Abstract:

[Objective] Glacier surges are abrupt episodes of rapid advance and mass redistribution on monthly-to-multi-year timescales. They are among the primary initiators of cascading natural hazards in high-mountain cold regions and a key topic in global climate research. To explore the predictability of surge events, building data-driven, intelligent models of glacier evolution based on spatiotemporally continuous observations is a promising approach. However, there is still a lack of detectable data features that are strongly correlated with the internal evolutionary processes of surge-type glaciers to support such data-driven models. [Methods] Within a data-driven framework, this study focuses on the feature and observation layers. We systematically review and analyze multi-source feature elements of glacier surges, covering apparent surface changes, observation/sensing methods, data manifestations, and model parameters to assess their detectability, mechanistic linkages, and applicability bounds, and on this basis establish a multi-source system of detectable features. [Results] Our analysis shows that most mainstream indicators are designed for mid- to post-event monitoring and lack the capacity to characterize the incubation process and critical triggering states. To address this gap, we propose a spectral surge index derived from electromagnetic-signal perspectives that are mechanistically related to surge initiation. Retrospective tests reveal clear precursor characteristics in this index. [Conclusion] As an illustrative example, the proposed spectral surge index demonstrates a practical pathway for transforming physical mechanisms into learnable features. Spectral-domain features can be integrated with velocity fields, InSAR coherence, and thickness change as key parameterized inputs to data-driven models.

Key words: glacier surges, measurable indicators, remote sensing detection, data-driven