Abstract: High resolution terrain Digital Elevation Model (DEM) and orthophoto bring severely load including data storage, schedule and real-time rendering, etc.. A high performance terrain data compression method is proposed based on lifting wavelet transform and parallel hybrid entropy codec, and combined with Graphics Process Unit (GPU) Ray-casting to achieve large-scale 3D terrain visualization. First, the multi-resolution wavelet transform model of terrain tile is constructed to map the refinement and simplification operation. Then the multi-resolution quadtree of DEM and terrain texture is built separately based on lifting wavelet transform, the sparse wavelet coefficient generated from quantization is compressed by a hybrid entropy codec which combined with parallel run-length coding and variable-length Huffman coding. The compressed data are organized into progressive stream to do real-time decoding and rendering. The present lifting wavelet transform and hybrid entropy codec is implemented by Compute Unified Device Architecture (CUDA) in GPU. Experiment results show that the data compression ratio is effective with this method, PSNR and code-decode data throughput. High Frames Per Second (FPS) in real-time rendering satisfied the demand of interactive visualization.