14 I Geometry Manipulation
To eﬀectively render large terrains, many dynamic multiresolution approaches
have been developed in the past years. These algorithms typically adapt the
terrain tessellation using local surface-roughness criteria together with the view
parameters. This allows dramatic reduction of the model complexity without sig-
niﬁcant loss of visual accuracy. New capabilities of DX11-class graphics hardware
enable a new approach, when adaptive terrain tessellation is built entirely on the
GPU using the dedicated hardware unit. This increases the triangle throughput
and reduces the memory-storage requirements together with the CPU load. It
also reduces the amount of data to be transferred from the main memory to the
GPU that improves rendering performance as well.
To reduce the storage requirements, several recent approaches exploit com-
pression schemes. Examples of such methods are geometry clipmaps [Losasso
and Hoppe 04] and C-BDAM [Gobbetti et al. 06]. Though these methods are
optimized for maximum GPU eﬃciency, they completely ignore local terrain sur-
face features, signiﬁcantly increasing the GPU load, and do not support real-time
Dynamic terrain modiﬁcations are an important feature for a number of ap-
plications such as games, where the action changes the terrain, landscape editors,
and other elements. It poses a number of problems such as the need to construct
new triangulation and update internal data structures, which are especially com-
plex when compressed representation is exploited. As a result, real-time terrain
deformations are usually not considered.
The new terrain-rendering technique presented in this chapter combines an
eﬃcient compression scheme with the GPU-accelerated triangulation construction
method and, at the same time, supports dynamic modiﬁcations. The technique
has the following advantages:
• The proposed multiresolution compressed height-map representation sub-
stantially reduces the storage requirements and enables direct control of a
• Height-map decompression is accelerated by the GPU, which reduces ex-
pensive CPU-GPU data-transfer overhead and eliminates data duplication
in main and GPU memory.
• Triangulation construction is completely done by the tessellation unit of
recent graphics hardware:
◦ Topology is not encoded and completely generated by the GPU for
each camera position.
◦ Strict screen space error bound of the rendered surface is provided.
◦ The topology is updated automatically as the terrain is modiﬁed.
• The technique is eﬃciently implemented using hardware-supported features
such as texture arrays and instancing.