gradient, mixed with a 3D grid texture), as well as render the image already existed in Imagine, so I was then easily able to render these:

I ended up ‘cheating’ slightly by not creating full voxel representations of stacks of cubes for per cell columns, instead just using single stretched cuboids to save geometry memory.

One issue with these renders is they’re just using the original source data projection in image space mapped to 3D space, which for places far from the equator like the UK, ends up squashing things quite a bit: I’d need to re-project the data in QGIS to rectify that, which maybe I’ll do in the future, as I do think they look pretty nice.

Some of the source data also seems to have artifacts (extra non-existent land-mass based off rasterising the image input data) in coastal areas, so to render these more nicely, Imagine’s unlikely to be the tool to do these image-space data touch-ups.