No, thats the wrong idea.
First of all. You are using an engine version that does not support large world, so forget about 100km.
You’d be lucky to reach 8km - and still probably get sub 30fps with that (landscape wise).
Second of all, the way to achieve it is by using meshes, not the landacape.
You have to section meshes off correctly.
Dynamically, and intelligently.
Using meshes also provides more control over looks, since you can merge meshes into one another instead of having seams/breaks along the road.
The second obvious improvement is that your meshes will be exact and analytical.
You won’t be bound by the number of tris, or have to maintain a static grid of tris for LOD0 any longer.
That’s particularly beneficial in desert areas (which seems to be your case).
A flat plane has 2 tris, regadless of size.
A landscape has 2tris to evey meter.
you likely aren’t tris locked, gpu wise, but it still somewhat matters. Less tris on the base = more objects with more detail you can add.
On sectioning meshes off here are some options.
- A. Go with a total number of tris per mesh regadless of size. Pick a number, say 5k tris, and section things off in a whay that makes sense so that every piece has a consitent count.
- B. Go with sizing. Each piece covers X square meters. This is similar to the components within the landscape, but I wouldn’t go over say 500m^2 per component.
- C. Go with what makes most sense occlusion wise. A circular area would be split into pizza slices. For instance.
Once you have the base, youll likely find that you are drowning in drawcalls when you render the whole landacape.
This is normal.
You need to create HLOD replacements that aggregate the landscape.
An alternative to the HLOD is to mesh merge at runtime based on character position.
Its fast enough that it works fine. Its however far more complex than just baking presets.
Another less effective option could be to toss all the terrains static meshes into an HISMC.
This will provide a cost reduction to the shader because of instancing, but it will preclude Hlods, acor merge, and runtime merge.
However, works ok for water with far more tris than what a landscape would call for, so in a pinch it could perform OK on a landscape too (if you ignore the fact its got bo nanite).
Also, because you are on an engine version where Nanite is a thing…
Chanches are you may not need to aggregate the meahes into HLODs at all. In theory (and remeber its a bad product in an alpha stage) Nanite should handle that for you when you set the terrain meshes to be part of it.
Downsides of meshes you have to work around:
No grass node.
No in editor sculpt.
No inneditor layer paint.
Best tool to use for developing is probably going to be houdini.
I bet that by now it even offers scripts or AI powered terrain segmentation…
Las thing to note.
You do need to make your own UCX colliders for the landscape to work as a mesh.
do not use complex as simple.
If you want to know why, look up what Ark Meshing is to see what happens when your developers are a bunch of shortcut taking idiots (who can’t be bothered to fix any bug in however many years of updates) 
.