Most terrain models are created based on a sampling of real-world terrain, and are represented using linearly-interpolated surfaces such as triangulated irregular networks or digital elevation models. The existing methods for the creation of such models and representations of real-world terrain lack a crucial analytical consideration of factors such as the errors introduced during sampling and geological variations between sample points. We present a volumetric representation of real-world terrain in which the volume encapsulates both sampling errors and geological variations and dynamically changes size based on such errors and variations. We define this volume using an octree, and demonstrate that when used within applications such as line-of-sight, the calculations are guaranteed to be within a user-defined confidence level of the real-world terrain.