Crust Macrofracturing as the Evidence of the Last Deglaciation

TL;DR

The study reexamines Finland's crustal structure by applying a uniform $k$-nearest neighbors approach to fuse three data types from receiver-function analyses: (A) LVSL presence, (B) $S$-velocity profiles, and (C) Moho depths. This data-driven framework yields a Moho-depth map and a three-lobed LVSL distribution, with the central LVSL plausibly arising from macrofracturing associated with the last deglaciation and post-glacial rebound. Hyperparameters are selected via cross-validation, and results align with previous seismic work while revealing new spatial patterns through a simple, interpretable method. The findings suggest that deglaciation-driven fracturing left a detectable imprint in the upper crust, with implications for regional deglaciation models and modern crustal rheology.

Abstract

Machine learning methods were applied to reconsider the results of several passive seismic experiments in Finland. We created datasets from different stages of the receiver function technique and processed them with one of basic machine learning algorithms. All the results were obtained uniformly with the $k$-nearest neighbors algorithm. The first result is the Moho depth map of the region. Another result is the delineation of the near-surface low $S$-wave velocity layer. There are three such areas in the Northern, Southern, and central parts of the region. The low $S$-wave velocity in the Northern and Southern areas can be linked to the geological structure. However, we attribute the central low $S$-wave velocity area to a large number of water-saturated cracks in the upper 1-5 km. Analysis of the structure of this area leads us to the conclusion that macrofracturing was caused by the last deglaciation.

Figures (8)

• Figure 1: A. Receiver function waveforms of two SVEKALAPKO stations, FH03 (thin line) and FA05 (thick line). FH03 waveform shows clear arrival of the converted phase near the first second after the $P$-wave arrival. This indicates the $S$-velocity jump immediately beneath the station. FA05 doesn’t show such a phase. The letters M$_1$ and M$_2$ pick to the Moho border phase. B. Corresponding 1D $S$-velocity profiles obtained through the receiver function inversion kozlovskaya2008. The Moho borders are pointed with the letters M$_1$ and M$_2$.
• Figure 2: The plot of the quality function (\ref{['eqn:hyperp']}) versus the number of nearest neighbors $K$ derived from the Moho boundary depth dataset.
• Figure 3: Depth to the Moho boundary. Colored triangles mark the positions of the input points. The color of the symbols corresponds to the source used: black — silvennoinen2014; white — kozlovskaya2008, green — dricker1996aleshin2006Aleshin2019. The abbreviation “CFGC” means Central Finnish Granitoid Complex.
• Figure 4: Depth to Moho along with the FD13-FJ01 profile. Black and red lines mark a cross-section of the Moho shape obtained with the kriging technique and the machine learning kNN method, correspondingly; red dashed lines show the kNN interpolation errors. The line oscillations at distances of 550-600 km are likely due to erratic data for station FG01.
• Figure 5: The plot of the ROC-AUC versus the number of nearest neighbors $K$ (left). The plot of ROC for the optimal value of $K=4$ (right).