An exploration of the use of machine learning to predict lateral spreading

The recent availability of large amounts of high-quality data from post-disaster field reconnaissance enables an exploration of the use of machine learning (ML) approaches to predict earthquake-induced damage. The 2011 Christchurch earthquake in New Zealand caused widespread liquefaction and lateral spreading, and the development of ML models to predict the lateral spreading was enabled by the availability of high-resolution data for lateral spreading displacements, ground shaking, and surface and subsurface features. A dataset of more than 7300 lateral spread observations from a single event in a single geologic setting were used to develop ML classification models using the Random Forest approach for the binary classification problem to identify lateral spread occurrence and a multiclass classification problem to predict the amount of displacement. The best ML models developed in this study accurately predict the lateral spread patterns with an overall accuracy of 80% for the lateral spread occurrence models and 70% for the multiclass displacement classification models. These models show that peak ground acceleration, distance to the river, ground elevation, and groundwater table contribute most to the accuracy of the lateral spread predictions for this dataset, and the inclusion of cone penetration test (CPT) features improves only the prediction of the largest displacement class (>1.0 m). Further research is needed to develop ML models that are generalizable to other earthquake events and geologic settings.

Automated summary

The study utilized machine learning models to accurately predict earthquake-induced lateral spreading using high-quality post-disaster data. It was found that factors such as peak ground acceleration, distance to the river, ground elevation, and groundwater table play crucial roles in accurate predictions of lateral spreading.