Editing a world map

[waldronate]

The expression field goes with the Type=Expression. This feature allows for direct computation of f(x,y) instead of a fractal function. Note that there is a fractal function available in the expression parser. The available functions are listed in the documentation.

The assumption of the 3D window is that it's viewing a unit cube centered at the origin. If you're using a whole-world top/left/right/bottom of 90/-180/180/-90, then you need to set the Vertical Scale value to something that will get into that range. For a world that's in the 3000 altitude range, try something like 0.01.

One way to get several mounds with the fractal settings on is to draw multiple disjoint selection parts. Draw a selection area. Then hold down the Shift key and draw another. And so on. You can then save that selection as an image file for later use if desired to avoid having to redo your selections.

For the fractal issues, consider increasing the area of the plane evaluated (increasing the Size values on the Height Field Computation dialog). This change will make the size of the fractal elements proportionately smaller and increase the apparent effect on the mound. To get an idea of how it will look as an unmodified noise function, try computing the height field without the mound operation using Filter>>Calculate Height Field. The resulting field will appear as modulation for the mound envelope.

I really should get around to updating things, especially those that predate the great UI reorganization that happened a few years back. You may have noticed that the colors on the height field in the continental shelves tutorial don't match those of Wilbur these days due to a decidedly purple cast. That tutorial is well over 10 years old now, I think. Amusingly, you're the first person to mention that the functions mentioned don't match those in the software.

[Deadshade]

Thanks. Your short comment is a very rich meal. I'll need a few hours of experimenting to get it all. I wish I knew this "shift" thingy earlier - it would have saved me a few hours. (And yes, I suspect it is in the "very old documentation" but I still didn't read it and especially comitted it to memory in its entirety.)

So in the meantime 2 very short questions :
- where do I find the units used ? I am wondering all the time if the heights are in feet, yards, meters or whatever.
- Do you mean that if i want to see a 3D view of a small part (f.ex a mountain) , I must put somewhere (where ?) the coordinates of a square containing the mountain ?

EDIT :
Could do nothing with : "(increasing the Size values on the Height Field Computation dialog)".
There are no Size values in this dialogue. Did you mean to put a high value (like 30) in the Scaling dialogue ? I tried that but saw no special change.

EDIT2 :

I still have great problems with porting Wilbur in FT3.
I took this image in Wilbur, saved it MDR (basically flat world with a few mounds)


Then I closed Wilbur and used the There and back in FT3 (New, Binary etc) inputting the above MDR file
And I obtained that :


It's just a mixture of strange pixels with all parameters with values in the trillions or more. Randomly positive or negative.

Have you an idea what is going wrong here ?

[waldronate]

The vertical units used are arbitrary. I don't think that there's anything in Wilbur that uses specific altitude units. The extents on Map Info dialog is interpreted as either arbitrary units for planar worlds or degrees for places where that matters like the map projection dialog.

Using the 3D preview window in Wilbur to view just part of a surface is pretty twitchy and difficult activity. The 3D preview window navigation uses the left mouse button to rotate the data around the origin, the middle mouse button (typically click and drag on the mouse wheel these days) to pan around the map, and the right mouse button to zoom in and out at the center of the view.

On the "Height Field Computation" dialog, there is a checkbox labeled "Spherical Evaluation". If checked, the labels for the controls of interest are under the heading "Sphere Radii"; if unchecked, the labels will read "Size". If you are instead using the "Fractal Noise" dialog, then the fields of interest are labeled "XY Scale".

To read an MDR file in FT3, make sure that you're using the "..." button on the "Binary Data" dialog in the new file wizard. On that file open dialog, ensure that selected the File Type drop list has "mdr files" type selected. from the picture you provided, it looks suspiciously like the import was done with the per-sample info set to "4 byte" instead of "4 byte, float". For a 512x512 MDR file, the dialog should look like:

[Deadshade]

To read an MDR file in FT3, make sure that you're using the "..." button on the "Binary Data" dialog in the new file wizard. On that file open dialog, ensure that selected the File Type drop list has "mdr files" type selected. from the picture you provided, it looks suspiciously like the import was done with the per-sample info set to "4 byte" instead of "4 byte, float". For a 512x512 MDR file, the dialog should look like:

It worked now. The 4 byte float is selected as default. But also Signed is selected by default. It took me a while to see the difference with your example but when I unchecked "Signed" it worked.

On the "Height Field Computation" dialog, there is a checkbox labeled "Spherical Evaluation". If checked, the labels for the controls of interest are under the heading "Sphere Radii"; if unchecked, the labels will read "Size". If you are instead using the "Fractal Noise" dialog, then the fields of interest are labeled "XY Scale".

Understood. I will experiment with that to check if I see any difference changing those settings..

A small question : when I have a multiple selection with the shift trick, is it possible to cancel only 1 selection keeping the rest active ?

A bigger question : every time I launch FT or Wilbur I am frustrated at seeing all those beautiful and realistic landscapes that no amount of working with a prescribed world can emulate. Not even near. Perhaps with hours of GIMP (or Paintshop) but I don't want to go there yet.
What would it take that exactly the same algoritm generates a world with the same realistic quality but with a small difference - the mountains and the coastlines are approximately at a prescribed place ?
I mean you would just handdraw an approximate blob representing a continent and Inside one or a few blobs representing mountains and the algorithm does the rest ?
After all when you define the ground-ocean boundary and the mountain boundaries, you have implicitely defined (almost) everything else necessary to get a realistic world (climate, rivers, winds, erosion and even plate tectonics).
All the rest of the work is then just adding some fine details like rivers and realistic colors for different terrain types (deserts, arctic, jungle etc).

I don't realise if this is a big or a small work.
Or was FT purposefully designed that it only generates high quality random worlds and any constraint on a placement of a feature is excluded if the same quality of generation is required ?
This was precisely the reason why I called this thread editing a (existing) world and not creating a (random) world when I was just entering this wonderful computer assisted cartography universe

[waldronate]

Shift adds to the current selection; ctrl should remove from the current selection.

If I understand correctly, you'd like to make a mask for the coastline and another mask representing mountains, correct? Then the fractal function would fill in details? That's pretty much the basic Wilbur processing chain of loading the coastline as a selection, filling it, loading the mountain mask as a selection, filling it, adding noise, and following all that up with post-processing effects like erosion. Most of the erosion operations are there to smooth the otherwise obvious transitions.

I think that what you really want, though, is a way to take the lowest-frequency octaves of the fractal function and replace them with some user-defined drawn areas. The simplest way to do that, of course, is simply to specify the fractal terrain as higher-frequency elements added to the base values. This process is trivial in Wilbur: draw your terrain and then do fractal noise with the operation set to "Add". However, there will still be steps in the mask. A more "correct" solution would be to band-limit the user inputs to low fequencies and go from there. Not a terribly hard task in absolute terms (transforming to and from the frequency domain is cheap enough computationally), but the transform would need to preserve the spherical topology that's mapped onto the plane area. Or reproject into pieces in an equal-area projection, use a blur on the pieces, and then reassemble the pieces. It's a fair amount of work to code up, but there's nothing inherently difficult about it.

The worlds in FT use the basic fractal functions that we discussed earlier. The overall shape and distribution of the largest features ("continents") is determined by the first one or two octaves of the fractal function. Mountains appear in the center of continents (a "feature" that has caused an unbelievable amount of whining among the "must use tectonics" crowd). I know a whole lot of things that would be useful for generating better world that I didn't know 15 years ago when I started on FT. My biggest single problem is that if a product is labeled "FT", then it needs to be able to read and generate the worlds that FT's users have created. There was one breaking version in FT1 to fix a timing sensitive multithreading bug and I promised not to do that again.

[Deadshade]

If I understand correctly, you'd like to make a mask for the coastline and another mask representing mountains, correct? Then the fractal function would fill in details? That's pretty much the basic Wilbur processing chain of loading the coastline as a selection, filling it, loading the mountain mask as a selection, filling it, adding noise, and following all that up with post-processing effects like erosion. Most of the erosion operations are there to smooth the otherwise obvious transitions.

Yes that's basically it. And because Wilbur (and FT in a way) do this very easily, this was one of the first things I tried. But unfortunately it was useless because I didn't get beyond the first easy step. What I did was :
- create a map
- load a coast selection. At this stage everything is flat and at 0 m.
- run fractal noise with default setting over the selection. It gives that :


While this looks quite fractal and realistic, 3 problems appear that make it seem useless.

- First for some reason the histogram shows that the height field is between 0.5 and some and 1. So the continents are full of mountains. Why there is a hole between 0 and 0.5 and some when I was supposed to add (or replace) some random number in [0,1] to 0 I have no idea. I can restore a full Spectrum without holes by an offset but this is still a problem.

- Second as the selection works just as a mask, there is no reason that the height field connects to the boundary exactly at 0. So there is no continuity and the continents are surrounded by cliffs with variable height everywhere. I have no idea how to establish continuity in order to have the heightfield = 0 everywhere at boundary.

- Third, of course, it is all wrong. Mountains are on places where they shouldn't be and are not on places where they should be. I could imagine that I might "sculpt" selections using scaling but there is potentialy an infinity of selections and I would create discontinuities all over the place. This seemed to be such an incredible amount of work that I rejected the idea immediately.

Being still a relative beginner, it is possible and even probable that I miss something important but I couldn't find what.

The overall shape and distribution of the largest features ("continents") is determined by the first one or two octaves of the fractal function. Mountains appear in the center of continents (a "feature" that has caused an unbelievable amount of whining among the "must use tectonics" crowd)..

Well the "must use tectonics" crowd that whined, misunderstood and misinterpreted everything about tectonics. They apparently didn't realize that the Earth is just a particular realization of the tectonic dynamics.
The latter is just a consequence of convective movements in the mantle. So on any planet with a convecting mantle there would be plate tectonics.
However the size, density and velocity vector of the plates is absolutely not an invariant.
Orogenesis will happen along collision boundaries of 2 plates and this is obviously not reserved to coastal areas even if it is approximately true for the particular case of the present Earth.
You can well imagine 2 continental plates moving against each other so that they first eliminate the ocean which is between them.
Then when they merge in a single land mass, they will start creating mountains and these mountains will be "in the middle" of the new continent.
This is btw the case of the Himalays - they are "in the middle of the continent" and the fact that they are relatively near to the coast is only due to the fact that the size of India is much smaller than Asia.
Of course there is no reason that it must always be so.

So for a fantasy planet with working tectonic dynamics and a given mountain distribution (excepting a few pathological cases) you can always establish a topology and velocities of the plates that would produce exactly this distribution. That it could be extremely different from the Earth's particular case is obvious but that's precisely the reason why we do fantasy planets and not Earth's copies.

[waldronate]

I think that the biggest difficulty with your example is that you used continental-scale features on a continental-scale mask. You need to greatly increase the area that your fractal parts occupy, perhaps by a factor of 20 or more. Then the noise will be more uniform over the surface. You'll still need post-processing things like erosion to smooth out the transitions, though. Or feather your selection before applying the fractal function.

The complaints from the tectonics crowd about the mountain placement are more due to the fact that the majority of people seem to want a world formed exactly like Earth, but with differently shaped continents. They want to see all of the tectonic features, not just one fairly rare case (mountain ranges are more likely at continental margins because ocean is more likely than continental crust). The most effective way to get tectonic features is with a tectonic simulation, of course. The big problem with random fractal forgeries (as Voss termed them back in the '80s) is that they are visually plausible, but don't hold up to detailed inspection. Whole categories of features tend to be missing. Statistics aren't quite right. Feature placement is odd. So while the first glance is visually pleasing, trying to use the things for much more than pretty pictures just doesn't work well.

[Deadshade]

I think that the biggest difficulty with your example is that you used continental-scale features on a continental-scale mask. You need to greatly increase the area that your fractal parts occupy, perhaps by a factor of 20 or more. Then the noise will be more uniform over the surface. You'll still need post-processing things like erosion to smooth out the transitions, though. Or feather your selection before applying the fractal function.

.

I am sorry but I don't understand that. My continents occupy about 1/4th of the map. So if I want to act on an area 20 times bigger, I'll need a map 5 times bigger than what I have and mine is already 2500x5000.
Of course I expect and am ready to do all the post treatment like erosion, rivers, canyons, proper coloring etc perhaps even a bit of GIMP. I think I know how to do all that already.
But before I start with it I need some proper mountains on right places first and this is what I am struggling with since I began this project. The only way that more or less worked was the height masks that created a terraced blobs that I can then erode but this is not satisfactory.

Btw in the problems I mentionned above, do you know why I get heights only between 0.5 and 1 when I apply the fractal ? Shouldn't it be between 0 and 1 ?

And is there a way to suppress all those cliffs so that the continents are continuous with the ocean ?


Regarding the tectonics, we don't have yet a correct solution of natural convection let alone of convection in planetary cores and mantles. So as we can't solve the physics of it, it is obvious that just a few mathematical graphics won't solve it either. But what I wanted to say was that because the physics are not quantitatively understood, no particular distribution of mountains is à priori preferred by the system.
Give me any (non pathologic) mountain distribution and I will draw the tectonic plates and their movements which will be qualitatively compatible with what we know about tectonics. It may be very different from the number, sizes and velocities on the Earth, but there exists no argument that would say that it is less probable than what we see on Earth . There would be problems only if the mountains area represented a very significant fraction of the sphere area - f.ex more than 40 or 50%.
Among others there is no reason that the mountains be more probable on coasts - sofar this is only observed in one particular case among billions - the Earth.

[waldronate]

The size of the features that the fractal function calculates is a function of the size of the evaluated area compared to the natural frequency of the noise function. Perlin noise is composed of gradient vectors evaluated at integer lattice points. This use of an integer grid (found in most fractal basis functions in common use) means that the natural frequency of the noise function is 1. If you view that function on a portion of a plane running from -1 to +1 in both X and Y, you would expect to see about 2 zero crossings in the field of view. If you instead view an area of the exact same running from -8 to +8 in both X and Y, you would expect to see about 16 zero crossings in the field of view. Applying the same logic to a fractal construction such as fBm, simply by changing the viewed area of the function by 2**3, we have effectively skipped the first few octaves of the fractal without changing any of the defining fractal parameters.

The ridged multifractal runs from about 0.5 to 2.5 in its basic definition, I think. It's a part of being defined as the absolute value of sum of values ranging from -1 to +1.

Feathering the selection before filling it will smooth out the edges of the resulting operation.


There actually is a case where mountains preferentially appear along the coast: when a continent is subducting oceanic crust. The best example here on Earth at the moment is the Andes. There also seems to be a need for significant water in the crust to make tectonics work because it dramatically lowers both the strength and melting point of the materials. Note that we have data for a number of other worlds and only Earth seems to have evidence of tectonics (moons like Europa have something similar, but not quite the same).

[Deadshade]

The size of the features that the fractal function calculates is a function of the size of the evaluated area compared to the natural frequency of the noise function. Perlin noise is composed of gradient vectors evaluated at integer lattice points. This use of an integer grid (found in most fractal basis functions in common use) means that the natural frequency of the noise function is 1.

OK. The good news is that I think that I now see what it's doing and why. The wave length domain has the fundamental wave length (L0) equal to the size of the whole map. Then every selection acts like a high pass filter and simply displays only a small part the first iterations L0,L1 ..Li.
Because it is not (apparently) possible to set the first wave length to a smaller size, the fractal is necessarily truncated for large wave lengths inside a selection.

Another good news is that during the Learning process I learned how to erode and do realistic canyons and rivers.

The bad news is that the answer on the very first question I asked in creating this thread "My target is to take the outlines of the continents&isles I created and put textures/elevations on it. Is Wilbur the right software ?" is no if I want the same realistic look that Wilbur creates for random full map generations and it is also no for FT3 that I bought when I unfortunately still had no idea about what it was good for.

So to reach my target, I'll have to focus on GIMP and learn it in detail because I already saw that it enables to create realistic mountains and to easily edit realistic color gradients on prescribed places by simulating fractals with filling a mask with random clouds.
I may also try to play with some cut (in Wilbur) and paste (in GIMP) if I find a workflow that allows to solve the continuity problem.

The ridged multifractal runs from about 0.5 to 2.5 in its basic definition, I think. It's a part of being defined as the absolute value of sum of values ranging from -1 to +1.

I didn't know that. By running a test, the result is in fact in [0.584 , 2.045]. I also tested that the interval was very different for every type of the fractal.

There actually is a case where mountains preferentially appear along the coast: when a continent is subducting oceanic crust.

Right. But there is no reason that a planet exhibits preferentially continents subducting oceanic crusts.
I can take a planet with a very hot core and low viscosity magma and there are certainly many among the billions that exist only in our Galaxy.
The consequence is small and active convecting cells. This planet would have mountains all over the place and an extremely active tectonics. So no, the Earth is in no way some kind of standard that every other planet in the Univers has to emulate. It is only a very particular realisation of a convecting planet.


So what's left is to thank you for your patience and explanations in this thread that allowed me to have now a fairly accurate idea about what Wilbur and FT3 can do and what they can't do.



EDIT
A small question.
I can't still erase a selection from a set of selections created with shift. When I keep ctrl pressed, I just draw another selection. But when this other selection crosses the old one, the result is a strange mixture of both.
What am I doing wrong ? I would like to make only disappear one selection among N, not to draw a new one.