Draping GIS Features on Terrain

A common need is to display conventional 2D GIS features (points, lines, polygons) on a 3D terrain.  The process of projecting a 2D feature on a 3D surface is known as "draping".  For very simple features, or a very flat terrain, this is almost trivial, but for complex features on uneven terrain, it is difficult to draw the features perfectly.

Points

• Point features are the simplest.  There are many ways to indicate a point on the ground: As a 3D object, as a 2D /3D text label, or as a 2D icon / 3D billboard.  Regardless of representation, the height of the terrain at the 2D point is used to compute the 3D point.

Lines (Polylines)

• The simplest approach is to convert each 2D point on the polyline to a 3D point using the height of the terrain, then draw the resulting 3D polyline as a series of line segments.  That is only sufficient if the terrain is flat, or the input polyline has vertices at each inflection point of the terrain.
• More sophisticated is to tessellate each segment of the incoming polyline, and drape each resulting vertex independently.  If draping on a regular grid, a reasonable approximation is to use the grid's horizontal spacing as the tessellation distance.  This handles most cases of uneven terrain.  The example to the right shows a single segment, tessellated and draped in this fashion.
• A small optimization of the above is to detect and omit vertices of collinear segments, so that they do not have to be drawn.

Polygons

• The simplest approach with polygons is again to naively drape each input vertex, resulting in a 3D polygon.  Since the result could be non-planar, it is advisable to decomposed the polygon, before or after the drape, into triangles.
• That is only sufficient if no polygon crosses an uneven part of the terrain.  Otherwise, they will intersect, with the polygon lost under the ground.
• In theory, it should be possible to split geometry of the 2D polygon into a large set of fragments, along the edges of the underlying geometry (grid or TIN).  Then, each fragment could be draped individually, conforming to the terrain.  However, this is complex and expensive in computation, memory and graphics bandwidth.  There are no known implementations of such an approach.

Problems with Draping Geometry

1. Z-buffer fighting.  If the feature (point, line, or polygon) is exactly on the terrain surface, it can conflict with the rendering of the terrain surface itself.
2. LOD problems.  For terrain systems which use LOD (of any kind), a feature draped on one level of detail may intersect the terrain whenever the LOD changes.

The most common solution to these problems is to raise the feature geometry by some offset, such as a few meters.  This is usually visually acceptable if the viewpoint will remain must higher above the terrain than the offset.  If the camera is very low, the offset must be small so that the feature does not appear to "float" above the terrain.  If the camera is very high, the offset must be large for Z-buffer precision.  Some software (such as Google Earth) deals with this by using a dynamic offset, raising the feature geometry as the viewpoint elevation increases.

Texture Draping

• An alternative to draping geometry is to draw the features into the 2D texture map of the terrain, either directly into the base texture, or separate with multitexturing.  This completely avoids much of the complexity and drawbacks of draping geometry, with no tessellation needed, no offsets and no Z-buffer concerns.
• However, the quality is limited to the resolution of the texture.  Sometimes this is acceptable, but often the texture resolution (due to memory or graphics card limitations) is much less than the screen resolution, resulting in blocky (aliased) or blurry feature edges.
• The example the right (from Enviro) shows a single triangle, rendered as a 1024*1024 'additive' texture.  Note how the polygon conforms exactly to the uneven terrain, since it is not independent of it.

Other Approaches

• Scheider,M., Klein,R.: Efficient and Accurate Rendering of Vector Data on Virtual Landscapes (WSCG 2007)
  • Proposes a solution which extrudes the vector data to polyhedra and uses them to create a mask in the stencil buffer, similar to how a stencil buffer can be used for shadow volumes.  It makes use of the OpenGL extensions EXT_stencil_wrap and EXT_stencil_two_side.  It is algorithmically complex, but claims to avoid the limitations of conventional geometry and texture approaches to feature draping.

Related Issues

• Draping roadways on terrain has similar challenges.  See geometry and z-buffer issues.