Many terrain rendering algorithms have been developed, which per-
form a variety of mesh optimizations to achieve the desired inter-
active frame rates. The most prominent approach is the so called
continuous level of detail technique, which approximates a terrain
by performing a view-dependent triangulation. One disadvantage
of this approach is the fact that the rendered terrain is of course just
an approximation of the original data set. By exploiting the capa-
bilities of todays mainstream PC graphics accelerators we propose
a new technique for the exact rendering of height fields. Due to
its relationship with volume slicing we call it adaptive terrain slic-
ing. This technique is adaptive in the sense that a bundle of slices
is used to sample the terrain, whereas the number of slices is de-
termined to assure sub-pixel accuracy. The core of our approach
is a hierarchical bounding box representation of the terrain, which
is traversed in a top-down order. During traversal we calculate the
actual and refined rendering costs of our adaptive terrain slicing
approach, which enabled us to decide whether it would be faster to
render the contents of the actual bounding box or to descend further
down the hierarchy. After that the minimized terrain slicing costs
are compared to the cost of polygonal rendering and then the faster
method of both is applied. Since the speed of our terrain slicing ap-
proach is limited only by the rasterization bandwidth of the graph-
ics hardware, we can efficiently decouple the rendering costs from
geometric complexity leading to high frame rates without compro-
mising image quality. In particular, our approach is well suited for
replacing the commonly used bump maps with the visually more
pleasing displacement maps.
