It seems that I am constantly asked this same question every week, so I thought I would make an explanatory blog post about it.
Where most people seem to fall down on terrain software, is a larger heightmap value directly equates to a larger area of land, and not to a higher detail version of the same terrain mesh.
It is a common misconception that changing the width and length resolution of the heightmap to a larger value somehow makes it shorter in altitude. It doesn’t. A larger heightmap size directly equates to a larger area of land, not a higher detail version of the same terrain mesh. That is simply an optical illusion of ratios. Basic math proves this.
What you are creating is a larger area of a constant altitude range. Many people seem to fall into this same false view of how 3D terrain worlds work. If I look at a 1024 heightmap versus a 4096 heightmap, both of them with the same height mountains on them, both of them at Zoom Extents in the editor, well of course the 4096 heightmap will visually appear to be one quarter the altitude height range, this is because I am zoomed out four times further. I am looking at four times as much width area with the 4096 heightmap, so of course the height will be one quarter as much.
If I take a digital elevation model that is 1 km by 1 km that contains mountains that are 1 km high (width:height ratio 1:1), then I take a larger piece of that same digital elevation model, say a 2 km by 2 km piece, well the mountains will still be 1 km high (ratio 2:1), the mountains will not somehow double in height just because I have taken a larger area piece of that same digital elevation model. Likewise if I go even larger, a 4 km by 4 km piece of that same digital elevation model, then I will be getting more mountains in my piece since the area is larger, but the mountains don’t become 4 times as high, the mountains with still be 1 km high (ratio 4:1). So the aspect ratio of width:height changes when I have a piece of heightmap that is getting larger in size, especially if I am Zoom Extents on all four terrains. If I Zoom Extents with differing heightmap resolutions, then the aspect will correctly change between them based on their resolution ratios. The mountains don’t somehow grow in height when the area of land mass that I am using increases. If the width:height ratio was always a constant, then when I had a 20 km by 20 km terrain, would I expect the mountains to all be 20 km high? Even Mount Everest is only 8 km in elevation, not 20 km.
This same false view also seems to get applied to the modifiers, especially the erosion modifiers. Most erosion simulation algorithms use cellular automata, where the algorithm loops over the surface of the heightmap examining 3×3 cells for altitude difference and moving water and soil down to lower elevations. The erosion cell is always a constant size, but if I look at a 1024 heightmap versus a 4096 heightmap, both at Zoom Extents, well of course the erosion on the 4096 heightmap will visually appear to be one quarter the size, this is because I am zoomed out four times further. But if I zoom up to the same heightmap feature so that it is the same size on the screen, such as walking on the terrain in a video game, then the erosion is going to be the identical size between the two heightmaps.
As another quick example, let’s look at the Rain Erosion algorithm in TerreSculptor. It uses a “brush” of a specific size to simulate a raindrop of flowing water. If I choose a 6 pixel brush, then it will be 6 pixels regardless of the size of the heightmap. But again, if I Zoom Extents on a 1024 heightmap and a 4096 heightmap, the brushes on the 4096 heightmap will look one quarter the size, this is because I am zoomed out four times further. But if I zoom up to the same heightmap feature so that it is the same size on the screen, such as walking on the terrain in a video game, then the erosion is going to be the identical size between the two heightmaps. A 6 pixel erosion brush is always a 6 pixel erosion brush, it doesn’t somehow change brush size when you change the resolution of the heightmap.
All of this discussion assumes that we are looking at a constant units per meter value for all 3D terrain meshes, which normally we are using constant values in a 3D application.
For example, the default 3D Sizing Units in TerreSculptor is 1 unit = 1 cm. This also happens to be the same in Unreal Engine.
So if I have a 1024 terrain and a 4096 terrain, the 4096 terrain is four times the total 3D world area, or four times larger in size, when both meshes are 1 unit = 1 cm.
I would have to change the 3D world Sizing Units to 1 unit = 0.25 cm for the 4096 terrain in order to render both terrains at the same 3D world area.
So if we instead look at a meter value for size measurement to make this easier, if the 1024 terrain is 1 km square in area, then the 4096 terrain is 4 km square in area.
They are not both 1 km in area unless the 4096 terrain is set to one quarter the Unit Sizing (1/4 Vertex Spacing) of the 1024 terrain.
Let’s look at Unreal Engine. It defaults to 1 grid unit = 1 cm.
The Landscape has a default Scale setting of 100. So each terrain polygon is 1 meter.
If I import a 4096 heightmap, it will be four times the land mass area of a 1024 heightmap.
It will NOT be four times greater in detail resolution and the same land mass area.
I would have to change the mesh Scale property to one quarter the amount to reduce the mesh vertex spacing.
But the underlying heightmap will still always be four times the area.
By default the vertex spacing in a terrain mesh is always a constant amount, such as 100 in Unreal Engine, and 256 in TerreSculptor.
So changing the number of vertices in any direction increases the total area of the terrain mesh, it does not increase the detail of the mesh.
So a 4096 vertex mesh is four times the area of a 1024 vertex mesh, NOT four times the amount of detail in the same area.
And in real-world use, in engines such as Unreal Engine, you MUST leave the terrain mesh vertex Spacing value fairly close to the default value of 100.
If you tried to import an 8192 heightmap and change the terrain mesh vertex Spacing to one eighth the amount (100/8=12.5), in order to increase the terrain detail to eight times as much, then you are going to be pushing so many polygons that you will probably get less than one frame per second.
So for all intents and purposes, the 3D mesh vertex Spacing must always be a fairly large value, so therefore, increasing the number of vertices always increases the terrain mesh area, and not the terrain detail level.
As you go from lower resolutions to higher resolutions you are in fact changing from hills to mountains to continents. Most people believe that it is an increase in detail when it is actually an increase in area.
Realistically you cannot increase the mesh detail in any 3D engine more than about half the vertex distance spatially, in other words, you can’t compress the vertexes more than about 50%, so on Unreal Engine you can adjust the Scale to 50.
You cannot go much higher in detail than that otherwise you are simply pushing too many polygons to get decent rendering performance.
And that increase in detail is not really much that you can notice (a one half meter increase).
That is why, instead of increasing vertices and polygons, you instead start using Normalmaps on the terrain textures.
Those allow you to get down to the renderer pixel/texel level of detail, which you simply cannot do with terrain polygons.
BUT to continue this lesson, we cannot confuse the 3D terrain mesh with the underlying heightmap image. They are two totally different things.
Choosing a larger heightmap ALWAYS chooses a larger “image” area, not a higher detail at the same area Sizing Units.
Sizing Units relates ONLY to the 3D mesh, it does not relate to the underlying heightmap at all.
If I am in Photoshop looking at a 1024×1024 image and a 4096×4096 image, the 4096 image is ALWAYS larger in area. It is always more pixels.
And unfortunately we cannot simply change the Units Scale to change the effects of such modifiers as Erosion.
ALL of the modifiers treat the underlying heightmap as a floating-point grayscale image, and ALL of the modifiers work directly based on pixels and groups of pixels.
None of the modifiers know anything about the Units Scale of the 3D world, they work only on the pixels of the underlying heightmap image.
So if I use a heightmap filter that uses a 3×3 cell, it will always be a 3×3 cell on the underlying heightmap image regardless of the 3D terrain mesh settings and 3D world settings.
And if I use the Rain Erosion with a 6 pixel brush, it will always be a 6 pixel brush on the underlying heightmap image regardless of the 3D terrain mesh settings and 3D world settings.
Where most people seem to fall down on terrain software, is a larger heightmap value directly equates to a larger area of land, and not to a higher detail version of the same terrain mesh.
I hope that this enlightens some people on how heightmaps and terrain mesh rendering work.
-eof-
