Mosaicking Andean morphostructure and seismic cycle crustal deformation patterns using GNSS velocities and machine learning

Abstract

Read online

We use unsupervised machine learning techniques to analyze continental-scale crustal motions in areas affected by the seismic cycle of large subduction earthquakes along the Chilean Trench. Specifically, we use the agglomerative clustering algorithm as an exploratory tool to investigate spatial patterns in GNSS regional velocities without the complexity of modeling a physical source. We present a continental-scale velocity field including all available GNSS data for two-time windows (pre-2014, 2018–2021) that represents two periods with different deformation patterns of the seismic cycle. We test two different pre-processing methodologies for the design of machine learning features from the GNSS-derived velocities. The first method uses the direction and magnitude of the secular rates as input features to the clustering algorithm. These results show a clustering spatially related to seismic cycle deformation, separating latitudinal segments with different velocities in the fore-arc and back-arc, as well as regions affected by postseismic relaxation. Thus, highlighting the effectiveness of this method for mapping first-order patterns of active deformation in a subduction zone, that are particularly related to variations on interplate coupling and postseismic transient deformation. In a more sophisticated approach, we use surface strain and rotational rates from GNSS velocities as features in the second methodology. Here, we develop a novel methodology to estimate strain and rotation rates accounting for the spatial heterogeneity of the GNSS-network. We determine the spatial scale at which these features are estimated by least squares inversions, by using a Bayesian model class selection method. The distribution of stations allows to identify heterogeneities in strain and rotation rates at spatial scales larger than 50 km, being particularly notorious the main features of regional deformation at scales > 100 km. Interestingly, the results show a spatial correlation between seismic segmentation in the fore-arc and geologic and structural domains in the arc and back-arc. Our results demonstrate the ability of the combination of inverse and machine learning methods to efficiently identify active deformation patterns and their relationship to the subduction seismic cycle and regional-scale geological structures. Furthermore, our analysis suggests that Andean geological structures influence the observed deformation field.

Keywords

• clustering
• machine learning (ML)
• GNSS crustal velocities
• continental strain
• seismic cycle deformation
• Andean segmentation