Statistical and Machine Learning Methods for Magnitude Scale Conversion and Gutenberg-Richter Analysis: Insights from Regional Seismicity of Morrocco

Abstract

This investigation presents a thorough examination of earthquake magnitude scales and seismicity parameters for Morocco using statistical and machine learning methodologies. This study developed robust magnitude conversion models using Multiple Linear Regression, Huber, and Elastic Net regression techniques, attaining R2 values of 0.691, 0.689, and 0.670, respectively, with residuals conforming to Gaussian distributions. The Gutenberg-Richter parameters were estimated using two complementary approaches: the Shi-Bolt method, yielding Mc = 4.2, b-value = 0.75, and a-value = 5.34, and an alternative approach similar to Mc95-Mc90-maximum curvature, resulting in Mc = 4.25, b-value = 0.80, and a-value = 5.58. The observed b-values (0.75-0.80), which fall below the global average of 1.0, indicate an elevated likelihood of higher-magnitude seismic events in the region. This aligns with Morocco's intricate tectonic setting at the Africa-Eurasia Plate boundary. The magnitude completeness threshold of approximately 4.2 underscores the necessity for enhanced seismic monitoring, given the region’s seismic hazard potential. These findings establish a robust framework for seismic hazard assessment in Morocco and may inform analogous analyses in regions with similar tectonic configurations. The integration of multiple analytical approaches offer novel insights into the reliability of magnitude conversion.

Keywords: Magnitude Conversion, Gutenberg-Richter Parameters, Machine Learning, Seismic Hazard, B-Value, Magnitude of Completeness