Rapid response maps and loss estimates that are used immediately after an earthquake to assess intensity and potential impact do not currently include effects from liquefaction hazard. Thus, there is a critical need to develop new methods of estimating the likelihood of liquefaction that can be rapidly and broadly derived from both earthquake-specific intensity estimates and simple geospatial features. A fundamental limitation of prior probabilistic liquefaction models is that the liquefaction datasets contain few non-liquefaction sites (a sampling bias). A second challenge to including liquefaction effects in loss estimation and rapid response maps is that most liquefaction models rely on site- and region-specific datasets (e.g. surficial geology maps) that are time and cost intensive to collect. The preliminary results for this project demonstrate how these problems have been solved by developing logistic regression models from representative datasets of liquefaction as a function of key input parameters that can easily be estimated from global datasets (e.g. digital elevation models or DEM) and standard earthquake-specific intensity data (e.g. peak ground acceleration). Candidate explanatory variables include those derived from the DEM as well as indexes for soil saturation, vegetation, climate, and hydrology. In preliminary work, a logistic regression model was developed using data from two regions (Kobe, Japan and Christchurch, New Zealand) which predicts probabilities of liquefaction based on peak ground acceleration, elevation, distance to coast, and a hydrologic parameter - compound topographic index - which is used as a proxy for soil saturation. The model has been tested in Port-au-Prince, Haiti and provides a consistent estimate of liquefaction probability. This demonstration shows that the proposed new method of estimating the probability of liquefaction can be rapidly and broadly derived from both earthquake-specific intensity estimates and simple geospatial features. However, in order to develop a geospatial liquefaction model that will be globally applicable, the database needs to be extended to more geologic and climatic environments so that the model can constrain regional variations in the geospatial proxies. In this project, a geospatial liquefaction database will be developed from global earthquakes, where the explanatory variables will be broadly available geospatial data. Sampling bias will be addressed by developing the database from observations that are representative of the true distribution of liquefaction. This marks a shift in liquefaction potential model development, which to date has focused on case history databases that are biased toward observations of liquefaction occurrence. The goals of the project are to: 1) Develop a global database of liquefaction observations with geospatial variables. 2) Test first-order proxies for saturation and soil density 3) Normalize proxies for different geomorphic and climatic regions 4) Develop a probabilistic geospatial liquefaction model 5) Train undergraduate civil engineers in seismic hazard and loss estimation
The broader impact and potentially transformative aspect of the proposed work is the global applicability of the model which will enable liquefaction effects to be included in future rapid response maps, loss estimates, and scenario simulations for any future event anywhere in the world and, therefore improve disaster response and reduce loss. In addition, undergraduate research and outreach within the geographic information systems class at Tufts will be used to introduce civil engineering undergraduate students to the importance of seismic hazard and loss estimation for earthquakes.
