Applied Computing and Geosciences (Feb 2025)
Developing ground motion prediction models for West Java: A machine learning approach to support Indonesia's earthquake early warning system
Abstract
Indonesia, one of the most earthquake-prone countries in the world, is currently developing an Earthquake Early Warning (EEW) system. A key component of this system, the Regional EEW, relies on Ground Motion Prediction models (GMPMs) to issue end-user alerts. However, in West Java, one of the pilot regions for this project, there is a lack of region-specific GMPMs essential for accurate early warnings. Traditionally, GMPMs are developed using linear regression based on complex, predefined mathematical equations and coefficients. However, Machine learning offers the advantages of bypassing the need for predefined equations and effectively capturing the nonlinear behavior present in ground motion data. To address this gap, we evaluated three machine learning algorithms (i.e. Artificial Neural Network [ANN], Gradient Boosting [GB], and Random Forest [RF]) to develop GMPMs for three tectonic categories: shallow-crustal, interface, and intraslab. These models were used to predict Peak Ground Acceleration (PGA) in West Java, utilizing 3116 strong ground motion records from 365 earthquakes with moment magnitude ranging from 2.4 to 7 and epicentral distance between 5.5 and 867 km, recorded since 2010. Our results show that The Gradient Boosting model outperformed the others across all three tectonic categories, with the lowest Mean Squared Error values (0.94, 0.60, 0.65), and Standard Deviation of Residuals (0.97, 0.77, 0.80), as well as the highest Pearson correlation coefficient-value (0.83, 0.88, 0.90) for shallow-crustal, interface, and intraslab events, respectively, demonstrating strong accuracy in predicting PGA. The model was further validated with recent earthquake data and from 2024 showing good agreement and confirming its robustness. Epicentral Distance and Moment Magnitude were the most influential in predicting PGA among the six explanatory variables used in this study. These findings highlight the potential of machine learning models to improve the accuracy of ground-shaking predictions, contributing to the success of Indonesia's Earthquake Early Warning System (EEWS).
