Projects | Xin Wei

Cover images are conceptual illustrations; the project pages present the underlying methods, data, and evidence.

Susceptibility

Physics-Guided AI for Regional Landslide Susceptibility

A three-paper first-author research program develops, tests, and interprets physics-guided models for cross-region landslide-susceptibility mapping, with explicit analysis of training-sample variability.

Conceptual illustration of fast-moving landslide runout combined with multiscale geospatial layers and a deep-learning representation.

Hazard

In development

Hazard Characterization of Fast-Moving Landslide Runout with Geospatial Deep Learning

Project materials are currently being prepared.

Conceptual illustration of an offshore mat foundation under cyclic wave loading, showing pore-pressure buildup at the soil-foundation interface and a schematic capacity response.

Infrastructure Response

Horizontal Capacity of Offshore Mat Foundations under Cyclic Waves

Coupled soil–water simulations show how cyclic loading, clay structure, and over-consolidation govern interface pore pressure and horizontal-capacity evolution in offshore mat foundations.

Conceptual illustration of submarine debris-flow runout intersecting a seabed pipeline network, with an alternative route outside the main impact corridor.

Risk

Submarine Debris-Flow Runout and Pipeline Risk

Regional runout simulations capture material softening over complex seabed terrain, then estimate pipeline impact, network damage, and the benefits of rerouting.

Conceptual illustration of slope failure, runout, and spatially distributed consequences across an exposed valley community.

Risk

Landslide Risk Assessment from Failure to Consequence

A subset-simulation and large-deformation framework propagates uncertain soil-strength parameters through slope failure and post-failure response, then links case-specific consequence measures to vulnerability, temporal exposure, and risk.

Conceptual illustration linking slope sensors and meteorological time series to recurrent-neural-network forecasts of pore-water pressure.

Monitoring

Pore-Water Pressure Forecasting with Recurrent Neural Networks

A field study compares MLP, RNN, LSTM, and GRU models for one-hour-ahead pore-water-pressure forecasting from monitored rainfall and PWP histories.

Conceptual illustration of a generative-AI evidence network synthesizing dispersed records of landslide consequences across regions.

Retrospective Analysis

Unveiling the Underappreciated Consequences of Landslides across the United States with Generative AI

An agentic generative-AI framework that extracts and synthesizes diverse web sources to capture the full range of direct and indirect landslide impacts.