Integrating Support Vector Regression and Kriging in Spatial Interpolation of Statistical Seismicity Parameters

Spatial interpolation methods, such as Inverse Distance Weighting (IDW) and kriging, are commonly used in various fields. In Kriging method, semivariogram fitting is an important step, where empirical data are used to derive a theoretical model. However, when the known theoretical semivariogram model does not provide a satisfactory fit, the bias in the estimated values is increased. To address this limitation, Support Vector Regression (SVR) can be used to model the empirical semivariogram with a machine-learning method. This method has been applied in ordinary kriging interpolation for semivariogram fitting to estimate parameters related to the potential occurrence of earthquake. Specifically, the calculated parameters, based on the Gutenberg-Richter law, include the seismic activity (a-value) and rock fragility (b-value) in the Sumatera region. The results showed that SVR can model the empirical semivariogram better than the theoretical. The integration of SVR-Ordinary Kriging provides the best performance compared to other methods, such as IDW, with the smallest RMSEP values for both the b-value and a-value measuring 0.1378 and 0.7423, respectively. Aceh and Mentawai Islands tend to show low a and b values, suggesting that these areas are more vulnerable to earthquake with large magnitudes.