Introducing: oilslick, a color elevation map layer designed to highlight the fine detail in terrain, and perform equally well throughout the entire range of Earth's elevation.
Typical color elevation (aka "hypsometric") maps suffer from several flaws that this map layer attempts to remediate:
Oilslick is designed to maximize contrast of small elevation differences
Oilslick is designed to work equally well at all elevations, and not favor any particular range over another
Oilslick is designed to faciliate narrow elevation comparisons, even over wide distances. Each elevation value has a unique and distinctive color.
Oilslick consciously breaks this convention
Human vision sees colors in three dimensions:
By far, the eye is most sensitive to lightness. Therefore, oilslick maximizes change in lightness per change in elevation. Small elevation differences lead to rapid changes in lightness.
But there are only so many levels of lightness ("dynamic range") to go through. So once we max out, we immediately reverse direction, decreasing lightness at the same rate per increase in elevation, until we hit the bottom of the range, and so on… This creates a zig-zagging pattern of the lightness level. Each cycle, or zig+zag from black → white → black again, covers 500m of elevation.
To differentiate among the lightness cycles, hue creeps forward more slowly with elevation. Hue makes one complete cycle — proceeding through every rainbow color — over the full range of land elevation on Earth (9,275m between the Dead Sea and Mt. Everest, rounded up to 9,500, or 19 lightness cycles). Thus every cycle is a slightly different color from the one preceding.
To distinguish the ascending and descending phases of the cycle (zig from zag), we vary saturation. The ascending phase uses saturated colors, and the descending phase desaturated. So when colors are getting darker, if the colors are saturated it means going down, whereas desaturated means going up.
Thus the oilslick palette assigns each elevation value a unique and fairly distinctive color, and employs a color range much larger than a typical elevation map.
There is a discontinuity in hue at sea level to make coastlines apparent. After already condemning overly-suggestive color choices, I feel slightly cheating that the two hues at sea level are orange and blue — evocative of land and water. But it is actually somewhat of an accident. My first draft used red as the base hue to completely divorce color and biome, but, given the organic appearance of landforms, everything looked a bit too much like raw meat or otherwise disturbingly biological.
The key to making this all work out is perceptual uniformity, i.e., how well your color scheme can isolate and map to those three axes of color perception — something most color schemes are notoriously terrible at. Version 1 used the HSL colorspace, which has awful variation in lightness depending on hue, but it was enough to validate the concept. Version 2 used the gold standard Lab colorspace, and everything seemed fine until… the rude surprise that Lab is not perceptually uniform in hue!
If you want true uniformity in all dimensions, you need to go all the way to the Munsell color system. Munsell is derived from extensive tests of actual human vision. As such, there is no easy formula; you must interpolate between samples of known reference colors. Oh, and did I mention the only public dataset of these samples was collected in the 1930's using illumination standards that are now obsolete? But… it works.
But even now, the hue of Munsell still does not appear entirely perceptually uniform to me. It seems to dwell in the teals and blues, while the transition from purple through green feels much more abrupt. But who am I to argue with hard data.
I am very pleased with the initial result. The map is lush, beautiful, and a bit trippy. It is not just incidental to some other map presentation, but stands as an artifact in its own right.
The map explodes in detail in a way that reminds me of the old fractal renders I used to make when I was 15.
The cycling and purity of colors looks a lot like a thin-film interference pattern, hence the name: oilslick. This could be a point of confusion for my more GIS-educated viewers: the pattern is easily mistaken for the raw result of radar interferometry elevation mapping (which was the method used to produce most of the data in this dataset). But this is not the case; what you see here is final, post-processed elevation data.
There is certainly a learning curve to truly understanding the map. You see small-scale changes in terms of lightness. You see large-scale changes in terms of hue. And it turns out the map is quite good at conveying texture. Note the starkly different appearance of rugged mountains, glaciers, high plateau, and dunes.
This is not to say an oilslick-style elevation map is strictly superior.
It is easy to lose the big picture. Without shaded relief, larger landforms don't instantly "pop" into view. Subtle changes in slope that shaded relief would highlight are swamped here by rapidly changing color. Oilslick definitely favors the flat.
In fact, for tall features, the color-banding starts to resemble a traditional topographic contour map — a type of map that people literally have to be trained to use.
But, while not a silver bullet, my hope is that oilslick illuminates terrain in a way that other maps have yet failed to do.
Coverage is global. All of Earth's landforms (excepting the immediate vicinity of the South Pole) are represented.
Data resolution is 3″ / 90m, or down to zoom level 11 (less at high latitudes). Actual data resolution for some areas may be lower, sometimes much lower, notably the Greenland and Antarctic icecaps.
Source data is the amazing work of Jonathan de Ferranti, who assembled a complete global digital elevation model, filling in the extensive gaps in SRTM data. Supplementary data in his dataset is pulled from myriad sources, including hand-digitizing of old Russian topographic maps!
This map layer is hosted as a public service, and map tiles are free to use under a Creative Commons 4.0 Attribution-NonCommercial-ShareAlike license.
