Rain reduces stability by changing pore-water pressure. In a slope above the water table, infiltration reduces matric suction and with it the part of the shear strength that suction provides (Fredlund and Rahardjo, 1993). Where a wetting front meets a less permeable layer, a perched water table can form and positive pressures develop. Iverson (2000) showed how the timing and depth of failure follow from the transient transmission of pore pressure, which depends on rainfall intensity and duration relative to the hydraulic properties of the ground. The response of a slope to a storm is thus governed by rainfall intensity, by the permeability of the soil and by how wet the ground was beforehand.

Empirical thresholds relate landsliding to rainfall intensity and duration. The original global threshold of Caine (1980) was updated with a much larger catalogue by Guzzetti et al. (2008). Thresholds describe when failures become likely across a region. They do not explain why one cut failed and its neighbour stood, and that is a question of geology.

In rock, the control is usually structural. A cut that daylights a persistent joint, bedding plane or foliation dipping out of the face at an angle steeper than its friction angle is kinematically free to slide, and the same is true of wedges formed by two intersecting sets (Hoek and Bray, 1981; Wyllie and Mah, 2004). In weathered profiles the joints of the parent rock survive as relict discontinuities and play the same part within what appears to be soil. In colluvium and residual soil, failure commonly follows the contact with underlying rock or a change in permeability. Hencher (2010) has drawn attention to preferential flow paths, such as soil pipes, open relict joints and permeable dykes, which deliver water rapidly to particular parts of a slope and account for failures that uniform-infiltration models do not predict.

A consistent description is the first product of an investigation. The classification of Varnes (1978), revised by Cruden and Varnes and updated by Hungr et al. (2014), names a landslide by its material and type of movement and provides terms for its state, rate and water content. Using it obliges the investigator to decide what moved and how, which is the beginning of a mechanism.

Clearing debris destroys evidence. Before plant moves in, the scar and deposit should be photographed from fixed points, with a drone if one is available, and the main dimensions measured. The record should capture the shape of the rupture surface, the position of tension cracks behind the crest, seepage points and their relation to geological features, and the state of drains and culverts above and below the cut. Temporary measures, such as diverting surface water away from the crest and sheeting open cracks, reduce the risk of further movement without prejudicing the investigation.

The rainfall record from the nearest gauges is compiled for the triggering storm and for the preceding weeks, so that both intensity and antecedent wetness can be judged against any regional threshold. Construction records, earlier failures, maintenance history and aerial photographs are gathered at the same time. Old photographs frequently show that a failed cut was sited in a hollow, on an older landslide or below a drainage line, facts that are hard to see once the road is in place.

The scar is a fresh exposure and should be logged as one. Materials are described and graded for weathering, and discontinuities are measured for orientation, spacing, persistence, roughness, aperture and infilling following the suggested methods of the ISRM (1978). The orientations are plotted on a stereonet with the cut face to test for planar, wedge and toppling mechanisms. Mapping is extended to the natural slope above and to adjacent cuts in the same materials, since these show whether the controlling feature is local or runs along the alignment.

Subsurface work is planned to answer questions left open by the mapping: the depth to rock, the position of the rupture surface beyond the scar, and the groundwater regime. Trial pits and a small number of cored boreholes are usually sufficient for a single cut. Samples are taken from the material in which the rupture surface formed, and shear strength is measured under drained conditions at stresses representative of the shallow depths involved. Piezometers or tensiometers installed at this stage begin a record that later steps depend on.

The observations are now combined into a ground model: one or more cross-sections showing the materials, structure, pre-failure and post-failure profiles, rupture surface and inferred water conditions, together with a classification of the movement. The model should state the geological control explicitly. Table 1 lists the controls most often met in road cuts, the evidence that points to each and the kind of remedy each calls for. More than one may be present.

Back-analysis asks what combination of strength and pore pressure gives a factor of safety of one on the observed rupture surface. For rotational and compound surfaces, limit equilibrium methods such as those of Bishop (1955) and Morgenstern and Price (1965) are used; for sliding on a discontinuity, the planar and wedge solutions of rock slope engineering apply (Wyllie and Mah, 2004). Strength and water pressure cannot both be found from one equation, so one must be constrained by measurement. The value of the exercise lies in testing the ground model: if failure can only be explained with implausible strengths, the model is wrong (Duncan et al., 2014).

Remedial options fall into four groups: changing the geometry, controlling water, adding structural support and protecting the surface. The ground model determines which will work. Regrading suits soil slopes with room at the crest and is ineffective against a persistent plane dipping at a shallower angle than any practical cut. Drainage addresses the trigger directly and is nearly always part of the solution, with surface channels to keep water off the slope and raking drains to lower pressures within it (Geotechnical Control Office, 1984). Soil nails and rock anchors are designed to cross the identified rupture surface or discontinuity set, following established guidance (Lazarte et al., 2015). Each candidate is checked by re-running the back-analysed model with the measure included and with the design water condition.

A repair based on a single failure rests on an inferred model, and the observational method offers a disciplined way of dealing with the uncertainty (Peck, 1969). The design records the pore pressures and movements it assumes, instruments are read through at least one wet season, and contingency measures, typically further drains, are defined in advance. If readings depart from the assumptions, the ground model is revised and the design revisited, as the return loop in Figure 1 indicates.