Terrain Generalization

[clhenrick]

We should experiment with generalizing the terrain using the DEM downsampling suggestions based on Tom Patterson's [Web Shaded Relief]() blog. Here is the DEM downsampling he recommends:

...you can resample DEMs to slightly coarser resolutions to produce more generalized shaded reliefs. This is especially useful at intermediate scales (zoom levels 5-13) that tend to look too noisy. The chart above offers suggested resampling amounts for these zoom levels.

We can also experiment with compositing a "slope shade" on top of the hill shade.

Relates to #74

[almccon]

I think here's an example of what we lost when we removed the slope shade. A user sent in this example of some terrain features that seem to have disappeared in the new version:

To my eye, I see two major differences. First, in the square area we see a plateau, and in the new hillshade you can't see the southwest edge of the plateau, but in the old hillshade you see it fairly strongly. Second, in the banana-shaped area, there is a long, deep valley that trends exactly NW-SE, which is quite visible in the old hillshade, but seems to disappear in the new version.

These are both cases where a terrain feature is lined up perfectly with the direction of the illumination (from the upper left), which causes the hillshade process to throw no shadow at all. But when a slope shade is also included, these inconvenient features at least have some type of definition, so you can still see that they exist.

Thoughts? Am I interpreting these images (and the process) correctly?

[clhenrick]

I'm thinking that the difference between the two is probably a combination of:

• difference in DEM source data
• variances in any DEM resampling (or lack there of)
• light angle
• lack of slope shade
• the "hard light" composite operation being used by Tessera

It's too bad that we didn't have enough time to experiment more with this part of the process before the v1 launch. I think it's definitely worth attempting to generate an alternative hillshade that's closer to the original stamen terrain tiles.

[clhenrick]

In a more positive light, it's worth noting that there aren't any visible tile seams in the new hillshade which are visible in the old sample above.