11:00 AM - 1:00 PM
[MGI34-P02] Regression analysis and variable selection that relate stagnation/penetration of subducting plates with subduction-zone parameters
Keywords:Subduction zones, Plate tectonics, Stagnant slab, Penetrating slab, Regression analysis, Exhaustive model selection
1. Introduction
It is widely accepted that the behavior of subducting plates has significant effects on geophysical and geochemical processes, including volcanism and material transportation into the deep mantle. Seismic observations have revealed that subducting plates behave in different ways at the depth of 660 km: some stagnate and extend horizontally as beneath Northeast Japan, whereas others penetrate into the lower mantle as beneath the Mariana Islands (e.g., Fukao & Obayashi, 2013). According to numerical experiments, multiple factors related to subduction tectonics control the behavior and the shape of subducting plates at the 660-km depth (e.g., Stegman et al., 2006; Nakao et al., 2016) and are still controversial. In this study, we performed regression analyses and exhaustive model selections to determine multiple subduction-zone parameters that control the behavior of subducting plates at the 660-km depth, using seismic, geodetic, and geological data observed along natural subduction zones.
2. Data & Methods
We referred to subduction-zone parameters recorded at 169 locations along global subduction zones from database SubMap 4.3 (e.g., Heuret & Lallemand, 2005). The data set includes the age of oceanic plates, seafloor roughness, plate velocities, the stress of the upper plate, the thickness of trench sediments, and so on. Logistic regressions were applied using the subduction-zone parameters as potential candidates for the explanatory variables of the behavior of subducted plates (i.e., probability of penetration). Among all possible combinations of the variables, we determined the three best models based on LOOCV (leave-one-out cross-validation), AIC (Akaike’s information criterion), and BIC (Bayesian information criterion).
3. Results & Discussion
In each best model, trench velocity, trench length, oceanic plate age, and trench sediment thickness were selected as the explanatory variables for the plate behavior at the 660-km depth. Our result is consistent with previous numerical studies that demonstrated that a retreating trench, a narrow oceanic plate, and an old oceanic plate contribute to roll-back and stagnation (e.g., Stegman et al., 2006; Nakao et al., 2016; Fuji et al., 2021). These results suggest that a so-called stagnant slab coincides with the plate roll-back associated with active toroidal flows driven by a heavy and/or narrow subducting plate. Meanwhile, the effects of the trench sediment on the subducting plate behavior, which our models suggest, is enigmatic and expected to be solved in the future.
It is widely accepted that the behavior of subducting plates has significant effects on geophysical and geochemical processes, including volcanism and material transportation into the deep mantle. Seismic observations have revealed that subducting plates behave in different ways at the depth of 660 km: some stagnate and extend horizontally as beneath Northeast Japan, whereas others penetrate into the lower mantle as beneath the Mariana Islands (e.g., Fukao & Obayashi, 2013). According to numerical experiments, multiple factors related to subduction tectonics control the behavior and the shape of subducting plates at the 660-km depth (e.g., Stegman et al., 2006; Nakao et al., 2016) and are still controversial. In this study, we performed regression analyses and exhaustive model selections to determine multiple subduction-zone parameters that control the behavior of subducting plates at the 660-km depth, using seismic, geodetic, and geological data observed along natural subduction zones.
2. Data & Methods
We referred to subduction-zone parameters recorded at 169 locations along global subduction zones from database SubMap 4.3 (e.g., Heuret & Lallemand, 2005). The data set includes the age of oceanic plates, seafloor roughness, plate velocities, the stress of the upper plate, the thickness of trench sediments, and so on. Logistic regressions were applied using the subduction-zone parameters as potential candidates for the explanatory variables of the behavior of subducted plates (i.e., probability of penetration). Among all possible combinations of the variables, we determined the three best models based on LOOCV (leave-one-out cross-validation), AIC (Akaike’s information criterion), and BIC (Bayesian information criterion).
3. Results & Discussion
In each best model, trench velocity, trench length, oceanic plate age, and trench sediment thickness were selected as the explanatory variables for the plate behavior at the 660-km depth. Our result is consistent with previous numerical studies that demonstrated that a retreating trench, a narrow oceanic plate, and an old oceanic plate contribute to roll-back and stagnation (e.g., Stegman et al., 2006; Nakao et al., 2016; Fuji et al., 2021). These results suggest that a so-called stagnant slab coincides with the plate roll-back associated with active toroidal flows driven by a heavy and/or narrow subducting plate. Meanwhile, the effects of the trench sediment on the subducting plate behavior, which our models suggest, is enigmatic and expected to be solved in the future.