The storage overhead and I/O bottleneck of supercomputers creates a challenge in efficiently analyzing and visualizing large-scale multivariate SAMR data. It is thus necessary to greatly reduce the data size on the premise of maintaining data accuracy. In this paper, we propose a feature-driven compact representation model to handle structurally complex, high-dimensional, and nonlinear structured adaptive mesh refinement (SAMR) data for efficient storage, analysis, and visualization. We combine information-guided domain partition, distance-based dimensionality reduction, and error-bounded data representation to form a coherent three-component framework, achieving high compression ratios while ensuring low accuracy loss. Our approach addresses the key bottleneck in the visualization of large-scale multivariate SAMR data generated by massively parallel scientific simulations, namely the mutual restraint relationship between compression efficiency and data fidelity. We validate the effectiveness of our method using four datasets, the largest of which contains 4 billion grid points. Experimental results demonstrate that, compared with the state-of-the-art methods, our approach reduces data storage costs by approximately an order of magnitude while improving data reconstruction accuracy by nearly two orders of magnitude.
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号