David O.S. Oakley’s research while affiliated with Pennsylvania State University and other places

The increasing availability of large geological datasets and modern methods of data analysis facilitate a data science approach to geology in which inferences are drawn from geological data using automated methods based on statistics and machine learning. Such methods offer the potential for faster and less subjective interpretations of geological data than are possible from a human interpreter, but translating the understanding of a trained geologist to an algorithm is not straightforward. In this paper, we present automated workflows for detecting geological folds from map data using both unsupervised and supervised machine learning. For the unsupervised case, we use regular expression matching to identify map patterns suggestive of folds along lines crossing the map. We then use the HDBSCAN clustering algorithm to cluster these possible fold identifications into a smaller number of distinct folds. This clustering algorithm is chosen because it does not require the number of clusters to be known a priori. For the supervised learning case, we use synthetic models of folds to train a convolutional neural network to identify folds using map and topographic data. We test both methods on synthetic and real datasets, where they both prove capable of identifying folds. We also find that distinguishing folds from similar map patterns produced by topography is a major issue that must be accounted for with both methods. The unsupervised method has advantages, including the explainability of its results, and provides clearly better results in one of the two real-world test datasets, while the supervised learning method is more fully automated and likely more easily extensible to other structures. Both methods demonstrate the ability of machine learning to interpret folds on geological maps and have potential for further development targeting a wider range of structures and datasets.

The increasing availability of large geological datasets together with modern methods of data analysis facilitate a data science approach to geology in which inferences are drawn from geological data using automated methods based on statistics and machine learning. Such methods offer the potential for faster and less subjective interpretations of geological data than are possible from a human interpreter, but translating the understanding of a trained geologist to an algorithm is not straightforward. In this paper, we present automated workflows for detecting geological folds from map data using both unsupervised and supervised machine learning. For the unsupervised case, we use regular expression matching to identify map patterns suggestive of folds along lines crossing the map. We then use the hdbscan clustering algorithm to cluster these possible fold identifications into a smaller number of distinct folds, the number of which is not known a priori. For the supervised learning case, we use synthetic models of folds to train a convolutional neural network to identify folds using map and topographic data. We test both methods on synthetic and real datasets.

The direct carbonate procedure for accelerator mass spectrometry radiocarbon (AMS ¹⁴ C) dating of submilligram samples of biogenic carbonate without graphitization is becoming widely used in a variety of studies. We compare the results of 153 paired direct carbonate and standard graphite ¹⁴ C determinations on single specimens of an assortment of biogenic carbonates. A reduced major axis regression shows a strong relationship between direct carbonate and graphite percent Modern Carbon (pMC) values (m = 0.996; 95% CI [0.991–1.001]). An analysis of differences and a 95% confidence interval on pMC values reveals that there is no significant difference between direct carbonate and graphite pMC values for 76% of analyzed specimens, although variation in direct carbonate pMC is underestimated. The difference between the two methods is typically within 2 pMC, with 61% of direct carbonate pMC measurements being higher than their paired graphite counterpart. Of the 36 specimens that did yield significant differences, all but three missed the 95% significance threshold by 1.2 pMC or less. These results show that direct carbonate ¹⁴ C dating of biogenic carbonates is a cost-effective and efficient complement to standard graphite ¹⁴ C dating.

K‐Ar ages of clay‐sized mineral grains are used to determine the timing of activity on fossil seismogenic faults within the Cretaceous‐Paleogene Shimanto accretionary complex, southwest Japan. Samples were collected from three regional faults that separate hanging wall coherent rocks from footwall subduction mélange: the Goshikinohama Fault that caps the Yokonami mélange, the roof thrust of the Okitsu mélange, and the Nobeoka Thrust that caps the Hyuga mélange. The K‐Ar ages of fault rocks decrease with decreasing 2M1 illite polytype component, indicating a mixture of 1Md and 2M1 illite polytypes. Based on illite dating analysis, the extrapolated ages of the pure 2M1 illite polytype from the Goshikinohama Fault, the roof thrust of the Okitsu mélange, and the Nobeoka Thrust are 79.3 ± 5.0, 66.1 ± 8.1, and 46.7 ± 8.2 Ma, respectively, similar to the depositional age of each host rock. Lower intercepts of regression lines, which correspond to samples containing 100% authigenic illite, are calculated as 50.7 ± 1.4, 18.4 ± 1.2, and 24.4 ± 1.4 Ma, respectively. These ages are significantly younger than both the depositional ages and the timing of accretion. These results indicate that authigenic illite associated with fault slip is not related to underthrusting along the subduction interface but rather formed during out‐of‐sequence thrusting in the upper plate. Early Miocene slip along faults of the northern Shimanto belt is coincident with major tectonic events along the convergent margin, including collision with elements of the Izu‐Bonin volcanic arc‐backarc system, and opening of the Japan Sea.

Despite the implementation of several kinematic and mechanical models for fault-related folding, and the incorporation of structural uncertainty in geological models, structural modeling is still too deterministic. The current focus is on forward modeling of a unique fit. We show the application of trishear inverse modeling, global optimization and Markov chain Monte Carlo (MCMC) methods, to a clay model of basement-involved compressional faulting, from which we know both total and incremental deformation. Global optimization and MCMC methods provide a range of possible models rather than a unique fit. Total inversions give an average of the model's deformation, while incremental inversions are more physically related to the model's evolution. Global optimization identifies the full range of possible models, while MCMC characterizes expected parameter values and their uncertainties. Structural inversions without sound geology are meaningless. A robust stratigraphy, geomorphic markers, mesoscopic structures, analogue and mechanical modeling can all greatly improve the inversions.

Marine terraces on growing fault-propagation folds provide valuable insight into the relationship between fold kinematics and uplift rates, providing a means to distinguish among otherwise non-unique kinematic model solutions. Here, we investigate this relationship at two locations in North Canterbury, New Zealand: the Kate anticline and Haumuri Bluff, at the northern end of the Hawkswood anticline. At both locations, we calculate uplift rates of previously dated marine terraces, using DGPS surveys to estimate terrace inner edge elevations. We then use Markov chain Monte Carlo methods to fit fault-propagation fold kinematic models to structural geologic data, and we incorporate marine terrace uplift into the models as an additional constraint. At Haumuri Bluff, we find that marine terraces, when restored to originally horizontal surfaces, can help to eliminate certain trishear models that would fit the geologic data alone. At Kate anticline, we compare uplift rates at different structural positions and find that the spatial pattern of uplift rates is more consistent with trishear than with a parallel-fault propagation fold kink-band model. Finally, we use our model results to compute new estimates for fault slip rates (~ 1–2 m/ka at Kate anticline and 1–4 m/ka at Haumuri Bluff) and ages of the folds (~ 1 Ma), which are consistent with previous estimates for the onset of folding in this region. These results are consistent with previous work on the age of onset of folding in this region, provide revised estimates of fault slip rates necessary to understand the seismic hazard posed by these faults, and demonstrate the value of incorporating marine terraces in inverse fold kinematic models as a means to distinguish among non-unique solutions.

The kinematics of deformation in the overriding plate of convergent margins may vary across timescales ranging from a single seismic cycle to many millions of years. In Northeast Japan, a network of active faults has accommodated contraction across the arc since the Pliocene, but several faults located along the inner forearc experienced extensional aftershocks following the 2011 Tohoku-oki earthquake, opposite that predicted from the geologic record. This observation suggests that forearc faults may be favorable for stress triggering and slip inversion, but the geometry and deformation history of these fault systems are poorly constrained. Here we document the Neogene kinematics and subsurface geometry of three prominent forearc faults in Tohoku, Japan. Geologic mapping and dating of growth strata provide evidence for a 5.6-2.2 Ma initiation of Plio-Quaternary contraction along the Oritsume, Noheji and Futaba faults, and an earlier phase of Miocene extension from 25 to 15 Ma along the Oritsume and Futaba faults associated with the opening of the Sea of Japan. Kinematic modeling indicates these faults have listric geometries, with ramps that dip ~40-65° W and sole into subhorizontal detachments at 6-10 km depth. These fault systems can experience both normal and thrust sense slip if they are mechanically weak relative to the surrounding crust. We suggest that the inversion history of Northeast Japan primed the forearc with a network of weak faults mechanically and geometrically favorable for slip inversion over geologic timescales, and in response to secular variations in stress state associated with the megathrust seismic cycle.