Coseismic river avulsion on surface rupturing faults: Assessing earthquake-induced flood hazard

Abstract

Surface-rupturing earthquakes can produce fault displacements that abruptly alter the established course of rivers. Several notable examples of fault rupture-induced river avulsions (FIRAs) have been documented, yet the factors influencing these phenomena have not been examined in detail. Here, we use a recent case study from New Zealand's 2016 Kaikōura earthquake to model the coseismic avulsion of a major braided river subjected to ~7-m vertical and ~4-m horizontal offset. We demonstrate that the salient characteristics of the avulsion can be reproduced with high accuracy by running a simple two-dimensional hydrodynamic model on synthetic (pre-earthquake) and "real" (post-earthquake) deformed lidar datasets. With adequate hydraulic inputs, deterministic and probabilistic hazard models can be precompiled for fault-river intersections to improve multihazard planning. Flood hazard models that ignore present and potential future fault deformation may underestimate the extent, frequency, and severity of inundation following large earthquakes.

Fig. 1.. FIRA scenario block diagrams.

(A) Visualization of a complex FIRA scenario, where an oblique strike-slip fault truncates multiple bends of a meandering river system. White arrows indicate flow direction, green stars demarcate the main avulsion nodes, and dotted white lines outline the edges of the submerged river channel. An initial avulsion node is created within the western meander bend, where a FSB obstructs flow from entering the parent channel located on the down-stream side of the bend. Water pools against the fault scarp until the existing channel (dotted lines) overtops, resulting in an avulsion. If the fault truncates multiple bends, then multiple avulsion nodes may be created at each displaced bend (green stars). An avulsion analogous to this example is the 2010 Greendale fault rupture in Canterbury, New Zealand, where a single bend of the Hororata River and an active paleochannel of the neighboring Selwyn River were laterally and vertically offset, causing avulsion along the Greendale fault scarp (9, 12). (B) Visualization of how a dip-slip FSB may influence river flow behavior. White arrows indicate flow direction and orange arrows demarcate relative displacements on the fault.

Fig. 2.. Tectonic Setting of the Papatea fault rupture.

(A) New Zealand tectonic setting. Red lines represent fault ruptures in the 2016 Kaikōura earthquake, with a green star marking the Papatea fault location (B) The Waiau Toa/Clarence River valley setting following the 2016 Papatea fault rupture. Aerial imagery taken 1 day following the fault rupture was used to constrain the general avulsion extent. The main geomorphic markers (i.e., Glen Alton Bridge and notable topographic or floodplain features) and fault displacement vectors are labeled, as per observations and data collected in the field (14). (C) Aerial imagery taken 1 day following the Papatea fault rupture showing the extent of the FIRA event. Image taken and provided by Environment Canterbury.

Fig. 3.. HEC-RAS benchmark and calibration model accuracy assessment.

(A) Benchmark HEC-RAS model results using lidar acquired within 2 months of the 2016 Papatea fault rupture. The flow regime is run as a forced steady-state plan over a period of 12 hours using event flow conditions of 187 m³ s⁻¹. (B) Accuracy assessment of the benchmark model when applying a contingency table confusion matrix approach. True-positive (TP) and true-negative (TN) values reflect areas where the model and classified aerial imagery agree that cells are wet (true positive) or dry (true negative) to provide an initial estimate of model accuracy, while false-positive (FP) and false-negative (FN) values reflect areas where the model results and classified aerial imagery do not align. The floodplain boundary is demarcated on the basis of the river valley geomorphology and generates many values in the “true-negative” category, creating a (floodplain) bias toward higher accuracy scores within the model. The true floodplain extent is inherently uncertain; therefore, revised accuracy scores address uncertainty within the model using the F 〈1〉 equation (see Materials and Methods), which equitably address both underprediction or overprediction within the model (59). (C) Calibration model results when an identical flow regime is applied to a pre-event (2012) DEM modified with a synthetic fault scarp reflective of Papatea fault displacement. (D) Contingency table accuracy assessment results for the calibration model.

Fig. 4.. HEC-RAS scenario modeling categories.

Twenty-five scenarios were modeled by combining five displacement regimes and five flow regimes. Results of the scenario models can be broadly partitioned into four main categories. (A; white) Little change in river behavior or morphology other than negligible water depth increases near the fault. (B; light gray) Mild to moderate changes in river behavior, with water depth values increasing near FSB localities as water begins to pool against the scarp and form a local backwater. Local antecedent anabranching braid channels are activated as flow values increase; however, no avulsion of flow occurs outside of the established braid plain. (C; dark gray) A partial avulsion of flow occurs, wherein some flow escapes from the main braid plain to flow along the fault scarp. Flow obstruction near FSB locations drives water depth increases as an extensive backwater forms upstream of the fault. (D; black) River flow is fully obstructed by the fault scarp, causing a full avulsion, accompanied by large water depth increases upstream of the FSB.

Fig. 5.. Waiau Toa|Clarence River HEC-RAS scenario model results.

Twenty-five Waiau Toa/Clarence River FIRA scenario models were produced in HEC-RAS using a pre-event (2012) DEM modified to reflect five different displacement values. Hanging-wall vertical uplift and lateral translation is determined by the vertical-to-horizontal (V:H) ratios listed at the top of each column. The five displaced DEM’s were combined with five individual flow regimes (found to the left of each row) to assess the interactions between variable offset scenarios and flow rates. Flow values were calculated using Glen Alton Bridge gauge data provided by Environment Canterbury, providing flow rate measurements from early 2014 to 1 hour before the 2016 Kaikōura earthquake (fig. S4). The exception is the mean annual flood (MAF) value, which was instead calculated during previous flood modeling work within the Waiau Toa/Clarence River valley (35). Results are partitioned into four main categories (white, gray, and black colored frames; see Fig. 4 for full descriptions).

Fig. 6.. Relative density of global fault-river intersections.

Contours reflect the density of points where active faults from the GEM Global Active Faults Database [black lines; (37)] intersect rivers (36) within a tolerance of 30 m. While not all fault-river intersections are capable of producing FIRA, the map shows that they are ubiquitous across plate boundaries and many stable continental regions. Intersections are particularly abundant in convergent plate settings, where rivers facilitate the transport of eroded material from actively uplifting, and in many cases, fault-bounded mountains. In New Zealand alone (inset), there are >2000 fault-river intersections (orange dots).

Fig. 7.. Plot comparing the potential for overbank flow (F) versus discharge for the 25 Waiau Toa | Clarence river scenario models.

In general, the likelihood of avulsion and parent channel abandonment rises with increasing F (Eq. 1). In reaches of equal slope, F may be a good indicator of relative FIRA hazard. Partial and full avulsions are associated with F > 2 in the Waiau Toa/Clarence River reach. Higher discharges seemingly sustain some flow in the pre-quake parent channel at increasing values of F, elevating the threshold at which a full avulsion occurs. This effect may be related to the backwater surface elevation increasing relative to flow depth in these scenarios, thus raising the water surface elevation to a point where flow can access the displaced pre-event parent channel (5).