Source and pattern identification of ground deformation based on non-negative matrix factorization: a case study

Ground deformation caused by mining affects the safety of underground mining and buildings. High-resolution characterization of ground deformation response to latent sources is a first step toward improved hazard forecasting. We rely on long-term Global Position System (GPS) displacement data and an improved non-negative matrix factorization (NMF) algorithm of blind source separation to identify the sources driving ground deformation. The NMF identifies three sources (S1, S2, and S3) with distinctly spatial and temporal patterns of ground deformation. S1 captures horizontal Y-displacement related to horizontal tectonic stress \({\upsigma }_{1}\). S2 dominates vertical Z-displacement with a widespread isotropous deformation driven by self-weight body force. S3 controls horizontal X-displacement related to horizontal tectonic stress \({\upsigma }_{2}\). Besides, the independent contributions of these three sources can be resolved from the GPS data. The results show that the sharp change of source contribution and even the transformation of dominant source are highly related to the severe deformation belt. This phenomenon is time-independent, and helps us to select sites and find the potential risk area at an early stage.