Contents lists available at ScienceDirect Engineering Geology journal homepage: www.elsevier.com/locate/enggeo Experimental study to identify premonitory factors of landslide dam failures Fawu Wang a, ⁎ , Zili Dai a , Chukwueloka Austin Udechukwu Okeke b , Yasuhiro Mitani a , Hufeng Yang c a Department of Earth Science, Faculty of Natural Science, Shimane University, 1060 Nishikawatsu-cho, Matsue, Shimane 690-8504, Japan b Department of Civil Engineering, College of Engineering, Covenant University Ota, Ogun State, Nigeria c Department of Geological Engineering, Faculty of Geosciences and Environmental Engineering, Southwest Jiaotong University, Chengdu, China ARTICLE INFO Keywords: Landslide dams Failure Settlement Turbidity Self-potential ABSTRACT Identifying premonitory factors before final failure for long-existing landslide dams is of high importance in disaster prevention and risk reduction. In this study, a series of large-scale (outdoor) experiments were designed and conducted to identify premonitory factors that may be used in failure prediction for actual landslide dams. Surface deformation, especially dam-crest settlement, dam seepage-water turbidity and self-potential across the dam crest were selected as the target parameters. Changes in these parameters showed apparent correlations between each other. Based on the monitoring data obtained and the observation performed during the tests, the deformation and failure sequence of the dam model can be separated into four time-sequential periods: 1) Emergence of seepage water and front wetting. In this period, the monitoring parameters did not show any obvious changes. However, wetting was observed in the downstream face. 2) Hyperconcentrated flow discharge. In this period, water flowed out of the drainage channel, and the vertical deformation of the dam body became obvious, while the turbidity of the seepage water increased. 3) Emergence and development of cracks on the dam crest. In this period, the dam-crest settlement also increased. 4) Sudden collapse and final failure. In this period, self-potential across the dam crest decreased rapidly, and the dam-crest settlement reached a peak value. Therefore, dam-crest settlement, seepage-water turbidity and self-potential changes can be regarded as pre- monitory factors of landslide dam failure. 1. Introduction Landslides and rock avalanches can result in natural damming of stream channels and gorges. Such events are common in many moun- tainous regions where several geomorphological and hydroclimatic factors favour the occurrence of geomorphic processes such as landslide dams (Hewitt, 1982; Hermanns et al., 2004). Once a landslide dam is formed, breaching can occur, resulting in the release of lake water impounded upstream of the dam. Failure of landslide dams often trigger outburst floods with potentially catastrophic effects in downstream areas (Evans, 1986; Costa and Schuster, 1991; Casagli and Ermini, 1999; Bovis and Jakob, 2000; Dai et al., 2005; Hancox et al., 2005; Korup and Tweed, 2007; O'Connor and Beebee, 2009). Therefore, a better understanding of premonitory factors, especially those that can be easily measured or observed in actual landslide dams at high risk of failure, is important for disaster reduction. The probability of landslide dam failure remains an integral part of flood-risk modelling and hazard-assessment studies. Costa and Schuster (1988) reported that the longevity of landslide dams depends on several factors including the rate of seepage through the dam; the internal structure and material properties of the dam; the size, shape and vo- lume of the blockage; and the rates of sediment and water flow into the upstream lake. Piping and internal erosion due to seepage flow are among the major failure modes of landslide dams. For example, the failure in 2004 of the Tsatichhu landslide dam in Bhutan was due to dam-face saturation and progressive seepage (Dunning et al., 2006), while seepage-induced instabilities were identified in the debris-ava- lanche dam at Castle Lake near Mount St Helens, Washington (Meyer et al., 1994), in the Bairaman landslide dam in Papua New Guinea (King et al., 1989) and in glacial moraine dams in Peru (Vilimek et al., 2005). Therefore, seepage flow can be regarded as one of the common trig- gering factors of landslide dam failure. Numerous attempts have been made to investigate the complex mechanisms of piping and internal erosion in landslide dams, including theoretical analyses and experimental studies (Ojha et al., 2008; Amaya et al., 2009; Vorogushyn et al., 2009). In their theoretical research, for instance, Bonelli and Benahmed (2010) proposed a simplified me- chanically-based approach for the prediction of piping failure in https://doi.org/10.1016/j.enggeo.2017.11.020 Received 31 October 2017; Received in revised form 23 November 2017; Accepted 23 November 2017 ⁎ Corresponding author. E-mail address: wangfw@riko.shimane-u.ac.jp (F. Wang). Engineering Geology 232 (2018) 123–134 Available online 26 November 2017 0013-7952/ © 2017 Elsevier B.V. All rights reserved. T