Elsevier

Engineering Geology

Volume 259, 4 September 2019, 105169
Engineering Geology

Investigating large landslides along a river valley using combined physical, statistical, and hydrologic modeling

https://doi.org/10.1016/j.enggeo.2019.105169Get rights and content

Highlights

  • A susceptibility map and trigger analysis were created for an unstable river valley.

  • Flooding caused by snowmelt and rain show correlation with landslide movement.

  • Combining physical and empirical methods improved accuracy of susceptibility map.

  • The full workflow is a powerful compliment to practical engineering investigations.

Abstract

Combined landslide susceptibility mapping and temporal early warning of failures can be a powerful method for mitigation and timely evacuation, but modeling must be well informed by the specific failure types and triggers unique to each climate and geomorphology. This paper describes the development of a landslide susceptibility map and threshold for riverbank erosion-triggered landslides in a northern climate with atypical landslide conditions. Located on the southern shore of Lake Superior, the Ontonagon River basin in northern Michigan receives an average of 4.8 m of snowfall annually, followed in the spring by a sharp warming trend and rain. Undercutting of the steep riverbanks causes large failures that continuously threaten bridges and a nearby hydroelectric facility. In this investigation, a landslide inventory was mapped using aerial imagery from 1992 to 2016. Landslide triggering factors were interpreted using temperature, cumulative precipitation, and river discharge data, demonstrating that river discharge is the primary predictor of landslides despite the source being either rainfall or snowmelt. A preliminary threshold was then created to determine the discharge characteristics likely to cause failures. A susceptibility map was created for the river system using a combination of Scoops3D with logistic regression, improving overall accuracy to 93%. Furthermore, Scoops3D proved valuable in constraining the model to failures of engineering significance (large volume and impact) and kinematic possibility. The threshold-susceptibility scheme is thus a powerful tool for assessing comprehensive slope stability along river channels.

Introduction

Landslides are one of the most common and costly natural hazards causing distress to many communities around the world. While it is widely known that a root cause of landslides is rapid rainfall events, landslides are also triggered by a multitude of unique and in some cases unusual circumstances, for which rainfall may only be an imperfect proxy (Bogaard and Greco, 2018). An example of this is the complex relationship between rainfall and long-term soil pore pressure (Kuriakose et al., 2008). Another is the effect of snowfall and snowmelt on landslide development and initiation (Martelloni et al., 2013; Okamoto et al., 2018). A particular landslide typology of interest is landslides occurring along steep river channels. In spatial susceptibility models, slope hydrology and river factors are commonly used as predictive factors (Regmi et al., 2010). Using a temporal analysis, Reichenbach et al. (1998) utilized river discharge as a proxy for rainfall to create early warning thresholds for floods and landslides in Italy. Successful temporal and spatial landslide management, however, depends on an understanding of climate factors and incorporating the knowledge of the landslide typology into the modeling process. In addition, it is increasingly important to acknowledge and plan for dynamic climate conditions in the face of global climate change (Barik et al., 2017).

Since research in rainfall thresholds and susceptibility mapping have developed independently (see Reichenbach et al., 2018 and Segoni et al., 2018a for recent reviews), an opportunity exists to quantitatively combine these approaches into a temporal-spatial warning system that then benefits from using multiple approaches. Temporal warning consists mainly of statistical thresholds or hydrodynamic modeling, combined with in-situ measurement (Intrieri et al., 2012), while spatial mapping is accomplished with traditional statistical methods, machine learning, or physical models. Recently, several systems have been built that effectively combine temporal and spatial information using various combinations of methods (Baum and Godt, 2010; de Souza and Ebecken, 2012; Naidu et al., 2017; Segoni et al., 2018b). Choosing from these methods depends on data availability, the types of landslide hazards encountered, their sizes and distribution, and the region's climate and geology. Our study region is of interest because there is little a priori knowledge of the triggering variables and the spatial extent of the hazard from which to assume appropriate methods. Primary field data and observations must therefore be used to make decisions regarding how to model the hazard. We find that when snow, rainfall, and river discharge variables are considered, discharge tends to give the best prediction in our study area.

It is important, however, to understand landslide triggers and initiation areas to inform the proactive design for mitigation to minimalize the danger to civilians. Susceptibility mapping provides targeted support for the design of erosion mitigation, road realignments, slope reinforcement, and water retaining structures (Lin et al., 2009; Mateo-Lázaro et al., 2016). Trigger analysis allows clients and decision makers to assess risks and allows engineers to design structures to withstand extreme events (Chang et al., 2011; Lee et al., 2013). This study demonstrates the combined use of several remote-sensing oriented investigation tools that are both free and easy to use, allowing for the enhancement of both desktop and field components of engineering site investigations.

The current project site is shown in Fig. 1 and was chosen primarily based on the occurrence of recent slope movements along the Ontonagon River with the potential to impact the roadway and hydroelectric dam. The area is located near US Highway 45 in eastern Ontonagon County, approximately 17 km north of Bruce Crossing in the Upper Peninsula of Michigan. Victoria Dam and the town of Rockland, Michigan are also indicated in Fig. 1.

The Ontonagon River watershed has been a recurring object of analysis for its unusual glacial soils and geomorphology. The location of past and current work in this area are shown in Fig. 1. Several Michigan Tech Master's theses have been performed on the area (Koons, 1969; Dyl, 1979; Smith, 2012), and geotechnical testing results from previous studies were compared with tests performed in the current study. In 2010, the US Army Corps of Engineers conducted a sediment modeling project identifying mass wasting as the primary source of sediment erosion and subsequent river loading. A rough average sediment load of 2.4 million tons per year was calculated by the US Army Corps of Engineers (2010).

The glacial geomorphology of the study area is illustrated in Fig. 2. The Military Hill section of the Ontonagon River encounters two major bedrock formations: the Portage Lake Volcanics (PLV) and the Jacobsville Sandstone. The PLV are a series of mostly basaltic lava flows interbedded with thinner sandstones and conglomerates (Hack, 1965). To the south and east of the PLV is the Jacobsville Sandstone. The two units contact in a regional NE-SW trending thrust fault known as the Keweenaw Fault. The inferred trace of the Keweenaw Fault crosses through the study area is buried under subsequent glacial deposits. Outcrops of the Jacobsville Sandstone are scarce but when present are at variable elevations, indicating the existence of significant bedrock topography and valleys (Hack, 1965).

The surface geology of the Ontonagon River is dominated by the final advance and retreat of the Wisconsin Glacial Episode that resulted in the formation of two distinct proglacial lakes between 10,000 and 11,800 BP in the study area. Sediments were deposited over an existing stiff till that forms the base of the Ontonagon River system. The first proglacial lake transitions from a fluvial till to a clean fluvial sand while the second proglacial lake deposited a highly varved lacustrine deposits over the fluvial sand. The Ontonagon River system cuts through these sediments forming the steep embankments up to 120 m in height and susceptible to mass wasting (Hack, 1965).

Section snippets

Field and lab work

We suspect the 2003 Military Hill landslide was one of the largest landslides in the US that year. The landslide temporarily blocked and rerouted the Ontonagon River and is critically located less than 300 m from the US Route 45 bridge. Smith (2012) created an idealized cross section of this slide, indicating that the units of soil included 50 m of varved clay above roughly 30 m of clean sand. Its volume was estimated to be 1.1 million cubic meters, and liquefaction of the sand layer has been

Discharge threshold

A total of 131 hydrologic events were identified in the 1999 to 2016 time period, providing the plot upon which a preliminary threshold was drawn. For each event, the total flood volume and the max daily discharge were recorded, as well as the year and month of the first day of the event (Fig. 10). Events plotting on the lower left portion of the graph indicate both low volume and low intensity events; these typically occur in the months of May through October. Other events of higher magnitude

Discussion

The combination of methods presented here are intended to answer the questions posed in a practical natural hazards investigation: where do the hazards exits (Fig. 12), and under what conditions are they set in motion (Fig. 10)? As summarized in Fig. 13, data and methods often exist for which both of these questions can be answered effectively in a time and resource efficient manner, and without the need for extensive fieldwork. Of equal importance is the fact that these data-driven methods

Conclusions

A temporal and spatial hazard investigation was performed to understand slope instability along a river channel in Michigan's Upper Peninsula. Potentially impacted infrastructure in this site included a highway bridge and the outlet works of a hydroelectric dam. In addition, the site's climate and glacially-derived soils make it a unique subject of study. A combination of methods are described, including a discharge threshold and susceptibility map, which together emphasize the utility of

References (35)

  • P. Reichenbach et al.

    A review of statistically-based landslide susceptibility models

    Earth Sci. Rev.

    (2018)
  • R.L. Baum et al.

    Early warning of rainfall-induced shallow landslides and debris flows in the USA

    Landslides

    (2010)
  • T. Bogaard et al.

    Invited perspectives: Hydrological perspectives on precipitation intensity-duration thresholds for landslide initiation: proposing hydro-meteorological thresholds

    Nat. Hazards Earth Syst. Sci.

    (2018)
  • C.T. Chang et al.

    Landslide susceptibility analysis of riverbank-a case study in upstream of Tai-An River

  • A.F. Chleborad

    Use of air temperature data to anticipate the onset of snowmelt-season landslides

  • D. Coduto et al.

    Geotechnical Engineering: Principles & Practices

    (2011)
  • G. Crosta et al.

    On slope instability induced by seepage erosion

    Can. Geotech. J.

    (1999)
  • Cited by (0)

    View full text