Finding the buried memory of past earthquakes with geophysical, GPR-based paleoseismology

We hypothesized that, in places where sedimentation and erosion compete at fast rates, part of the memory of past earthquakes on faults may be buried, hence hidden, in the first tens meters of the ground. We test that hypothesis on a fast slipping, large, strike-slip fault (Hope, New Zealand), at a site where marked alluvial conditions prevail (Terako). We first use LiDAR data to analyze the ground surface morphology of the 2 km2 site at the greatest resolution. About twenty clear, distinct, morphological markers are observed -mainly alluvial terrace risers and small stream channels, all are laterally offset by the fault. The measured offsets range between 3 and 200 m, yet are discrete and showing several large slip gaps. The measurements are well-constrained and allow estimating the mean slip per event amplitude to 3.9 ± 1.4 m, and the last earthquake slip to 3 ± 0.5 m. About 10 past earthquakes are well documented in the surface data, while about 50 are requested to account for the 200 m largest cumulative slip. We then investigate the zone (on smaller area, 400 x 600 m2) with dense, pseudo-3D Ground Penetrating Radar (GPR) data. We measured 56, ~400 m-long, 5-10 m spaced GPR profiles (250 MHz), parallel to the fault and evenly distributed on either sides. Their analysis reveals the existence of several tens morphological markers buried in the first 3 m of the ground, most of them do not imprint the ground surface as they are blanketed with a 0.1-3 m-thick poorly reflective layer. A few buried markers exhibit however surface expressions. All buried markers are laterally offset by the fault. Based on a number of evidence, we interpret these buried markers as stream channels, most were decapitated by the repeated fault slips and abandoned. We measured ~50 lateral offsets in the buried channel network, almost three times more than at the surface. These offsets range between 2.5 and 106 m, as observed at the surface, yet provide a more continuous record of the fault slip. The similarity of the successive slip increments suggests a slip per event averaging 3.9 ± 1.9 m, similar to that estimated from surface data. From the total 'surface and buried' offset collection, we infer that a minimum of 30 large earthquakes have broken the Hope fault at the Terako site in the last 5 kyrs, with an average slip per event of 3.8 ± 1.3 m, an average recurrence time of 100-250 yrs, and a likely magnitude of at least Mw 7.2-7.7. The last major earthquake likely occurred at 1875 ± 15 AD, in agreement with previous suggestions. Our study therefore confirms that part of the memory of past earthquakes may indeed reside in the first tens meters of the ground, where it may be explored with a novel type, geophysical and GPR-based, paleoseismology. We emphasize that developing such a new paleoseismology will provide a rich information complementary to surface observation, and help documenting the past earthquakes on faults.