Weiwei Zhan

Weiwei Zhan

Assistant Professor at University of Central Florida

ML-based Ground-Motion Modeling

less than 1 minute read

Ground-motion models (GMM) estimate the ground shaking intensity measures given a set of the earthquake source, path, and site variables, which provides the basis for any seismic risk assessment. In several ongoing works, I improved the accuracy of GMMs and discovered new site variables by integrating nonparametric machine learning and feature selection techniques. I observed that geospatial environmental data contributed to ground-motion modeling.


You May Also Enjoy

Accelerogram-based Real-time Liquefaction Hazard Detection

I developed a new accelerogram-based method and a worldwide database with paired liquefaction observations and ground motion records. The new method consisted of a logistic regression model and an automated signal processing procedure that extracts two physics- informed features for liquefaction detection from the raw accelerograms using different signal processing techniques. I also evaluated the performance and sensitivity of the accelerogram-based liquefaction detection method to different earthquake source, path, and site conditions.

Geomorphological Precursors for Landslide Early Detection

Early detection of rockslides at high-elevation and well-vegetated slopes remains challenging. This study used satellite and unmanned aerial vehicle (UAV) optical remote-sensing (ORS) images to track evidence of slope deformation and examine potential geomorphological precursors of five large rockslides in China. All the surveyed landslides had cracks or scarps and rockfalls within the landslide source area before the onset of rapid sliding. These precursors can be identified in ORS images taken several years or decades before the rapid slope failure, which provides sufficient time for the landslide early detection in practice. Local topography affects the spatial locations of cracks or scarps. Rockfalls within the landslide source area tend to locate at “key blocks” where slope mass provides forces resisting sliding. The rockfall area ratio, defined as the accumulated area of rockfalls over the landslide source area, ranged from 0.33 to 0.92 before rapid slope failure. The landslides developed on anti-dip and igneous rock slopes show a more significant rise of rockfall area ratio before the slope failure than the landslides on dip slopes. Given the broad availability of ORS data, this study could shed light on the ORS- based landslide early detection and landslide kinematics study.

Open-source Geospatial Liquefaction Database and Associated Tools

OpenLIQ is a geospatial database of global seismic liquefaction hazards. The OpenLIQ database currently includes ~290,000 samples (~63% of them are liquefaction points) and 19 geospatial predictors for 51 worldwide earthquakes. In addition, the OpenLIQ project includes a semi-automatic GIS workflow that can automatically conduct geospatial sampling of liquefaction and non- liquefaction points from liquefaction inventories (georeferenced liquefaction points and/or polygons), extract candidate geospatial predictors, and generate maps for quality check and visualization purposes. This workflow only requires limited manual inputs (e.g., incomplete or complete liquefaction inventory, and raster data for geospatial predictors). The GIS workflow will be used to process new liquefaction inventories, and the OpenLIQ database will be used to advance applications of different machine learning and uncertainty quantification methods in global geospatial liquefaction modeling.