Visualize 3D terrain instead of the sphere only.

Note: this involves preparing and hosting of a separate tileset which will represent the geometric height by the values of pixels (greyscale). Triangles in the WebGL Earth can be then adjusted by the relevant height (pixel value) loaded from the texture (tile) at given lat/lon position.

Possible datasets to use:
- GTOPO30: global digital elevation model (DEM) ~ 1km x 1km - http://eros.usgs.gov/#/Find_Data/Products_and_Data_Available/gtopo30_info
- SRTM3: Shuttle Radar Topography Mission ~ 90m x 90m - http://www2.jpl.nasa.gov/srtm/
- CGIAR: 90m
http://www.cgiar-csi.org/data/elevation/item/45-srtm-90m-digital-elevation-database-v41
http://developmentseed.org/blog/2010/may/04/srtm-data-amazons-cloud
- ASTER GDEM: Aster Global Digital Elevation Model ~ 30m x 30m -
http://asterweb.jpl.nasa.gov/gdem.asp

Rendering of such tiles can be done with GDAL / GDAL2Tiles / MapTiler / MapTiler Cluster. Ideally into spherical mercator tiles with 16bit grayscale PNG.

Such functionality is not going to be supported by every graphic card, because WebGL vertex shader needs to read a texture:

"The WebGL specification says that you can query WebGL to see how many vertex textures are supported - and the answer can legally be (and often is) zero...in which case you don't have vertex textures and you have to figure another way around the problem."

n = gl.getParameter(MAX_VERTEX_TEXTURE_IMAGE_UNITS);

Demo of similar terrain LoD streaming:
http://www.spidergl.org/example.php?id=8

Display houses and buildings available in the Collada format.

Notes:
More efficient then direct support for Collada format would be to convert the models in advance into JSON files:
http://code.google.com/p/o3d/wiki/ColladaConverter
http://code.google.com/p/opencollada/

Follow: http://learningwebgl.com/blog/?p=1658

The initial implementation should probably load the model and locate it in a correct place on Earth programmatically.

For efficient displaying of whole town composed of larger amount of models a server side preprocessing or dynamic streaming must be implemented - with several level of details (generalized version of model for distant view, detailed for close view once it is loaded).

I we need something like: http://www.youtube.com/watch?v=rGLInTXTdNI#t=07s then prerendering of descreet areas of town should be done. Think of mercator tiles, but with 3d content instead of raster data.