Resolving Data Conflicts in Multi-Source Landslide Monitoring: A Critical Review Towards Trustworthy Early Warning

Abstract

Multi-source monitoring has become the default configuration for landslide early warning, yet inter-sensor divergence is often treated primarily as measurement uncertainty or noise to be reduced, while its potential diagnostic value receives comparatively less attention. This review examines the transition from data fusion to conflict-aware monitoring through a bibliometric analysis of 264 Web of Science records (2015–2026) and a critical appraisal of methodological advances. The bibliometric results reveal a three-stage evolution—from instrument-driven to data-driven and now to fusion-driven research—and expose a structural blind spot: the literature concentrates on how to fuse data successfully while largely neglecting how to diagnose and resolve conflicts when sensors disagree. To address this gap, we propose a conflict typology that distinguishes datum conflicts, scale conflicts, and response conflicts according to their physical origins rather than their numerical symptoms. Unlike traditional uncertainty inventories, sensor fault classification schemes, and fusion-stage taxonomies, the classification system proposed in this paper links each type of discrepancy to a falsifiable physical cause, the necessary metadata, and a corresponding resolution pathway. We further argue that conflict detection should move beyond cross-sensor threshold comparison toward physics-informed consistency checks, because the divergence of trend and phase among monitoring quantities often encodes mechanistic transitions that statistical smoothing would discard. A critical comparison of five resolution paradigms—mathematical weighting, adaptive deep learning, physics-informed neural networks, Dempster–Shafer evidence fusion, and knowledge-graph-based reasoning—shows that no single approach dominates; rather, the choice depends on the match between data endowment, theoretical reserve, and site complexity. Finally, we outline an auditable closed-loop framework in which every conflict is recorded with its sensor metadata, environmental driver, resolution paradigm, and verification outcome, converting disagreement from an operational nuisance into a reusable knowledge asset. These findings suggest that trustworthy early warning does not require the elimination of inter-sensor divergence, but rather transparent, traceable, and physically grounded interpretation of such divergence.