Terrain data pose modeling challenges due to their high inherent redundancy and difficulty of identifying features at different levels of detail. We propose a multiresolution analysis framework based on a graph-based wavelet construction. Our approach produces a sequence of intermediate resolution approximations of the terrain model. The details pertaining to each resolution reveal scale-specific features. Using a guiding heuristic, the proposed wavelet construction also conserves salient features. Furthermore, the proposed framework allows both geometric and attribute vertex information and can be used for modeling tasks sharing the same characteristics and constraints with terrain modeling. In particular, our graph-based wavelet framework is an option for multiresolution filtering and feature classification or clustering.