Problem Statement¶
Landslides after atmospheric rivers, liquefaction after storms, catastrophic runoff after fires—the next generation of geodisaster risk emerges from interacting climate and solid-Earth processes. But our models cannot yet keep pace with these nonlinear cascades. We are building a platform that merges multimodal Earth observations in real-time , reduced-order physics models, and hazard assessment and forecast in a digtal twin framework.
Natural hazards pose significant risks to communities, infrastructure, and ecosystems worldwide. The GAIA HazLab platform addresses the critical need for advanced, data-driven approaches to hazard assessment and mitigation.
Earth System Science Nexus¶
We treat the critical zone, the shallow subsurface from weathered rock to the surface, as the dynamic skin of the Earth that regulate water infiltrating down to the water table, or water evaporating from soils or transpiration and modulating lower atmosphere thermo and convective dynamics. Understanding the hydromechanical and hydrological structure and dynamics of soils is central to the severity of geohazards and land-atmosphere coupling. We investigate and test hypothesis around the contributors to the severity of these geohazards (e.g., extreme meterological events vs soil conditions etc). Extreme weather events are modulated by atmosphere and oceanographic coupling, which defines the cascades of events from ocean -> atmosphere -> geohazards.
Use Case Events - Validations
2025 Western Washington Landslides and Floods (Nicoleta Cristea, Shuyi Chen, Richard Zhang, Brendan Jerns, Berkan Mertan, Akash Kharita, Marine Denolle, Brad Lipovsky, Nate Stevens, Alex Hutko)
2001 Nisqually Earthquake (Morgan Sanger, Yiyu Ni, Manuela Köpfli, Brett Maurer, Marine Denolle)
2025 Postfire Debris Flow at Stehekin (Abdullah Al Mehedi, Erkan Istanbulluoglu, Marine Denolle).
2021 Skagit River Floods (Nicoleta Cristea, +team)
Technological Development¶
1. Data Hub¶
Hazard data is often scattered across multiple sources and formats
Lack of standardized data formats makes integration difficult
Limited access to high-quality, up-to-date hazard datasets
2. Model Hub¶
Traditional hazard models may not capture complex spatial and temporal patterns
Need for machine learning approaches that can handle multi-scale hazard processes
Difficulty in validating models across different hazard types and geographic regions
3. Eval Hub¶
Develop a series of task and evaluation metrics for model evaluation
4. Research Software Agent¶
Developing a RSE agent to support multi-disciplinary science, in collaboration with eScience Instiute (Vani Mandava, lead at Scientific Software Engineering Center) and supported by the Paros Center.