15 January 2026 in IEEE Transactions on Geoscience and Remote Sensing
Time-series interferometric synthetic aperture radar (TS-InSAR) has been extensively applied to ground deformation monitoring and disaster risk assessment. In this process, coherent point (CP) selection and connection are two essential steps. Typically, CPs are initially connected using methods such as Delaunay triangulation, followed by the estimation of deformation parameters for each edge. However, the removal of low-quality edges often fragments the initial network into multiple subnetworks, and reliable deformation estimation requires reconnecting these subnetworks into a single, fully connected network. Despite its importance, this problem remains challenging and has not been fully addressed. In this article, we propose an efficient subnetwork connection method, termed KD-tree with iterative expansion (KDIE), which integrates a K-dimension (KD)-tree-based distance searching strategy with an iterative expansion scheme. KDIE efficiently identifies connection edges using the KD-tree and progressively extends the search distance (SD) up to the atmospheric correlation distance (ACD), thereby avoids missing any connectable subnetworks and preferentially connects nearly all subnetworks with short edges. Compared with the existing multilayer subnetwork connection (MLSC) method, two experiments concerning C-band Sentinel-1A images over a mountainous landscape and X-band TerraSAR-X images over an urban area demonstrate that KDIE connects 1.84% and 0.25% more CPs than the MLSC method, while reducing computational time to only 5%–8% of that required by the MLSC method.