for example, if I had a forest can I have a zone far away that would have 2d trees and then a closer zone would have a higher detail of tree(3d) when moving around?
have a geezer at this http://www.irisa.fr/bunraku/GENS/kboulang/grass.html
to get a better idea
Yes, you can.
Check this out:
http://www.blender.org/documentation/pydoc_gameengine/PyDoc-Gameengine-2.34/KX_SCA_ReplaceMeshActuator.KX_SCA_ReplaceMeshActuator-class.html
It contains a script to perform LOD (Level of detail).
Thanks Monster 
I made an LoD thing myself, it’s not as good as that other one, but it’s simpler.
You’re thinking too big if you’re going to try and incorporate LOD. LOD means you have to develop more art junk which means more time. Ofcourse there are some cheap and dirty methods but with blender ge think smallish. lot of opinion
Jessegp, what are your suggestions?
Creating easy art asset: You can use renders from your orginal tree from slightly varying angles and map the rend to distant billboard poly’s, then do some random scaling. I haven’t had a chance to play with the lod script so I’m not sure if it and can swap objects from different layers.
Alternative to LOD: None I was just suggesting not going through the additional work. I don’t think you can come up with any good work around that will allow you to capture your world the way you want, especially with a low level camera that views large areas or the game, without LOD.
Using different lod selection methods: If you know the general path that your character is going to be walking you can group your objjects into ideal sets. Maybe quarter circles, if you re walking in a forest and the character path is restricted, or with a grid, grass on a field. For grass on a field or even trees you could model a large set then just join them, delete half the tree and create a new object. That would allow you densify the forest as you walk around or keep the grass thick where you are an thin outside.
Not using lod: If you are a one man show and you want to do really good separate lod models you will need to make varying polycount models. Typically lod models just try to capture the silohuette and shading characteristics (<atleast thats what I think would be best) Then you’d have adapt the UV and texture to the different lod models. With the number of BGE games that get finished I just think that having these extra time eating steps will hinder most people.
I was just trying to offer an opinion on not using lod, but I don’t know what jhop is actually working on: just playing with the idea, or has a project that he’s been devoting many an hour to. No project + just playing + wants to complete a game = Me suggesing no lod, otherwise go for it. I’m kind of guilty of looking at post count and letting my thinker say “count equals time spent”…I should know better, but its too late can’t take back my previous post.
Don’t worry, I agree that setting up LOD meshes is more work to do. I was just wondering as it sounds you have other options in mind just not telling.
According to the first post, I would suggest to use a (billbord) plane with a tree texture as low detailed mesh and a tree mesh as high detailed mesh. So you end up with 2 meshes. One is simple, one is complex. This shouldn’t be to much overhead on workload but increase performance really much. I think performance is the main reason for using LOD.
I guess there’s mist, too. Its not lod but same lines to an extent.