Landslide Risk Assessment and Prediction

Problem Statement

Rainfall-induced landslides pose a persistent hazard to California’s mountainous terrain and critical infrastructure. After wildfires, destabilized slopes combined with intense rainfall events significantly increase the risk of debris flows and shallow landslides, threatening lives, transportation corridors, and downstream communities. With climate change intensifying storm events, prolonging droughts, and expanding wildfire frequency and severity, landslide susceptibility is expected to rise further. With traditional susceptibility models, we have gained important insights by leveraging static factors such as slope, geology, and land cover. By adding evolving conditions such as rainfall and vegetation variability, we aim to extend these approaches and provide fine-scale tools that better capture the spatiotemporal variability of landslide risk and support more effective mitigation planning.

Approach

This research develops a multi-stage, data-driven pipeline for landslide susceptibility mapping. We integrate high-resolution slope and geology data with multi-year time series of rainfall and vegetation greenness. Advanced statistical and machine learning methods — including spatiotemporal generative adversarial networks (GANs), graph neural networks, and ensemble classifiers — are used to capture both static susceptibility and dynamic triggering processes. The workflow leverages high-performance computing (HPC) resources to process billions of grid cells, enabling scalable statewide analysis. Outputs include susceptibility maps, fragility curves, and road-network vulnerability assessments that inform infrastructure resilience planning.

Ongoing Work

Current efforts focus on:
Data fusion and preprocessing: Cleaning, aligning, and integrating a high volume of slope, rainfall, NDVI data, and landslide inventory datasets into scalable formats.
Model development: Testing state-of-the-art susceptibility approaches
Dynamic risk modeling: Injecting rainfall and NDVI time series into landslide triggering models to move beyond static maps.
Infrastructure resilience: Coupling landslide hazard outputs with road-network models to evaluate vulnerability and prioritize retrofitting.
Interactive tools: Prototyping a web-based interactive landslide map to make findings accessible to stakeholders and agencies.

Relevant Publications

Under Preparation

Students

Melis Fidansoy