Graph-Based Spatiotemporal Multi-Hazard Risk Assessment: Case Studies from India

Over the years, India has experienced numerous rainfall-triggered landslides that initiate multi-hazard events, resulting in substantial human loss. This study presents a graph-based risk assessment of multi-hazards for two case studies in India: The North Sikkim Glacial Lake Outburst Flood in October 2023 (NS-GLOF) and the Wayanad Landslides in July 2024 (MCW-Landslide), which collectively claimed over 600 lives. Together, these events caused extensive loss of life, infrastructure damage, and long-lasting disruption across fragile mountain catchments. The framework integrates a multidimensional approach that uniquely combines dynamic rainfall and discharge thresholds, stakeholder-informed hazard sequence identification, spatiotemporal hazard progression, and elements at risk. Heterogeneous data sources, including remote sensing, field surveys, and gray literature (non-peer-reviewed sources such as government reports, technical documents, and official situation bulletins), are synthesized to construct weighted, directed hazard networks. Graph-theoretic metrics, such as degree centrality, betweenness centrality, and cascade depth, are then used to compute dynamic sub-basin-level risk scores.

Empirical threshold analysis using rain gauge and discharge data showed consistent exceedance across multiple antecedent rainfall models, confirming their applicability for the 2023 NS-GLOF and 2024 MCW-Landslide events. Both the NS-GLOF and MCW-Landslide were triggered by extreme rainfall events, which played a pivotal role in their initiation, progression, and impact.  Additionally, in the case of the NS-GLOF, critical discharge thresholds also played a role.

Hazard sequences for both regions were reconstructed using gray literature, scientific reports, and stakeholder consultations to establish how primary hazards evolve into secondary and tertiary outcomes. Rather than treating hazards as isolated events, the synthesis revealed consistent pathways in which rainfall-driven processes form the initiating trigger in both locations. Stakeholder engagement further validated these patterns: in Wayanad, the dominant sequences—rainfall → landslides and rainfall → floods—reflect a tightly coupled system where hydrometeorological forcing rapidly translates into slope and channel instability. In Sikkim, stakeholders highlighted rainfall-triggered landslides but also identified earthquake-linked cascades, indicating a more diverse and compound-trigger environment. 

These sequences are mapped onto ~5 km² sub-basins using ALOS PALSAR DEM-based discretisation and multi-temporal satellite imagery to capture spatial impact footprints, runout lengths, and intersections between multiple hazards. Directed, weighted hazard networks are then constructed, with edge weights combining stakeholder-reported frequencies and observed occurrences, and node importance quantified using degree centrality, betweenness centrality, and cascade depth. The resulting weighted directed graphs reveal that Wayanad’s risk is dominated by a small number of highly connected hazards, namely landslides and floods. North Sikkim exhibits a longer, multi-hazard failure chain, with earthquakes, landslides, and GLOF-related dam collapse each playing comparable roles in propagating risk. Spatial integration of network scores with sub-basin characteristics further highlights downstream districts in Sikkim and upstream failure zones in Wayanad as critical amplification nodes. 

The usability of these results and the methods employed provides a foundation for initial trigger analysis that can serve as downstream early warning and targeted risk mitigation. The mapped hazard progression, which identifies where cascades originate and how they propagate through interconnected sub-basins, offers actionable guidance for designing sub-basin–specific warning thresholds that reflect the actual sequence and timing of hazard escalation in both regions.