Daily Summary

Direct geohazard studies in this set are notably state-aware. Landslide and slope papers converge on moisture history, spatial heterogeneity, and submerged loading as primary controls rather than secondary modifiers: Mediterranean projections indicate that landslide frequency can rise even under overall drying when rainfall timing, intensity, and antecedent wetness shift; controlled detachment tests in tropical dispersive soils likewise center threshold behavior on antecedent moisture; and site-specific knowledge-informed learning embeds sparse investigations within physics-based slope simulations to retain spatially variable soil properties. Marine and submerged slope studies extend this mechanistic emphasis below water, showing that elastoplastic soil-water coupling can reproduce liquefaction and reconsolidation, micropile reinforcement must be limited by structural capacity to avoid overstated safety gains, hydrate dissociation drives distinct creep-to-instability pathways, and seismic failure of submerged silty clay and coral sand slopes depends on hydrodynamic pressure, excess pore-water pressure, and submergence state.

A second cluster strengthens hazard diagnosis across seismic, volcanic, coastal, flood, and infrastructure settings. Megathrust coupling estimates are shown to be systematically biased when material heterogeneity is ignored, with direct implications for earthquake and tsunami interpretation. Volcanic studies extract source and stress information from sparse but physically constrained observations, using joint seismic inversion to characterize the 23 November 2025 Hayli Gubbi eruption sequence and focal-mechanism-based modelling to resolve depth-dependent stress rotations before the 1998 Piton de la Fournaise eruption. Coastal and cryospheric datasets broaden the empirical baseline for low-frequency hazards through a global fatal coastal-cliff inventory and a worldwide glacier-failure database, while flood and coastal-risk papers move toward actionable diagnosis through training-free post-event floodwater-depth retrieval, inequality-explicit socio-hydrology, and integrated coastal hazard mapping. Parallel engineering studies on flysch tunnels, gravel-foundation dams, and multi-hazard urban settings show the same shift toward monitored, scale-aware mitigation rather than static classification alone.

Separate from these direct findings, the selected transferable remote-sensing and AI papers concentrate on observation quality, calibration, and physical coherence. They improve geospatial representation learning, SAR denoising and matching, optical-SAR soil-moisture fusion, river-ice extraction, active reconstruction, LiDAR representation, adverse-weather restoration, event-time detection, and long-horizon forecasting with predictive uncertainty. Several contributions explicitly target robustness under missing modalities, scene shift, temporal correlation volatility, or sparse labels, while physics-oriented operator and PINN papers address heterogeneous coefficients and localized gradients. These are methodologically relevant to future geohazard systems, but the papers themselves should be read as transferable advances rather than as demonstrated geohazard validations.