An Efficient Compressed Sensing Model for Spatial Interpolation with Sparse Geospatial Data via Total Variation Regularization and ADMM

Transport in Porous Media · Springer Nature · 2026-09-18 · DOI: 10.1007/s11004-026-10343-4

Аннотация

Bridging the gap between required and available geospatial data is paramount across diverse geoscience disciplines. To achieve stable and accurate spatial estimation from sparse measurements, this article proposes an efficient compressed sensing (CS) model termed CSTV. It leverages a dual regularization of sparsity and total variation (TV), for which a customized alternating direction method of multipliers algorithm was developed to accelerate the optimization process. Mechanism analysis reveals that the combination of a truncated search space and small-magnitude regularization parameters generally contributes to achieving favorable optimization results. When evaluated on two artificially designed fields and a soil electrical conductivity dataset, CSTV outperformed another CS-based model ( l 1 -TV), four representative geostatistical and non-geostatistical methods, and Gaussian process regression. At the same time, the model exhibited a pronounced trend of variance compression, which requires the use of raw measurements as inputs to mitigate extreme value misestimation. In the future, CSTV could be evaluated with broader types of spatial fields and sampling patterns, while alternative sparse representation methods, higher-dimensional applications, and uncertainty quantification techniques can be explored for further framework enhancements.

Читать у издателя →

Похожие статьи