A new Graph Neural Network (GNN) based model for the evaluation of lateral spreading displacement in New Zealand

The increased availability of high quality data from post disaster field reconnaissance, enabled the use of deep learning algorithms in the field of geotechnical earthquake engineering. The 2010-2011 Canterbury earthquake sequence in New Zealand caused significant damage due to abundant manifestation of liquefaction induced lateral spreading. The data available from this sequence is an ideal case study for deep learning analyses due to the amount and quality of information available through the New Zealand Geotechnical Database (NZGD). A dataset of about 7500 datapoints was collected and organized by the authors to develop a new Graph Neural Network (GNN) algorithm for lateral spreading in the Canterbury area. The comparison between predicted and observed data is performed using feed forward Neural Network. Several GNN models with different hyperparameters are explored and the best model is presented in this paper, and Explainable Artificial Intelligence is applied to the model that provides the best performance. These computationally expensive analyses were carried out utilizing cloud based computing capabilities offered by the Texas Advanced Computing Center (TACC) available to the natural hazard community through the cyberinfrastructure DesignSafe.