Spatiotemporal Clustering and Seismic Gap Identification in Nepal Using Density-Based Machine Learning (1990–2026)

Background: Nepal lies along the Himalayan collision zone, generating recurrent seismic activity. Traditional seismic hazard assessments identify gaps through narrative geophysical interpretations, lacking reproducible computational frameworks. Objectives: To apply density-based clustering and Monte Carlo-tested Kernel Density Estimation (KDE) to a long-duration earthquake catalog in Nepal to identify statistically significant low-density zones, assess their temporal persistence, and evaluate their spatial overlap with previously hypothesized seismic gaps. Methods: This study applies density-based unsupervised machine learning to a 36-year earthquake catalog (January 1, 1990, to December 31, 2026). Merging USGS and local network data, we applied Gardner-Knopoff declustering and Maximum Curvature completeness screening. Density-based clustering and KDE were utilized to identify spatial voids. To address spatial biases, analysis was restricted to the seismogenic zone, and an inhomogeneous Monte Carlo null model with False Discovery Rate (FDR) correction was implemented. Results: The pipeline retained 736 independent events, with 351 falling within the defined seismogenic box. DBSCAN identified 16 localized clusters. After FDR correction (alpha = 0.05), 26 statistically significant gap cells were identified. Temporal analysis across sub-periods of 10, 15, and 11 years showed 11 cells persisting across all periods. Validation against the Western Nepal reference box yielded a 15.4% overlap rate, underperforming the 29.4% random chance probability. Conclusion: The method serves as an exploratory, reproducible screening tool for identifying low-density zones. It does not definitively validate previously hypothesized gap locations based on small-to-moderate seismicity alone. Novelty: This study presents a reproducible pipeline applying FDR-corrected inhomogeneous spatial null models to Himalayan seismicity, moving toward auditable statistical screening.