Mass movements Open Access

In this chapter, we consider key advances in the understanding of hillslopes and mass movements between c. 1965 and c. 2000. During this time, ‘hillslope geomorphology’ developed from an arguably secondary branch of geomorphology, characterized by a range of disparate studies undertaken in a range of related subdisciplines, to a unified, cross-disciplinary and fundamental component of studies of Earth surface systems. This overarching shift in disciplinary focus was, in part, in response to growing societal and economic need, but was facilitated by significant technological advances that permitted the collection of high-quality data, allowing hillslope processes and change to be considered at unprecedented levels of detail. In turn, this allowed the development of quantitative, dynamic and testable models of hillslope change, building on descriptive and qualitative studies typical of the preceding century, during which the foundations for the key developments in mass movements and hillslope geomorphology between c. 1965 and c. 2000 (and, indeed, beyond) were firmly laid, but which prior to c. 1965 had lacked full empirical support, quantitative rigour, and broader spatial and temporal context.

The considerable progress seen in hillslope geomorphology in the second half of the twentieth century was driven by two factors. First, there was growing economic need and societal pressure for increased understanding and management of slope stability hazards. For example, rapid expansion of civil infrastructure in Europe, North America and Australasia led to large-scale projects involving slope cutting and embankment construction (Brunsden et al. 1975; Fookes et al. 1985). This resulted in the creation of man-made hillslopes that varied greatly in terms of lithology, size, morphology and, hence, propensity to fail. This was particularly the case along, for example, highway or pipeline routes that crossed several physiographical settings through varied and difficult terrain (Jones et al. 1983; e.g. Howes and Kenk 1988).

The late 1950s–early 1970s also saw a number of high-profile hillslope failures that resulted in considerable loss of life and damage to infrastructure, emphasizing the need to improve understanding of controls on hillslope stability. For example, the construction and filling of the Vaiont Reservoir (Italy) resulted in the reactivation of a pre-existing rock slide on the slopes of the adjacent Mount Toc on 9 October 1963 (Müller 1964). The destabilized rock mass travelled downslope at 20–30 m s⁻¹, displacing the water in the reservoir and creating a c. 100 m-high wave that overtopped the Vaiont Dam, causing the destruction of Longarone village downstream and the deaths of more than 2000 people (Hendron and Patton 1985). In South Wales in 1966, the Aberfan tip mudflow/slide resulted in 144 deaths, including 116 children, when 110 000 m³ of colliery tailings spoil placed hazardously on a natural spring destabilized following heavy rainfall and flowed downhill, engulfing Pantglas Junior School and an adjacent row of houses (Bishop et al. 1969). The ensuing litigation to establish whether mismanagement had directly contributed to the Vaiont and Aberfan landslide events, and the intensive investigations that followed, marked the beginning of the era of the professional engineering geologist and geomorphologist, and standardized ground investigation protocols (Coates 1984). Large and often devastating landslides occurring in natural slope settings in the 1960s and 1970s were also influential. For example, two large debris avalanches in the Cordillera Blanca of Peru in 1962 and 1972 killed c. 4000–5000 and 18 000 people, respectively (Schuster et al. 2002). The c. 13 Mm³ 1962 event was triggered by the collapse of the hanging glacier on the north peak of Mount Huascarán (Morales 1966), whereas the 50–100 Mm³ 1970 event was caused by the M_s = 7.7 Peru earthquake (Plafker et al. 1971). In contrast, the causes of the 47 Mm³ 1965 Hope Slide in British Columbia, Canada were less clear, with limited evidence of triggering events sufficient to result in instability (Bruce and Cruden 1977). As the global catalogue of landslides and their impacts grew, so too did recognition of the vast range of landslide types, each with a specific set of causal processes, failure mechanisms and effects in urgent need of classification and detailed scientific investigation (Crozier 1986).

The second key factor that facilitated the rapid proliferation of landslide studies related to the exceptional pace of technological development and, critically, how rapidly new technologies became available for direct application, or adaptation for use, in geomorphic studies of hillslopes. This permitted the collection of datasets about hillslope processes and forms that span an unprecedented range of spatial and temporal scales and resolutions. Use of ground-based, airborne and satellite remote-sensing platforms and image-processing techniques became increasingly prevalent throughout the 1970s and 1980s (Verstappen 1977; Townshend 1981). Bolstered by the advent of Geographic Information Systems (GISs), large inventories of mass movements could be interrogated by sophisticated spatial analyses on increasingly affordable personal computers. This provided the first synoptic and quantitative assessment of patterns of, and key controls on, landslide occurrence at the landscape scale (e.g. Carrara 1983; Keefer 1984).

At the site scale, classic field surveying and mapping methods describing hillslope morphology were complemented by innovative instrumentation and automated data collection (Brunsden et al. 1975; Angeli et al. 2000). This allowed rates of hillslope deformation to be recorded near-continuously, capturing both longer-term, subtle changes as well as more episodic, short-lived and catastrophic movements (Armstrong and Whalley 1985). Simultaneous collection of data describing environmental variables, such as temperature and precipitation, provided new insights into process–response relationships, the existence of any threshold behaviour and lagged responses (Caine 1980; Allison and Brunsden 1990). Ground investigations of landslides, conventionally carried out via exploratory and/or instrumented boreholes, started to provide a combined view of, for example, groundwater conditions and subsurface strain development (Hutchinson 1983). By the late 1990s, nascent use of geophysical survey techniques in hillslope geomorphology promised ever-greater detail in assessments of the subsurface structure of landslides (e.g. see Schrott and Sass 2008 for a summary).

A growing appreciation in hillslope geomorphology of the significance of the strength properties and the behavioural response of slope-forming materials also resulted in the adoption of a range of field-testing devices more typically used in engineering studies (Whalley 1976). For example, use of hand-held penetrometers and shear vanes provided a rapid means of obtaining data to understand how, for example, soil strength varies across a landscape (Chorley 1959). Similarly, information on spatial and temporal variations in intact rock strength could be estimated through use of a Schmidt hammer (Day and Goudie 1977), providing insight into rock-strength control on topographical relief and landsliding (Day 1981). More detailed laboratory investigations into the rheological properties of soils and rocks developed in parallel as hillslope geomorphologists used geotechnical testing equipment to explain hillslope behaviour. For example, use of the triaxial cell allowed accurate and tightly controlled simulation of stress conditions experienced by hillslope materials and, in turn, permitted detailed observation of how hillslope-forming materials deform and/or fail in response to changes in stress state (e.g. Bishop and Henkel 1962; Bishop and Wesley 1975). This added a key experimental approach to explain behaviour not easily observed in more complex field settings (Petley and Allison 1997). Laboratory geotechnical testing also provided an accurate determination of shear strength, yielding fundamental inputs into burgeoning numerical studies of hillslope stability, made possible by rapidly maturing computing capabilities (Morgenstern and Price 1965; Lee 1978; Augustinus and Selby 1990; Kimber et al. 1998).

Conventional ‘hillslope geomorphology’ as an academic subdiscipline ostensibly suffered something of a crisis of confidence in the second half of the twentieth century and beyond (Crozier 2010). Whilst this reflected wider discussions and concerns about the value and disciplinary ‘home’ of geomorphology (Worsley 1979), Strahler (1950) observed that quantitative assessments of hillslope processes lagged behind developments seen in fluvial geomorphology, following pioneering work by, for example, Gilbert and Murphy (1914). Strahler (1950, p. 209) described a general inertia in academic circles to question, if not wholly abandon, the descriptive and qualitative methods of analysis central to Davisian geomorphology and similar assessments of landscape evolution that centred on the (at least perceived) dominance of fluvial processes (Playfair 1802; Penck 1953).

The shift in academic conceptual framework, from the Davisian Cycle of Erosion to quantitative studies of rates and processes, began in the 1960s. By the early to mid-1970s, the increasingly quantitative nature of hillslope studies was evident in the literature. Young (1978) documented a decline between 1970 and 1975 in theoretical and qualitative studies of hillslope evolution models, and a growth in studies of hillslopes processes that used instrumentation to constrain rates of hillslope processes, with additional focus on landsliding mechanisms and quantitative modelling. Despite this, hillslope geomorphology arguably remained a secondary element of general, overarching geomorphological discourse. In this sense, the delayed application of quantitative techniques to hillslopes proved costly, resulting in a commensurate delay in the consideration of hillslopes in modelling work focused largely on fluvial systems (e.g. Schumm 1977, 1979). In such models, hillslopes were often considered passive elements of the fluvial system, responsible for sediment production, rather than significant components worthy of independent, quantitative and process-based investigation (Crozier 2010). Prior (1978) noted the fundamental contributions of engineers in understanding hillslope and landsliding processes, ultimately questioning whether geomorphologists had any role to play in such studies using conventional, and largely qualitative, approaches. Prior (1978) also called for geomorphologists to continuously (re-)appraise and (re-)define their contributions to studies of mass movements, with increasing emphasis on broadening the traditional skills base to include the fundamentals of soil and rock mechanics.

This intradisciplinary crisis of confidence amongst some hillslope geomorphologists did not, however, reflect all perspectives. The importance of geomorphic input into civil engineering projects was widely recognized in cognate subdisciplines (Peck 1969; e.g. Henkel 1982) and significant value was placed on the key skills conventionally possessed by geomorphologists. For example, Brunsden (2002, p. 103) noted that:

[T]he ‘peculiarly geographical’ skills which became valuable in the early days of subject application can be summarized as being able to appreciate the significance of spatial and temporal patterns, events, correlations; to appreciate how variables change in importance with the scale of the problem and an ability to use maps, geographical information systems and information data banks to map and analyse the development of the landforms … Geomorphologists in the early 1970s were trained as geographers because geomorphology was primarily carried out in those departments. The nature of geography, which is to synthesize data from many sources, to group, classify, describe and understand the surface of the earth, made them ideal scientists to contribute to the feasibility, walkover and site investigation phases of [engineering] projects.

Brunsden (2002) viewed geomorphology as an indispensable component of the Core Geo-Team required to understand a landscape from an engineering perspective, placing the unique insight and context provided by geomorphologists (e.g. Hutchinson 2001) on an equal footing with disciplines that are traditionally associated with civil construction work (Fig. 1). Hillslope geomorphologists responded to Prior's (1978) call to diversify the traditional geomorphic skill set, as seen, for example, by the development of expertise in soil and rock mechanics that became fundamental elements of geomorphic investigation (Yatsu 1962; Whalley 1976). By the 1990s, subdisciplinary divides were blurred, and ‘hillslope geomorphology’ was a melting pot of experience, expertise and techniques.

Our focus here follows the multidisciplinary approach advocated by Brunsden (2002) that encapsulates modern hillslope geomorphology. We summarize the key developments in the understanding of hillslopes and mass movements between the mid-1960s and c. 2000, regardless of apparent (sub-)disciplinary origin and with less emphasis on somewhat arbitrary (sub-)disciplinary divides. Our summary is inevitably limited in scope, but we focus on key contributions to understanding, explaining and forecasting mass movements across a broad range of spatial and temporal scales.

Classification of mass movements is an important step in geomorphological investigation, allowing rapid appraisal of the key form and deformation features of a particular hillslope (Crozier 1986). The 1950s–1970s saw a rapid proliferation in the number and scope of mass-movement classification schemes (Varnes 1958, 1978; Skempton and Hutchinson 1969; e.g. Crozier 1973), in part reflecting the isolated consideration of hillslopes, both geographically and in terms of disciplinary focus. Direct equivalence between classification schemes was rare, hindered by overlapping and divergent uses of key descriptive terms. This ‘unintentionally defeated one of the principal purposes of the [classification] exercise; that is, the provision of clear and unambiguous terminology’ (Crozier 1986, p. 3). The degree of complexity, interpretation and analysis required for classification also varied greatly. Crozier (1986) highlighted that the design of a specific scheme reflected the underpinning objective: from those that are purely descriptive to those that required a considerable degree of field investigation, observation and interpretation. For example, Varnes (1958, 1978) developed a descriptive landslide classification scheme based on two factors: the type of movement (fall, topple slide, spread, flow or ‘complex’: Fig. 2) and the material (rock, debris or earth) (Table 1) (see Hungr et al. 2014 for a recent update). In contrast, Crozier (1973) classified the type and form of mass movements based on morphometric indices. Subsequent analyses permitted a greater inference to be drawn about potential failure mechanisms, culminating in the derivation of mass-movement ‘process groups’ – an approach that Prior (1978) noted to be prone to issues of equifinality, whereby similar hillslope forms can result from a range of causal processes. Similarly, classification schemes developed in geotechnical engineering (e.g. Skempton and Hutchinson 1969) and rock mechanics (e.g. Hoek and Bray 1981) focused on consistent interpretations of site conditions leading to potential first-time failures and/or reactivation along pre-existing failure surfaces.

General, descriptive classifications published in the 1960s and 1970s were further refined throughout the 1980s and 1990s. As data became available, this included, for example, the velocity of movement (Cruden and Varnes 1996). The relative simplicity of these schemes, along with a growing emphasis on international standardization, have ensured that these earlier approaches endure, are well used and permit straightforward comparison between sites (Keefer 1984; Crozier 1986). Critically, classification schemes alone could not be used to identify the ultimate cause, location or timing of new, first-time failures. This greater degree of spatial and temporal forecasting required: (1) improved data summarizing the location of the hillslope failures; (2) comparison of these data with key ‘environmental’ datasets; and (3) greater process-based understanding of the mechanisms of instability. These themes formed the focus of hillslope geomorphology throughout the latter half of the twentieth century.

During the period up to the end of the twentieth century, geomorphic research began to explore the potential of remote sensing for understanding mass movements. Until the mid–late 1990s, aerial photography dominated remote-sensing analysis of mass movements, with photographic interpretation a central pillar of most site-based investigations of landslides (Chandler and Brunsden 1995). The number, type, resolution and temporal frequency of data readily available from different remote-sensing platforms remained somewhat limited (see Metternicht et al. 2005). In a review of the use of remote sensing for landslide studies, Mantovani et al. (1996) argued that the application of remote sensing during the 1980s and early 1990s remained largely haphazard, attributable in part to: the limited awareness of the potential of different kinds of remote-sensing techniques and data available within and beyond the discipline; a lack of funding; and limited experience in interdisciplinarity working, such that the experience of remote-sensing specialists was not often drawn upon.

Despite these initial limitations, the very first generations of widely available and systemically collected satellite data, such as the LandSat Thematic Mapper launched in 1984, demonstrated the potential that a regional synoptic perspective offered for understanding landslides and mass movements. Remote-sensing data fed into landslide research at different junctures, including: landslide detection and classification; monitoring existing landslides; and in the spatio-temporal analysis and prediction of future landslide hazard and risk. Many of the techniques originally developed in early studies, notably around inventory generation and, more recently, change detection, remain central to how such data are used today. Since the mid-1970s and until the early 2000s, satellite data were used in increasingly coherent and standardized ways, although initially within research-facing rather than more applied (e.g. hazard assessment) studies of landslides (Mantovani et al. 1996).

The recognition of landslides from remotely sensed imagery remained traditionally heavily reliant on skilled operators recognizing characteristic spectral contrast and morphology. Early approaches developed a suite of common tools, including determining minimum detectable landslide sizes, typical spectral signatures and key morphological indices, as summarized by Rengers et al. (1992). However, image-resolution constraints limited early data analysis to assessments of the areal extent of a landslide; this was often only reliable for those landslides that showed a clear chromatic contrast to the surrounding landscape. This resulted in some inevitable bias in data mapped to, for example, only the largest landslides. In turn, these issues hindered full realization of the potential of remote sensing for understanding mass movements. Critical methodological advances in the 1970s and 1980s helped to overcome some of these problems. For example, stereo approaches were key to providing greater confidence in mapping and in the recognition of key morphological features, such as the location of head scarps or areas of ground-surface disruption (e.g. Chandler and Brunsden 1995). Similarly, data from LandSat (90 m ground resolution) and SPOT HRV (High Resolution Visible; 10 m ground resolution) were initially used to analyse larger regions prone to mass movement, rather than focusing solely on individual landslides. Notable studies that documented fundamental methodological development for the regional-scale assessment of mass movements include the assessment of the potential for landslide mapping in Colorado, USA (Sauchyn and Trench 1978), and the generation of inventories of mass movements in Bolivia (Scanvic and Girault 1989). The instantaneous capture of imagery across a region enabled by robust analytical remote-sensing tools resulted in, for the first time, the potential for rapid and even responsive data collection after landslide-triggering events, such as heavy rainfall or large earthquakes (Keefer 1984; Harp and Jibson 1996). This work paved the way for use of remote sensing and GIS in landslide mapping that are now central to humanitarian and aid efforts following significant landslide-triggering events.

Building on work undertaken using monochrome aerial photography, the growth of data in landslide inventories provided what, at first glance, may seem a straightforward approach to populating mass-movement classification schemes with a wide range of case studies. However, most mass-movement classification systems remained reliant on information such as failure mode and mechanism that fundamentally could not be readily mapped from remotely sensed imagery. This led to simpler diagnostic classifications based solely on features that were directly visible in remotely sensed images, and from which inferences around movement style, activity and depth could be made (van Westen 1993). This complemented the range of existing classification schemes, adding regional context but also highlighting the continued need for field investigation to ‘ground truth’ remotely sensed phenomena (see e.g. Rott et al. 1999).

The generation of landslide inventories also led to a growing body of research on the emergent statistical properties of landslide populations, including the scaling characteristics of landslide magnitude and frequency, their inherent biases (Stark and Hovius 2001), and the degree to which statistical patterns in landsliding mirrored those observed in other fields of geoscience and beyond, with key implications for: (1) the underlying physics of geomorphic change; and (2) the geomorphic effectiveness of mass movements of varying size and frequency of occurrence (Crozier and Glade 1999; Malamud and Turcotte 2000). A key feature revealed within multiple regional landslide inventories was the existence of a negative power law that defines the relationship between the magnitude and frequency of landslides (Whitehouse and Griffiths 1983; Pelletier et al. 1997). This was a fundamental finding that ultimately led to statistical appraisal of the ‘completeness’ of landslide inventories (Malamud et al. 2004), and permitted robust estimation of erosion rates resulting from landsliding in mountainous landscapes (Hovius et al. 1997). Subsequent comparison of the volumetric flux resulting from bedrock landsliding with those derived from catchment-scale measurements of sediment discharge revealed the dominance of bedrock landsliding in the erosion of mountain landscapes (Hovius et al. 1997). In turn, and in stark contrast to geomorphic thought in the first half of the twentieth century, this finding further cemented hillslope geomorphology as a fundamental component of Earth surface systems (Playfair 1802; Strahler 1950; cf. Penck 1953).

For successful landslide monitoring, assessment of changes in landslide footprint and topography, plus derivatives such as velocity, sediment flux and/or head-scarp retreat rate through time, data are required from successive monitoring epochs. This introduced the complexities of accessing and handling large-scale, multi-temporal data and increased the need for inter-survey consistency to enable real change to be isolated from uncertainty. A key step in this field of research for landslides was the growing accessibility of the Global Positioning System (GPS) (Gili et al. 2000; Malet et al. 2002) to complement a wider range of developments in ground instruments such as extensometers, tacheometers, electronic distance meters (EDM) and borehole-based instruments such as tilt-meters (see Malet et al. 2002). GPS offered reoccupation of key landslide features (‘geodetic monuments’) to achieve millimetric precision, and so was complementary to conventional geodetic methods (Jackson et al. 1996; Gili et al. 2000); continuous GPS monitoring was yet to be applied to capturing landslide movement, largely due to the high costs of field equipment. The late 1990s saw the first application of satellite-based radar interferometry for landslide monitoring (Scanvic and Girault 1989; Rott et al. 1999), achieving centimetre-level accuracy using ERS-1 imagery, building on advances made in the study of active crustal deformation. The second half of the twentieth century also saw the start of work on more advanced ground-based remote-sensing data collection including, for example, the application of thermal infrared techniques (Bison et al. 1990), and terrestrial LiDAR (Light Detection and Ranging) and photogrammetry (Ray and Fisher 1960; Chandler and Cooper 1989; Adams and Chandler 2002), to monitor deformation in unstable slopes.

The ability to repeatedly monitor landslides in their entirety, rather than on the basis of a limited number of point measurements, permitted identification of spatial and temporal variability in behaviour, including details on the sequencing and variability in deformation, highlighting the existence of transient and steady-state mass-movement phases (Rott et al. 1999) (Fig. 3). Comparison with regional-scale datasets describing, for example, annual precipitation provided insights into underlying mechanisms responsible for varying rates of movement. However, this approach considered general movement patterns only, and lacked the explanatory power needed to explain the spatial and temporal detail in landslide deformation patterns (cf. Allison and Brunsden 1990).

The application of remote sensing in landslide studies has been heavily motivated by a need to map landslide susceptibility and hazard, where the objective is the division of the land surface according to degrees of actual and/or potential mass-movement hazard (Varnes 1984). Inevitably, such a task necessitated trade-offs between available data and resolution, whereby the inputs influence what is feasible to achieve (van Westen 1993). Early work in the 1980s identified the ideal information needed for hazard maps in terms of spatial probability, timing of landslide occurrence, and the direction and velocity of runout and head-scarp regression (Varnes 1984; Hartlen and Viberg 1988). However, these aspirations were initially rarely obtainable from remotely sensed data alone, leading to many early landslide hazards maps being relative rather than absolute, or qualitative rather than quantitative (Mantovani et al. 1996). Extensive work on landslide hazard and zonation was conducted during the late 1970s and early 1980s (e.g. Cotecchia 1978; see the review by Brabb 1984). These studies tended to focus on site-scale problems to derive data to feed into deterministic modelling of slope stability (Hoek and Bray 1981; e.g. Anderson and Richards 1987). Upscaling was immensely challenging, where findings were heavily contingent on very local values of geotechnical properties and groundwater conditions, which is challenging to integrate into large(r)-scale deterministic models. Several approaches were developed to tackle this indifference. This included the generation of maps of landslide inventories, capturing the spatial distribution and footprint of landslides across often extensive areas (Wieczorek 1984), which formed the empirical basis for most other hazard zonation techniques. Irrespective of how such data have been compiled, these approaches now form the mainstay of much research on landsliding and its role in generating hazard (Hungr et al. 1999).

A further key step was in adding utility to such maps, specifically around the identification of the hazard posed by mass movements, notably via geomorphological mapping. Kienholz (1978) outlined the basis for this geomorphic approach, which involves field surveys to undertake a thorough investigation of the full range of geomorphic field evidence of historical slope movements to define the degree of hazard across a mapped region. Key examples of this often highly subjective approach include work by Brunsden et al. (1975) and by Ives and Messerli (1981). To achieve a higher degree of objectivity, and to also reduce reliance on ‘expert judgement’ and hence to enable wider application, statistical techniques were developed to aid the assessment of landslide hazard, benefitting from the proliferation of digital data and access to computers. Brabb et al. (1972) presented one of the first frameworks for ‘quantitative landslide susceptibility analysis’, with the intention of operation at a regional scale. Such approaches required the introduction of proxy variables (such as categories of rock type, rather than measurements of rock strength) that homogenized localized controls on slope stability when considered at larger spatial scales. This was a marked departure from detailed, site-specific measurements more typical in engineering geology and geotechnical studies. Building on insights provided by the spatial distribution of landslides in mapped inventories, Carrara (1983) introduced multivariate analysis of mass-movement data that queried local conditions for large numbers of landslides to identify patterns in causative factors. This approach has subsequently been applied widely (Lessing et al. 1983; e.g. Corominas et al. 1992). As the ease with which computers enabled such datasets to be explored (Carrara et al. 1982), a wide range of methods has been tested to analyse landslide hazard (Guzzetti et al. 1999). In addition, considerable progress was made in regional deterministic modelling, often with a physical basis rooted in a codification of the simplified slope-stability models for translational landslides (Skempton and DeLory 1957; Ward et al. 1982; Jibson 1993). Whilst these approaches had the advantage of having a physical basis, the high degree of simplification involved limited their scale of application to regions, rather than offering the potential for site-specific hazard assessment across multiple locations (Jibson et al. 2000).

Although this considerable progress began to map out the potential enabled by newly available landslide data, the majority of work still fell short of the true hazard maps defined by Varnes (1984). Fundamentally, the temporal probability of mass movements remained an elusive parameter to constrain, whereby knowing within what time period and with what temporal uncertainty a given slope will rather than might fail was not described. The underpinning causative mechanisms were also often difficult to study and capture at an appropriate scale (Crozier 1986).

Finlayson and Statham (1980) proposed a conceptual model (Fig. 4), later refined and exemplified by Julian and Anthony (1996), that forms a useful framework to allow us to contextualize the breadth of research carried out in the latter half of the twentieth century into the causes and mechanisms of mass movements, and the relative timescales over which they operate. Whilst landsliding is the end result of the operation of a wide range of influences, a distinction can be made between: (1) longer-term preparatory factors that reduce slope stability without causing failure; and (2) transient triggering factors that result in hillslope deformation or failure (Julian and Anthony 1996). In Figure 4, the continuous black line represents the shear strength (τ_f) of the slope-forming materials and the dashed black line is the baseline shear stress (τ) acting at depth within the slope. Points 1–5 indicate locations where the two lines meet, and so where τ_f = τ (F_s = 1) failure is likely to occur. Line A represents continuous but gradual reductions in shear strength resulting from, for example, weathering processes. Superimposed on this longer-term trend is Line B, which illustrates shorter-term oscillations in soil strength driven by, for example, seasonal fluctuations in groundwater resulting from precipitation. Line C represents the critical shear stress that causes slope failure if it exceeds slope strength. Critical shear stress can remain constant but might also be punctuated by transient shear stress during earthquakes. Gradual increases in shear stress may also result from, for example, basal erosion and resultant slope steepening and/or loss of toe support (Line D). Although we do not explicitly deal with this element in this chapter, it serves to illustrate how slopes, and how they change through time, became increasingly viewed as an integral (e.g. Pearce and Watson 1986) and diagnostic (e.g. Carson and Petley 1970) component of geomorphic systems.

Terzaghi's (1950) principle of effective stress states that the frictional resistance of soils is, in part, controlled by the magnitude of porewater pressure (u), which, in turn, is a function of depth below groundwater level. Over diurnal to monthly timescales, precipitation-driven increases in groundwater level (and hence a reduction in frictional resistance of slope materials) are a common trigger of landslides (points 2 and 4 in Fig. 4). Establishing critical groundwater levels that caused slope instability in a range of settings became possible through the use of limit equilibrium assessments based on equation (2) and a range of derivatives that consider failure surfaces of varying geometry (Bishop 1955; Morgenstern and Price 1965). However, these physically based, yet simple, geotechnical models were not able to relate quantitatively groundwater changes to the underlying trigger, such as rainfall-driven groundwater-level increases.

Several key empirical studies throughout the 1980s defined rainfall thresholds for slope failure. Perhaps the most influential study was carried out by Caine (1980), who compiled published data for 73 locations across varied climatic, geological and topographical characteristics, slope configurations, and, implicitly, antecedent moisture conditions. For each location, the intensity (in mm h⁻¹) and duration (in h) of rainfall that triggered shallow landslides and debris flows were considered and compared (Fig. 5). Using this approach, Caine (1980) defined a limiting (lower) intensity-duration threshold for landslide initiation. Critically, a common threshold was evident despite the considerable differences in the characteristics of the locations in the underpinning dataset, showing great promise for the regional-scale prediction of landslide occurrence based on rainfall forecasting alone.

Caine's (1980) empirical approach was subsequently refined in several studies to create more tailored regional landslide-triggering rainfall thresholds, with greater consideration of the role of antecedent precipitation (e.g. Cannon and Ellen 1985). These studies improved assessment of the types of failure that occur in response to differing combinations of rainfall intensity and duration (e.g. Larsen and Simon 1993), sometimes coupled with geotechnical analysis to add a physical-based, predictive element to the general nature and configuration of hillslopes prone to failure (e.g. Moser and Hohensinn 1983). Building on concurrent developments in hillslope hydrology (Iverson and Major 1986; Calver and Binning 1990), coupled models subsequently linked rainfall with local to regional slope properties, infiltration capacity and, hence, hillslope groundwater level (Buchanan and Savigny 1990; Crosta 1998). This helped to overcome the limitations of using broad empirical relationships alone and contributed to the development of the first generation of regional-scale landslide early-warning systems (Keefer et al. 1987). Ultimately, close synergies with fundamental geotechnical research provided geomorphologists with the impetus to develop field monitoring to appraise more fully the role of, for example, concepts and processes such as undrained loading (Hutchinson and Bhandari 1971; Allison and Brunsden 1990).

Transient shear stresses resulting from ground shaking during earthquakes cause landslides in steep landscapes (point 5 in Fig. 4). The effects of earthquakes on natural slopes were documented in a range of disparate historical accounts (e.g. Hamilton 1783). Throughout the 1960s and 1970s, great progress was made in understanding seismic ground motions (Arias 1970), alongside engineering assessments of the seismic stability of embankments and dams (e.g. Goodman and Seed 1966; Seed 1979). However, general understanding of the regional-scale effects of earthquakes on the stability of slopes remained poorly constrained until the mid-1980s. In a pioneering study that demonstrated the value of detailed landslide inventories, Keefer (1984) compiled a database of landslides triggered in 40 historical earthquakes. Keefer (1984) explored and defined broad relationships between earthquake magnitude and the maximum area (km²) affected by landslides. Critically, this analysis indicated that: (1) earthquake magnitudes greater than M 4–5 are required to trigger landslides in natural slopes; (2) the extent of earthquake impacts are limited, and that these vary for different types of landslides (cf. Varnes 1978); and (3) local factors, such as variations in slope, lithology and strength, control the variability in the overall patterns of coseismic landsliding.

Although Keefer's (1984) study was subsequently critiqued and refined, its fundamental conclusions continue to provide previously unparalleled insight into the spatial character of earthquake-triggered landslides and, hence, a reliable first-order appraisal of the susceptibility of slopes to failure during future earthquakes. It also provided a framework for subsequent research to improve predictive models of earthquake-triggered landslides across a range of scales, building on developments in seismology and slope engineering. For example, the use of geophysical quantities that better quantified coseismic energy release and attenuation offered a tighter constraint on models of regional patterns of ground shaking that influence slope stability (Ambraseys and Menu 1988). Early pseudostatic stability models focused on the ‘critical accelerations’ required to cause slope failure (e.g. Terzaghi 1950). Whilst very simple in form and straightforward to implement, these models assumed full, catastrophic slope failure if coseismic ground accelerations exceed the calculated critical acceleration at any point during an earthquake – an assumption at odds with field observations (Wilson and Keefer 1983). Stress-deformation analysis offered a more accurate alternative, but the underpinning techniques (e.g. finite-element modelling) were in their infancy (Clough and Chopra 1966), and required considerable data input and computation that precluded application beyond critical infrastructure engineering projects (Jibson 2012).

Newmark's (1965) sliding block model, originally developed for assessment of the seismic stability of dams and embankments, provided an effective and pragmatic compromise between the simplicity of pseudostatic analysis and the complexities of stress-deformation analysis. Newmark's (1965) approach models a potential landslide mass as a rigid block on an inclined plane. The first steps involve calculation of: (1) the static factor of safety of the modelled hillslope; and (2) the calculated critical seismic ground acceleration required to destabilize the landslide mass. Cumulative landslide displacement can then be estimated through two successive integrations (with respect to time) of the elements of the slope-parallel acceleration time history (or some approximation thereof) that are greater than the calculated critical acceleration. Despite a range of underpinning assumptions (Harp and Jibson 1996), Newmark's (1965) model provided an appropriately accurate and quantitative assessment of coseismic slope displacements (e.g. Goodman and Seed 1966; Wilson and Keefer 1983). Using the increasing availability of remotely sensed data describing slope geometry and regional estimates of the strength properties of slopes, Newmark analysis could be implemented in GIS models to perform regional-scale assessments of coseismic slope stability (e.g. Jibson 1993). GIS-based Newmark analyses complemented Keefer's (1984) initial empirical overview of the broad distance-from-epicentre limits to coseismic landsliding by providing a finer-grained spatial evaluation of specific slopes prone to failure during future earthquakes. This was key step in the development of parsimonious regional-scale seismic hazard assessment (Jibson et al. 2000).

Throughout the second half of the twentieth century, field, modelling and laboratory studies continued to provide insight into the nature and mechanisms of hillslope response to earthquakes, adding context and explanatory power to regional-scale assessments. For example, Boore (1972) and Ashford and Sitar (1997) observed and demonstrated the significance of local amplification of incident seismic waves in steep hillslopes and how this can result in increased susceptibility of surface fissuring and/or landsliding at ridge crests (e.g. Çelebi 1991). Laboratory testing of the response of shear surfaces to dynamic loading demonstrated the fundamental roles of lithology and the magnitude of ground accelerations in controlling shear strength during earthquakes (e.g. Crawford and Curran 1981). Such work highlighted the importance of variable dynamic shear strength and how this might differ from the static strength assumed in conventional Newmark analysis. Although widely recognized, dynamic and/or variable shear strength was rarely fully incorporated into models of earthquake-triggered landslides (e.g. see Lin and Whitman 1986). Nevertheless, the insight provided by these first appraisals of event-scale earthquake impacts on landsliding now underpins current research into the role of hillslopes in the evolution of topography, in the generation of risk and in the transfer of organic carbon from land to sea.

In addition to the advances made in understanding hillslope response to meteorological and seismic triggers of mass movements, there was a growing recognition that slope deformation and landsliding can occur with no apparent external trigger and with no perceptible change in slope geometry and resultant driving stress (e.g. Wieczorek and Jäger 1996). Although failures with no apparent trigger could be an artefact of limited availability of data, increased remote monitoring and mapping capability, and an improved understanding of why mechanisms may cause a lagged slope response to a trigger (Hutchinson and Bhandari 1971; Allison and Brunsden 1990), have reduced the possibility of misinterpretation of ostensibly ‘unexplained’ mass-movement events.

From the early 1960s, a growing body of research began to consider slope movements driven by a reduction in slope strength through time (Line A in Fig. 4), ultimately to the point of slope failure (Point 1 in Fig. 4) if external triggers do not first result in instability. For example, early (pre-1960) studies of rock weathering documented rock breakdown and cohesion loss with remarkable experimental and mechanistic insight (e.g. see Goudie and Viles 2008). The latter half of the twentieth century saw an increasingly quantitative dimension in weathering studies, permitting weathering effects to be related to measurements of rock strength, or approximations thereof (e.g. Ballantyne et al. 1989; McCarroll 1991). The significance of the degree weathering on slope strength was incorporated into rock-mass-strength classification schemes (e.g. Selby 1982), and also used to explain differences in slope geometry and stability (e.g. Durgin 1977). However, relating mechanisms and rates of weathering to rates and styles of slope deformation was notoriously difficult to constrain beyond qualitative and general statements that slopes weaken through time.

Cognate research in experimental rock mechanics considered intrinsic strength degradation in brittle materials (e.g. Attewell and Farmer 1973; Cruden 1974). Although a first approximation of full slope stability could be obtained from equation (2) and the underpinning estimates of slope strength, any smaller-scale (microscopic) brittle failure and/or strain-weakening behaviour were hypothesized to occur at stress values that are significantly below the ultimate failure strength of slope-forming materials, even under conditions where the overall Factor of Safety of the slope suggests stability (F_s > 1) (Bjerrum 1967). These microscale failures occur, for example, at stress concentrations in and around heterogeneities within rock masses and soils. As these failures occur, the ability of the material to resist stress drops, causing a transfer in stress to the surroundings and resulting in additional local failures, strain accumulation and further stress transfer (Bjerrum 1967). The resultant positive feedback loop ultimately results in the formation of a shear surface that, critically, does not form in response to a discrete external trigger. Bjerrum (1967) termed this overall cascading process ‘progressive failure’. This became a key concept in subsequent decades, guiding research towards improved understanding and validation of the mechanisms, movement patterns and rates associated with progressive failure (Palmer et al. 1973; e.g. Chowdhury 1978; Chowdhury and A-Grivas 1982; Martin and Chandler 1994; Cooper et al. 1998), and ultimately strain-rate based methods of slope-failure prediction. This insight opened the door to retrospective analysis of previous collected movement data (e.g. Vaiont: Kilburn and Petley 2003), and also a mechanical interpretation of previously observed linearity in velocity–time space observed by Saito (1965, 1969) and later formalized by Voight (1988, 1989). Such models are only more recently being fully exploited as a means of providing prediction and early warning of landslide occurrence using high-resolution monitoring data (e.g. Loew et al. 2017).

In this chapter, we have highlighted progress in research on mass movements undertaken between c. 1965 and c. 2000. This period saw a burgeoning need for a greater level of understanding of mass-movement processes in response to a series of high-impact events, and because of the need to develop infrastructure safely. A series of step changes were enabled through hillslope geomorphology (broadly defined) being open to overarching and consolidating concepts, methods, and models from cognate disciplines, whilst seizing opportunities to gain insights from rapidly advancing methods increasingly focused at a scale of investigation relevant to landsliding. Here, we suggest that these key conceptual advances included: bridging across scales of hillslope investigation; linking and developing understanding on process, form and mechanisms of slope stability; and articulating the temporal characteristics of slope stability and movement. The potential that hillslope geomorphology offered for making conceptual and methodological advances in this period remains recognized today, as illustrated by the continued citation of many seminal papers that remain at the core of the discipline. As such, this period left hillslope geomorphology well positioned to contribute to rapidly evolving debates around landscape evolution, to feed into work on hazard and risk management and mitigation, and to position hillslope geomorphologists as key members of the ‘Geo-Team’ in more applied work. The research and progress made in hillslope geomorphology on mass movements in this period clearly laid the conceptual foundations for much work undertaken since the turn of the millennium, where hillslope geomorphology has continued to develop as a central pillar in the study of Earth surface processes.

We thank the Editors for their support in writing this book chapter. We also thank Bob Allison and an anonymous reviewer for their helpful comments that contributed to the clarity and focus of the chapter. We gratefully acknowledge permission to reproduce figures, tables (captions for details) and text excerpts, as follows: from Crozier (1986): ©1986 From Landslides: Causes, Consequences & Environment by M.J. Crozier. Reproduced by permission of Taylor and Francis Group, LLC, a division of Informa plc. From Brunsden (2002), with permission from the Geological Society of London and reproduced from a previous GSL publication (https://www.geolsoc.org.uk/en/Publications/Permissions). We thank Nel Caine for permitting us to reproduce figure 5 from Caine (1980).

MJB: writing – original draft (lead), writing – review & editing (equal); NJR: writing – original draft (supporting), writing – review & editing (equal).

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data sharing is not applicable to this article as no datasets were generated or analysed during the current study.