Analogue modelling of pre-failure strain accumulation for landslide failure prediction

Landslides are one of the most destructive geological hazards, resulting in excess of 100,000 fatalities between 1990 and 2006. There is now considerable evidence that landslide impact is increasing. These effects are focussed upon mountainous areas in developing countries, where losses can be 2 / 3% of GDP per year. To undertake a full assessment of landslide hazard, an understanding of the likely location, style, size, speed and timing of the failure is needed, but the last of these remains poorly understood; a problem well illustrated by the 17th February '06 landslide at Leyte, Philippines. This event killed more than 1,100 people and occurred on slopes previously identified as potentially unstable. Prior to the landslide ten days of intense rainfall occurred, a period during which cracks on the slopes above the village were noted. After five days of rainfall the town was evacuated, but after the cessation of rain five days later, the town was re-occupied. Disastrously, the landslide occurred shortly after. The potential for a large landslide was known, precursory signs observed, but a poor understanding of landslide timing, resulted in catastrophe. One approach to understand failure timing has been developed after observations of movements, or strains, on unstable embankments by Japanese engineer, Saito. He observed that when the inverse of velocity of deformation prior to failure was plotted against time, a straight line was observed (Saito, 1965), termed 'Saito linearity'. The implication was that the linear trend could be extrapolated to a point at which velocity reached infinity, predicting of failure. More recently, others noted characteristic movement patterns in landslides, termed 'three phase creep' Varnes (1978), in which a failing slope exhibits three distinct periods of movement. Recently attempts to understand the failure mechanisms which control this pattern have used monitoring of failing slopes and novel geotechnical testing. Questions remain surrounding the application of this model to real slopes particularly with respect to the influence of slope geometry on the evolution of failure, and the complexity of interpreting strain state from real-world monitoring data. Research is proposed to observe the influence of variations in slope geometry and strength on the mechanisms of pre-failure deformation, using a newly developed technique named a vertical 'gravity accelerator' table. In this model a slope, formed from a material with elasto-brittle-plastic properties, that at scale behaves comparably to natural slope materials, is subject to cyclic-loading by dropping onto a hydraulic ram imposing a rapid acceleration in the same direction as gravity. Hence, a 2,000 m slope is simulated using a 0.4 m scale analogue. In a series of model configurations the influence of slope angle, slope length, curvature and strength will be tested. Deformation will be monitored using a high-precision 3D structured light scanner, which measures the surface deformation. In a final set of experiments the processes acting within three well-instrumented real-world landslides will be replicated. These will be the Selbourne cutting stability experiment, UK, the Ota-Mura Landslide, Japan, and the Pos Selim Landslide, Malaysia. Using 3D printing each slope will be recreated, failure simulated and deformation results compared to monitoring data, to establish the controls on the mechanisms and evolution of failure. These data will then be compared to a new database of landslide movement collated from the literature to exmaine landslide mechanisms and movement in a large number of failures. Determining the controls of slope form and material strength on landslide mechanisms and resulting pre-failure surface deformation will give enhanced understanding of deformation trends observed in landslides, mechanisms for landslide triggering and ovement, and will have implications for the use of Saito methods for slope failure prediction.