1. Introduction

1.1. Preamble

The production and safe storage of tailings is an integral waste management procedure of standard mining projects as well as some industrial and power plant operations. Tailings are fine-grained waste rocks that are conventionally discharged as slurries into long-term storage facilities (Vick, 1983; Blight, 2010; Kossoff et al., 2014; Jefferies and Been, 2016). Given that tailings disposal is necessary insofar as operations continue, tailings dams are perpetually evolving structures that are raised in stages over the span of several years to decades, with some impoundments being among the largest man-made structures in the world (Vick, 1983; ICOLD, 2001). When unfavourable conditions persist in a storage facility (i.e. when a large volume of tailings are very wet, contractive and/or overlain by ponded water), a triggering mechanism may initiate a dam breach resulting in a mass movement of failed materials into the downstream environment (Dobry and Alvarez, 1967; Chandler and Tosatti, 1995; Blight, 2010; Morgenstern et al., 2015, Morgenstern et al., 2016; Robertson et al., 2019).

Voluminous tailings flows are capable of catastrophic, highly mobile behaviour, with some events rendering considerable infrastructural damage, geomorphological impact, environmental contamination and socio-economic cost (Troncoso et al., 1993; Kossoff et al., 2014; Takahashi, 2014; Macías et al., 2015; Bowker and Chambers, 2015, Bowker and Chambers, 2017; Cuervo et al., 2017; de Lima et al., 2020). Some tailings flows have also led to mass casualties when travelling through inhabited areas, with a few examples including 1937 Los Cedros in Mexico (300+ deaths; Macías et al., 2015), 1965 El Cobre in Chile (200+ deaths; Dobry and Alvarez, 1967), 1985 Stava in Italy (268 deaths; Chandler and Tosatti, 1995; Takahashi, 2014) and 2019 Feijão in Brazil (272 deaths; Robertson et al., 2019; de Lima et al., 2020). The occurrence of three high-profile incidents (Mt. Polley in Canada and Fundão and Feijão in Brazil) over a span of just five years (2014–2019) has further raised concerns about large at-risk facilities (Morgenstern et al., 2015, Morgenstern et al., 2016; Bowker and Chambers, 2017; Cuervo et al., 2017; Morgenstern, 2018; Robertson et al., 2019; Santamarina et al., 2019; de Lima et al., 2020; ICMM et al., 2020; Franks et al., 2021; Chovan et al., 2021). Two recent examples of catastrophic tailings flows are presented in Fig. 1 (Petley, 2020a, Petley, 2020b).

1. Download : Download high-res image (1MB)
2. Download : Download full-size image

Fig. 1

A seemingly high frequency of such incidents has generated public scrutiny of the tailings management industry (Azam and Li, 2010; Morgenstern, 2018; Santamarina et al., 2019; ICMM et al., 2020; WISE, 2020; Franks et al., 2021; Chovan et al., 2021). These events continue to add to a growing list of tailings dam failures that includes over 350 reported incidents (ICOLD, 2001; Azam and Li, 2010; Lyu et al., 2019; Ghahramani et al., 2020; WISE, 2020). These failure cases also occur at a time of increasing mine waste production commensurate with declining ore grades and increasing demand for metals (ICOLD, 1996; IIED, 2002; Mudd, 2007; Jones and Boger, 2012; Mudd and Boger, 2013; Adiansyah et al., 2015; Owen et al., 2020), thus raising an urgent need to reduce the frequency and consequences of these incidents.

2. Tailings storage facilities

Mining operations involve the development of a geological ore body to extract minerals while producing vast quantities of mine waste. The nature of this waste varies from dry, angular, very coarse (sandy gravel to boulder-sized) waste rocks stripped from the overburden to very wet, fine-grained (sandy to silty) tailings that comprise the residuum of the comminution process (Blight, 2010). Although predominantly produced from mining activities, tailings are also generated from some industrial and power plant operations (e.g. bauxite from alumina and fly ash from coal). These materials are also considered as tailings in this work, whereby we define “tailings” to include both tailings solids and interstitial (pore) water (Table 2). The types of tailings may be generally grouped as either hard rock (e.g. lead, zinc, copper, gold, silver, molybdenum, iron, nickel and uranium) or soft rock (e.g. coal, fly ash, phosphate and bauxite) depending on the specific gravity (Vick, 1983; Small et al., 2017; Ghahramani et al., 2020). The use of hard and soft rock tailings as a qualitative indicator variable (as in Section 7 of this paper) has previously been recommended as a simpler way to distinguish the type of ore due to the numerous distinct types of tailings and the data sparsity corresponding to each type (Small et al., 2017; Ghahramani et al., 2020). As reflected in our database of tailings flows, the number of data points still remains very low for soft rock tailings (n = 6) compared to hard rock tailings (n = 33).

To protect against environmental contamination, tailings are conventionally impounded behind a compacted dam that is typically built out of the coarse (sandy) fraction of tailings, foundation materials and/or waste rockfill/earthfill (Blight, 2010; Kossoff et al., 2014). Tailings are typically hydraulically discharged as a low-density slurry into the impoundment, with the finer materials (slimes) settling further into the central core while the coarser materials make up the beach abutting the dam. To prevent the generation of surface dusts and acid mine drainage through oxidation processes, the tailings slimes may form the base of a supernatant pond consisting of process water used during mineral extraction (Fig. 2; Blight, 2010; Kossoff et al., 2014). Tailings storage facilities may be constructed as partially enclosed impoundments on a slope, across a valley or as fully enclosed impoundments on relatively flat terrain. In keeping with the rising volume of the impounded material, tailings dams may be raised in the upstream direction (i.e. over the impounded tailings), in the downstream direction or vertically upwards using the centerline method (Fig. 2; Vick, 1983; ICOLD, 2001; Blight, 2010). It is worth noting that the rate of tailings deposition, consolidation and dam construction varies considerably depending on the site, hence impoundment age does not necessarily correlate with the total impounded volume (and by extension total outflow volume and flow mobility), as reflected in our database of tailings flows. Nevertheless, over time, tailings dams may impound tens of M m3 of tailings, and in many cases supernatant ponds, over heights of tens of m, thus creating a long-term anthropogenic footprint on a previously natural landscape. Large impoundments thereby represent sources of significant potential energy capable of extreme downstream geomorphological, geochemical and socio-economic consequences in the event of catastrophic failure (Dobry and Alvarez, 1967; Takahashi, 1991; Kossoff et al., 2014; Macías et al., 2015; Cuervo et al., 2017; de Lima et al., 2020).

1. Download : Download high-res image (141KB)
2. Download : Download full-size image

Fig. 2

3. Susceptibility to mass flow behaviour

The practice of hydraulically disposing tailings into a large wet impoundment, without allowing for the necessary time for settlement and consolidation, may promote a loose, saturated tailings structure that is susceptible to brittle stress-strain behaviour and liquefaction following a triggering mechanism (Martin and McRoberts, 1999; Blight, 2010; Jefferies and Been, 2016). Numerous case studies (e.g. Dobry and Alvarez, 1967; Fourie et al., 2001; Morgenstern et al., 2016; Robertson et al., 2019) indicate that a high propensity for tailings to liquefy en masse – whether induced statically or seismically – is typically congruent with high moisture contents approaching or exceeding the liquid limit (i.e. liquidity index >1.0) and positive state parameters indicating contractive properties. The state parameter (ψ) is the difference between the existing void ratio and the critical void ratio (i.e. the void ratio at which no volume change occurs during shear as determined from reconstituted samples) of the tailings at the same mean effective stress (Jefferies and Been, 2016). In saturated conditions, loose, contractive materials (+ψ) are potentially liquefiable upon undrained shearing whereas dilative materials (−ψ) are more compact and demonstrate greater resistance to liquefaction (Vick, 1983; Martin and McRoberts, 1999; Fourie et al., 2001; Blight, 2010; Morgenstern et al., 2016). During flow liquefaction, impounded tailings undergo rapid strain-softening (i.e. progressively lower shear stress is needed to induce the same level of strain) as the interstitial water begins to carry part of the load during material contraction until residual strength is attained. This initiates rapid flow-like movement of the liquefied mass as it escapes through the breach into the downstream environment (Morgenstern et al., 2016).

Several tailings flows have also initiated due to the uncontrolled or accidental release of a supernatant pond, irrespective of the liquefaction susceptibility of the impounded tailings. Cases such as 1937 Los Cedros and 2014 Mt. Polley have involved the prolonged, rapid discharge of free water with eroded non-liquefied tailings, displaying field behaviour akin to a sediment-rich outburst flood (Macías et al., 2015; Morgenstern et al., 2015; Cuervo et al., 2017). When combining the effects of flow liquefaction with the availability of a voluminous pond, the resulting outflow may be typified by additional fluidization, catastrophic discharge rates, multiple flow surges, larger outflow volumes and widespread downstream inundation (CDA, 2020). Still, the 2015 Fundão, 2016 Luoyang and 2019 Feijão runouts have demonstrated that the occurrence of liquefaction is a sufficient condition to initiate catastrophic mass flows, even without the influence of free water (Reid and Fourie, 2017; Palu and Julien, 2019; de Lima et al., 2020).

Given this knowledge, the Canadian Dam Association (CDA) introduced a classification matrix that attempts to categorize breach events into one of four distinct classes based on the involvement of free water within the outflow and the occurrence of tailings liquefaction (Fig. 3; Small et al., 2017; Martin et al., 2019; CDA, 2020; Gildeh et al., 2020). This sets a useful conceptual framework that may also be adopted as an indicator variable for tailings dam breach-runout prediction studies. The present work applies the CDA framework purely as a qualitative indicator variable to observe the influence of tailings liquefaction and supernatant pond on tailings flow behaviour. It should be noted, however, that the task of assigning CDA classes is prone to subjectivity particularly for complex cases where, for instance, (i) tailings liquefaction only occurred locally, (ii) the volume of free water was small enough to question the corresponding influence on runout behaviour, (iii) the failure occurred in multiple stages of different CDA classes or (iv) rather than a standard dam breach, the failure manifested due to a deficient drainage system that produced a prolonged single-stage discharge into the downstream terrain.

1. Download : Download high-res image (295KB)
2. Download : Download full-size image

Fig. 3

4. Causative variables

A significant percentage of failed tailings dams have been those that were raised in the upstream direction (Lyu et al., 2019; Franks et al., 2021), including 36 of the 43 tailings flow cases with reported dam raising methods in our database. The upstream design has thus received scrutiny, including a prohibition in Chile since 1970 after a series of earthquake-triggered failures (Dobry and Alvarez, 1967; Villavicencio et al., 2014; Valenzuela, 2016; Troncoso et al., 2017). Failures of some upstream-raised dams have been attributed to partial founding on soft, compressible, poorly consolidated slimes that consequently underwent lateral extrusion, undrained creep and development of excess pore water pressures to the point of failure (Dobry and Alvarez, 1967; Troncoso et al., 1993; AECOM, 2009; Morgenstern et al., 2016). However, as counter-argued by Vick (1992), Martin and McRoberts (1999) and Morgenstern (2018), the prominent causes of these instabilities may lie in improper design, construction and operation of the facility – including, but not limited to, the overabundance of contractive materials, insufficient drainage and seepage control and a loose embankment core – rather than the general upstream method itself.

Fig. 4 presents a temporal bar chart distribution of major causative variables for 46 tailings impoundment failures that produced mass flows. We note that assigning causative variables to complex case histories can be a subjective task. For example, as summarized in our database, some failures (Bafokeng, Arcturus, Stava, Karamken, Ajka and Fundão) resulted from a complex interplay of multiple causative variables (e.g. Table 3). In addition, Fig. 4 excludes deep-rooted, mainly anthropogenic predisposal variables related to engineering, operational or regulatory issues (Morgenstern, 2018). The causative variables in Fig. 4 are grouped as follows:

• •

  “Weather hazards” include heavy rainfall, snowfall, storms and freeze-thaw cycles. These factors may lead to a number of failure mechanisms such as overtopping, internal erosion (piping), seepage, excess pore pressures and structural drainage issues.

• •

  “Embankment deficiency” highlights structural issues of the tailings dam core (i.e. lack of compaction, use of improper construction materials and/or oversteepened slopes) that may promote internal erosion, piping, excess seepage and/or slope instability.

• •

  “Seismic activity” refers to earthquake-induced liquefaction failures or seismic tremors that accelerate the impoundment towards failure.

• •

  “Unstable foundation” refers to impoundment foundations that were hydrogeologically, geotechnically and/or topographically unsuitable to sustain high, rapid vertical loading.

• •

  “Structural drainage deficiency” refers to deficiencies within the interstitial and supernatant pond drainage system, resulting in static liquefaction, overtopping or structural failure.

 

1. Download : Download high-res image (104KB)
2. Download : Download full-size image

Fig. 4

Weather hazards have contributed to 39% of the 46 cases, followed by seismic activity (26%), structural drainage deficiency (24%), embankment deficiency (20%) and unstable foundation (15%). Tailings flows triggered by weather hazards have steadily increased in frequency over time, which is consistent with the findings of Rico et al. (2008a), Azam and Li (2010) and Lyu et al. (2019) in relation to tailings dam failures in general. The meteorological effect on tailings dam stability may be manifested in several ways. For instance, heavy rainfall may trigger erosional failures of embankments through pond overtopping (Midgley, 1979; Wagener, 1997). Long-term increases in regional rainfall may trigger loss of suction in unsaturated zones of impounded tailings and lead to undrained conditions (Robertson et al., 2019). Freeze-thaw cycling may cause structural drainage deficiency and loosening of the embankment core, while spring-time melting of snow and ice covers on the impoundment may reduce the freeboard between the dam crest and impounded materials and exacerbate undrained conditions (Chandler and Tosatti, 1995; UNEP and OCHA, 2000). Extensive ice formation contributed to the recent catastrophic failure of the decant drainage system at the Luming tailings facility in China (Fig. 1A; Petley, 2020a).

The stability of geological foundations underneath large impoundments has become of critical concern following the 1998 Aznalcóllar, 2014 Mt. Polley and 2018 Cadia incidents – all of which occurred with minimal advance warning (Gens and Alonso, 2006; Morgenstern et al., 2015; Jefferies et al., 2019). Such experiences have underscored the importance of identifying the weakest units in the foundation for consideration in tailings impoundment design. Moreover, the setting of large impoundments on mountain or valley slopes may yield significant changes to the local hydrogeologic regime, while inducing a perpetual state of high gravitational driving stresses against the embankment. The disasters of 1966 Sgorigrad, 1985 Stava and 2019 Feijão were all preconditioned in part by consistent groundwater seepage into the sloping foundations (Chandler and Tosatti, 1995; Lucchi, 2009; Robertson et al., 2019). Undetected groundwater upwelling was also a key contributory variable in the 1994 Tapo Canyon and 2010 Las Palmas dam breaches (Harder Jr. and Stewart, 1996; Villavicencio et al., 2014).

Tailings flows triggered by seismic activity have steadily reduced in frequency over time. A majority of early seismic liquefaction failure cases were in Chile and Japan (Torres and Brito, 1966; Dobry and Alvarez, 1967; Okusa et al., 1980; Ishihara, 1984); since then, both nations have advanced the state of operational practice such that tailings facilities have withstood powerful earthquakes (Yasuda et al., 2013; Villavicencio et al., 2014; Ishihara et al., 2015; Valenzuela, 2016; Troncoso et al., 2017). A novel concern, however, is the potential role of sequential seismic tremors in accelerating the failure of marginally stable facilities (Agurto-Detzel et al., 2016; Morgenstern et al., 2016).

5. Magnitude–frequency distribution

The documented total outflow volumes (VF) in our database range from 20,000 m3 to 32 M m3, summing to a cumulative total of 136 M m3. Based on our review of pre-existing datasets of tailings dam breaches and tailings flows (e.g. ICOLD, 2001; Rico et al., 2008b; Bowker and Chambers, 2015; Small et al., 2017; Larrauri and Lall, 2018; Ghahramani et al., 2020; WISE, 2020), we conclude that all tailings flows with VF ≥ 200,000 m3 over the period 1965–2020 (totalling to 34 cases) have been publicly documented and are thus included in our database. Fig. 5A illustrates the temporal cumulative VF distribution of the 34 tailings flows that fall within this “complete interval”. An estimated 40% of the cumulative VF (~50 M m3) occurred in the 49-year period between 1965 and 2013, whereas the remainder (60% or ~ 75 M m3) occurred since 2014. This recent spike is dominated by the combined VF of ~67 M m3 from the 2014 Mt. Polley, 2015 Fundão and 2019 Feijão incidents.

1. Download : Download high-res image (222KB)
2. Download : Download full-size image

Fig. 5

A cumulative magnitude-frequency distribution of the complete interval of 34 cases is shown in Fig. 5B. A rollover effect is observed whereby the slope of the power-law relation [Log F = −0.88 (log VF) + 4.93] flattens below a VF threshold of ~1 M m3. This statistical rollover phenomenon has also characterized natural debris flows and is attributed to either data bias (undersampling) at smaller volumes or the manifestation of limiting physical conditions under which mass flows occur (Guthrie and Evans, 2004; Hungr et al., 2008). Using the power-law formula, we calculate that tailings dam breaches that have produced catastrophic mass flows with VF ≥ 1 M m3 have occurred at a mean recurrence interval of 2–3 years over the period 1965–2020. For context and comparison, we also draw in Fig. 5B a minimum-estimate magnitude-frequency curve of the 5-year planned volumes of 1260 tailings impoundments constructed globally over the period 1965–2019. This data is derived from a detailed but incomplete record in Franks et al. (2021) that was collated through voluntary survey responses from extractive companies. By extrapolation of the curve, it is estimated that a tailings facility with a planned 5-year storage volume of ≥200 M m3 was constructed at least once every year over the period 1965–2019. The magnitude-frequency curves of tailings flows and constructed tailings impoundments are noted to be similar in shape and form.

6. Initiation and field behaviour

6.2. Outflow process

The post-breach outflow behaviour of impounded materials is highly complex and depends on the failure mechanism and site conditions (Martin et al., 2015; CDA, 2020). Some insight into outflow processes can be gained from performing a case review that is categorized into CDA classes (Fig. 3). As remarked earlier in Section 3, the effort of assigning CDA classes to breach events is prone to subjectivity depending on the application objective and case complexity. In this work, we apply the CDA framework purely as a qualitative indicator variable to observe the influence of tailings liquefaction and supernatant pond on tailings flow behaviour.

6.2.1. 1A cases (flow of liquefied tailings with a supernatant pond)

During 1A erosional breach cases, the mass flow may be preceded by a steady outflow of excess free water which may serve as a premonitory sign of impending failure (Midgley, 1979; Blight et al., 1981; Wagener, 1997; Blight and Fourie, 2005). In the 1994 Merriespruit case, where the total outflow volume was ~630,000 m3, residents reported observing a strong flow of water passing through the village for ~2 h prior to the catastrophic surge (Wagener, 1997). A significant proportion of the 70,000–100,000 m3 of free water from the facility (30,000–50,000 m3 of rainwater and 40,000–50,000 m3 of supernatant pond) was incorporated within this initial steady flow (Wagener, 1997). The subsequent rapid surge consisted of statically liquefied tailings and residual free water. In the case of 1974 Bafokeng, following ~1.5 h of water seeping along the dam face, ~3 M m3 of statically liquefied tailings (18% of Dam 1) were mobilized over 40–60 min with a mean velocity and discharge of ~10 m/s and ~ 1000 m3/s, respectively, through the ~110 m wide breach (Blight et al., 1981).

During rapid 1A breach events, the total outflow ratios and peak discharges may vary widely depending on site conditions and failure modes. As a consequence of the 1965 Chilean earthquake, 45% of the El Cobre Old Dam (1.9 M m3) escaped the facility whereas almost 100% of the freshly deposited, poorly consolidated tailings in the adjacent New Dam were mobilized (0.5 M m3), resulting in a combined outflow of 2.4 M m3. Eyewitnesses reported “waves of liquid columns” (interpreted as sand blows) on the New Dam tailings surface as a consequence of rapid pore pressure relief (Torres and Brito, 1966; Dobry and Alvarez, 1967). The 1985 Stava outflow, predisposed by inadequate drainage and the sloping foundation (12°–16°), achieved a catastrophic peak discharge of 28,160 m3/s, indicating that it took just 13 s for 63% of the Upper and Lower Basins (~180,000 m3) to be released (Takahashi, 1991, Takahashi, 2014). During the outflow sequence in such catastrophic 1A breaches, liquefaction flowsliding may continue for hours as a series of slope failures that regress backwards until stronger scarp materials are encountered, typically rendering steep, semi-circular scarps surrounding sub-horizontal (1° to 5°) wet terraces (Dobry and Alvarez, 1967; Shakesby and Whitlow, 1991; Blight and Fourie, 2005; Blight, 2010; Robertson et al., 2019; CDA, 2020).

A contrasting example was the 2009 Karamken incident (Russia) that was caused by heavy rainfall and lack of maintenance of the decommissioned facility (Glotov et al., 2018). Approximately 1.1 Mm3 of floodwaters and 0.3 Mm3 of liquefied tailings and dam materials outflowed through the ~50 m wide breach at a relatively low peak discharge of ~10 m3/s for ~30 min before reducing to ~1.4 m3/s over the subsequent three days (Glotov et al., 2018).

6.2.2. 1B cases (flow of non-liquefied tailings with a supernatant pond)

In catastrophic 1B cases, the mass flow may occur in one or multiple surges lasting for several hours to a few days depending on free water availability (i.e. volume and distribution of the pond and, if applicable, the intensity and duration of rainfall). For example, the 1937 Los Cedros flow involved at least three major flow surges over a span of 48 h with a total outflow volume of ~1.5 M m3 (Macías et al., 2015). The flow pulses were triggered and sustained by a torrential rainstorm that lasted several days and provided a sufficient supply of free water to induce progressive softening and erosion of the impounded tailings. The peak discharge through the breach was retrospectively modelled to be ~8000 m3/s with a solids concentration of ~50% (Macías et al., 2015).

The Mt. Polley event also involved at least three flow surges (Morgenstern et al., 2015; Cuervo et al., 2017). The peak discharge through the ~160 m wide breach was calculated to be 5130 m3/s – about a thousand times higher than the probable maximum 200-year flood for Hazeltine Creek (Cuervo et al., 2017). Of the total outflow volume of 24.4 M m3 (a third of the impoundment), 30% (7.3 M m3) consisted of tailings solids, 27% (6.5 M m3) was interstitial water and 43% (10.6 M m3) comprised the supernatant pond (Morgenstern et al., 2015; Cuervo et al., 2017). Helicopter videos of the impoundment captured 8 h after failure showed active sheet erosion of tailings, dendritic drainage channels and cirque-like arcuate headscarps (Fig. 6; Morgenstern et al., 2015). As the steep scarps kept migrating headward due to fluvial erosion, the height drop in the headscarp enhanced pond drainage from farther ends of the facility, which in turn further promoted headscarp retreat akin to a geomorphic cascade sequence. In a natural landscape, this process may take hundreds to thousands of years whereas the pond supply was exhausted within a few hours following the Mt. Polley breach (Morgenstern et al., 2015).

1. Download : Download high-res image (861KB)
2. Download : Download full-size image

Fig. 6

Occasionally in 1B cases, dam failure may occur not due to a breach but instead as a result of a deficient drainage system, triggering a prolonged single-stage discharge into the downstream environment. An example is the 1995 Omai case that was caused by blockage and damage of drainpipes that in turn induced internal erosion of the improperly constructed dam core (Vick, 1996). The resulting cyanide-laden leakage lasted for almost five days, with the initial peak discharge to the adjacent Omai River estimated to be ~50 m3/s (Vick, 1996).

6.2.3. 2A cases (flow of liquefied tailings without a supernatant pond)

The 2015 Fundão and 2019 Feijão events (Morgenstern et al., 2016; Palu and Julien, 2019; Robertson et al., 2019; de Lima et al., 2020) are archetypes of catastrophic 2A cases. The outflow behaviours of 2A cases depend on the proportional volume of liquefiable tailings, which was atypically high in both the Fundão and Feijão cases despite the absence (or negligible volume) of impounded free water. Both sets of tailings contained high iron contents, corresponding to strong inter-particle bonding between oxidized grains which in turn elevated stiffness and specific gravity. Global instability of both facilities subsequently occurred as a sudden brittle response to the high peak strength being exceeded (Morgenstern et al., 2016; Robertson et al., 2019). The Fundão Dam collapse released 32 Mm3 of liquefied tailings (61% of the impoundment), whereas 75% (9.7 M m3) of the Feijão tailings underwent flow failure. Palu and Julien (2019) report an initial volumetric solids concentration of 57.5% in the Fundão outflow, while the peak discharge remains unknown. The Feijão outflow occurred as a singular surge with a peak height of 30 m that was completed in less than 5 mins (Robertson et al., 2019). Despite the decommissioned state of the Feijão facility, the contractive tailings were very wet owing to excess groundwater seepage and intense antecedent rainfall, leading to an interstitial water concentration of ~50% (Robertson et al., 2019). Fig. 7 shows a post-failure view of the Feijão impoundment.

1. Download : Download high-res image (233KB)
2. Download : Download full-size image

Fig. 7

6.2.4. 2B cases (flow of non-liquefied tailings without a supernatant pond)

2B events are relatively under-reported in available literature. In 2B cases, the tailings are typically characterized by higher solids concentrations with a lower susceptibility to flow liquefaction. Failures of these tailings may be mobilized as slope failures or slumps with relatively limited downstream impact (Small et al., 2017; Martin et al., 2019; CDA, 2020).

6.3. Runout behaviour

Large volumes of failed tailings are capable of exhibiting highly mobile, destructive behaviour in the downstream environment. In addition to the properties of the flowing mass (i.e. contractive tendency and volumetric proportions of free water, pore water and solids), the downstream topography also exerts a strong control over the runout behaviour. A case review is presented below that provides insight into the observed effects of topographic confinement, slope and substrate on the velocity, geomorphic impact and mobility of tailings flows.

Following the approach of Ghahramani et al. (2020), we note that our case review focuses on the primary impact zone (“Zone 1” extent) of tailings flows, which is defined by the extent of the main solid tailings deposit characterized by remotely visible or field-confirmed sedimentation, and above mean bankfull elevations if extending into downstream river channels (Table 2 and Fig. 8). The secondary impact zone (Zone 2) is defined as the area downstream of Zone 1 that is further impacted by the tailings flow in some form. Secondary impacts may include flood or displacement wave impacts (i.e. fluid impacts above typical downstream water levels) and sediment plume impacts (i.e. below typical downstream water levels). If no water bodies intercept the released tailings, then the entire runout is classified as Zone 1. Our focus on Zone 1 is therefore due to the geomorphological and potential human consequences associated with the primary impact zones of tailings flows. This approach is consistent with how analogous mass movements have historically been mapped and characterized, while providing value from a life loss risk management perspective. Moreover, as indicated by Ghahramani et al. (2020), Zone 2 extents are typically more challenging to evaluate due to the variability of downstream flow mixing conditions, the relatively transient nature of secondary impacts and the inherent limitations of the adopted GIS-remote sensing methods (e.g. image resolution). Our GIS-calculated values of Zone 1 runout distances, inundation areas and travel path angles (as reported in our database) are also inevitably subject to uncertainties associated with satellite image resolution constraints and measurement subjectivity, further details of which are provided by Rana et al. (2021).

1. Download : Download high-res image (149KB)
2. Download : Download full-size image

Fig. 8

6.3.1. Channelized runouts

Extremely rapid flow velocities (>5 m/s; Cruden and Varnes, 1996; Hungr et al., 2001, Hungr et al., 2014) and high kinetic energy have typically been exhibited by voluminous tailings flows meandering along relatively narrow creek or stream channels and/or relatively steep bed slopes. The seismically triggered Barahona flow in 1928 is among the first documented cases. The flow achieved a maximum superelevation of almost 60 m along laterally incised bends (indicating peak velocity of ~30 m/s) when flowing down the narrow Barahona creek channel with a bed slope of 4°–7° (Troncoso et al., 1993, Troncoso et al., 2017). The surge resulted in the deaths of 54 people and the destruction of a bridge, railway station and campsite (Troncoso et al., 1993, Troncoso et al., 2017). Nine years later, at least 300 people were killed during the Los Cedros disaster (Macías et al., 2015). The flow sustained velocities of over 20 m/s (peak of ~25 m/s) and depths of over 5 m (peak of ~20 m) for ~2 km (Macías et al., 2015). The extremely rapid field behaviour was attributed to the location of the tailings facility that rested on a natural incline of ~22° in mountainous terrain, resulting in high gravitational driving stresses. Moreover, the mass flow travelled along the Tlalpujahua stream channel that had been further wettened by the triggering torrential rainstorm. Evidence of high mechanical energy included significant volumes of entrained debris including trees, roof tiles and detached building blocks (Macías et al., 2015).

The catastrophic field behaviour of the Stava flow (268 deaths) is well-documented (Takahashi, 1991, Takahashi, 2014; Chandler and Tosatti, 1995; Colombo and Colleselli, 2003; Luino and De Graff, 2012). The initial flowslide travelled 600 m and crashed into the left bank of the Stava valley at a velocity of 8.3 m/s, destroying 31 buildings in the process (Chandler and Tosatti, 1995). The flowslide then accelerated along Stava Creek (bed slope of 2°–12°), sustaining a mean velocity of 25 m/s (peak of 31 m/s) until Tesero village 3.3 km downstream of the breach (Chandler and Tosatti, 1995; Takahashi, 1991, Takahashi, 2014). The flow depth progressively decreased from an initial high of 20 m to 5 m near Tesero (Takahashi, 1991, Takahashi, 2014). Almost 50,000 m3 of debris – including soils, trees, and buildings – were entrained from both sides and the bottom of the Stava Creek valley which indicates an entrainment ratio of 0.28 (Hungr and Evans, 2004; Luino and De Graff, 2012). The flow entered the Rio Avisio 4.2 km downstream and subsequently dammed the river and formed a 500 m long lake upstream (Luino and De Graff, 2012).

A high erosional capacity was also demonstrated by the Mt. Polley flow along Hazeltine Creek valley (Cuervo et al., 2017). Although the tailings were not liquefied, a high volumetric proportion of free and interstitial water (70% or 17 M m3) enabled a 9 km long runout over a travel path angle of 1.3°. The mass would have travelled much further had it not been terminally confined by Quesnel Lake. Multiple knickpoint features up to 10 m in height were observed along the valley (Cuervo et al., 2017). Up to 0.6 M m3 of creek sediments and vegetation were eroded and transported during the flow – almost equal to the entire outflow volume following the Merriespruit Dam breach. Steeper channel gradients along the creek corresponded to higher velocities and kinetic energy, which in turn exacerbated erosion (scouring depths of up to 7 m) while limiting deposition (less than 0.1 Mm3) (Cuervo et al., 2017).

Catastrophic tailings flows may be capable of overwhelming creek channel banks to render widespread floodplain damage. An example is the seismically triggered El Cobre flow that killed over 200 people in 1965 (Fig. 9; Torres and Brito, 1966; Dobry and Alvarez, 1967). A peak velocity of ~13 m/s was estimated from arrival times and run-up heights (Torres and Brito, 1966). The tailings travelled along the El Cobre creek channel but carried sufficient energy and discharge to overwhelm both banks along the 11 km long runout, achieving a maximum width of 1 km. Similarly, the failure of the Plakalnitsa Dam (caused by intense groundwater seepage and rainfall) just a year later triggered a 6 km long runout along a creek channel through the village of Sgorigrad (Bulgaria). The tailings overwhelmed both banks and caused the destruction of 156 homes and the deaths of 107–488 people (Lucchi, 2009; Mossa and James, 2013).