Tinkering with Wilbur to achieve "above average" mountains

[Peter Toth]

Hello Guild,

I remember Morne saying that realistic topography is difficult to achieve manually, without using a pre-existing satellite DEM. So here I am again, attempting to do just that, as preparation for a huge project that demands realism. I did this differently by selecting only particular altitudes in Wilbur (below 175), feathering the selection with a sigma of 1.5, then going ahead with incise flow. (Of course, my exact process is too long and complicated to describe here).

For all you Guild artisans, let me know if I'm anywhere near my goal.

Thank you for any feedback.

Peter

Mountains.png

[Eusebio Ptolomeu]

I'm not an expert on this, so maybe other users will have a better answer to your question, but for me, they're looking really nice! I guess the only thing looking weird to me is that huge flat plateau on the lower left of the image, which is a result I get often when using wilbur, I think it's just a matter of doing more erosion passes to carve it away. Near the top-center there's a river that gets a bit orthogonal and straight, but these are very minor details. Looking at the rest of the picture, namely the carved mountains and lakes as weel as the coast in lower-right, I really like the resuts, to the point that I'm surprised this was done without any real topographic map. As I said, I'm looking at this as a non-experienced person about this subject, so other users might give a deeper and more meaningful look here.
Overall, I really liked this result, and would love to know more about your process to achieve this.

[Harrg]

Looks nice. But if you really plan to make a project with a realistic relief in the future, then unfortunately the wilbur will not be able to satisfy you. Wilbur works on a fairly simple system that gives good results. But he will not be able to make glacial eorosis or weathering, he does not see the difference between alpine mountains and old cratons.
Wilbur also likes to create artifacts in the form of parallel rivers or crow's feet. In my opinion, for a project that really claims to be realistic, it is better to use the wilbur as a kind of equalizer. It smooths out strange joints well and adds nice river crossings between heights. I usually use Wilbur a to match the terrain with the coast or to make transitions between sharp changes in elevation. Already on top of this map, you can add new details, masking the wilbur artifacts.
There are not many techniques for creating realistic relief. Or you draw everything with your hands and control each pixel (this is real, but very loooooooong and sometimes boring).
Either you use srtm and fit the pieces. Nothing prevents you from combining these methods.
Wilbur, world machine and other programs work well either for small, local maps or as a fine erosion generator.
in any case, you do not have to listen to me, you may be able to come up with and hone a new method

[samuel.harven]

There are not many techniques for creating realistic relief. Or you draw everything with your hands and control each pixel (this is real, but very loooooooong and sometimes boring).
Either you use srtm and fit the pieces. Nothing prevents you from combining these methods.
Wilbur, world machine and other programs work well either for small, local maps or as a fine erosion generator.
in any case, you do not have to listen to me, you may be able to come up with and hone a new method

I've never really worked with realistic terrain before but I'm still very interested. If worldmachine/wilbur do not work, then do you know of any applications that are capable of handling large scale continental maps?

[waldronate]

That was too much text. I'll try not to do that again, sorry. Somebody called my child ugly and I had to join in.

I'm not sure what you're trying to approximate with the high-level plateaus (the lack of a scale certainly doesn't help). I think that you missed the mark if you're trying for something like the Tibetan plateau in the Himalayas or the Andean Altiplano because both of those have far more interior detail. The Altiplano also has a whole lot more volcanic features, which you'd need to paint by hand.

The rivers also show some typical Wilbur artifacts: axis-aligned flow, parallel flow, and crow's foot elements. Avoiding those would have needed more noise in the early stages of processing. The noise that would make it though to the final height field really needs something like a median filter to get rid of it, but Wilbur doesn't have one of those.

You're getting pretty good with the overall terrain processing, and with the rendering of terrain as your final image. One effect that could add a touch more realism to the apparent lighting is to add a texture shading layer to your final composite if you're not already doing so ( see http://www.shadedrelief.com/ for some beautiful examples of adding texture shading to an image).

[Harrg]

That was too much text. I'll try not to do that again, sorry. Somebody called my child ugly and I had to join in.

I really like Wilbur and I am really grateful to you for it. This is a cool program. My only thought was that it is not very efficient for realistic continental scale maps.

I've never really worked with realistic terrain before but I'm still very interested. If worldmachine/wilbur do not work, then do you know of any applications that are capable of handling large scale continental maps?

Depends on the goals. Some people use gis. You can check out Morne's tutorials, this is the most current guide on how to work with fake maps at the moment.
I only use gis to steal high resolution height maps from there. I do the rest of the process in Photoshop. There I also do shading, texturing, etc.
P.S.
My point is that there is no generator program that can immediately make a geologically correct continental scale map. You can use a combination of all the existing methods to achieve this.
It is similar to clay sculpting or marble processing. At first, it's a big, shapeless chunk, and with each cycle you pick out more and more details until you get to the level of detail you want.
If you are assembling a fake map from a puzzle of real maps, then the wilbur will handle all your details well together. It will destroy most of the small parts, but it will weld all the fragments together.

When I work on a heightmap, I first type in the "clay" to show the dramatic changes in height and shape. I don't really care at this stage about the consistency of all the details. (Rivers can create chaos, a lot of closed river systems, etc.) When you sculpt the main rough relief. Load it into Wilbur. Run over erosion cycles. As I said before, the Wilbur will destroy most of the small details. But this is exactly what you need. In return, the Wilbur will give smooth transitions between the pieces.
You can find many videos on the internet that explain where and how different landforms are formed.
In order to understand how certain complex zones appeared on the ground. For example the Himalayas, you need to look at the paleogeography.
Perhaps the most difficult thing is to achieve not the "erosion" effect, but the geological plausibility. You need to pay attention to which plate is immersed. Where we have subduction, and where is the rift. Show results from previous geological eras, etc.

https://www.youtube.com/watch?v=bI5r4cs74w8
https://www.youtube.com/watch?v=Ly78AxI9gaE
https://www.youtube.com/watch?v=d9bKXY0OMxc
https://www.geokniga.org/bookfiles/g...-tectonics.pdf

[waldronate]

As Harrg points out, Wilbur has some pretty serious limitations, as do many tools that have erosion simulators. Wilbur's limitations stem from a few critical underlying simplifications: a constant-resolution fixed square grid representation for altitudes (it's a basic image raster), limited connectivity between samples (it's either a plus or star), constant-hardness terrain (it's effectively a a big pile of mud), and erosivity proportional purely to altitude differences between cells (there is no water layer to allow for ponds and their effect on surface geometry). There's really nothing that can be done about the first one and it's a fairly common problem with most erosion simulators that's dictated by the machine architecture that the software run on. The connectivity problem is also common because the calculations are typically purely local (the nine or so cells at and directly adjacent to the point under consideration are what's used to determine flow). Wilbur's constant hardness problem is due mostly to programmer laziness: changing it would require having additional layers in the software to allow for things like a hardness definition layer and an alluvium layer and water/sediment transport layers, all of which are needed for mostly-complete models. The pure erosivity model leads to a number of problems, including terrain getting much shorter over time than would be expected with a more complete model.

Because Wilbur doesn't have a good model for erosion, it has several pieces that can be used to get plausible (not physically correct, just visually plausible) results within certain constraints.
Fill Basins establishes surface-wide connectivity. However, the nearly-flat surfaces from filled basins with the default settings lead directly to crow's feet across those basins because it's a striaght downslope shot from everywhere in the basin to the outlet.
Incise Flow does a global connectivity analysis across the surface and does erosion proportional to a function of the number of upstream cells at a point. Limited connectivity rears its ugly head here because the downstream directions are computed from purely local neighborhoods. Incise Flow without Fill Basins gives a very ugly effect, filling the surface with little pits due to disconnected flow. The calculated flow assumes that flow would be proportional purely to the number of cells "upstream" of a point: in the real world, flow is as much proportional to the amount of water incoming upstream as it is to the catchment area and incoming water is proportional to a whole lot of factors, including rainfall, orographic effects, freeze/thaw cycle, and so on. Again, Wilbur has a constant assumption: rainfall is the same everywhere.
Precipiton Erosion drops an agent on the surface at a random location and it pushes a fraction of the steepest altitude difference in the local area to the downstream point. When the agent runs off the edge or hits a pit or gets too old, it stops and respawns somewhere else on the surface. Notice that terms like "precipitation" or "erosivity" never appear: Everything is equally soft and altitude that's been moved has exactly the same characteristics as the place it came from.
Wilbur also has some mathematical tools like adding noise to the surface and performing a local minimum or maximum operation.

The general loop that I use with Wilbur combines tools to get better results together than any of them can get alone. The results have a tendency to be fairly uniform, which can be desirable or not. I construct a fractal manually by starting with a simple, low-resolution surface, running it through a filter bank to modify the surface, scaling it upwards, and then repeating until the result is large enough to be useful. A little post-processing will improve the plausibility of the result. A few artifacts (mostly spiky single-point noise) tend to creep in along the way, and a simple local minimum gets rid of those. See https://www.cartographersguild.com/a...chmentid=80066 for a painfully-detailed example.
There are lots of other ways to go about generating a surface, however. Wilbur can do a wonder job of elaborating basic contour maps (I abused mbartelsm's thread here at https://www.cartographersguild.com/s...t=33087&page=2 to show how to turn a hand-drawn contour map into a more elaborate version). It still requires a light hand and careful attention to artifacts (you'll probably want way more noise early on than you're comfortable with).
The general idea, though, is to always apply mitigations for the classes of implausible artifacts that appear. Connectivity artifacts can be usually be mitigated with noise. Once an artifact is established in the system, though, it tends to grow (too much incise flow on a relatively noiseless surface will give networks of dual axis-aligned artifacts that persist forever). Selections can be made to substitute for things like hardness maps, but the underlying nature of the 8-bit image that is a selection can itself leak through into the terrain image. A selection can be used to simulate differential mountain uplift between passes. A selection can limit exponential operations so that only some parts of the map end up with flat-bottomed glacial-style valleys. Calculating a selection, saving it, then restoring it later can be useful (e.g. calculating a selection of basins and using that selection later on for flattening crow's feet or for restoring lakes is a lesser-used trick; keeping a mask of where important rivers HAVE to be and using that as a selection to chop the rivers back into the terrain is also useful).

At the end of everything, though, Wilbur terrains will still end up plausible over the range of maybe 10 meters per cell to maybe 1000 meters per cell with a sweet spot right around 100 meters or so. Continental-scale things have a tendency to be little wormy things with huge nearly-flat spots, both of which aren't very amenable to Wilbur-style processing. Wilbur can be helpful for doing a little river calculation or such, but it's not going to hallucinate much detail into a mountain range that's two-pixel-wide worms a couple of pixels apart and rivers over flatlands just aren't that interesting. Remember that what you see in the real world is caused by physics at the micrometer scale and that the best that something like Wilbur can do is a really bad approximation of physics at the pixel scale (which is often 100+ meters across) with a huge unknown void between each pixel.

[Peter Toth]

Thank you Eusebio Ptolomeu, Harrg, Samuel Harven, and Waldronate for all the critiques and guidance on how I could take my topography to the next level.

The e-book on plate tectonics is somewhat above my level of expertise, but I appreciate you including it, Harrg. A quick question for Waldronate: what exactly is "connectivity" and what are "dual axis-aligned artifacts?"

My process involves something similar to what you've described: I start with a low resolution image and keep up ramping up the resolution until my computer is on the threshold of crashing (usually above 5,000 pixels by 5,000 pixels). Each step of the way I add some fractal noise, erode using the precipitation function, and incise flow. Yes, Wilbur does tend to annihilate all the smaller details if you run the precipitation function enough times.

I ran my procedure another time with somewhat higher noise levels and other modifications, and got the following image:

Attachment 130383


I think I'm beginning to like this result much better, but I'm far from being completely satisfied. Next time, however, I'm going to aim for that example you included (Waldronate) in which a member was attempting to emulate Alpine topography.

Thanks again for the tips.

Peter

[Peter Toth]

Sorry if the link above doesn't work. Here are two views, with and without texture.

Better.png

Better2.png